[PATCH] vfio powerpc: enabled and supported on powernv platform

[PATCH] vfio powerpc: enabled and supported on powernv platform

[Alexey Kardashevskiy]

VFIO implements platform independent stuff such as
a PCI driver, BAR access (via read/write on a file descriptor
or direct mapping when possible) and IRQ signaling.
The platform dependent part includes IOMMU initialization
and handling.

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan, only POWERNV
platform is supported at the moment.

Also the patch implements an VFIO-IOMMU driver which
manages DMA mapping/unmapping requests coming from
the client (now QEMU). It also returns a DMA window
information to let the guest initialize the device tree
for a guest OS properly. Although this driver has been
tested only on POWERNV, it should work on any platform
supporting TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |    6 +
 arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 drivers/vfio/Kconfig                 |    6 +
 drivers/vfio/Makefile                |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c  |  247 ++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                 |   20 +++
 8 files changed, 563 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..5ba66cb 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -64,30 +64,33 @@ struct iommu_pool {
 } ____cacheline_aligned_in_smp;
 
 struct iommu_table {
 	unsigned long  it_busno;     /* Bus number this table belongs to */
 	unsigned long  it_size;      /* Size of iommu table in entries */
 	unsigned long  it_offset;    /* Offset into global table */
 	unsigned long  it_base;      /* mapped address of tce table */
 	unsigned long  it_index;     /* which iommu table this is */
 	unsigned long  it_type;      /* type: PCI or Virtual Bus */
 	unsigned long  it_blocksize; /* Entries in each block (cacheline) */
 	unsigned long  poolsize;
 	unsigned long  nr_pools;
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
 
 static inline void set_iommu_table_base(struct device *dev, void *base)
 {
 	dev->archdata.dma_data.iommu_table_base = base;
 }
 
 static inline void *get_iommu_table_base(struct device *dev)
 {
 	return dev->archdata.dma_data.iommu_table_base;
 }
 
 /* Frees table for an individual device node */
@@ -135,17 +138,20 @@ static inline void pci_iommu_init(void) { }
 extern void alloc_dart_table(void);
 #if defined(CONFIG_PPC64) && defined(CONFIG_PM)
 static inline void iommu_save(void)
 {
 	if (ppc_md.iommu_save)
 		ppc_md.iommu_save();
 }
 
 static inline void iommu_restore(void)
 {
 	if (ppc_md.iommu_restore)
 		ppc_md.iommu_restore();
 }
 #endif
 
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
+		enum dma_data_direction direction, unsigned long pages);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..94f614b 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -32,30 +32,31 @@
 #include <linux/dma-mapping.h>
 #include <linux/bitmap.h>
 #include <linux/iommu-helper.h>
 #include <linux/crash_dump.h>
 #include <linux/hash.h>
 #include <linux/fault-inject.h>
 #include <linux/pci.h>
 #include <asm/io.h>
 #include <asm/prom.h>
 #include <asm/iommu.h>
 #include <asm/pci-bridge.h>
 #include <asm/machdep.h>
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
 static int novmerge;
 
 static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned int);
 
 static int __init setup_iommu(char *str)
 {
 	if (!strcmp(str, "novmerge"))
 		novmerge = 1;
 	else if (!strcmp(str, "vmerge"))
 		novmerge = 0;
 	return 1;
 }
@@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev, struct iommu_table *tbl,
 }
 
 void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 			 void *vaddr, dma_addr_t dma_handle)
 {
 	if (tbl) {
 		unsigned int nio_pages;
 
 		size = PAGE_ALIGN(size);
 		nio_pages = size >> IOMMU_PAGE_SHIFT;
 		iommu_free(tbl, dma_handle, nio_pages);
 		size = PAGE_ALIGN(size);
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+static struct page *free_tce(struct iommu_table *tbl, unsigned long entry)
+{
+	struct page *page = NULL;
+	unsigned long oldtce;
+
+	oldtce = ppc_md.tce_get(tbl, entry);
+
+	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
+		return NULL;
+
+	page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+	WARN_ON(!page);
+	if (page && (oldtce & TCE_PCI_WRITE))
+		SetPageDirty(page);
+	ppc_md.tce_free(tbl, entry, 1);
+
+	return page;
+}
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset;
+
+	/* Map new TCE */
+	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret < 1) {
+		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		if (!ret)
+			ret = -EFAULT;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	entry += tbl->it_offset; /* Offset into real TCE table */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	return 0;
+}
+
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
+		enum dma_data_direction direction, unsigned long pages)
+{
+	int i, ret = 0, pages_to_put = 0;
+	struct page *page;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+	struct page **oldpages;
+	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+
+	/* Handle a single page request without allocation
+	   of pages-to-release array */
+	if (pages == 1) {
+		spin_lock(&(pool->lock));
+		page = free_tce(tbl, entry);
+
+		if (direction != DMA_NONE)
+			ret = put_tce(tbl, entry, tce, direction);
+
+		tce_flush(tbl);
+
+		if (page)
+			put_page(page);
+
+		spin_unlock(&(pool->lock));
+		return ret;
+	}
+
+	/* Releasing multiple pages */
+	/* Allocate an array for pages to be released after TCE table
+	   is updated */
+	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!oldpages)
+		return -ENOMEM;
+
+	spin_lock(&(pool->lock));
+
+	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce += IOMMU_PAGE_SIZE) {
+		page = free_tce(tbl, entry);
+		if (page) {
+			oldpages[pages_to_put] = page;
+			++pages_to_put;
+		}
+
+		if (direction != DMA_NONE)
+			ret = put_tce(tbl, entry, tce, direction);
+
+		/* Release old pages if we reached the end of oldpages[] or
+		   it is the last page or we are about to exit the loop */
+		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret) {
+			tce_flush(tbl);
+
+			/* Release pages after removing them from TCE table */
+			while (pages_to_put) {
+				--pages_to_put;
+				put_page(oldpages[pages_to_put]);
+			}
+		}
+	}
+
+	spin_unlock(&(pool->lock));
+	kfree(oldpages);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..676f4d9 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -8,30 +8,31 @@
  * This program is free software; you can redistribute it and/or
  * modify it under the terms of the GNU General Public License
  * as published by the Free Software Foundation; either version
  * 2 of the License, or (at your option) any later version.
  */
 
 #include <linux/kernel.h>
 #include <linux/pci.h>
 #include <linux/delay.h>
 #include <linux/string.h>
 #include <linux/init.h>
 #include <linux/bootmem.h>
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
 #include <asm/prom.h>
 #include <asm/pci-bridge.h>
 #include <asm/machdep.h>
 #include <asm/ppc-pci.h>
 #include <asm/opal.h>
 #include <asm/iommu.h>
 #include <asm/tce.h>
 #include <asm/abs_addr.h>
 #include <asm/firmware.h>
 
 #include "powernv.h"
 #include "pci.h"
@@ -601,15 +602,149 @@ void __init pnv_pci_init(void)
 	/* Configure IOMMU DMA hooks */
 	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
 	ppc_md.tce_build = pnv_tce_build;
 	ppc_md.tce_free = pnv_tce_free;
 	ppc_md.tce_get = pnv_tce_get;
 	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
 	set_pci_dma_ops(&dma_iommu_ops);
 
 	/* Configure MSIs */
 #ifdef CONFIG_PCI_MSI
 	ppc_md.msi_check_device = pnv_msi_check_device;
 	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev->kobj.name,
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev->kobj.name);
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev->kobj.name, iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
+				dev->kobj.name, ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			printk(KERN_INFO "tce_vfio: cannot create "
+					"new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -175,16 +175,24 @@ config EXYNOS_IOMMU
 	  processor family. This enables H/W multimedia accellerators to see
 	  non-linear physical memory chunks as a linear memory in their
 	  address spaces
 
 	  If unsure, say N here.
 
 config EXYNOS_IOMMU_DEBUG
 	bool "Debugging log for Exynos IOMMU"
 	depends on EXYNOS_IOMMU
 	help
 	  Select this to see the detailed log message that shows what
 	  happens in the IOMMU driver
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
index 7cd5dec..b464687 100644
--- a/drivers/vfio/Kconfig
+++ b/drivers/vfio/Kconfig
@@ -1,16 +1,22 @@
 config VFIO_IOMMU_TYPE1
 	tristate
 	depends on VFIO
 	default n
 
+config VFIO_IOMMU_SPAPR_TCE
+	tristate
+	depends on VFIO && SPAPR_TCE_IOMMU
+	default n
+
 menuconfig VFIO
 	tristate "VFIO Non-Privileged userspace driver framework"
 	depends on IOMMU_API
 	select VFIO_IOMMU_TYPE1 if X86
+	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
 	help
 	  VFIO provides a framework for secure userspace device drivers.
 	  See Documentation/vfio.txt for more details.
 
 	  If you don't know what to do here, say N.
 
 source "drivers/vfio/pci/Kconfig"
diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
index 2398d4a..72bfabc 100644
--- a/drivers/vfio/Makefile
+++ b/drivers/vfio/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_VFIO) += vfio.o
 obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
+obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
 obj-$(CONFIG_VFIO_PCI) += pci/
diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
new file mode 100644
index 0000000..ac72c74d
--- /dev/null
+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
@@ -0,0 +1,247 @@
+/*
+ * VFIO: IOMMU DMA mapping support for TCE on POWER
+ *
+ * Copyright (C) 2012 IBM Corp.  All rights reserved.
+ *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Derived from original vfio_iommu_type1.c:
+ * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
+ *     Author: Alex Williamson <alex.williamson@redhat.com>
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/err.h>
+#include <linux/vfio.h>
+#include <asm/iommu.h>
+
+#define DRIVER_VERSION  "0.1"
+#define DRIVER_AUTHOR   "aik@ozlabs.ru"
+#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group);
+
+/*
+ * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
+ */
+
+/*
+ * The container descriptor supports only a single group per container.
+ * Required by the API as the container is not supplied with the IOMMU group
+ * at the moment of initialization.
+ */
+struct tce_container {
+	struct mutex lock;
+	struct iommu_table *tbl;
+};
+
+static void *tce_iommu_open(unsigned long arg)
+{
+	struct tce_container *container;
+
+	if (arg != VFIO_SPAPR_TCE_IOMMU) {
+		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	container = kzalloc(sizeof(*container), GFP_KERNEL);
+	if (!container)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&container->lock);
+
+	return container;
+}
+
+static void tce_iommu_release(void *iommu_data)
+{
+	struct tce_container *container = iommu_data;
+
+	WARN_ON(container->tbl && !container->tbl->it_group);
+	if (container->tbl && container->tbl->it_group)
+		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
+
+	mutex_destroy(&container->lock);
+
+	kfree(container);
+}
+
+static long tce_iommu_ioctl(void *iommu_data,
+				 unsigned int cmd, unsigned long arg)
+{
+	struct tce_container *container = iommu_data;
+	unsigned long minsz;
+
+	switch (cmd) {
+	case VFIO_CHECK_EXTENSION: {
+		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
+	}
+	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
+		struct vfio_iommu_spapr_tce_info info;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
+				dma64_window_size);
+
+		if (copy_from_user(&info, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (info.argsz < minsz)
+			return -EINVAL;
+
+		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
+		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
+		info.dma64_window_start = 0;
+		info.dma64_window_size = 0;
+		info.flags = 0;
+
+		if (copy_to_user((void __user *)arg, &info, minsz))
+			return -EFAULT;
+
+		return 0;
+	}
+	case VFIO_IOMMU_MAP_DMA: {
+		vfio_iommu_spapr_tce_dma_map par;
+		struct iommu_table *tbl = container->tbl;
+		enum dma_data_direction direction = DMA_NONE;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
+
+		if (copy_from_user(&par, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (par.argsz < minsz)
+			return -EINVAL;
+
+		if ((par.flags & VFIO_DMA_MAP_FLAG_READ) &&
+				(par.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
+			direction = DMA_BIDIRECTIONAL;
+		} else if (par.flags & VFIO_DMA_MAP_FLAG_READ) {
+			direction = DMA_TO_DEVICE;
+		} else if (par.flags & VFIO_DMA_MAP_FLAG_WRITE) {
+			direction = DMA_FROM_DEVICE;
+		}
+
+		par.size += par.iova & ~IOMMU_PAGE_MASK;
+		par.size = _ALIGN_UP(par.size, IOMMU_PAGE_SIZE);
+
+		return iommu_put_tces(tbl, par.iova >> IOMMU_PAGE_SHIFT,
+				par.vaddr & IOMMU_PAGE_MASK, direction,
+				par.size >> IOMMU_PAGE_SHIFT);
+	}
+	case VFIO_IOMMU_UNMAP_DMA: {
+		vfio_iommu_spapr_tce_dma_unmap par;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
+
+		if (copy_from_user(&par, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (par.argsz < minsz)
+			return -EINVAL;
+
+		par.size += par.iova & ~IOMMU_PAGE_MASK;
+		par.size = _ALIGN_UP(par.size, IOMMU_PAGE_SIZE);
+
+		return iommu_put_tces(tbl, par.iova >> IOMMU_PAGE_SHIFT,
+				0, DMA_NONE, par.size >> IOMMU_PAGE_SHIFT);
+	}
+	default:
+		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
+	}
+
+	return -ENOTTY;
+}
+
+static int tce_iommu_attach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
+			iommu_group_id(iommu_group), iommu_group);
+	if (container->tbl) {
+		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
+				iommu_group_id(container->tbl->it_group),
+				iommu_group_id(iommu_group));
+		mutex_unlock(&container->lock);
+		return -EBUSY;
+	}
+
+	container->tbl = tbl;
+	mutex_unlock(&container->lock);
+
+	return 0;
+}
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	if (tbl != container->tbl) {
+		printk(KERN_WARNING "tce_vfio: detaching group #%u, expected group is #%u\n",
+				iommu_group_id(iommu_group),
+				iommu_group_id(tbl->it_group));
+	} else {
+
+		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
+				iommu_group_id(iommu_group), iommu_group);
+
+		iommu_put_tces(tbl, tbl->it_offset, 0, DMA_NONE, tbl->it_size);
+		container->tbl = NULL;
+	}
+	mutex_unlock(&container->lock);
+}
+
+const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
+	.name		= "iommu-vfio-powerpc",
+	.owner		= THIS_MODULE,
+	.open		= tce_iommu_open,
+	.release	= tce_iommu_release,
+	.ioctl		= tce_iommu_ioctl,
+	.attach_group	= tce_iommu_attach_group,
+	.detach_group	= tce_iommu_detach_group,
+};
+
+static int __init tce_iommu_init(void)
+{
+	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+
+MODULE_VERSION(DRIVER_VERSION);
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR(DRIVER_AUTHOR);
+MODULE_DESCRIPTION(DRIVER_DESC);
+
diff --git a/include/linux/vfio.h b/include/linux/vfio.h
index 0a4f180..3ecd65c 100644
--- a/include/linux/vfio.h
+++ b/include/linux/vfio.h
@@ -87,30 +87,31 @@ extern void vfio_unregister_iommu_driver(
  * Simple helper macro for dealing with variable sized structures passed
  * from user space.  This allows us to easily determine if the provided
  * structure is sized to include various fields.
  */
 #define offsetofend(TYPE, MEMBER) ({				\
 	TYPE tmp;						\
 	offsetof(TYPE, MEMBER) + sizeof(tmp.MEMBER); })		\
 
 #endif /* __KERNEL__ */
 
 /* Kernel & User level defines for VFIO IOCTLs. */
 
 /* Extensions */
 
 #define VFIO_TYPE1_IOMMU		1
+#define VFIO_SPAPR_TCE_IOMMU		2
 
 /*
  * The IOCTL interface is designed for extensibility by embedding the
  * structure length (argsz) and flags into structures passed between
  * kernel and userspace.  We therefore use the _IO() macro for these
  * defines to avoid implicitly embedding a size into the ioctl request.
  * As structure fields are added, argsz will increase to match and flag
  * bits will be defined to indicate additional fields with valid data.
  * It's *always* the caller's responsibility to indicate the size of
  * the structure passed by setting argsz appropriately.
  */
 
 #define VFIO_TYPE	(';')
 #define VFIO_BASE	100
 
@@ -430,16 +431,35 @@ struct vfio_iommu_type1_dma_map {
 /**
  * VFIO_IOMMU_UNMAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 14, struct vfio_dma_unmap)
  *
  * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
  * Caller sets argsz.
  */
 struct vfio_iommu_type1_dma_unmap {
 	__u32	argsz;
 	__u32	flags;
 	__u64	iova;				/* IO virtual address */
 	__u64	size;				/* Size of mapping (bytes) */
 };
 
 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 
+/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
+
+struct vfio_iommu_spapr_tce_info {
+	__u32 argsz;
+	__u32 flags;
+	__u32 dma32_window_start;
+	__u32 dma32_window_size;
+	__u64 dma64_window_start;
+	__u64 dma64_window_size;
+};
+
+#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
+
+/* Reuse type1 map/unmap structs as they are the same at the moment */
+typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
+typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
+
+/* ***************************************************************** */
+
 #endif /* VFIO_H */
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Alex Williamson]

On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
> VFIO implements platform independent stuff such as
> a PCI driver, BAR access (via read/write on a file descriptor
> or direct mapping when possible) and IRQ signaling.
> The platform dependent part includes IOMMU initialization
> and handling.
> 
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan, only POWERNV
> platform is supported at the moment.
> 
> Also the patch implements an VFIO-IOMMU driver which
> manages DMA mapping/unmapping requests coming from
> the client (now QEMU). It also returns a DMA window
> information to let the guest initialize the device tree
> for a guest OS properly. Although this driver has been
> tested only on POWERNV, it should work on any platform
> supporting TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |    6 +
>  arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
>  drivers/iommu/Kconfig                |    8 ++
>  drivers/vfio/Kconfig                 |    6 +
>  drivers/vfio/Makefile                |    1 +
>  drivers/vfio/vfio_iommu_spapr_tce.c  |  247 ++++++++++++++++++++++++++++++++++
>  include/linux/vfio.h                 |   20 +++
>  8 files changed, 563 insertions(+)
>  create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..5ba66cb 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -64,30 +64,33 @@ struct iommu_pool {
>  } ____cacheline_aligned_in_smp;
>  
>  struct iommu_table {
>  	unsigned long  it_busno;     /* Bus number this table belongs to */
>  	unsigned long  it_size;      /* Size of iommu table in entries */
>  	unsigned long  it_offset;    /* Offset into global table */
>  	unsigned long  it_base;      /* mapped address of tce table */
>  	unsigned long  it_index;     /* which iommu table this is */
>  	unsigned long  it_type;      /* type: PCI or Virtual Bus */
>  	unsigned long  it_blocksize; /* Entries in each block (cacheline) */
>  	unsigned long  poolsize;
>  	unsigned long  nr_pools;
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
>  
>  static inline void set_iommu_table_base(struct device *dev, void *base)
>  {
>  	dev->archdata.dma_data.iommu_table_base = base;
>  }
>  
>  static inline void *get_iommu_table_base(struct device *dev)
>  {
>  	return dev->archdata.dma_data.iommu_table_base;
>  }
>  
>  /* Frees table for an individual device node */
> @@ -135,17 +138,20 @@ static inline void pci_iommu_init(void) { }
>  extern void alloc_dart_table(void);
>  #if defined(CONFIG_PPC64) && defined(CONFIG_PM)
>  static inline void iommu_save(void)
>  {
>  	if (ppc_md.iommu_save)
>  		ppc_md.iommu_save();
>  }
>  
>  static inline void iommu_restore(void)
>  {
>  	if (ppc_md.iommu_restore)
>  		ppc_md.iommu_restore();
>  }
>  #endif
>  
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
> +		enum dma_data_direction direction, unsigned long pages);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..94f614b 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -32,30 +32,31 @@
>  #include <linux/dma-mapping.h>
>  #include <linux/bitmap.h>
>  #include <linux/iommu-helper.h>
>  #include <linux/crash_dump.h>
>  #include <linux/hash.h>
>  #include <linux/fault-inject.h>
>  #include <linux/pci.h>
>  #include <asm/io.h>
>  #include <asm/prom.h>
>  #include <asm/iommu.h>
>  #include <asm/pci-bridge.h>
>  #include <asm/machdep.h>
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
>  static int novmerge;
>  
>  static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned int);
>  
>  static int __init setup_iommu(char *str)
>  {
>  	if (!strcmp(str, "novmerge"))
>  		novmerge = 1;
>  	else if (!strcmp(str, "vmerge"))
>  		novmerge = 0;
>  	return 1;
>  }
> @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev, struct iommu_table *tbl,
>  }
>  
>  void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  			 void *vaddr, dma_addr_t dma_handle)
>  {
>  	if (tbl) {
>  		unsigned int nio_pages;
>  
>  		size = PAGE_ALIGN(size);
>  		nio_pages = size >> IOMMU_PAGE_SHIFT;
>  		iommu_free(tbl, dma_handle, nio_pages);
>  		size = PAGE_ALIGN(size);
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +static struct page *free_tce(struct iommu_table *tbl, unsigned long entry)
> +{
> +	struct page *page = NULL;

NULL initialization doesn't appear to be necessary

> +	unsigned long oldtce;
> +
> +	oldtce = ppc_md.tce_get(tbl, entry);
> +
> +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> +		return NULL;
> +
> +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +	WARN_ON(!page);
> +	if (page && (oldtce & TCE_PCI_WRITE))
> +		SetPageDirty(page);
> +	ppc_md.tce_free(tbl, entry, 1);
> +
> +	return page;
> +}
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset;
> +
> +	/* Map new TCE */
> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);
> +	if (ret < 1) {
> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> +		if (!ret)
> +			ret = -EFAULT;

Missing return ret?  Otherwise we've got some bogus uses of page below
and we're setting ret for no reason here.

> +	}
> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	entry += tbl->it_offset; /* Offset into real TCE table */
> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	return 0;
> +}
> +
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
> +		enum dma_data_direction direction, unsigned long pages)
> +{
> +	int i, ret = 0, pages_to_put = 0;
> +	struct page *page;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +	struct page **oldpages;
> +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +
> +	/* Handle a single page request without allocation
> +	   of pages-to-release array */
> +	if (pages == 1) {
> +		spin_lock(&(pool->lock));
> +		page = free_tce(tbl, entry);
> +
> +		if (direction != DMA_NONE)
> +			ret = put_tce(tbl, entry, tce, direction);
> +
> +		tce_flush(tbl);
> +
> +		if (page)
> +			put_page(page);
> +
> +		spin_unlock(&(pool->lock));
> +		return ret;
> +	}
> +
> +	/* Releasing multiple pages */
> +	/* Allocate an array for pages to be released after TCE table
> +	   is updated */
> +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
> +	if (!oldpages)
> +		return -ENOMEM;
> +
> +	spin_lock(&(pool->lock));
> +
> +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce += IOMMU_PAGE_SIZE) {
> +		page = free_tce(tbl, entry);
> +		if (page) {
> +			oldpages[pages_to_put] = page;
> +			++pages_to_put;
> +		}
> +
> +		if (direction != DMA_NONE)
> +			ret = put_tce(tbl, entry, tce, direction);
> +
> +		/* Release old pages if we reached the end of oldpages[] or
> +		   it is the last page or we are about to exit the loop */
> +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret) {
> +			tce_flush(tbl);

Avoiding tce_flush() is the reason for all this extra overhead, right?
I wonder if it'd be cleaner separating map vs unmap, where the map case
can avoid the oldpages array... but that means inserting new mappings on
top of old ones wouldn't put the pages.

> +
> +			/* Release pages after removing them from TCE table */
> +			while (pages_to_put) {
> +				--pages_to_put;
> +				put_page(oldpages[pages_to_put]);
> +			}
> +		}
> +	}
> +
> +	spin_unlock(&(pool->lock));
> +	kfree(oldpages);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..676f4d9 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -8,30 +8,31 @@
>   * This program is free software; you can redistribute it and/or
>   * modify it under the terms of the GNU General Public License
>   * as published by the Free Software Foundation; either version
>   * 2 of the License, or (at your option) any later version.
>   */
>  
>  #include <linux/kernel.h>
>  #include <linux/pci.h>
>  #include <linux/delay.h>
>  #include <linux/string.h>
>  #include <linux/init.h>
>  #include <linux/bootmem.h>
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
>  #include <asm/prom.h>
>  #include <asm/pci-bridge.h>
>  #include <asm/machdep.h>
>  #include <asm/ppc-pci.h>
>  #include <asm/opal.h>
>  #include <asm/iommu.h>
>  #include <asm/tce.h>
>  #include <asm/abs_addr.h>
>  #include <asm/firmware.h>
>  
>  #include "powernv.h"
>  #include "pci.h"
> @@ -601,15 +602,149 @@ void __init pnv_pci_init(void)
>  	/* Configure IOMMU DMA hooks */
>  	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
>  	ppc_md.tce_build = pnv_tce_build;
>  	ppc_md.tce_free = pnv_tce_free;
>  	ppc_md.tce_get = pnv_tce_get;
>  	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
>  	set_pci_dma_ops(&dma_iommu_ops);
>  
>  	/* Configure MSIs */
>  #ifdef CONFIG_PCI_MSI
>  	ppc_md.msi_check_device = pnv_msi_check_device;
>  	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev->kobj.name,

dev_name(dev)

> +				iommu_group_id(dev->iommu_group));
> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev->kobj.name);
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev->kobj.name, iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> +				dev->kobj.name, ret);
> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);

There's already a notifier in the iommu code if you were to register an
iommu_ops with the add/remove_device entries.  That would allow you to
remove the notifier block and notifier function below and the second
loop below.  Are you avoiding that to avoid the rest of iommu_ops?

Also, shouldn't this notifier only be registered after the first loop
below?  Otherwise ADD_DEVICE could race with setting up groups, which we
assume are present in the add_device() above.

> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			printk(KERN_INFO "tce_vfio: cannot create "
> +					"new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));
> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);
> +	}
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -175,16 +175,24 @@ config EXYNOS_IOMMU
>  	  processor family. This enables H/W multimedia accellerators to see
>  	  non-linear physical memory chunks as a linear memory in their
>  	  address spaces
>  
>  	  If unsure, say N here.
>  
>  config EXYNOS_IOMMU_DEBUG
>  	bool "Debugging log for Exynos IOMMU"
>  	depends on EXYNOS_IOMMU
>  	help
>  	  Select this to see the detailed log message that shows what
>  	  happens in the IOMMU driver
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

How are you planning to split this up among maintainers?  A powerpc
patch, an iommu kconfig patch, then the vfio changes below for me?

> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> index 7cd5dec..b464687 100644
> --- a/drivers/vfio/Kconfig
> +++ b/drivers/vfio/Kconfig
> @@ -1,16 +1,22 @@
>  config VFIO_IOMMU_TYPE1
>  	tristate
>  	depends on VFIO
>  	default n
>  
> +config VFIO_IOMMU_SPAPR_TCE
> +	tristate
> +	depends on VFIO && SPAPR_TCE_IOMMU
> +	default n
> +
>  menuconfig VFIO
>  	tristate "VFIO Non-Privileged userspace driver framework"
>  	depends on IOMMU_API
>  	select VFIO_IOMMU_TYPE1 if X86
> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>  	help
>  	  VFIO provides a framework for secure userspace device drivers.
>  	  See Documentation/vfio.txt for more details.
>  
>  	  If you don't know what to do here, say N.
>  
>  source "drivers/vfio/pci/Kconfig"
> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> index 2398d4a..72bfabc 100644
> --- a/drivers/vfio/Makefile
> +++ b/drivers/vfio/Makefile
> @@ -1,3 +1,4 @@
>  obj-$(CONFIG_VFIO) += vfio.o
>  obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>  obj-$(CONFIG_VFIO_PCI) += pci/
> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> new file mode 100644
> index 0000000..ac72c74d
> --- /dev/null
> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> @@ -0,0 +1,247 @@
> +/*
> + * VFIO: IOMMU DMA mapping support for TCE on POWER
> + *
> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + *
> + * Derived from original vfio_iommu_type1.c:
> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> + *     Author: Alex Williamson <alex.williamson@redhat.com>
> + */
> +
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +#include <linux/slab.h>
> +#include <linux/uaccess.h>
> +#include <linux/err.h>
> +#include <linux/vfio.h>
> +#include <asm/iommu.h>
> +
> +#define DRIVER_VERSION  "0.1"
> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group);
> +
> +/*
> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> + */
> +
> +/*
> + * The container descriptor supports only a single group per container.
> + * Required by the API as the container is not supplied with the IOMMU group
> + * at the moment of initialization.
> + */
> +struct tce_container {
> +	struct mutex lock;
> +	struct iommu_table *tbl;
> +};
> +
> +static void *tce_iommu_open(unsigned long arg)
> +{
> +	struct tce_container *container;
> +
> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> +		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
> +	if (!container)
> +		return ERR_PTR(-ENOMEM);
> +
> +	mutex_init(&container->lock);
> +
> +	return container;
> +}
> +
> +static void tce_iommu_release(void *iommu_data)
> +{
> +	struct tce_container *container = iommu_data;
> +
> +	WARN_ON(container->tbl && !container->tbl->it_group);
> +	if (container->tbl && container->tbl->it_group)
> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> +
> +	mutex_destroy(&container->lock);
> +
> +	kfree(container);
> +}
> +
> +static long tce_iommu_ioctl(void *iommu_data,
> +				 unsigned int cmd, unsigned long arg)
> +{
> +	struct tce_container *container = iommu_data;
> +	unsigned long minsz;
> +
> +	switch (cmd) {
> +	case VFIO_CHECK_EXTENSION: {
> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> +	}
> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> +		struct vfio_iommu_spapr_tce_info info;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> +				dma64_window_size);
> +
> +		if (copy_from_user(&info, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (info.argsz < minsz)
> +			return -EINVAL;
> +
> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> +		info.dma64_window_start = 0;
> +		info.dma64_window_size = 0;
> +		info.flags = 0;
> +
> +		if (copy_to_user((void __user *)arg, &info, minsz))
> +			return -EFAULT;
> +
> +		return 0;
> +	}
> +	case VFIO_IOMMU_MAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_map par;

What does "par" stand for?

> +		struct iommu_table *tbl = container->tbl;
> +		enum dma_data_direction direction = DMA_NONE;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> +
> +		if (copy_from_user(&par, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (par.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((par.flags & VFIO_DMA_MAP_FLAG_READ) &&
> +				(par.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
> +			direction = DMA_BIDIRECTIONAL;
> +		} else if (par.flags & VFIO_DMA_MAP_FLAG_READ) {
> +			direction = DMA_TO_DEVICE;
> +		} else if (par.flags & VFIO_DMA_MAP_FLAG_WRITE) {
> +			direction = DMA_FROM_DEVICE;
> +		}
> +
> +		par.size += par.iova & ~IOMMU_PAGE_MASK;
> +		par.size = _ALIGN_UP(par.size, IOMMU_PAGE_SIZE);
> +
> +		return iommu_put_tces(tbl, par.iova >> IOMMU_PAGE_SHIFT,
> +				par.vaddr & IOMMU_PAGE_MASK, direction,
> +				par.size >> IOMMU_PAGE_SHIFT);
> +	}
> +	case VFIO_IOMMU_UNMAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_unmap par;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> +
> +		if (copy_from_user(&par, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (par.argsz < minsz)
> +			return -EINVAL;
> +
> +		par.size += par.iova & ~IOMMU_PAGE_MASK;
> +		par.size = _ALIGN_UP(par.size, IOMMU_PAGE_SIZE);
> +
> +		return iommu_put_tces(tbl, par.iova >> IOMMU_PAGE_SHIFT,
> +				0, DMA_NONE, par.size >> IOMMU_PAGE_SHIFT);
> +	}
> +	default:
> +		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
> +	}
> +
> +	return -ENOTTY;
> +}
> +
> +static int tce_iommu_attach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> +			iommu_group_id(iommu_group), iommu_group);
> +	if (container->tbl) {
> +		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
> +				iommu_group_id(container->tbl->it_group),
> +				iommu_group_id(iommu_group));
> +		mutex_unlock(&container->lock);
> +		return -EBUSY;
> +	}
> +
> +	container->tbl = tbl;
> +	mutex_unlock(&container->lock);
> +
> +	return 0;
> +}
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	if (tbl != container->tbl) {
> +		printk(KERN_WARNING "tce_vfio: detaching group #%u, expected group is #%u\n",
> +				iommu_group_id(iommu_group),
> +				iommu_group_id(tbl->it_group));
> +	} else {
> +
> +		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
> +				iommu_group_id(iommu_group), iommu_group);
> +
> +		iommu_put_tces(tbl, tbl->it_offset, 0, DMA_NONE, tbl->it_size);
> +		container->tbl = NULL;
> +	}
> +	mutex_unlock(&container->lock);
> +}
> +
> +const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
> +	.name		= "iommu-vfio-powerpc",
> +	.owner		= THIS_MODULE,
> +	.open		= tce_iommu_open,
> +	.release	= tce_iommu_release,
> +	.ioctl		= tce_iommu_ioctl,
> +	.attach_group	= tce_iommu_attach_group,
> +	.detach_group	= tce_iommu_detach_group,
> +};
> +
> +static int __init tce_iommu_init(void)
> +{
> +	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +
> +MODULE_VERSION(DRIVER_VERSION);
> +MODULE_LICENSE("GPL v2");
> +MODULE_AUTHOR(DRIVER_AUTHOR);
> +MODULE_DESCRIPTION(DRIVER_DESC);
> +
> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
> index 0a4f180..3ecd65c 100644
> --- a/include/linux/vfio.h
> +++ b/include/linux/vfio.h
> @@ -87,30 +87,31 @@ extern void vfio_unregister_iommu_driver(
>   * Simple helper macro for dealing with variable sized structures passed
>   * from user space.  This allows us to easily determine if the provided
>   * structure is sized to include various fields.
>   */
>  #define offsetofend(TYPE, MEMBER) ({				\
>  	TYPE tmp;						\
>  	offsetof(TYPE, MEMBER) + sizeof(tmp.MEMBER); })		\
>  
>  #endif /* __KERNEL__ */
>  
>  /* Kernel & User level defines for VFIO IOCTLs. */
>  
>  /* Extensions */
>  
>  #define VFIO_TYPE1_IOMMU		1
> +#define VFIO_SPAPR_TCE_IOMMU		2
>  
>  /*
>   * The IOCTL interface is designed for extensibility by embedding the
>   * structure length (argsz) and flags into structures passed between
>   * kernel and userspace.  We therefore use the _IO() macro for these
>   * defines to avoid implicitly embedding a size into the ioctl request.
>   * As structure fields are added, argsz will increase to match and flag
>   * bits will be defined to indicate additional fields with valid data.
>   * It's *always* the caller's responsibility to indicate the size of
>   * the structure passed by setting argsz appropriately.
>   */
>  
>  #define VFIO_TYPE	(';')
>  #define VFIO_BASE	100
>  
> @@ -430,16 +431,35 @@ struct vfio_iommu_type1_dma_map {
>  /**
>   * VFIO_IOMMU_UNMAP_DMA - _IOW(VFIO_TYPE, VFIO_BASE + 14, struct vfio_dma_unmap)
>   *
>   * Unmap IO virtual addresses using the provided struct vfio_dma_unmap.
>   * Caller sets argsz.
>   */
>  struct vfio_iommu_type1_dma_unmap {
>  	__u32	argsz;
>  	__u32	flags;
>  	__u64	iova;				/* IO virtual address */
>  	__u64	size;				/* Size of mapping (bytes) */
>  };
>  
>  #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
>  
> +/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
> +
> +struct vfio_iommu_spapr_tce_info {
> +	__u32 argsz;
> +	__u32 flags;
> +	__u32 dma32_window_start;
> +	__u32 dma32_window_size;
> +	__u64 dma64_window_start;
> +	__u64 dma64_window_size;
> +};
> +
> +#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
> +
> +/* Reuse type1 map/unmap structs as they are the same at the moment */
> +typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
> +typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
> +
> +/* ***************************************************************** */
> +
>  #endif /* VFIO_H */

I'm glad you were able to reuse these, after this gets merged we can
rename the structure to something more common and typedef for both type1
and spapr_tce so we don't forget it's shared.  Thanks,

Alex

RE: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Sethi Varun-B16395]

[-- Warning: decoded text below may be mangled, UTF-8 assumed --]
[-- Attachment #1: Type: text/plain; charset="utf-8", Size: 14871 bytes --]



> -----Original Message-----
> From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
> owner@vger.kernel.org] On Behalf Of Alex Williamson
> Sent: Tuesday, November 20, 2012 11:50 PM
> To: Alexey Kardashevskiy
> Cc: Benjamin Herrenschmidt; Paul Mackerras; linuxppc-
> dev@lists.ozlabs.org; linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
> David Gibson
> Subject: Re: [PATCH] vfio powerpc: enabled and supported on powernv
> platform
> 
> On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
> > VFIO implements platform independent stuff such as a PCI driver, BAR
> > access (via read/write on a file descriptor or direct mapping when
> > possible) and IRQ signaling.
> > The platform dependent part includes IOMMU initialization and
> > handling.
> >
> > This patch initializes IOMMU groups based on the IOMMU configuration
> > discovered during the PCI scan, only POWERNV platform is supported at
> > the moment.
> >
> > Also the patch implements an VFIO-IOMMU driver which manages DMA
> > mapping/unmapping requests coming from the client (now QEMU). It also
> > returns a DMA window information to let the guest initialize the
> > device tree for a guest OS properly. Although this driver has been
> > tested only on POWERNV, it should work on any platform supporting TCE
> > tables.
> >
> > To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config option.
> >
> > Cc: David Gibson <david@gibson.dropbear.id.au>
> > Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> > ---
> >  arch/powerpc/include/asm/iommu.h     |    6 +
> >  arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
> >  arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
> >  drivers/iommu/Kconfig                |    8 ++
> >  drivers/vfio/Kconfig                 |    6 +
> >  drivers/vfio/Makefile                |    1 +
> >  drivers/vfio/vfio_iommu_spapr_tce.c  |  247
> ++++++++++++++++++++++++++++++++++
> >  include/linux/vfio.h                 |   20 +++
> >  8 files changed, 563 insertions(+)
> >  create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> >
> > diff --git a/arch/powerpc/include/asm/iommu.h
> > b/arch/powerpc/include/asm/iommu.h
> > index cbfe678..5ba66cb 100644
> > --- a/arch/powerpc/include/asm/iommu.h
> > +++ b/arch/powerpc/include/asm/iommu.h
> > @@ -64,30 +64,33 @@ struct iommu_pool {  }
> > ____cacheline_aligned_in_smp;
> >
> >  struct iommu_table {
> >  	unsigned long  it_busno;     /* Bus number this table belongs to */
> >  	unsigned long  it_size;      /* Size of iommu table in entries */
> >  	unsigned long  it_offset;    /* Offset into global table */
> >  	unsigned long  it_base;      /* mapped address of tce table */
> >  	unsigned long  it_index;     /* which iommu table this is */
> >  	unsigned long  it_type;      /* type: PCI or Virtual Bus */
> >  	unsigned long  it_blocksize; /* Entries in each block (cacheline)
> */
> >  	unsigned long  poolsize;
> >  	unsigned long  nr_pools;
> >  	struct iommu_pool large_pool;
> >  	struct iommu_pool pools[IOMMU_NR_POOLS];
> >  	unsigned long *it_map;       /* A simple allocation bitmap for now
> */
> > +#ifdef CONFIG_IOMMU_API
> > +	struct iommu_group *it_group;
> > +#endif
> >  };
> >
> >  struct scatterlist;
> >
> >  static inline void set_iommu_table_base(struct device *dev, void
> > *base)  {
> >  	dev->archdata.dma_data.iommu_table_base = base;  }
> >
> >  static inline void *get_iommu_table_base(struct device *dev)  {
> >  	return dev->archdata.dma_data.iommu_table_base;
> >  }
> >
> >  /* Frees table for an individual device node */ @@ -135,17 +138,20 @@
> > static inline void pci_iommu_init(void) { }  extern void
> > alloc_dart_table(void);  #if defined(CONFIG_PPC64) &&
> > defined(CONFIG_PM)  static inline void iommu_save(void)  {
> >  	if (ppc_md.iommu_save)
> >  		ppc_md.iommu_save();
> >  }
> >
> >  static inline void iommu_restore(void)  {
> >  	if (ppc_md.iommu_restore)
> >  		ppc_md.iommu_restore();
> >  }
> >  #endif
> >
> > +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long
> entry, uint64_t tce,
> > +		enum dma_data_direction direction, unsigned long pages);
> > +
> >  #endif /* __KERNEL__ */
> >  #endif /* _ASM_IOMMU_H */
> > diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> > index ff5a6ce..94f614b 100644
> > --- a/arch/powerpc/kernel/iommu.c
> > +++ b/arch/powerpc/kernel/iommu.c
> > @@ -32,30 +32,31 @@
> >  #include <linux/dma-mapping.h>
> >  #include <linux/bitmap.h>
> >  #include <linux/iommu-helper.h>
> >  #include <linux/crash_dump.h>
> >  #include <linux/hash.h>
> >  #include <linux/fault-inject.h>
> >  #include <linux/pci.h>
> >  #include <asm/io.h>
> >  #include <asm/prom.h>
> >  #include <asm/iommu.h>
> >  #include <asm/pci-bridge.h>
> >  #include <asm/machdep.h>
> >  #include <asm/kdump.h>
> >  #include <asm/fadump.h>
> >  #include <asm/vio.h>
> > +#include <asm/tce.h>
> >
> >  #define DBG(...)
> >
> >  static int novmerge;
> >
> >  static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned
> > int);
> >
> >  static int __init setup_iommu(char *str)  {
> >  	if (!strcmp(str, "novmerge"))
> >  		novmerge = 1;
> >  	else if (!strcmp(str, "vmerge"))
> >  		novmerge = 0;
> >  	return 1;
> >  }
> > @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev,
> > struct iommu_table *tbl,  }
> >
> >  void iommu_free_coherent(struct iommu_table *tbl, size_t size,
> >  			 void *vaddr, dma_addr_t dma_handle)  {
> >  	if (tbl) {
> >  		unsigned int nio_pages;
> >
> >  		size = PAGE_ALIGN(size);
> >  		nio_pages = size >> IOMMU_PAGE_SHIFT;
> >  		iommu_free(tbl, dma_handle, nio_pages);
> >  		size = PAGE_ALIGN(size);
> >  		free_pages((unsigned long)vaddr, get_order(size));
> >  	}
> >  }
> > +
> > +#ifdef CONFIG_IOMMU_API
> > +/*
> > + * SPAPR TCE API
> > + */
> > +static struct page *free_tce(struct iommu_table *tbl, unsigned long
> > +entry) {
> > +	struct page *page = NULL;
> 
> NULL initialization doesn't appear to be necessary
> 
> > +	unsigned long oldtce;
> > +
> > +	oldtce = ppc_md.tce_get(tbl, entry);
> > +
> > +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> > +		return NULL;
> > +
> > +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
> > +
> > +	WARN_ON(!page);
> > +	if (page && (oldtce & TCE_PCI_WRITE))
> > +		SetPageDirty(page);
> > +	ppc_md.tce_free(tbl, entry, 1);
> > +
> > +	return page;
> > +}
> > +
> > +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> > +		uint64_t tce, enum dma_data_direction direction) {
> > +	int ret;
> > +	struct page *page = NULL;
> > +	unsigned long kva, offset;
> > +
> > +	/* Map new TCE */
> > +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> > +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> > +			direction != DMA_TO_DEVICE, &page);
> > +	if (ret < 1) {
> > +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed
> tce=%llx ioba=%lx ret=%d\n",
> > +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> > +		if (!ret)
> > +			ret = -EFAULT;
> 
> Missing return ret?  Otherwise we've got some bogus uses of page below
> and we're setting ret for no reason here.
> 
> > +	}
> > +
> > +	kva = (unsigned long) page_address(page);
> > +	kva += offset;
> > +
> > +	/* tce_build receives a virtual address */
> > +	entry += tbl->it_offset; /* Offset into real TCE table */
> > +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> > +
> > +	/* tce_build() only returns non-zero for transient errors */
> > +	if (unlikely(ret)) {
> > +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx
> ioba=%lx kva=%lx ret=%d\n",
> > +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> > +		put_page(page);
> > +		return -EIO;
> > +	}
> > +
> > +	return 0;
> > +}
> > +
> > +static void tce_flush(struct iommu_table *tbl) {
> > +	/* Flush/invalidate TLB caches if necessary */
> > +	if (ppc_md.tce_flush)
> > +		ppc_md.tce_flush(tbl);
> > +
> > +	/* Make sure updates are seen by hardware */
> > +	mb();
> > +}
> > +
> > +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> uint64_t tce,
> > +		enum dma_data_direction direction, unsigned long pages) {
> > +	int i, ret = 0, pages_to_put = 0;
> > +	struct page *page;
> > +	struct iommu_pool *pool = get_pool(tbl, entry);
> > +	struct page **oldpages;
> > +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
> > +
> > +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> > +
> > +	/* Handle a single page request without allocation
> > +	   of pages-to-release array */
> > +	if (pages == 1) {
> > +		spin_lock(&(pool->lock));
> > +		page = free_tce(tbl, entry);
> > +
> > +		if (direction != DMA_NONE)
> > +			ret = put_tce(tbl, entry, tce, direction);
> > +
> > +		tce_flush(tbl);
> > +
> > +		if (page)
> > +			put_page(page);
> > +
> > +		spin_unlock(&(pool->lock));
> > +		return ret;
> > +	}
> > +
> > +	/* Releasing multiple pages */
> > +	/* Allocate an array for pages to be released after TCE table
> > +	   is updated */
> > +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
> > +	if (!oldpages)
> > +		return -ENOMEM;
> > +
> > +	spin_lock(&(pool->lock));
> > +
> > +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce +=
> IOMMU_PAGE_SIZE) {
> > +		page = free_tce(tbl, entry);
> > +		if (page) {
> > +			oldpages[pages_to_put] = page;
> > +			++pages_to_put;
> > +		}
> > +
> > +		if (direction != DMA_NONE)
> > +			ret = put_tce(tbl, entry, tce, direction);
> > +
> > +		/* Release old pages if we reached the end of oldpages[] or
> > +		   it is the last page or we are about to exit the loop */
> > +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret)
> {
> > +			tce_flush(tbl);
> 
> Avoiding tce_flush() is the reason for all this extra overhead, right?
> I wonder if it'd be cleaner separating map vs unmap, where the map case
> can avoid the oldpages array... but that means inserting new mappings on
> top of old ones wouldn't put the pages.
> 
> > +
> > +			/* Release pages after removing them from TCE table */
> > +			while (pages_to_put) {
> > +				--pages_to_put;
> > +				put_page(oldpages[pages_to_put]);
> > +			}
> > +		}
> > +	}
> > +
> > +	spin_unlock(&(pool->lock));
> > +	kfree(oldpages);
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(iommu_put_tces);
> > +#endif /* CONFIG_IOMMU_API */
> > diff --git a/arch/powerpc/platforms/powernv/pci.c
> > b/arch/powerpc/platforms/powernv/pci.c
> > index 05205cf..676f4d9 100644
> > --- a/arch/powerpc/platforms/powernv/pci.c
> > +++ b/arch/powerpc/platforms/powernv/pci.c
> > @@ -8,30 +8,31 @@
> >   * This program is free software; you can redistribute it and/or
> >   * modify it under the terms of the GNU General Public License
> >   * as published by the Free Software Foundation; either version
> >   * 2 of the License, or (at your option) any later version.
> >   */
> >
> >  #include <linux/kernel.h>
> >  #include <linux/pci.h>
> >  #include <linux/delay.h>
> >  #include <linux/string.h>
> >  #include <linux/init.h>
> >  #include <linux/bootmem.h>
> >  #include <linux/irq.h>
> >  #include <linux/io.h>
> >  #include <linux/msi.h>
> > +#include <linux/iommu.h>
> >
> >  #include <asm/sections.h>
> >  #include <asm/io.h>
> >  #include <asm/prom.h>
> >  #include <asm/pci-bridge.h>
> >  #include <asm/machdep.h>
> >  #include <asm/ppc-pci.h>
> >  #include <asm/opal.h>
> >  #include <asm/iommu.h>
> >  #include <asm/tce.h>
> >  #include <asm/abs_addr.h>
> >  #include <asm/firmware.h>
> >
> >  #include "powernv.h"
> >  #include "pci.h"
> > @@ -601,15 +602,149 @@ void __init pnv_pci_init(void)
> >  	/* Configure IOMMU DMA hooks */
> >  	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
> >  	ppc_md.tce_build = pnv_tce_build;
> >  	ppc_md.tce_free = pnv_tce_free;
> >  	ppc_md.tce_get = pnv_tce_get;
> >  	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
> >  	set_pci_dma_ops(&dma_iommu_ops);
> >
> >  	/* Configure MSIs */
> >  #ifdef CONFIG_PCI_MSI
> >  	ppc_md.msi_check_device = pnv_msi_check_device;
> >  	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
> >  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;  #endif  }
> > +
> > +#ifdef CONFIG_IOMMU_API
> > +/*
> > + * IOMMU groups support required by VFIO  */ static int
> > +add_device(struct device *dev) {
> > +	struct iommu_table *tbl;
> > +	int ret = 0;
> > +
> > +	if (WARN_ON(dev->iommu_group)) {
> > +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu
> group %d, skipping\n",
> > +				dev->kobj.name,
> 
> dev_name(dev)
> 
> > +				iommu_group_id(dev->iommu_group));
> > +		return -EBUSY;
> > +	}
> > +
> > +	tbl = get_iommu_table_base(dev);
> > +	if (!tbl) {
> > +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> > +				dev->kobj.name);
> > +		return 0;
> > +	}
> > +
> > +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> > +			dev->kobj.name, iommu_group_id(tbl->it_group));
> > +
> > +	ret = iommu_group_add_device(tbl->it_group, dev);
> > +	if (ret < 0)
> > +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> > +				dev->kobj.name, ret);
> > +
> > +	return ret;
> > +}
> > +
> > +static void del_device(struct device *dev) {
> > +	iommu_group_remove_device(dev);
> > +}
> > +
> > +static int iommu_bus_notifier(struct notifier_block *nb,
> > +			      unsigned long action, void *data) {
> > +	struct device *dev = data;
> > +
> > +	switch (action) {
> > +	case BUS_NOTIFY_ADD_DEVICE:
> > +		return add_device(dev);
> > +	case BUS_NOTIFY_DEL_DEVICE:
> > +		del_device(dev);
> > +		return 0;
> > +	default:
> > +		return 0;
> > +	}
> > +}
> > +
> > +static struct notifier_block tce_iommu_bus_nb = {
> > +	.notifier_call = iommu_bus_notifier, };
> > +
> > +static void group_release(void *iommu_data) {
> > +	struct iommu_table *tbl = iommu_data;
> > +	tbl->it_group = NULL;
> > +}
> > +
> > +static int __init tce_iommu_init(void) {
> > +	struct pci_dev *pdev = NULL;
> > +	struct iommu_table *tbl;
> > +	struct iommu_group *grp;
> > +
> > +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> 
> There's already a notifier in the iommu code if you were to register an
> iommu_ops with the add/remove_device entries.  That would allow you to
> remove the notifier block and notifier function below and the second loop
> below.  Are you avoiding that to avoid the rest of iommu_ops?
> 
[Sethi Varun-B16395] Could be one reason, also they are associating the iommu group with the tce table entry and not the device.

> Also, shouldn't this notifier only be registered after the first loop
> below?  Otherwise ADD_DEVICE could race with setting up groups, which we
> assume are present in the add_device() above.
[Sethi Varun-B16395] Isn't this similar to how how the notifier is registered in iommu_bus_init? First a notifier is registered and then we check for devices that have already been probed.

-Varun
ÿôèº{.nÇ+‰·Ÿ®‰­†+%ŠËÿ±éݶ\x17¥Šwÿº{.nÇ+‰·¥Š{±þG«éÿŠ{ayº\x1dʇڙë,j\a­¢f£¢·hšïêÿ‘êçz_è®\x03(­éšŽŠÝ¢j"ú\x1a¶^[m§ÿÿ¾\a«þG«éÿ¢¸?™¨è­Ú&£ø§~á¶iO•æ¬z·švØ^\x14\x04\x1a¶^[m§ÿÿÃ\fÿ¶ìÿ¢¸?–I¥

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Alexey Kardashevskiy]

On 22/11/12 22:56, Sethi Varun-B16395 wrote:
>
>
>> -----Original Message-----
>> From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
>> owner@vger.kernel.org] On Behalf Of Alex Williamson
>> Sent: Tuesday, November 20, 2012 11:50 PM
>> To: Alexey Kardashevskiy
>> Cc: Benjamin Herrenschmidt; Paul Mackerras; linuxppc-
>> dev@lists.ozlabs.org; linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
>> David Gibson
>> Subject: Re: [PATCH] vfio powerpc: enabled and supported on powernv
>> platform
>>
>> On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
>>> VFIO implements platform independent stuff such as a PCI driver, BAR
>>> access (via read/write on a file descriptor or direct mapping when
>>> possible) and IRQ signaling.
>>> The platform dependent part includes IOMMU initialization and
>>> handling.
>>>
>>> This patch initializes IOMMU groups based on the IOMMU configuration
>>> discovered during the PCI scan, only POWERNV platform is supported at
>>> the moment.
>>>
>>> Also the patch implements an VFIO-IOMMU driver which manages DMA
>>> mapping/unmapping requests coming from the client (now QEMU). It also
>>> returns a DMA window information to let the guest initialize the
>>> device tree for a guest OS properly. Although this driver has been
>>> tested only on POWERNV, it should work on any platform supporting TCE
>>> tables.
>>>
>>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config option.
>>>
>>> Cc: David Gibson <david@gibson.dropbear.id.au>
>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>> ---
>>>   arch/powerpc/include/asm/iommu.h     |    6 +
>>>   arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
>>>   arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
>>>   drivers/iommu/Kconfig                |    8 ++
>>>   drivers/vfio/Kconfig                 |    6 +
>>>   drivers/vfio/Makefile                |    1 +
>>>   drivers/vfio/vfio_iommu_spapr_tce.c  |  247
>> ++++++++++++++++++++++++++++++++++
>>>   include/linux/vfio.h                 |   20 +++
>>>   8 files changed, 563 insertions(+)
>>>   create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
>>>
>>> diff --git a/arch/powerpc/include/asm/iommu.h
>>> b/arch/powerpc/include/asm/iommu.h
>>> index cbfe678..5ba66cb 100644
>>> --- a/arch/powerpc/include/asm/iommu.h
>>> +++ b/arch/powerpc/include/asm/iommu.h
>>> @@ -64,30 +64,33 @@ struct iommu_pool {  }
>>> ____cacheline_aligned_in_smp;
>>>
>>>   struct iommu_table {
>>>   	unsigned long  it_busno;     /* Bus number this table belongs to */
>>>   	unsigned long  it_size;      /* Size of iommu table in entries */
>>>   	unsigned long  it_offset;    /* Offset into global table */
>>>   	unsigned long  it_base;      /* mapped address of tce table */
>>>   	unsigned long  it_index;     /* which iommu table this is */
>>>   	unsigned long  it_type;      /* type: PCI or Virtual Bus */
>>>   	unsigned long  it_blocksize; /* Entries in each block (cacheline)
>> */
>>>   	unsigned long  poolsize;
>>>   	unsigned long  nr_pools;
>>>   	struct iommu_pool large_pool;
>>>   	struct iommu_pool pools[IOMMU_NR_POOLS];
>>>   	unsigned long *it_map;       /* A simple allocation bitmap for now
>> */
>>> +#ifdef CONFIG_IOMMU_API
>>> +	struct iommu_group *it_group;
>>> +#endif
>>>   };
>>>
>>>   struct scatterlist;
>>>
>>>   static inline void set_iommu_table_base(struct device *dev, void
>>> *base)  {
>>>   	dev->archdata.dma_data.iommu_table_base = base;  }
>>>
>>>   static inline void *get_iommu_table_base(struct device *dev)  {
>>>   	return dev->archdata.dma_data.iommu_table_base;
>>>   }
>>>
>>>   /* Frees table for an individual device node */ @@ -135,17 +138,20 @@
>>> static inline void pci_iommu_init(void) { }  extern void
>>> alloc_dart_table(void);  #if defined(CONFIG_PPC64) &&
>>> defined(CONFIG_PM)  static inline void iommu_save(void)  {
>>>   	if (ppc_md.iommu_save)
>>>   		ppc_md.iommu_save();
>>>   }
>>>
>>>   static inline void iommu_restore(void)  {
>>>   	if (ppc_md.iommu_restore)
>>>   		ppc_md.iommu_restore();
>>>   }
>>>   #endif
>>>
>>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long
>> entry, uint64_t tce,
>>> +		enum dma_data_direction direction, unsigned long pages);
>>> +
>>>   #endif /* __KERNEL__ */
>>>   #endif /* _ASM_IOMMU_H */
>>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
>>> index ff5a6ce..94f614b 100644
>>> --- a/arch/powerpc/kernel/iommu.c
>>> +++ b/arch/powerpc/kernel/iommu.c
>>> @@ -32,30 +32,31 @@
>>>   #include <linux/dma-mapping.h>
>>>   #include <linux/bitmap.h>
>>>   #include <linux/iommu-helper.h>
>>>   #include <linux/crash_dump.h>
>>>   #include <linux/hash.h>
>>>   #include <linux/fault-inject.h>
>>>   #include <linux/pci.h>
>>>   #include <asm/io.h>
>>>   #include <asm/prom.h>
>>>   #include <asm/iommu.h>
>>>   #include <asm/pci-bridge.h>
>>>   #include <asm/machdep.h>
>>>   #include <asm/kdump.h>
>>>   #include <asm/fadump.h>
>>>   #include <asm/vio.h>
>>> +#include <asm/tce.h>
>>>
>>>   #define DBG(...)
>>>
>>>   static int novmerge;
>>>
>>>   static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned
>>> int);
>>>
>>>   static int __init setup_iommu(char *str)  {
>>>   	if (!strcmp(str, "novmerge"))
>>>   		novmerge = 1;
>>>   	else if (!strcmp(str, "vmerge"))
>>>   		novmerge = 0;
>>>   	return 1;
>>>   }
>>> @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev,
>>> struct iommu_table *tbl,  }
>>>
>>>   void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>>>   			 void *vaddr, dma_addr_t dma_handle)  {
>>>   	if (tbl) {
>>>   		unsigned int nio_pages;
>>>
>>>   		size = PAGE_ALIGN(size);
>>>   		nio_pages = size >> IOMMU_PAGE_SHIFT;
>>>   		iommu_free(tbl, dma_handle, nio_pages);
>>>   		size = PAGE_ALIGN(size);
>>>   		free_pages((unsigned long)vaddr, get_order(size));
>>>   	}
>>>   }
>>> +
>>> +#ifdef CONFIG_IOMMU_API
>>> +/*
>>> + * SPAPR TCE API
>>> + */
>>> +static struct page *free_tce(struct iommu_table *tbl, unsigned long
>>> +entry) {
>>> +	struct page *page = NULL;
>>
>> NULL initialization doesn't appear to be necessary
>>
>>> +	unsigned long oldtce;
>>> +
>>> +	oldtce = ppc_md.tce_get(tbl, entry);
>>> +
>>> +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
>>> +		return NULL;
>>> +
>>> +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
>>> +
>>> +	WARN_ON(!page);
>>> +	if (page && (oldtce & TCE_PCI_WRITE))
>>> +		SetPageDirty(page);
>>> +	ppc_md.tce_free(tbl, entry, 1);
>>> +
>>> +	return page;
>>> +}
>>> +
>>> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
>>> +		uint64_t tce, enum dma_data_direction direction) {
>>> +	int ret;
>>> +	struct page *page = NULL;
>>> +	unsigned long kva, offset;
>>> +
>>> +	/* Map new TCE */
>>> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
>>> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
>>> +			direction != DMA_TO_DEVICE, &page);
>>> +	if (ret < 1) {
>>> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed
>> tce=%llx ioba=%lx ret=%d\n",
>>> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
>>> +		if (!ret)
>>> +			ret = -EFAULT;
>>
>> Missing return ret?  Otherwise we've got some bogus uses of page below
>> and we're setting ret for no reason here.
>>
>>> +	}
>>> +
>>> +	kva = (unsigned long) page_address(page);
>>> +	kva += offset;
>>> +
>>> +	/* tce_build receives a virtual address */
>>> +	entry += tbl->it_offset; /* Offset into real TCE table */
>>> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
>>> +
>>> +	/* tce_build() only returns non-zero for transient errors */
>>> +	if (unlikely(ret)) {
>>> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx
>> ioba=%lx kva=%lx ret=%d\n",
>>> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
>>> +		put_page(page);
>>> +		return -EIO;
>>> +	}
>>> +
>>> +	return 0;
>>> +}
>>> +
>>> +static void tce_flush(struct iommu_table *tbl) {
>>> +	/* Flush/invalidate TLB caches if necessary */
>>> +	if (ppc_md.tce_flush)
>>> +		ppc_md.tce_flush(tbl);
>>> +
>>> +	/* Make sure updates are seen by hardware */
>>> +	mb();
>>> +}
>>> +
>>> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
>> uint64_t tce,
>>> +		enum dma_data_direction direction, unsigned long pages) {
>>> +	int i, ret = 0, pages_to_put = 0;
>>> +	struct page *page;
>>> +	struct iommu_pool *pool = get_pool(tbl, entry);
>>> +	struct page **oldpages;
>>> +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
>>> +
>>> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
>>> +
>>> +	/* Handle a single page request without allocation
>>> +	   of pages-to-release array */
>>> +	if (pages == 1) {
>>> +		spin_lock(&(pool->lock));
>>> +		page = free_tce(tbl, entry);
>>> +
>>> +		if (direction != DMA_NONE)
>>> +			ret = put_tce(tbl, entry, tce, direction);
>>> +
>>> +		tce_flush(tbl);
>>> +
>>> +		if (page)
>>> +			put_page(page);
>>> +
>>> +		spin_unlock(&(pool->lock));
>>> +		return ret;
>>> +	}
>>> +
>>> +	/* Releasing multiple pages */
>>> +	/* Allocate an array for pages to be released after TCE table
>>> +	   is updated */
>>> +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
>>> +	if (!oldpages)
>>> +		return -ENOMEM;
>>> +
>>> +	spin_lock(&(pool->lock));
>>> +
>>> +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce +=
>> IOMMU_PAGE_SIZE) {
>>> +		page = free_tce(tbl, entry);
>>> +		if (page) {
>>> +			oldpages[pages_to_put] = page;
>>> +			++pages_to_put;
>>> +		}
>>> +
>>> +		if (direction != DMA_NONE)
>>> +			ret = put_tce(tbl, entry, tce, direction);
>>> +
>>> +		/* Release old pages if we reached the end of oldpages[] or
>>> +		   it is the last page or we are about to exit the loop */
>>> +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret)
>> {
>>> +			tce_flush(tbl);
>>
>> Avoiding tce_flush() is the reason for all this extra overhead, right?
>> I wonder if it'd be cleaner separating map vs unmap, where the map case
>> can avoid the oldpages array... but that means inserting new mappings on
>> top of old ones wouldn't put the pages.


Yes, we do not want to loose pages if the guest forgot to unmap them.


>>> +
>>> +			/* Release pages after removing them from TCE table */
>>> +			while (pages_to_put) {
>>> +				--pages_to_put;
>>> +				put_page(oldpages[pages_to_put]);
>>> +			}
>>> +		}
>>> +	}
>>> +
>>> +	spin_unlock(&(pool->lock));
>>> +	kfree(oldpages);
>>> +
>>> +	return ret;
>>> +}
>>> +EXPORT_SYMBOL_GPL(iommu_put_tces);
>>> +#endif /* CONFIG_IOMMU_API */
>>> diff --git a/arch/powerpc/platforms/powernv/pci.c
>>> b/arch/powerpc/platforms/powernv/pci.c
>>> index 05205cf..676f4d9 100644
>>> --- a/arch/powerpc/platforms/powernv/pci.c
>>> +++ b/arch/powerpc/platforms/powernv/pci.c
>>> @@ -8,30 +8,31 @@
>>>    * This program is free software; you can redistribute it and/or
>>>    * modify it under the terms of the GNU General Public License
>>>    * as published by the Free Software Foundation; either version
>>>    * 2 of the License, or (at your option) any later version.
>>>    */
>>>
>>>   #include <linux/kernel.h>
>>>   #include <linux/pci.h>
>>>   #include <linux/delay.h>
>>>   #include <linux/string.h>
>>>   #include <linux/init.h>
>>>   #include <linux/bootmem.h>
>>>   #include <linux/irq.h>
>>>   #include <linux/io.h>
>>>   #include <linux/msi.h>
>>> +#include <linux/iommu.h>
>>>
>>>   #include <asm/sections.h>
>>>   #include <asm/io.h>
>>>   #include <asm/prom.h>
>>>   #include <asm/pci-bridge.h>
>>>   #include <asm/machdep.h>
>>>   #include <asm/ppc-pci.h>
>>>   #include <asm/opal.h>
>>>   #include <asm/iommu.h>
>>>   #include <asm/tce.h>
>>>   #include <asm/abs_addr.h>
>>>   #include <asm/firmware.h>
>>>
>>>   #include "powernv.h"
>>>   #include "pci.h"
>>> @@ -601,15 +602,149 @@ void __init pnv_pci_init(void)
>>>   	/* Configure IOMMU DMA hooks */
>>>   	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
>>>   	ppc_md.tce_build = pnv_tce_build;
>>>   	ppc_md.tce_free = pnv_tce_free;
>>>   	ppc_md.tce_get = pnv_tce_get;
>>>   	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
>>>   	set_pci_dma_ops(&dma_iommu_ops);
>>>
>>>   	/* Configure MSIs */
>>>   #ifdef CONFIG_PCI_MSI
>>>   	ppc_md.msi_check_device = pnv_msi_check_device;
>>>   	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
>>>   	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;  #endif  }
>>> +
>>> +#ifdef CONFIG_IOMMU_API
>>> +/*
>>> + * IOMMU groups support required by VFIO  */ static int
>>> +add_device(struct device *dev) {
>>> +	struct iommu_table *tbl;
>>> +	int ret = 0;
>>> +
>>> +	if (WARN_ON(dev->iommu_group)) {
>>> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu
>> group %d, skipping\n",
>>> +				dev->kobj.name,
>>
>> dev_name(dev)
>>
>>> +				iommu_group_id(dev->iommu_group));
>>> +		return -EBUSY;
>>> +	}
>>> +
>>> +	tbl = get_iommu_table_base(dev);
>>> +	if (!tbl) {
>>> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
>>> +				dev->kobj.name);
>>> +		return 0;
>>> +	}
>>> +
>>> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
>>> +			dev->kobj.name, iommu_group_id(tbl->it_group));
>>> +
>>> +	ret = iommu_group_add_device(tbl->it_group, dev);
>>> +	if (ret < 0)
>>> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
>>> +				dev->kobj.name, ret);
>>> +
>>> +	return ret;
>>> +}
>>> +
>>> +static void del_device(struct device *dev) {
>>> +	iommu_group_remove_device(dev);
>>> +}
>>> +
>>> +static int iommu_bus_notifier(struct notifier_block *nb,
>>> +			      unsigned long action, void *data) {
>>> +	struct device *dev = data;
>>> +
>>> +	switch (action) {
>>> +	case BUS_NOTIFY_ADD_DEVICE:
>>> +		return add_device(dev);
>>> +	case BUS_NOTIFY_DEL_DEVICE:
>>> +		del_device(dev);
>>> +		return 0;
>>> +	default:
>>> +		return 0;
>>> +	}
>>> +}
>>> +
>>> +static struct notifier_block tce_iommu_bus_nb = {
>>> +	.notifier_call = iommu_bus_notifier, };
>>> +
>>> +static void group_release(void *iommu_data) {
>>> +	struct iommu_table *tbl = iommu_data;
>>> +	tbl->it_group = NULL;
>>> +}
>>> +
>>> +static int __init tce_iommu_init(void) {
>>> +	struct pci_dev *pdev = NULL;
>>> +	struct iommu_table *tbl;
>>> +	struct iommu_group *grp;
>>> +
>>> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
>>
>> There's already a notifier in the iommu code if you were to register an
>> iommu_ops with the add/remove_device entries.  That would allow you to
>> remove the notifier block and notifier function below and the second loop
>> below.  Are you avoiding that to avoid the rest of iommu_ops?

Yes. I need to implement either a small part of iommu_ops (especially the 
part which I think should not be there at all) or notifier, cannot how how 
the first is simpler.


> [Sethi Varun-B16395] Could be one reason, also they are associating the iommu group with the tce table entry and not the device.

Also true.
I would actually allocate IOMMU groups right after we discovered the new 
one but this is done during PCI scan which works before 
subsys_initcall(iommu_init) is called so I added this first loop.


>> Also, shouldn't this notifier only be registered after the first loop
>> below?  Otherwise ADD_DEVICE could race with setting up groups, which we
>> assume are present in the add_device() above.
> [Sethi Varun-B16395] Isn't this similar to how how the notifier is registered in iommu_bus_init? First a notifier is registered and then we check for devices that have already been probed.

Yep. Not very familiar with this stuff but if it is done one way and it is 
already upstream, I cannot see why I should go another way :)



-- 
Alexey

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Alex Williamson]

On Fri, 2012-11-23 at 13:02 +1100, Alexey Kardashevskiy wrote:
> On 22/11/12 22:56, Sethi Varun-B16395 wrote:
> >
> >
> >> -----Original Message-----
> >> From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
> >> owner@vger.kernel.org] On Behalf Of Alex Williamson
> >> Sent: Tuesday, November 20, 2012 11:50 PM
> >> To: Alexey Kardashevskiy
> >> Cc: Benjamin Herrenschmidt; Paul Mackerras; linuxppc-
> >> dev@lists.ozlabs.org; linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
> >> David Gibson
> >> Subject: Re: [PATCH] vfio powerpc: enabled and supported on powernv
> >> platform
> >>
> >> On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
> >>> VFIO implements platform independent stuff such as a PCI driver, BAR
> >>> access (via read/write on a file descriptor or direct mapping when
> >>> possible) and IRQ signaling.
> >>> The platform dependent part includes IOMMU initialization and
> >>> handling.
> >>>
> >>> This patch initializes IOMMU groups based on the IOMMU configuration
> >>> discovered during the PCI scan, only POWERNV platform is supported at
> >>> the moment.
> >>>
> >>> Also the patch implements an VFIO-IOMMU driver which manages DMA
> >>> mapping/unmapping requests coming from the client (now QEMU). It also
> >>> returns a DMA window information to let the guest initialize the
> >>> device tree for a guest OS properly. Although this driver has been
> >>> tested only on POWERNV, it should work on any platform supporting TCE
> >>> tables.
> >>>
> >>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config option.
> >>>
> >>> Cc: David Gibson <david@gibson.dropbear.id.au>
> >>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >>> ---
> >>>   arch/powerpc/include/asm/iommu.h     |    6 +
> >>>   arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
> >>>   arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
> >>>   drivers/iommu/Kconfig                |    8 ++
> >>>   drivers/vfio/Kconfig                 |    6 +
> >>>   drivers/vfio/Makefile                |    1 +
> >>>   drivers/vfio/vfio_iommu_spapr_tce.c  |  247
> >> ++++++++++++++++++++++++++++++++++
> >>>   include/linux/vfio.h                 |   20 +++
> >>>   8 files changed, 563 insertions(+)
> >>>   create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> >>>
> >>> diff --git a/arch/powerpc/include/asm/iommu.h
> >>> b/arch/powerpc/include/asm/iommu.h
> >>> index cbfe678..5ba66cb 100644
> >>> --- a/arch/powerpc/include/asm/iommu.h
> >>> +++ b/arch/powerpc/include/asm/iommu.h
> >>> @@ -64,30 +64,33 @@ struct iommu_pool {  }
> >>> ____cacheline_aligned_in_smp;
> >>>
> >>>   struct iommu_table {
> >>>   	unsigned long  it_busno;     /* Bus number this table belongs to */
> >>>   	unsigned long  it_size;      /* Size of iommu table in entries */
> >>>   	unsigned long  it_offset;    /* Offset into global table */
> >>>   	unsigned long  it_base;      /* mapped address of tce table */
> >>>   	unsigned long  it_index;     /* which iommu table this is */
> >>>   	unsigned long  it_type;      /* type: PCI or Virtual Bus */
> >>>   	unsigned long  it_blocksize; /* Entries in each block (cacheline)
> >> */
> >>>   	unsigned long  poolsize;
> >>>   	unsigned long  nr_pools;
> >>>   	struct iommu_pool large_pool;
> >>>   	struct iommu_pool pools[IOMMU_NR_POOLS];
> >>>   	unsigned long *it_map;       /* A simple allocation bitmap for now
> >> */
> >>> +#ifdef CONFIG_IOMMU_API
> >>> +	struct iommu_group *it_group;
> >>> +#endif
> >>>   };
> >>>
> >>>   struct scatterlist;
> >>>
> >>>   static inline void set_iommu_table_base(struct device *dev, void
> >>> *base)  {
> >>>   	dev->archdata.dma_data.iommu_table_base = base;  }
> >>>
> >>>   static inline void *get_iommu_table_base(struct device *dev)  {
> >>>   	return dev->archdata.dma_data.iommu_table_base;
> >>>   }
> >>>
> >>>   /* Frees table for an individual device node */ @@ -135,17 +138,20 @@
> >>> static inline void pci_iommu_init(void) { }  extern void
> >>> alloc_dart_table(void);  #if defined(CONFIG_PPC64) &&
> >>> defined(CONFIG_PM)  static inline void iommu_save(void)  {
> >>>   	if (ppc_md.iommu_save)
> >>>   		ppc_md.iommu_save();
> >>>   }
> >>>
> >>>   static inline void iommu_restore(void)  {
> >>>   	if (ppc_md.iommu_restore)
> >>>   		ppc_md.iommu_restore();
> >>>   }
> >>>   #endif
> >>>
> >>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long
> >> entry, uint64_t tce,
> >>> +		enum dma_data_direction direction, unsigned long pages);
> >>> +
> >>>   #endif /* __KERNEL__ */
> >>>   #endif /* _ASM_IOMMU_H */
> >>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> >>> index ff5a6ce..94f614b 100644
> >>> --- a/arch/powerpc/kernel/iommu.c
> >>> +++ b/arch/powerpc/kernel/iommu.c
> >>> @@ -32,30 +32,31 @@
> >>>   #include <linux/dma-mapping.h>
> >>>   #include <linux/bitmap.h>
> >>>   #include <linux/iommu-helper.h>
> >>>   #include <linux/crash_dump.h>
> >>>   #include <linux/hash.h>
> >>>   #include <linux/fault-inject.h>
> >>>   #include <linux/pci.h>
> >>>   #include <asm/io.h>
> >>>   #include <asm/prom.h>
> >>>   #include <asm/iommu.h>
> >>>   #include <asm/pci-bridge.h>
> >>>   #include <asm/machdep.h>
> >>>   #include <asm/kdump.h>
> >>>   #include <asm/fadump.h>
> >>>   #include <asm/vio.h>
> >>> +#include <asm/tce.h>
> >>>
> >>>   #define DBG(...)
> >>>
> >>>   static int novmerge;
> >>>
> >>>   static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned
> >>> int);
> >>>
> >>>   static int __init setup_iommu(char *str)  {
> >>>   	if (!strcmp(str, "novmerge"))
> >>>   		novmerge = 1;
> >>>   	else if (!strcmp(str, "vmerge"))
> >>>   		novmerge = 0;
> >>>   	return 1;
> >>>   }
> >>> @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev,
> >>> struct iommu_table *tbl,  }
> >>>
> >>>   void iommu_free_coherent(struct iommu_table *tbl, size_t size,
> >>>   			 void *vaddr, dma_addr_t dma_handle)  {
> >>>   	if (tbl) {
> >>>   		unsigned int nio_pages;
> >>>
> >>>   		size = PAGE_ALIGN(size);
> >>>   		nio_pages = size >> IOMMU_PAGE_SHIFT;
> >>>   		iommu_free(tbl, dma_handle, nio_pages);
> >>>   		size = PAGE_ALIGN(size);
> >>>   		free_pages((unsigned long)vaddr, get_order(size));
> >>>   	}
> >>>   }
> >>> +
> >>> +#ifdef CONFIG_IOMMU_API
> >>> +/*
> >>> + * SPAPR TCE API
> >>> + */
> >>> +static struct page *free_tce(struct iommu_table *tbl, unsigned long
> >>> +entry) {
> >>> +	struct page *page = NULL;
> >>
> >> NULL initialization doesn't appear to be necessary
> >>
> >>> +	unsigned long oldtce;
> >>> +
> >>> +	oldtce = ppc_md.tce_get(tbl, entry);
> >>> +
> >>> +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> >>> +		return NULL;
> >>> +
> >>> +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
> >>> +
> >>> +	WARN_ON(!page);
> >>> +	if (page && (oldtce & TCE_PCI_WRITE))
> >>> +		SetPageDirty(page);
> >>> +	ppc_md.tce_free(tbl, entry, 1);
> >>> +
> >>> +	return page;
> >>> +}
> >>> +
> >>> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> >>> +		uint64_t tce, enum dma_data_direction direction) {
> >>> +	int ret;
> >>> +	struct page *page = NULL;
> >>> +	unsigned long kva, offset;
> >>> +
> >>> +	/* Map new TCE */
> >>> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> >>> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> >>> +			direction != DMA_TO_DEVICE, &page);
> >>> +	if (ret < 1) {
> >>> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed
> >> tce=%llx ioba=%lx ret=%d\n",
> >>> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> >>> +		if (!ret)
> >>> +			ret = -EFAULT;
> >>
> >> Missing return ret?  Otherwise we've got some bogus uses of page below
> >> and we're setting ret for no reason here.
> >>
> >>> +	}
> >>> +
> >>> +	kva = (unsigned long) page_address(page);
> >>> +	kva += offset;
> >>> +
> >>> +	/* tce_build receives a virtual address */
> >>> +	entry += tbl->it_offset; /* Offset into real TCE table */
> >>> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> >>> +
> >>> +	/* tce_build() only returns non-zero for transient errors */
> >>> +	if (unlikely(ret)) {
> >>> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx
> >> ioba=%lx kva=%lx ret=%d\n",
> >>> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> >>> +		put_page(page);
> >>> +		return -EIO;
> >>> +	}
> >>> +
> >>> +	return 0;
> >>> +}
> >>> +
> >>> +static void tce_flush(struct iommu_table *tbl) {
> >>> +	/* Flush/invalidate TLB caches if necessary */
> >>> +	if (ppc_md.tce_flush)
> >>> +		ppc_md.tce_flush(tbl);
> >>> +
> >>> +	/* Make sure updates are seen by hardware */
> >>> +	mb();
> >>> +}
> >>> +
> >>> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> >> uint64_t tce,
> >>> +		enum dma_data_direction direction, unsigned long pages) {
> >>> +	int i, ret = 0, pages_to_put = 0;
> >>> +	struct page *page;
> >>> +	struct iommu_pool *pool = get_pool(tbl, entry);
> >>> +	struct page **oldpages;
> >>> +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
> >>> +
> >>> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> >>> +
> >>> +	/* Handle a single page request without allocation
> >>> +	   of pages-to-release array */
> >>> +	if (pages == 1) {
> >>> +		spin_lock(&(pool->lock));
> >>> +		page = free_tce(tbl, entry);
> >>> +
> >>> +		if (direction != DMA_NONE)
> >>> +			ret = put_tce(tbl, entry, tce, direction);
> >>> +
> >>> +		tce_flush(tbl);
> >>> +
> >>> +		if (page)
> >>> +			put_page(page);
> >>> +
> >>> +		spin_unlock(&(pool->lock));
> >>> +		return ret;
> >>> +	}
> >>> +
> >>> +	/* Releasing multiple pages */
> >>> +	/* Allocate an array for pages to be released after TCE table
> >>> +	   is updated */
> >>> +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
> >>> +	if (!oldpages)
> >>> +		return -ENOMEM;
> >>> +
> >>> +	spin_lock(&(pool->lock));
> >>> +
> >>> +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce +=
> >> IOMMU_PAGE_SIZE) {
> >>> +		page = free_tce(tbl, entry);
> >>> +		if (page) {
> >>> +			oldpages[pages_to_put] = page;
> >>> +			++pages_to_put;
> >>> +		}
> >>> +
> >>> +		if (direction != DMA_NONE)
> >>> +			ret = put_tce(tbl, entry, tce, direction);
> >>> +
> >>> +		/* Release old pages if we reached the end of oldpages[] or
> >>> +		   it is the last page or we are about to exit the loop */
> >>> +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret)
> >> {
> >>> +			tce_flush(tbl);
> >>
> >> Avoiding tce_flush() is the reason for all this extra overhead, right?
> >> I wonder if it'd be cleaner separating map vs unmap, where the map case
> >> can avoid the oldpages array... but that means inserting new mappings on
> >> top of old ones wouldn't put the pages.
> 
> 
> Yes, we do not want to loose pages if the guest forgot to unmap them.

Hmm, does that mean we're not actively clearing tce entries or somehow
disabling the iommu window when the iommu is released through vfio?

> >>> +
> >>> +			/* Release pages after removing them from TCE table */
> >>> +			while (pages_to_put) {
> >>> +				--pages_to_put;
> >>> +				put_page(oldpages[pages_to_put]);
> >>> +			}
> >>> +		}
> >>> +	}
> >>> +
> >>> +	spin_unlock(&(pool->lock));
> >>> +	kfree(oldpages);
> >>> +
> >>> +	return ret;
> >>> +}
> >>> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> >>> +#endif /* CONFIG_IOMMU_API */
> >>> diff --git a/arch/powerpc/platforms/powernv/pci.c
> >>> b/arch/powerpc/platforms/powernv/pci.c
> >>> index 05205cf..676f4d9 100644
> >>> --- a/arch/powerpc/platforms/powernv/pci.c
> >>> +++ b/arch/powerpc/platforms/powernv/pci.c
> >>> @@ -8,30 +8,31 @@
> >>>    * This program is free software; you can redistribute it and/or
> >>>    * modify it under the terms of the GNU General Public License
> >>>    * as published by the Free Software Foundation; either version
> >>>    * 2 of the License, or (at your option) any later version.
> >>>    */
> >>>
> >>>   #include <linux/kernel.h>
> >>>   #include <linux/pci.h>
> >>>   #include <linux/delay.h>
> >>>   #include <linux/string.h>
> >>>   #include <linux/init.h>
> >>>   #include <linux/bootmem.h>
> >>>   #include <linux/irq.h>
> >>>   #include <linux/io.h>
> >>>   #include <linux/msi.h>
> >>> +#include <linux/iommu.h>
> >>>
> >>>   #include <asm/sections.h>
> >>>   #include <asm/io.h>
> >>>   #include <asm/prom.h>
> >>>   #include <asm/pci-bridge.h>
> >>>   #include <asm/machdep.h>
> >>>   #include <asm/ppc-pci.h>
> >>>   #include <asm/opal.h>
> >>>   #include <asm/iommu.h>
> >>>   #include <asm/tce.h>
> >>>   #include <asm/abs_addr.h>
> >>>   #include <asm/firmware.h>
> >>>
> >>>   #include "powernv.h"
> >>>   #include "pci.h"
> >>> @@ -601,15 +602,149 @@ void __init pnv_pci_init(void)
> >>>   	/* Configure IOMMU DMA hooks */
> >>>   	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
> >>>   	ppc_md.tce_build = pnv_tce_build;
> >>>   	ppc_md.tce_free = pnv_tce_free;
> >>>   	ppc_md.tce_get = pnv_tce_get;
> >>>   	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
> >>>   	set_pci_dma_ops(&dma_iommu_ops);
> >>>
> >>>   	/* Configure MSIs */
> >>>   #ifdef CONFIG_PCI_MSI
> >>>   	ppc_md.msi_check_device = pnv_msi_check_device;
> >>>   	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
> >>>   	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;  #endif  }
> >>> +
> >>> +#ifdef CONFIG_IOMMU_API
> >>> +/*
> >>> + * IOMMU groups support required by VFIO  */ static int
> >>> +add_device(struct device *dev) {
> >>> +	struct iommu_table *tbl;
> >>> +	int ret = 0;
> >>> +
> >>> +	if (WARN_ON(dev->iommu_group)) {
> >>> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu
> >> group %d, skipping\n",
> >>> +				dev->kobj.name,
> >>
> >> dev_name(dev)
> >>
> >>> +				iommu_group_id(dev->iommu_group));
> >>> +		return -EBUSY;
> >>> +	}
> >>> +
> >>> +	tbl = get_iommu_table_base(dev);
> >>> +	if (!tbl) {
> >>> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> >>> +				dev->kobj.name);
> >>> +		return 0;
> >>> +	}
> >>> +
> >>> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> >>> +			dev->kobj.name, iommu_group_id(tbl->it_group));
> >>> +
> >>> +	ret = iommu_group_add_device(tbl->it_group, dev);
> >>> +	if (ret < 0)
> >>> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> >>> +				dev->kobj.name, ret);
> >>> +
> >>> +	return ret;
> >>> +}
> >>> +
> >>> +static void del_device(struct device *dev) {
> >>> +	iommu_group_remove_device(dev);
> >>> +}
> >>> +
> >>> +static int iommu_bus_notifier(struct notifier_block *nb,
> >>> +			      unsigned long action, void *data) {
> >>> +	struct device *dev = data;
> >>> +
> >>> +	switch (action) {
> >>> +	case BUS_NOTIFY_ADD_DEVICE:
> >>> +		return add_device(dev);
> >>> +	case BUS_NOTIFY_DEL_DEVICE:
> >>> +		del_device(dev);
> >>> +		return 0;
> >>> +	default:
> >>> +		return 0;
> >>> +	}
> >>> +}
> >>> +
> >>> +static struct notifier_block tce_iommu_bus_nb = {
> >>> +	.notifier_call = iommu_bus_notifier, };
> >>> +
> >>> +static void group_release(void *iommu_data) {
> >>> +	struct iommu_table *tbl = iommu_data;
> >>> +	tbl->it_group = NULL;
> >>> +}
> >>> +
> >>> +static int __init tce_iommu_init(void) {
> >>> +	struct pci_dev *pdev = NULL;
> >>> +	struct iommu_table *tbl;
> >>> +	struct iommu_group *grp;
> >>> +
> >>> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> >>
> >> There's already a notifier in the iommu code if you were to register an
> >> iommu_ops with the add/remove_device entries.  That would allow you to
> >> remove the notifier block and notifier function below and the second loop
> >> below.  Are you avoiding that to avoid the rest of iommu_ops?
> 
> Yes. I need to implement either a small part of iommu_ops (especially the 
> part which I think should not be there at all) or notifier, cannot how how 
> the first is simpler.
> 
> 
> > [Sethi Varun-B16395] Could be one reason, also they are associating the iommu group with the tce table entry and not the device.
> 
> Also true.
> I would actually allocate IOMMU groups right after we discovered the new 
> one but this is done during PCI scan which works before 
> subsys_initcall(iommu_init) is called so I added this first loop.
> 
> 
> >> Also, shouldn't this notifier only be registered after the first loop
> >> below?  Otherwise ADD_DEVICE could race with setting up groups, which we
> >> assume are present in the add_device() above.
> > [Sethi Varun-B16395] Isn't this similar to how how the notifier is registered in iommu_bus_init? First a notifier is registered and then we check for devices that have already been probed.
> 
> Yep. Not very familiar with this stuff but if it is done one way and it is 
> already upstream, I cannot see why I should go another way :)

The existing notifier callback and loop should be able to operate in
parallel... of course they don't because we're not actively adding new
devices at the point where it's setup.  IIRC, the notifier callback
blindly uses something that's not setup at the point it's registered.
That's a bit sloppy.  Maybe I'm mis-remembering, I'll verify in your new
version.  Thanks,

Alex

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Alex Williamson]

On Mon, 2012-11-26 at 08:18 -0700, Alex Williamson wrote:
> On Fri, 2012-11-23 at 13:02 +1100, Alexey Kardashevskiy wrote:
> > On 22/11/12 22:56, Sethi Varun-B16395 wrote:
> > >
> > >
> > >> -----Original Message-----
> > >> From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
> > >> owner@vger.kernel.org] On Behalf Of Alex Williamson
> > >> Sent: Tuesday, November 20, 2012 11:50 PM
> > >> To: Alexey Kardashevskiy
> > >> Cc: Benjamin Herrenschmidt; Paul Mackerras; linuxppc-
> > >> dev@lists.ozlabs.org; linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
> > >> David Gibson
> > >> Subject: Re: [PATCH] vfio powerpc: enabled and supported on powernv
> > >> platform
> > >>
> > >> On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
> > >>> VFIO implements platform independent stuff such as a PCI driver, BAR
> > >>> access (via read/write on a file descriptor or direct mapping when
> > >>> possible) and IRQ signaling.
> > >>> The platform dependent part includes IOMMU initialization and
> > >>> handling.
> > >>>
> > >>> This patch initializes IOMMU groups based on the IOMMU configuration
> > >>> discovered during the PCI scan, only POWERNV platform is supported at
> > >>> the moment.
> > >>>
> > >>> Also the patch implements an VFIO-IOMMU driver which manages DMA
> > >>> mapping/unmapping requests coming from the client (now QEMU). It also
> > >>> returns a DMA window information to let the guest initialize the
> > >>> device tree for a guest OS properly. Although this driver has been
> > >>> tested only on POWERNV, it should work on any platform supporting TCE
> > >>> tables.
> > >>>
> > >>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config option.
> > >>>
> > >>> Cc: David Gibson <david@gibson.dropbear.id.au>
> > >>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> > >>> ---
> > >>>   arch/powerpc/include/asm/iommu.h     |    6 +
> > >>>   arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
> > >>>   arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
> > >>>   drivers/iommu/Kconfig                |    8 ++
> > >>>   drivers/vfio/Kconfig                 |    6 +
> > >>>   drivers/vfio/Makefile                |    1 +
> > >>>   drivers/vfio/vfio_iommu_spapr_tce.c  |  247
> > >> ++++++++++++++++++++++++++++++++++
> > >>>   include/linux/vfio.h                 |   20 +++
> > >>>   8 files changed, 563 insertions(+)
> > >>>   create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> > >>>
> > >>> diff --git a/arch/powerpc/include/asm/iommu.h
> > >>> b/arch/powerpc/include/asm/iommu.h
> > >>> index cbfe678..5ba66cb 100644
> > >>> --- a/arch/powerpc/include/asm/iommu.h
> > >>> +++ b/arch/powerpc/include/asm/iommu.h
> > >>> @@ -64,30 +64,33 @@ struct iommu_pool {  }
> > >>> ____cacheline_aligned_in_smp;
> > >>>
> > >>>   struct iommu_table {
> > >>>   	unsigned long  it_busno;     /* Bus number this table belongs to */
> > >>>   	unsigned long  it_size;      /* Size of iommu table in entries */
> > >>>   	unsigned long  it_offset;    /* Offset into global table */
> > >>>   	unsigned long  it_base;      /* mapped address of tce table */
> > >>>   	unsigned long  it_index;     /* which iommu table this is */
> > >>>   	unsigned long  it_type;      /* type: PCI or Virtual Bus */
> > >>>   	unsigned long  it_blocksize; /* Entries in each block (cacheline)
> > >> */
> > >>>   	unsigned long  poolsize;
> > >>>   	unsigned long  nr_pools;
> > >>>   	struct iommu_pool large_pool;
> > >>>   	struct iommu_pool pools[IOMMU_NR_POOLS];
> > >>>   	unsigned long *it_map;       /* A simple allocation bitmap for now
> > >> */
> > >>> +#ifdef CONFIG_IOMMU_API
> > >>> +	struct iommu_group *it_group;
> > >>> +#endif
> > >>>   };
> > >>>
> > >>>   struct scatterlist;
> > >>>
> > >>>   static inline void set_iommu_table_base(struct device *dev, void
> > >>> *base)  {
> > >>>   	dev->archdata.dma_data.iommu_table_base = base;  }
> > >>>
> > >>>   static inline void *get_iommu_table_base(struct device *dev)  {
> > >>>   	return dev->archdata.dma_data.iommu_table_base;
> > >>>   }
> > >>>
> > >>>   /* Frees table for an individual device node */ @@ -135,17 +138,20 @@
> > >>> static inline void pci_iommu_init(void) { }  extern void
> > >>> alloc_dart_table(void);  #if defined(CONFIG_PPC64) &&
> > >>> defined(CONFIG_PM)  static inline void iommu_save(void)  {
> > >>>   	if (ppc_md.iommu_save)
> > >>>   		ppc_md.iommu_save();
> > >>>   }
> > >>>
> > >>>   static inline void iommu_restore(void)  {
> > >>>   	if (ppc_md.iommu_restore)
> > >>>   		ppc_md.iommu_restore();
> > >>>   }
> > >>>   #endif
> > >>>
> > >>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long
> > >> entry, uint64_t tce,
> > >>> +		enum dma_data_direction direction, unsigned long pages);
> > >>> +
> > >>>   #endif /* __KERNEL__ */
> > >>>   #endif /* _ASM_IOMMU_H */
> > >>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> > >>> index ff5a6ce..94f614b 100644
> > >>> --- a/arch/powerpc/kernel/iommu.c
> > >>> +++ b/arch/powerpc/kernel/iommu.c
> > >>> @@ -32,30 +32,31 @@
> > >>>   #include <linux/dma-mapping.h>
> > >>>   #include <linux/bitmap.h>
> > >>>   #include <linux/iommu-helper.h>
> > >>>   #include <linux/crash_dump.h>
> > >>>   #include <linux/hash.h>
> > >>>   #include <linux/fault-inject.h>
> > >>>   #include <linux/pci.h>
> > >>>   #include <asm/io.h>
> > >>>   #include <asm/prom.h>
> > >>>   #include <asm/iommu.h>
> > >>>   #include <asm/pci-bridge.h>
> > >>>   #include <asm/machdep.h>
> > >>>   #include <asm/kdump.h>
> > >>>   #include <asm/fadump.h>
> > >>>   #include <asm/vio.h>
> > >>> +#include <asm/tce.h>
> > >>>
> > >>>   #define DBG(...)
> > >>>
> > >>>   static int novmerge;
> > >>>
> > >>>   static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned
> > >>> int);
> > >>>
> > >>>   static int __init setup_iommu(char *str)  {
> > >>>   	if (!strcmp(str, "novmerge"))
> > >>>   		novmerge = 1;
> > >>>   	else if (!strcmp(str, "vmerge"))
> > >>>   		novmerge = 0;
> > >>>   	return 1;
> > >>>   }
> > >>> @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev,
> > >>> struct iommu_table *tbl,  }
> > >>>
> > >>>   void iommu_free_coherent(struct iommu_table *tbl, size_t size,
> > >>>   			 void *vaddr, dma_addr_t dma_handle)  {
> > >>>   	if (tbl) {
> > >>>   		unsigned int nio_pages;
> > >>>
> > >>>   		size = PAGE_ALIGN(size);
> > >>>   		nio_pages = size >> IOMMU_PAGE_SHIFT;
> > >>>   		iommu_free(tbl, dma_handle, nio_pages);
> > >>>   		size = PAGE_ALIGN(size);
> > >>>   		free_pages((unsigned long)vaddr, get_order(size));
> > >>>   	}
> > >>>   }
> > >>> +
> > >>> +#ifdef CONFIG_IOMMU_API
> > >>> +/*
> > >>> + * SPAPR TCE API
> > >>> + */
> > >>> +static struct page *free_tce(struct iommu_table *tbl, unsigned long
> > >>> +entry) {
> > >>> +	struct page *page = NULL;
> > >>
> > >> NULL initialization doesn't appear to be necessary
> > >>
> > >>> +	unsigned long oldtce;
> > >>> +
> > >>> +	oldtce = ppc_md.tce_get(tbl, entry);
> > >>> +
> > >>> +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> > >>> +		return NULL;
> > >>> +
> > >>> +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
> > >>> +
> > >>> +	WARN_ON(!page);
> > >>> +	if (page && (oldtce & TCE_PCI_WRITE))
> > >>> +		SetPageDirty(page);
> > >>> +	ppc_md.tce_free(tbl, entry, 1);
> > >>> +
> > >>> +	return page;
> > >>> +}
> > >>> +
> > >>> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> > >>> +		uint64_t tce, enum dma_data_direction direction) {
> > >>> +	int ret;
> > >>> +	struct page *page = NULL;
> > >>> +	unsigned long kva, offset;
> > >>> +
> > >>> +	/* Map new TCE */
> > >>> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> > >>> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> > >>> +			direction != DMA_TO_DEVICE, &page);
> > >>> +	if (ret < 1) {
> > >>> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed
> > >> tce=%llx ioba=%lx ret=%d\n",
> > >>> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> > >>> +		if (!ret)
> > >>> +			ret = -EFAULT;
> > >>
> > >> Missing return ret?  Otherwise we've got some bogus uses of page below
> > >> and we're setting ret for no reason here.
> > >>
> > >>> +	}
> > >>> +
> > >>> +	kva = (unsigned long) page_address(page);
> > >>> +	kva += offset;
> > >>> +
> > >>> +	/* tce_build receives a virtual address */
> > >>> +	entry += tbl->it_offset; /* Offset into real TCE table */
> > >>> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> > >>> +
> > >>> +	/* tce_build() only returns non-zero for transient errors */
> > >>> +	if (unlikely(ret)) {
> > >>> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx
> > >> ioba=%lx kva=%lx ret=%d\n",
> > >>> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> > >>> +		put_page(page);
> > >>> +		return -EIO;
> > >>> +	}
> > >>> +
> > >>> +	return 0;
> > >>> +}
> > >>> +
> > >>> +static void tce_flush(struct iommu_table *tbl) {
> > >>> +	/* Flush/invalidate TLB caches if necessary */
> > >>> +	if (ppc_md.tce_flush)
> > >>> +		ppc_md.tce_flush(tbl);
> > >>> +
> > >>> +	/* Make sure updates are seen by hardware */
> > >>> +	mb();
> > >>> +}
> > >>> +
> > >>> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> > >> uint64_t tce,
> > >>> +		enum dma_data_direction direction, unsigned long pages) {
> > >>> +	int i, ret = 0, pages_to_put = 0;
> > >>> +	struct page *page;
> > >>> +	struct iommu_pool *pool = get_pool(tbl, entry);
> > >>> +	struct page **oldpages;
> > >>> +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
> > >>> +
> > >>> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> > >>> +
> > >>> +	/* Handle a single page request without allocation
> > >>> +	   of pages-to-release array */
> > >>> +	if (pages == 1) {
> > >>> +		spin_lock(&(pool->lock));
> > >>> +		page = free_tce(tbl, entry);
> > >>> +
> > >>> +		if (direction != DMA_NONE)
> > >>> +			ret = put_tce(tbl, entry, tce, direction);
> > >>> +
> > >>> +		tce_flush(tbl);
> > >>> +
> > >>> +		if (page)
> > >>> +			put_page(page);
> > >>> +
> > >>> +		spin_unlock(&(pool->lock));
> > >>> +		return ret;
> > >>> +	}
> > >>> +
> > >>> +	/* Releasing multiple pages */
> > >>> +	/* Allocate an array for pages to be released after TCE table
> > >>> +	   is updated */
> > >>> +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
> > >>> +	if (!oldpages)
> > >>> +		return -ENOMEM;
> > >>> +
> > >>> +	spin_lock(&(pool->lock));
> > >>> +
> > >>> +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce +=
> > >> IOMMU_PAGE_SIZE) {
> > >>> +		page = free_tce(tbl, entry);
> > >>> +		if (page) {
> > >>> +			oldpages[pages_to_put] = page;
> > >>> +			++pages_to_put;
> > >>> +		}
> > >>> +
> > >>> +		if (direction != DMA_NONE)
> > >>> +			ret = put_tce(tbl, entry, tce, direction);
> > >>> +
> > >>> +		/* Release old pages if we reached the end of oldpages[] or
> > >>> +		   it is the last page or we are about to exit the loop */
> > >>> +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret)
> > >> {
> > >>> +			tce_flush(tbl);
> > >>
> > >> Avoiding tce_flush() is the reason for all this extra overhead, right?
> > >> I wonder if it'd be cleaner separating map vs unmap, where the map case
> > >> can avoid the oldpages array... but that means inserting new mappings on
> > >> top of old ones wouldn't put the pages.
> > 
> > 
> > Yes, we do not want to loose pages if the guest forgot to unmap them.
> 
> Hmm, does that mean we're not actively clearing tce entries or somehow
> disabling the iommu window when the iommu is released through vfio?

Ok, I see tces are put on shutdown via tce_iommu_detach_group, so you're
more concerned about the guest simply mapping over top of it's own
mappings.  Is that common?  Is it common enough for every multi-page
mapping to assume it will happen?  I know this is a performance
sensitive path for you and it seems like a map-only w/ fallback to
unmap, remap would be better in the general case.

On x86 we do exactly that, but we do the unmap, remap from userspace
when we get an EBUSY.  Thanks,

Alex

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Benjamin Herrenschmidt]

On Mon, 2012-11-26 at 11:04 -0700, Alex Williamson wrote:
> Ok, I see tces are put on shutdown via tce_iommu_detach_group, so you're
> more concerned about the guest simply mapping over top of it's own
> mappings.  Is that common?  Is it common enough for every multi-page
> mapping to assume it will happen?  I know this is a performance
> sensitive path for you and it seems like a map-only w/ fallback to
> unmap, remap would be better in the general case.
> 
> On x86 we do exactly that, but we do the unmap, remap from userspace
> when we get an EBUSY.  Thanks, 

Right, Linux as guest at least will never map "over" an existing
mapping. It will always unmap first. IE. The only transition we do on
H_PUT_TCE are 0 -> valid and valid -> 0.

So it would be fine to simplify the code and keep the "map over map" as
a slow fallback. I can't tell for other operating systems but we don't
care about those at this point :-)

Cheers,
Ben.

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Alexey Kardashevskiy]

On 27/11/12 05:04, Alex Williamson wrote:
> On Mon, 2012-11-26 at 08:18 -0700, Alex Williamson wrote:
>> On Fri, 2012-11-23 at 13:02 +1100, Alexey Kardashevskiy wrote:
>>> On 22/11/12 22:56, Sethi Varun-B16395 wrote:
>>>>
>>>>
>>>>> -----Original Message-----
>>>>> From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
>>>>> owner@vger.kernel.org] On Behalf Of Alex Williamson
>>>>> Sent: Tuesday, November 20, 2012 11:50 PM
>>>>> To: Alexey Kardashevskiy
>>>>> Cc: Benjamin Herrenschmidt; Paul Mackerras; linuxppc-
>>>>> dev@lists.ozlabs.org; linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
>>>>> David Gibson
>>>>> Subject: Re: [PATCH] vfio powerpc: enabled and supported on powernv
>>>>> platform
>>>>>
>>>>> On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
>>>>>> VFIO implements platform independent stuff such as a PCI driver, BAR
>>>>>> access (via read/write on a file descriptor or direct mapping when
>>>>>> possible) and IRQ signaling.
>>>>>> The platform dependent part includes IOMMU initialization and
>>>>>> handling.
>>>>>>
>>>>>> This patch initializes IOMMU groups based on the IOMMU configuration
>>>>>> discovered during the PCI scan, only POWERNV platform is supported at
>>>>>> the moment.
>>>>>>
>>>>>> Also the patch implements an VFIO-IOMMU driver which manages DMA
>>>>>> mapping/unmapping requests coming from the client (now QEMU). It also
>>>>>> returns a DMA window information to let the guest initialize the
>>>>>> device tree for a guest OS properly. Although this driver has been
>>>>>> tested only on POWERNV, it should work on any platform supporting TCE
>>>>>> tables.
>>>>>>
>>>>>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config option.
>>>>>>
>>>>>> Cc: David Gibson <david@gibson.dropbear.id.au>
>>>>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>>>>> ---
>>>>>>    arch/powerpc/include/asm/iommu.h     |    6 +
>>>>>>    arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
>>>>>>    arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
>>>>>>    drivers/iommu/Kconfig                |    8 ++
>>>>>>    drivers/vfio/Kconfig                 |    6 +
>>>>>>    drivers/vfio/Makefile                |    1 +
>>>>>>    drivers/vfio/vfio_iommu_spapr_tce.c  |  247
>>>>> ++++++++++++++++++++++++++++++++++
>>>>>>    include/linux/vfio.h                 |   20 +++
>>>>>>    8 files changed, 563 insertions(+)
>>>>>>    create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
>>>>>>
>>>>>> diff --git a/arch/powerpc/include/asm/iommu.h
>>>>>> b/arch/powerpc/include/asm/iommu.h
>>>>>> index cbfe678..5ba66cb 100644
>>>>>> --- a/arch/powerpc/include/asm/iommu.h
>>>>>> +++ b/arch/powerpc/include/asm/iommu.h
>>>>>> @@ -64,30 +64,33 @@ struct iommu_pool {  }
>>>>>> ____cacheline_aligned_in_smp;
>>>>>>
>>>>>>    struct iommu_table {
>>>>>>    	unsigned long  it_busno;     /* Bus number this table belongs to */
>>>>>>    	unsigned long  it_size;      /* Size of iommu table in entries */
>>>>>>    	unsigned long  it_offset;    /* Offset into global table */
>>>>>>    	unsigned long  it_base;      /* mapped address of tce table */
>>>>>>    	unsigned long  it_index;     /* which iommu table this is */
>>>>>>    	unsigned long  it_type;      /* type: PCI or Virtual Bus */
>>>>>>    	unsigned long  it_blocksize; /* Entries in each block (cacheline)
>>>>> */
>>>>>>    	unsigned long  poolsize;
>>>>>>    	unsigned long  nr_pools;
>>>>>>    	struct iommu_pool large_pool;
>>>>>>    	struct iommu_pool pools[IOMMU_NR_POOLS];
>>>>>>    	unsigned long *it_map;       /* A simple allocation bitmap for now
>>>>> */
>>>>>> +#ifdef CONFIG_IOMMU_API
>>>>>> +	struct iommu_group *it_group;
>>>>>> +#endif
>>>>>>    };
>>>>>>
>>>>>>    struct scatterlist;
>>>>>>
>>>>>>    static inline void set_iommu_table_base(struct device *dev, void
>>>>>> *base)  {
>>>>>>    	dev->archdata.dma_data.iommu_table_base = base;  }
>>>>>>
>>>>>>    static inline void *get_iommu_table_base(struct device *dev)  {
>>>>>>    	return dev->archdata.dma_data.iommu_table_base;
>>>>>>    }
>>>>>>
>>>>>>    /* Frees table for an individual device node */ @@ -135,17 +138,20 @@
>>>>>> static inline void pci_iommu_init(void) { }  extern void
>>>>>> alloc_dart_table(void);  #if defined(CONFIG_PPC64) &&
>>>>>> defined(CONFIG_PM)  static inline void iommu_save(void)  {
>>>>>>    	if (ppc_md.iommu_save)
>>>>>>    		ppc_md.iommu_save();
>>>>>>    }
>>>>>>
>>>>>>    static inline void iommu_restore(void)  {
>>>>>>    	if (ppc_md.iommu_restore)
>>>>>>    		ppc_md.iommu_restore();
>>>>>>    }
>>>>>>    #endif
>>>>>>
>>>>>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long
>>>>> entry, uint64_t tce,
>>>>>> +		enum dma_data_direction direction, unsigned long pages);
>>>>>> +
>>>>>>    #endif /* __KERNEL__ */
>>>>>>    #endif /* _ASM_IOMMU_H */
>>>>>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
>>>>>> index ff5a6ce..94f614b 100644
>>>>>> --- a/arch/powerpc/kernel/iommu.c
>>>>>> +++ b/arch/powerpc/kernel/iommu.c
>>>>>> @@ -32,30 +32,31 @@
>>>>>>    #include <linux/dma-mapping.h>
>>>>>>    #include <linux/bitmap.h>
>>>>>>    #include <linux/iommu-helper.h>
>>>>>>    #include <linux/crash_dump.h>
>>>>>>    #include <linux/hash.h>
>>>>>>    #include <linux/fault-inject.h>
>>>>>>    #include <linux/pci.h>
>>>>>>    #include <asm/io.h>
>>>>>>    #include <asm/prom.h>
>>>>>>    #include <asm/iommu.h>
>>>>>>    #include <asm/pci-bridge.h>
>>>>>>    #include <asm/machdep.h>
>>>>>>    #include <asm/kdump.h>
>>>>>>    #include <asm/fadump.h>
>>>>>>    #include <asm/vio.h>
>>>>>> +#include <asm/tce.h>
>>>>>>
>>>>>>    #define DBG(...)
>>>>>>
>>>>>>    static int novmerge;
>>>>>>
>>>>>>    static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned
>>>>>> int);
>>>>>>
>>>>>>    static int __init setup_iommu(char *str)  {
>>>>>>    	if (!strcmp(str, "novmerge"))
>>>>>>    		novmerge = 1;
>>>>>>    	else if (!strcmp(str, "vmerge"))
>>>>>>    		novmerge = 0;
>>>>>>    	return 1;
>>>>>>    }
>>>>>> @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev,
>>>>>> struct iommu_table *tbl,  }
>>>>>>
>>>>>>    void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>>>>>>    			 void *vaddr, dma_addr_t dma_handle)  {
>>>>>>    	if (tbl) {
>>>>>>    		unsigned int nio_pages;
>>>>>>
>>>>>>    		size = PAGE_ALIGN(size);
>>>>>>    		nio_pages = size >> IOMMU_PAGE_SHIFT;
>>>>>>    		iommu_free(tbl, dma_handle, nio_pages);
>>>>>>    		size = PAGE_ALIGN(size);
>>>>>>    		free_pages((unsigned long)vaddr, get_order(size));
>>>>>>    	}
>>>>>>    }
>>>>>> +
>>>>>> +#ifdef CONFIG_IOMMU_API
>>>>>> +/*
>>>>>> + * SPAPR TCE API
>>>>>> + */
>>>>>> +static struct page *free_tce(struct iommu_table *tbl, unsigned long
>>>>>> +entry) {
>>>>>> +	struct page *page = NULL;
>>>>>
>>>>> NULL initialization doesn't appear to be necessary
>>>>>
>>>>>> +	unsigned long oldtce;
>>>>>> +
>>>>>> +	oldtce = ppc_md.tce_get(tbl, entry);
>>>>>> +
>>>>>> +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
>>>>>> +		return NULL;
>>>>>> +
>>>>>> +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
>>>>>> +
>>>>>> +	WARN_ON(!page);
>>>>>> +	if (page && (oldtce & TCE_PCI_WRITE))
>>>>>> +		SetPageDirty(page);
>>>>>> +	ppc_md.tce_free(tbl, entry, 1);
>>>>>> +
>>>>>> +	return page;
>>>>>> +}
>>>>>> +
>>>>>> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
>>>>>> +		uint64_t tce, enum dma_data_direction direction) {
>>>>>> +	int ret;
>>>>>> +	struct page *page = NULL;
>>>>>> +	unsigned long kva, offset;
>>>>>> +
>>>>>> +	/* Map new TCE */
>>>>>> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
>>>>>> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
>>>>>> +			direction != DMA_TO_DEVICE, &page);
>>>>>> +	if (ret < 1) {
>>>>>> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed
>>>>> tce=%llx ioba=%lx ret=%d\n",
>>>>>> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
>>>>>> +		if (!ret)
>>>>>> +			ret = -EFAULT;
>>>>>
>>>>> Missing return ret?  Otherwise we've got some bogus uses of page below
>>>>> and we're setting ret for no reason here.
>>>>>
>>>>>> +	}
>>>>>> +
>>>>>> +	kva = (unsigned long) page_address(page);
>>>>>> +	kva += offset;
>>>>>> +
>>>>>> +	/* tce_build receives a virtual address */
>>>>>> +	entry += tbl->it_offset; /* Offset into real TCE table */
>>>>>> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
>>>>>> +
>>>>>> +	/* tce_build() only returns non-zero for transient errors */
>>>>>> +	if (unlikely(ret)) {
>>>>>> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx
>>>>> ioba=%lx kva=%lx ret=%d\n",
>>>>>> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
>>>>>> +		put_page(page);
>>>>>> +		return -EIO;
>>>>>> +	}
>>>>>> +
>>>>>> +	return 0;
>>>>>> +}
>>>>>> +
>>>>>> +static void tce_flush(struct iommu_table *tbl) {
>>>>>> +	/* Flush/invalidate TLB caches if necessary */
>>>>>> +	if (ppc_md.tce_flush)
>>>>>> +		ppc_md.tce_flush(tbl);
>>>>>> +
>>>>>> +	/* Make sure updates are seen by hardware */
>>>>>> +	mb();
>>>>>> +}
>>>>>> +
>>>>>> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
>>>>> uint64_t tce,
>>>>>> +		enum dma_data_direction direction, unsigned long pages) {
>>>>>> +	int i, ret = 0, pages_to_put = 0;
>>>>>> +	struct page *page;
>>>>>> +	struct iommu_pool *pool = get_pool(tbl, entry);
>>>>>> +	struct page **oldpages;
>>>>>> +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
>>>>>> +
>>>>>> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
>>>>>> +
>>>>>> +	/* Handle a single page request without allocation
>>>>>> +	   of pages-to-release array */
>>>>>> +	if (pages == 1) {
>>>>>> +		spin_lock(&(pool->lock));
>>>>>> +		page = free_tce(tbl, entry);
>>>>>> +
>>>>>> +		if (direction != DMA_NONE)
>>>>>> +			ret = put_tce(tbl, entry, tce, direction);
>>>>>> +
>>>>>> +		tce_flush(tbl);
>>>>>> +
>>>>>> +		if (page)
>>>>>> +			put_page(page);
>>>>>> +
>>>>>> +		spin_unlock(&(pool->lock));
>>>>>> +		return ret;
>>>>>> +	}
>>>>>> +
>>>>>> +	/* Releasing multiple pages */
>>>>>> +	/* Allocate an array for pages to be released after TCE table
>>>>>> +	   is updated */
>>>>>> +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
>>>>>> +	if (!oldpages)
>>>>>> +		return -ENOMEM;
>>>>>> +
>>>>>> +	spin_lock(&(pool->lock));
>>>>>> +
>>>>>> +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce +=
>>>>> IOMMU_PAGE_SIZE) {
>>>>>> +		page = free_tce(tbl, entry);
>>>>>> +		if (page) {
>>>>>> +			oldpages[pages_to_put] = page;
>>>>>> +			++pages_to_put;
>>>>>> +		}
>>>>>> +
>>>>>> +		if (direction != DMA_NONE)
>>>>>> +			ret = put_tce(tbl, entry, tce, direction);
>>>>>> +
>>>>>> +		/* Release old pages if we reached the end of oldpages[] or
>>>>>> +		   it is the last page or we are about to exit the loop */
>>>>>> +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret)
>>>>> {
>>>>>> +			tce_flush(tbl);
>>>>>
>>>>> Avoiding tce_flush() is the reason for all this extra overhead, right?
>>>>> I wonder if it'd be cleaner separating map vs unmap, where the map case
>>>>> can avoid the oldpages array... but that means inserting new mappings on
>>>>> top of old ones wouldn't put the pages.
>>>
>>>
>>> Yes, we do not want to loose pages if the guest forgot to unmap them.
>>
>> Hmm, does that mean we're not actively clearing tce entries or somehow
>> disabling the iommu window when the iommu is released through vfio?
>
> Ok, I see tces are put on shutdown via tce_iommu_detach_group, so you're
> more concerned about the guest simply mapping over top of it's own
> mappings.  Is that common?  Is it common enough for every multi-page
> mapping to assume it will happen?  I know this is a performance
> sensitive path for you and it seems like a map-only w/ fallback to
> unmap, remap would be better in the general case.


I do not get it. Where exactly does the performance suffer?
iommu_put_tces() with non zero "tce" (i.e. "map") has to check if the entry 
is not used, at least to return EBUSY when it is, and this check is 
performed. If it is zero, there is no overhead at all. And it is going to 
be the 99.(9)% case as the guest (un)maps one page per call.

Generally speaking we want to move "put tce" completely to the kernel for 
the (much) better performance and vfio won't be dealing with it all.

We already agreed that SPAPR TCE driver uses x86 (aka type1) API but I do 
not see why the powerpc implementation should look x86 alike as it still 
operates with powerpc machine dependent callbacks so the reader has to have 
some powerpc knowledge.


> On x86 we do exactly that, but we do the unmap, remap from userspace
> when we get an EBUSY.  Thanks,



-- 
Alexey

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Alex Williamson]

On Tue, 2012-11-27 at 14:28 +1100, Alexey Kardashevskiy wrote:
> On 27/11/12 05:04, Alex Williamson wrote:
> > On Mon, 2012-11-26 at 08:18 -0700, Alex Williamson wrote:
> >> On Fri, 2012-11-23 at 13:02 +1100, Alexey Kardashevskiy wrote:
> >>> On 22/11/12 22:56, Sethi Varun-B16395 wrote:
> >>>>
> >>>>
> >>>>> -----Original Message-----
> >>>>> From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
> >>>>> owner@vger.kernel.org] On Behalf Of Alex Williamson
> >>>>> Sent: Tuesday, November 20, 2012 11:50 PM
> >>>>> To: Alexey Kardashevskiy
> >>>>> Cc: Benjamin Herrenschmidt; Paul Mackerras; linuxppc-
> >>>>> dev@lists.ozlabs.org; linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
> >>>>> David Gibson
> >>>>> Subject: Re: [PATCH] vfio powerpc: enabled and supported on powernv
> >>>>> platform
> >>>>>
> >>>>> On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
> >>>>>> VFIO implements platform independent stuff such as a PCI driver, BAR
> >>>>>> access (via read/write on a file descriptor or direct mapping when
> >>>>>> possible) and IRQ signaling.
> >>>>>> The platform dependent part includes IOMMU initialization and
> >>>>>> handling.
> >>>>>>
> >>>>>> This patch initializes IOMMU groups based on the IOMMU configuration
> >>>>>> discovered during the PCI scan, only POWERNV platform is supported at
> >>>>>> the moment.
> >>>>>>
> >>>>>> Also the patch implements an VFIO-IOMMU driver which manages DMA
> >>>>>> mapping/unmapping requests coming from the client (now QEMU). It also
> >>>>>> returns a DMA window information to let the guest initialize the
> >>>>>> device tree for a guest OS properly. Although this driver has been
> >>>>>> tested only on POWERNV, it should work on any platform supporting TCE
> >>>>>> tables.
> >>>>>>
> >>>>>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config option.
> >>>>>>
> >>>>>> Cc: David Gibson <david@gibson.dropbear.id.au>
> >>>>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >>>>>> ---
> >>>>>>    arch/powerpc/include/asm/iommu.h     |    6 +
> >>>>>>    arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
> >>>>>>    arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
> >>>>>>    drivers/iommu/Kconfig                |    8 ++
> >>>>>>    drivers/vfio/Kconfig                 |    6 +
> >>>>>>    drivers/vfio/Makefile                |    1 +
> >>>>>>    drivers/vfio/vfio_iommu_spapr_tce.c  |  247
> >>>>> ++++++++++++++++++++++++++++++++++
> >>>>>>    include/linux/vfio.h                 |   20 +++
> >>>>>>    8 files changed, 563 insertions(+)
> >>>>>>    create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> >>>>>>
> >>>>>> diff --git a/arch/powerpc/include/asm/iommu.h
> >>>>>> b/arch/powerpc/include/asm/iommu.h
> >>>>>> index cbfe678..5ba66cb 100644
> >>>>>> --- a/arch/powerpc/include/asm/iommu.h
> >>>>>> +++ b/arch/powerpc/include/asm/iommu.h
> >>>>>> @@ -64,30 +64,33 @@ struct iommu_pool {  }
> >>>>>> ____cacheline_aligned_in_smp;
> >>>>>>
> >>>>>>    struct iommu_table {
> >>>>>>    	unsigned long  it_busno;     /* Bus number this table belongs to */
> >>>>>>    	unsigned long  it_size;      /* Size of iommu table in entries */
> >>>>>>    	unsigned long  it_offset;    /* Offset into global table */
> >>>>>>    	unsigned long  it_base;      /* mapped address of tce table */
> >>>>>>    	unsigned long  it_index;     /* which iommu table this is */
> >>>>>>    	unsigned long  it_type;      /* type: PCI or Virtual Bus */
> >>>>>>    	unsigned long  it_blocksize; /* Entries in each block (cacheline)
> >>>>> */
> >>>>>>    	unsigned long  poolsize;
> >>>>>>    	unsigned long  nr_pools;
> >>>>>>    	struct iommu_pool large_pool;
> >>>>>>    	struct iommu_pool pools[IOMMU_NR_POOLS];
> >>>>>>    	unsigned long *it_map;       /* A simple allocation bitmap for now
> >>>>> */
> >>>>>> +#ifdef CONFIG_IOMMU_API
> >>>>>> +	struct iommu_group *it_group;
> >>>>>> +#endif
> >>>>>>    };
> >>>>>>
> >>>>>>    struct scatterlist;
> >>>>>>
> >>>>>>    static inline void set_iommu_table_base(struct device *dev, void
> >>>>>> *base)  {
> >>>>>>    	dev->archdata.dma_data.iommu_table_base = base;  }
> >>>>>>
> >>>>>>    static inline void *get_iommu_table_base(struct device *dev)  {
> >>>>>>    	return dev->archdata.dma_data.iommu_table_base;
> >>>>>>    }
> >>>>>>
> >>>>>>    /* Frees table for an individual device node */ @@ -135,17 +138,20 @@
> >>>>>> static inline void pci_iommu_init(void) { }  extern void
> >>>>>> alloc_dart_table(void);  #if defined(CONFIG_PPC64) &&
> >>>>>> defined(CONFIG_PM)  static inline void iommu_save(void)  {
> >>>>>>    	if (ppc_md.iommu_save)
> >>>>>>    		ppc_md.iommu_save();
> >>>>>>    }
> >>>>>>
> >>>>>>    static inline void iommu_restore(void)  {
> >>>>>>    	if (ppc_md.iommu_restore)
> >>>>>>    		ppc_md.iommu_restore();
> >>>>>>    }
> >>>>>>    #endif
> >>>>>>
> >>>>>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long
> >>>>> entry, uint64_t tce,
> >>>>>> +		enum dma_data_direction direction, unsigned long pages);
> >>>>>> +
> >>>>>>    #endif /* __KERNEL__ */
> >>>>>>    #endif /* _ASM_IOMMU_H */
> >>>>>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> >>>>>> index ff5a6ce..94f614b 100644
> >>>>>> --- a/arch/powerpc/kernel/iommu.c
> >>>>>> +++ b/arch/powerpc/kernel/iommu.c
> >>>>>> @@ -32,30 +32,31 @@
> >>>>>>    #include <linux/dma-mapping.h>
> >>>>>>    #include <linux/bitmap.h>
> >>>>>>    #include <linux/iommu-helper.h>
> >>>>>>    #include <linux/crash_dump.h>
> >>>>>>    #include <linux/hash.h>
> >>>>>>    #include <linux/fault-inject.h>
> >>>>>>    #include <linux/pci.h>
> >>>>>>    #include <asm/io.h>
> >>>>>>    #include <asm/prom.h>
> >>>>>>    #include <asm/iommu.h>
> >>>>>>    #include <asm/pci-bridge.h>
> >>>>>>    #include <asm/machdep.h>
> >>>>>>    #include <asm/kdump.h>
> >>>>>>    #include <asm/fadump.h>
> >>>>>>    #include <asm/vio.h>
> >>>>>> +#include <asm/tce.h>
> >>>>>>
> >>>>>>    #define DBG(...)
> >>>>>>
> >>>>>>    static int novmerge;
> >>>>>>
> >>>>>>    static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned
> >>>>>> int);
> >>>>>>
> >>>>>>    static int __init setup_iommu(char *str)  {
> >>>>>>    	if (!strcmp(str, "novmerge"))
> >>>>>>    		novmerge = 1;
> >>>>>>    	else if (!strcmp(str, "vmerge"))
> >>>>>>    		novmerge = 0;
> >>>>>>    	return 1;
> >>>>>>    }
> >>>>>> @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev,
> >>>>>> struct iommu_table *tbl,  }
> >>>>>>
> >>>>>>    void iommu_free_coherent(struct iommu_table *tbl, size_t size,
> >>>>>>    			 void *vaddr, dma_addr_t dma_handle)  {
> >>>>>>    	if (tbl) {
> >>>>>>    		unsigned int nio_pages;
> >>>>>>
> >>>>>>    		size = PAGE_ALIGN(size);
> >>>>>>    		nio_pages = size >> IOMMU_PAGE_SHIFT;
> >>>>>>    		iommu_free(tbl, dma_handle, nio_pages);
> >>>>>>    		size = PAGE_ALIGN(size);
> >>>>>>    		free_pages((unsigned long)vaddr, get_order(size));
> >>>>>>    	}
> >>>>>>    }
> >>>>>> +
> >>>>>> +#ifdef CONFIG_IOMMU_API
> >>>>>> +/*
> >>>>>> + * SPAPR TCE API
> >>>>>> + */
> >>>>>> +static struct page *free_tce(struct iommu_table *tbl, unsigned long
> >>>>>> +entry) {
> >>>>>> +	struct page *page = NULL;
> >>>>>
> >>>>> NULL initialization doesn't appear to be necessary
> >>>>>
> >>>>>> +	unsigned long oldtce;
> >>>>>> +
> >>>>>> +	oldtce = ppc_md.tce_get(tbl, entry);
> >>>>>> +
> >>>>>> +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> >>>>>> +		return NULL;
> >>>>>> +
> >>>>>> +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
> >>>>>> +
> >>>>>> +	WARN_ON(!page);
> >>>>>> +	if (page && (oldtce & TCE_PCI_WRITE))
> >>>>>> +		SetPageDirty(page);
> >>>>>> +	ppc_md.tce_free(tbl, entry, 1);
> >>>>>> +
> >>>>>> +	return page;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> >>>>>> +		uint64_t tce, enum dma_data_direction direction) {
> >>>>>> +	int ret;
> >>>>>> +	struct page *page = NULL;
> >>>>>> +	unsigned long kva, offset;
> >>>>>> +
> >>>>>> +	/* Map new TCE */
> >>>>>> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> >>>>>> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> >>>>>> +			direction != DMA_TO_DEVICE, &page);
> >>>>>> +	if (ret < 1) {
> >>>>>> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed
> >>>>> tce=%llx ioba=%lx ret=%d\n",
> >>>>>> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> >>>>>> +		if (!ret)
> >>>>>> +			ret = -EFAULT;
> >>>>>
> >>>>> Missing return ret?  Otherwise we've got some bogus uses of page below
> >>>>> and we're setting ret for no reason here.
> >>>>>
> >>>>>> +	}
> >>>>>> +
> >>>>>> +	kva = (unsigned long) page_address(page);
> >>>>>> +	kva += offset;
> >>>>>> +
> >>>>>> +	/* tce_build receives a virtual address */
> >>>>>> +	entry += tbl->it_offset; /* Offset into real TCE table */
> >>>>>> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> >>>>>> +
> >>>>>> +	/* tce_build() only returns non-zero for transient errors */
> >>>>>> +	if (unlikely(ret)) {
> >>>>>> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx
> >>>>> ioba=%lx kva=%lx ret=%d\n",
> >>>>>> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> >>>>>> +		put_page(page);
> >>>>>> +		return -EIO;
> >>>>>> +	}
> >>>>>> +
> >>>>>> +	return 0;
> >>>>>> +}
> >>>>>> +
> >>>>>> +static void tce_flush(struct iommu_table *tbl) {
> >>>>>> +	/* Flush/invalidate TLB caches if necessary */
> >>>>>> +	if (ppc_md.tce_flush)
> >>>>>> +		ppc_md.tce_flush(tbl);
> >>>>>> +
> >>>>>> +	/* Make sure updates are seen by hardware */
> >>>>>> +	mb();
> >>>>>> +}
> >>>>>> +
> >>>>>> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> >>>>> uint64_t tce,
> >>>>>> +		enum dma_data_direction direction, unsigned long pages) {
> >>>>>> +	int i, ret = 0, pages_to_put = 0;
> >>>>>> +	struct page *page;
> >>>>>> +	struct iommu_pool *pool = get_pool(tbl, entry);
> >>>>>> +	struct page **oldpages;
> >>>>>> +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
> >>>>>> +
> >>>>>> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> >>>>>> +
> >>>>>> +	/* Handle a single page request without allocation
> >>>>>> +	   of pages-to-release array */
> >>>>>> +	if (pages == 1) {
> >>>>>> +		spin_lock(&(pool->lock));
> >>>>>> +		page = free_tce(tbl, entry);
> >>>>>> +
> >>>>>> +		if (direction != DMA_NONE)
> >>>>>> +			ret = put_tce(tbl, entry, tce, direction);
> >>>>>> +
> >>>>>> +		tce_flush(tbl);
> >>>>>> +
> >>>>>> +		if (page)
> >>>>>> +			put_page(page);
> >>>>>> +
> >>>>>> +		spin_unlock(&(pool->lock));
> >>>>>> +		return ret;
> >>>>>> +	}
> >>>>>> +
> >>>>>> +	/* Releasing multiple pages */
> >>>>>> +	/* Allocate an array for pages to be released after TCE table
> >>>>>> +	   is updated */
> >>>>>> +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
> >>>>>> +	if (!oldpages)
> >>>>>> +		return -ENOMEM;
> >>>>>> +
> >>>>>> +	spin_lock(&(pool->lock));
> >>>>>> +
> >>>>>> +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce +=
> >>>>> IOMMU_PAGE_SIZE) {
> >>>>>> +		page = free_tce(tbl, entry);
> >>>>>> +		if (page) {
> >>>>>> +			oldpages[pages_to_put] = page;
> >>>>>> +			++pages_to_put;
> >>>>>> +		}
> >>>>>> +
> >>>>>> +		if (direction != DMA_NONE)
> >>>>>> +			ret = put_tce(tbl, entry, tce, direction);
> >>>>>> +
> >>>>>> +		/* Release old pages if we reached the end of oldpages[] or
> >>>>>> +		   it is the last page or we are about to exit the loop */
> >>>>>> +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret)
> >>>>> {
> >>>>>> +			tce_flush(tbl);
> >>>>>
> >>>>> Avoiding tce_flush() is the reason for all this extra overhead, right?
> >>>>> I wonder if it'd be cleaner separating map vs unmap, where the map case
> >>>>> can avoid the oldpages array... but that means inserting new mappings on
> >>>>> top of old ones wouldn't put the pages.
> >>>
> >>>
> >>> Yes, we do not want to loose pages if the guest forgot to unmap them.
> >>
> >> Hmm, does that mean we're not actively clearing tce entries or somehow
> >> disabling the iommu window when the iommu is released through vfio?
> >
> > Ok, I see tces are put on shutdown via tce_iommu_detach_group, so you're
> > more concerned about the guest simply mapping over top of it's own
> > mappings.  Is that common?  Is it common enough for every multi-page
> > mapping to assume it will happen?  I know this is a performance
> > sensitive path for you and it seems like a map-only w/ fallback to
> > unmap, remap would be better in the general case.
> 
> 
> I do not get it. Where exactly does the performance suffer?
> iommu_put_tces() with non zero "tce" (i.e. "map") has to check if the entry 
> is not used, at least to return EBUSY when it is, and this check is 
> performed. If it is zero, there is no overhead at all. And it is going to 
> be the 99.(9)% case as the guest (un)maps one page per call.

I was mostly concerned about the kmalloc in your mapping path.  If you
had a map-only path it could scan the whole range to verify it's clear
and return EBUSY w/o allocating a buffer.  A second pass could do the
actual mappings.  Maybe it's not much of a win if you expect 99% of the
mappings to be single pages but since you effectively have a pv iommu
interface I wouldn't be surprised if they get batched in the guest.

> Generally speaking we want to move "put tce" completely to the kernel for 
> the (much) better performance and vfio won't be dealing with it all.

Right, but let's not use that as an excuse to be lazy and not ask if we
can do better here.

> We already agreed that SPAPR TCE driver uses x86 (aka type1) API but I do 
> not see why the powerpc implementation should look x86 alike as it still 
> operates with powerpc machine dependent callbacks so the reader has to have 
> some powerpc knowledge.

I'm only using x86 as an example because it's the only one we have.  I
don't think anything we're talking about here is x86-ish or powerpc-ish.
There's a kmalloc in a performance path and I'm asking if we can get rid
of it.  I'm also nervous that we're silently doing fixups on user
parameters to adjust mapping sizes and clear overlaps without any
warning to the user.  Thanks,

Alex

Re: [PATCH] vfio powerpc: enabled and supported on powernv platform

[Alex Williamson]

On Thu, 2012-11-22 at 11:56 +0000, Sethi Varun-B16395 wrote:
> 
> > -----Original Message-----
> > From: linux-kernel-owner@vger.kernel.org [mailto:linux-kernel-
> > owner@vger.kernel.org] On Behalf Of Alex Williamson
> > Sent: Tuesday, November 20, 2012 11:50 PM
> > To: Alexey Kardashevskiy
> > Cc: Benjamin Herrenschmidt; Paul Mackerras; linuxppc-
> > dev@lists.ozlabs.org; linux-kernel@vger.kernel.org; kvm@vger.kernel.org;
> > David Gibson
> > Subject: Re: [PATCH] vfio powerpc: enabled and supported on powernv
> > platform
> > 
> > On Tue, 2012-11-20 at 11:48 +1100, Alexey Kardashevskiy wrote:
> > > VFIO implements platform independent stuff such as a PCI driver, BAR
> > > access (via read/write on a file descriptor or direct mapping when
> > > possible) and IRQ signaling.
> > > The platform dependent part includes IOMMU initialization and
> > > handling.
> > >
> > > This patch initializes IOMMU groups based on the IOMMU configuration
> > > discovered during the PCI scan, only POWERNV platform is supported at
> > > the moment.
> > >
> > > Also the patch implements an VFIO-IOMMU driver which manages DMA
> > > mapping/unmapping requests coming from the client (now QEMU). It also
> > > returns a DMA window information to let the guest initialize the
> > > device tree for a guest OS properly. Although this driver has been
> > > tested only on POWERNV, it should work on any platform supporting TCE
> > > tables.
> > >
> > > To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config option.
> > >
> > > Cc: David Gibson <david@gibson.dropbear.id.au>
> > > Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> > > ---
> > >  arch/powerpc/include/asm/iommu.h     |    6 +
> > >  arch/powerpc/kernel/iommu.c          |  140 +++++++++++++++++++
> > >  arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
> > >  drivers/iommu/Kconfig                |    8 ++
> > >  drivers/vfio/Kconfig                 |    6 +
> > >  drivers/vfio/Makefile                |    1 +
> > >  drivers/vfio/vfio_iommu_spapr_tce.c  |  247
> > ++++++++++++++++++++++++++++++++++
> > >  include/linux/vfio.h                 |   20 +++
> > >  8 files changed, 563 insertions(+)
> > >  create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> > >
> > > diff --git a/arch/powerpc/include/asm/iommu.h
> > > b/arch/powerpc/include/asm/iommu.h
> > > index cbfe678..5ba66cb 100644
> > > --- a/arch/powerpc/include/asm/iommu.h
> > > +++ b/arch/powerpc/include/asm/iommu.h
> > > @@ -64,30 +64,33 @@ struct iommu_pool {  }
> > > ____cacheline_aligned_in_smp;
> > >
> > >  struct iommu_table {
> > >  	unsigned long  it_busno;     /* Bus number this table belongs to */
> > >  	unsigned long  it_size;      /* Size of iommu table in entries */
> > >  	unsigned long  it_offset;    /* Offset into global table */
> > >  	unsigned long  it_base;      /* mapped address of tce table */
> > >  	unsigned long  it_index;     /* which iommu table this is */
> > >  	unsigned long  it_type;      /* type: PCI or Virtual Bus */
> > >  	unsigned long  it_blocksize; /* Entries in each block (cacheline)
> > */
> > >  	unsigned long  poolsize;
> > >  	unsigned long  nr_pools;
> > >  	struct iommu_pool large_pool;
> > >  	struct iommu_pool pools[IOMMU_NR_POOLS];
> > >  	unsigned long *it_map;       /* A simple allocation bitmap for now
> > */
> > > +#ifdef CONFIG_IOMMU_API
> > > +	struct iommu_group *it_group;
> > > +#endif
> > >  };
> > >
> > >  struct scatterlist;
> > >
> > >  static inline void set_iommu_table_base(struct device *dev, void
> > > *base)  {
> > >  	dev->archdata.dma_data.iommu_table_base = base;  }
> > >
> > >  static inline void *get_iommu_table_base(struct device *dev)  {
> > >  	return dev->archdata.dma_data.iommu_table_base;
> > >  }
> > >
> > >  /* Frees table for an individual device node */ @@ -135,17 +138,20 @@
> > > static inline void pci_iommu_init(void) { }  extern void
> > > alloc_dart_table(void);  #if defined(CONFIG_PPC64) &&
> > > defined(CONFIG_PM)  static inline void iommu_save(void)  {
> > >  	if (ppc_md.iommu_save)
> > >  		ppc_md.iommu_save();
> > >  }
> > >
> > >  static inline void iommu_restore(void)  {
> > >  	if (ppc_md.iommu_restore)
> > >  		ppc_md.iommu_restore();
> > >  }
> > >  #endif
> > >
> > > +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long
> > entry, uint64_t tce,
> > > +		enum dma_data_direction direction, unsigned long pages);
> > > +
> > >  #endif /* __KERNEL__ */
> > >  #endif /* _ASM_IOMMU_H */
> > > diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> > > index ff5a6ce..94f614b 100644
> > > --- a/arch/powerpc/kernel/iommu.c
> > > +++ b/arch/powerpc/kernel/iommu.c
> > > @@ -32,30 +32,31 @@
> > >  #include <linux/dma-mapping.h>
> > >  #include <linux/bitmap.h>
> > >  #include <linux/iommu-helper.h>
> > >  #include <linux/crash_dump.h>
> > >  #include <linux/hash.h>
> > >  #include <linux/fault-inject.h>
> > >  #include <linux/pci.h>
> > >  #include <asm/io.h>
> > >  #include <asm/prom.h>
> > >  #include <asm/iommu.h>
> > >  #include <asm/pci-bridge.h>
> > >  #include <asm/machdep.h>
> > >  #include <asm/kdump.h>
> > >  #include <asm/fadump.h>
> > >  #include <asm/vio.h>
> > > +#include <asm/tce.h>
> > >
> > >  #define DBG(...)
> > >
> > >  static int novmerge;
> > >
> > >  static void __iommu_free(struct iommu_table *, dma_addr_t, unsigned
> > > int);
> > >
> > >  static int __init setup_iommu(char *str)  {
> > >  	if (!strcmp(str, "novmerge"))
> > >  		novmerge = 1;
> > >  	else if (!strcmp(str, "vmerge"))
> > >  		novmerge = 0;
> > >  	return 1;
> > >  }
> > > @@ -844,15 +845,154 @@ void *iommu_alloc_coherent(struct device *dev,
> > > struct iommu_table *tbl,  }
> > >
> > >  void iommu_free_coherent(struct iommu_table *tbl, size_t size,
> > >  			 void *vaddr, dma_addr_t dma_handle)  {
> > >  	if (tbl) {
> > >  		unsigned int nio_pages;
> > >
> > >  		size = PAGE_ALIGN(size);
> > >  		nio_pages = size >> IOMMU_PAGE_SHIFT;
> > >  		iommu_free(tbl, dma_handle, nio_pages);
> > >  		size = PAGE_ALIGN(size);
> > >  		free_pages((unsigned long)vaddr, get_order(size));
> > >  	}
> > >  }
> > > +
> > > +#ifdef CONFIG_IOMMU_API
> > > +/*
> > > + * SPAPR TCE API
> > > + */
> > > +static struct page *free_tce(struct iommu_table *tbl, unsigned long
> > > +entry) {
> > > +	struct page *page = NULL;
> > 
> > NULL initialization doesn't appear to be necessary
> > 
> > > +	unsigned long oldtce;
> > > +
> > > +	oldtce = ppc_md.tce_get(tbl, entry);
> > > +
> > > +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> > > +		return NULL;
> > > +
> > > +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
> > > +
> > > +	WARN_ON(!page);
> > > +	if (page && (oldtce & TCE_PCI_WRITE))
> > > +		SetPageDirty(page);
> > > +	ppc_md.tce_free(tbl, entry, 1);
> > > +
> > > +	return page;
> > > +}
> > > +
> > > +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> > > +		uint64_t tce, enum dma_data_direction direction) {
> > > +	int ret;
> > > +	struct page *page = NULL;
> > > +	unsigned long kva, offset;
> > > +
> > > +	/* Map new TCE */
> > > +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> > > +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> > > +			direction != DMA_TO_DEVICE, &page);
> > > +	if (ret < 1) {
> > > +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed
> > tce=%llx ioba=%lx ret=%d\n",
> > > +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> > > +		if (!ret)
> > > +			ret = -EFAULT;
> > 
> > Missing return ret?  Otherwise we've got some bogus uses of page below
> > and we're setting ret for no reason here.
> > 
> > > +	}
> > > +
> > > +	kva = (unsigned long) page_address(page);
> > > +	kva += offset;
> > > +
> > > +	/* tce_build receives a virtual address */
> > > +	entry += tbl->it_offset; /* Offset into real TCE table */
> > > +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> > > +
> > > +	/* tce_build() only returns non-zero for transient errors */
> > > +	if (unlikely(ret)) {
> > > +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx
> > ioba=%lx kva=%lx ret=%d\n",
> > > +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> > > +		put_page(page);
> > > +		return -EIO;
> > > +	}
> > > +
> > > +	return 0;
> > > +}
> > > +
> > > +static void tce_flush(struct iommu_table *tbl) {
> > > +	/* Flush/invalidate TLB caches if necessary */
> > > +	if (ppc_md.tce_flush)
> > > +		ppc_md.tce_flush(tbl);
> > > +
> > > +	/* Make sure updates are seen by hardware */
> > > +	mb();
> > > +}
> > > +
> > > +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> > uint64_t tce,
> > > +		enum dma_data_direction direction, unsigned long pages) {
> > > +	int i, ret = 0, pages_to_put = 0;
> > > +	struct page *page;
> > > +	struct iommu_pool *pool = get_pool(tbl, entry);
> > > +	struct page **oldpages;
> > > +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
> > > +
> > > +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> > > +
> > > +	/* Handle a single page request without allocation
> > > +	   of pages-to-release array */
> > > +	if (pages == 1) {
> > > +		spin_lock(&(pool->lock));
> > > +		page = free_tce(tbl, entry);
> > > +
> > > +		if (direction != DMA_NONE)
> > > +			ret = put_tce(tbl, entry, tce, direction);
> > > +
> > > +		tce_flush(tbl);
> > > +
> > > +		if (page)
> > > +			put_page(page);
> > > +
> > > +		spin_unlock(&(pool->lock));
> > > +		return ret;
> > > +	}
> > > +
> > > +	/* Releasing multiple pages */
> > > +	/* Allocate an array for pages to be released after TCE table
> > > +	   is updated */
> > > +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
> > > +	if (!oldpages)
> > > +		return -ENOMEM;
> > > +
> > > +	spin_lock(&(pool->lock));
> > > +
> > > +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce +=
> > IOMMU_PAGE_SIZE) {
> > > +		page = free_tce(tbl, entry);
> > > +		if (page) {
> > > +			oldpages[pages_to_put] = page;
> > > +			++pages_to_put;
> > > +		}
> > > +
> > > +		if (direction != DMA_NONE)
> > > +			ret = put_tce(tbl, entry, tce, direction);
> > > +
> > > +		/* Release old pages if we reached the end of oldpages[] or
> > > +		   it is the last page or we are about to exit the loop */
> > > +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret)
> > {
> > > +			tce_flush(tbl);
> > 
> > Avoiding tce_flush() is the reason for all this extra overhead, right?
> > I wonder if it'd be cleaner separating map vs unmap, where the map case
> > can avoid the oldpages array... but that means inserting new mappings on
> > top of old ones wouldn't put the pages.
> > 
> > > +
> > > +			/* Release pages after removing them from TCE table */
> > > +			while (pages_to_put) {
> > > +				--pages_to_put;
> > > +				put_page(oldpages[pages_to_put]);
> > > +			}
> > > +		}
> > > +	}
> > > +
> > > +	spin_unlock(&(pool->lock));
> > > +	kfree(oldpages);
> > > +
> > > +	return ret;
> > > +}
> > > +EXPORT_SYMBOL_GPL(iommu_put_tces);
> > > +#endif /* CONFIG_IOMMU_API */
> > > diff --git a/arch/powerpc/platforms/powernv/pci.c
> > > b/arch/powerpc/platforms/powernv/pci.c
> > > index 05205cf..676f4d9 100644
> > > --- a/arch/powerpc/platforms/powernv/pci.c
> > > +++ b/arch/powerpc/platforms/powernv/pci.c
> > > @@ -8,30 +8,31 @@
> > >   * This program is free software; you can redistribute it and/or
> > >   * modify it under the terms of the GNU General Public License
> > >   * as published by the Free Software Foundation; either version
> > >   * 2 of the License, or (at your option) any later version.
> > >   */
> > >
> > >  #include <linux/kernel.h>
> > >  #include <linux/pci.h>
> > >  #include <linux/delay.h>
> > >  #include <linux/string.h>
> > >  #include <linux/init.h>
> > >  #include <linux/bootmem.h>
> > >  #include <linux/irq.h>
> > >  #include <linux/io.h>
> > >  #include <linux/msi.h>
> > > +#include <linux/iommu.h>
> > >
> > >  #include <asm/sections.h>
> > >  #include <asm/io.h>
> > >  #include <asm/prom.h>
> > >  #include <asm/pci-bridge.h>
> > >  #include <asm/machdep.h>
> > >  #include <asm/ppc-pci.h>
> > >  #include <asm/opal.h>
> > >  #include <asm/iommu.h>
> > >  #include <asm/tce.h>
> > >  #include <asm/abs_addr.h>
> > >  #include <asm/firmware.h>
> > >
> > >  #include "powernv.h"
> > >  #include "pci.h"
> > > @@ -601,15 +602,149 @@ void __init pnv_pci_init(void)
> > >  	/* Configure IOMMU DMA hooks */
> > >  	ppc_md.pci_dma_dev_setup = pnv_pci_dma_dev_setup;
> > >  	ppc_md.tce_build = pnv_tce_build;
> > >  	ppc_md.tce_free = pnv_tce_free;
> > >  	ppc_md.tce_get = pnv_tce_get;
> > >  	ppc_md.pci_probe_mode = pnv_pci_probe_mode;
> > >  	set_pci_dma_ops(&dma_iommu_ops);
> > >
> > >  	/* Configure MSIs */
> > >  #ifdef CONFIG_PCI_MSI
> > >  	ppc_md.msi_check_device = pnv_msi_check_device;
> > >  	ppc_md.setup_msi_irqs = pnv_setup_msi_irqs;
> > >  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;  #endif  }
> > > +
> > > +#ifdef CONFIG_IOMMU_API
> > > +/*
> > > + * IOMMU groups support required by VFIO  */ static int
> > > +add_device(struct device *dev) {
> > > +	struct iommu_table *tbl;
> > > +	int ret = 0;
> > > +
> > > +	if (WARN_ON(dev->iommu_group)) {
> > > +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu
> > group %d, skipping\n",
> > > +				dev->kobj.name,
> > 
> > dev_name(dev)
> > 
> > > +				iommu_group_id(dev->iommu_group));
> > > +		return -EBUSY;
> > > +	}
> > > +
> > > +	tbl = get_iommu_table_base(dev);
> > > +	if (!tbl) {
> > > +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> > > +				dev->kobj.name);
> > > +		return 0;
> > > +	}
> > > +
> > > +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> > > +			dev->kobj.name, iommu_group_id(tbl->it_group));
> > > +
> > > +	ret = iommu_group_add_device(tbl->it_group, dev);
> > > +	if (ret < 0)
> > > +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> > > +				dev->kobj.name, ret);
> > > +
> > > +	return ret;
> > > +}
> > > +
> > > +static void del_device(struct device *dev) {
> > > +	iommu_group_remove_device(dev);
> > > +}
> > > +
> > > +static int iommu_bus_notifier(struct notifier_block *nb,
> > > +			      unsigned long action, void *data) {
> > > +	struct device *dev = data;
> > > +
> > > +	switch (action) {
> > > +	case BUS_NOTIFY_ADD_DEVICE:
> > > +		return add_device(dev);
> > > +	case BUS_NOTIFY_DEL_DEVICE:
> > > +		del_device(dev);
> > > +		return 0;
> > > +	default:
> > > +		return 0;
> > > +	}
> > > +}
> > > +
> > > +static struct notifier_block tce_iommu_bus_nb = {
> > > +	.notifier_call = iommu_bus_notifier, };
> > > +
> > > +static void group_release(void *iommu_data) {
> > > +	struct iommu_table *tbl = iommu_data;
> > > +	tbl->it_group = NULL;
> > > +}
> > > +
> > > +static int __init tce_iommu_init(void) {
> > > +	struct pci_dev *pdev = NULL;
> > > +	struct iommu_table *tbl;
> > > +	struct iommu_group *grp;
> > > +
> > > +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> > 
> > There's already a notifier in the iommu code if you were to register an
> > iommu_ops with the add/remove_device entries.  That would allow you to
> > remove the notifier block and notifier function below and the second loop
> > below.  Are you avoiding that to avoid the rest of iommu_ops?
> > 
> [Sethi Varun-B16395] Could be one reason, also they are associating
> the iommu group with the tce table entry and not the device.

That's fine, the tce is the level they claim to have isolation.

> 
> > Also, shouldn't this notifier only be registered after the first loop
> > below?  Otherwise ADD_DEVICE could race with setting up groups, which we
> > assume are present in the add_device() above.
> [Sethi Varun-B16395] Isn't this similar to how how the notifier is
> registered in iommu_bus_init? First a notifier is registered and then
> we check for devices that have already been probed.

It's not quite the same because the existing notifier callbacks to add
devices also creates groups as necessary.  My point here was that we
register a notifier that assumes a group prior to setting up the groups.
In the existing code the order doesn't matter so much because the system
isn't susceptible to hot device adds at that point.  That's likely the
case here too, but registering a notifier before setting up the data the
callback references seems unnecessary.  Thanks,

Alex

[PATCH 0/2] vfio powerpc: implemented and enabled

[Alexey Kardashevskiy]

The series includes IOMMU implementation and necessary IOMMU groups initialization.

Alexey Kardashevskiy (2):
  vfio powerpc: implemented IOMMU driver for VFIO
  vfio powerpc: enabled on powernv platform

 arch/powerpc/include/asm/iommu.h     |    6 +
 arch/powerpc/kernel/iommu.c          |  141 +++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 drivers/vfio/Kconfig                 |    6 +
 drivers/vfio/Makefile                |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c  |  247 ++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                 |   20 +++
 8 files changed, 564 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

-- 
1.7.10.4

[PATCH 1/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

VFIO implements platform independent stuff such as
a PCI driver, BAR access (via read/write on a file descriptor
or direct mapping when possible) and IRQ signaling.

The platform dependent part includes IOMMU initialization
and handling. This patch implements an IOMMU driver for VFIO
which does mapping/unmapping pages for the guest IO and
provides information about DMA window (required by a POWERPC
guest).

The counterpart in QEMU is required to support this functionality.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/vfio/Kconfig                |    6 +
 drivers/vfio/Makefile               |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c |  247 +++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                |   20 +++
 4 files changed, 274 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
index 7cd5dec..b464687 100644
--- a/drivers/vfio/Kconfig
+++ b/drivers/vfio/Kconfig
@@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
 	depends on VFIO
 	default n
 
+config VFIO_IOMMU_SPAPR_TCE
+	tristate
+	depends on VFIO && SPAPR_TCE_IOMMU
+	default n
+
 menuconfig VFIO
 	tristate "VFIO Non-Privileged userspace driver framework"
 	depends on IOMMU_API
 	select VFIO_IOMMU_TYPE1 if X86
+	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
 	help
 	  VFIO provides a framework for secure userspace device drivers.
 	  See Documentation/vfio.txt for more details.
diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
index 2398d4a..72bfabc 100644
--- a/drivers/vfio/Makefile
+++ b/drivers/vfio/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_VFIO) += vfio.o
 obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
+obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
 obj-$(CONFIG_VFIO_PCI) += pci/
diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
new file mode 100644
index 0000000..46a6298
--- /dev/null
+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
@@ -0,0 +1,247 @@
+/*
+ * VFIO: IOMMU DMA mapping support for TCE on POWER
+ *
+ * Copyright (C) 2012 IBM Corp.  All rights reserved.
+ *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Derived from original vfio_iommu_type1.c:
+ * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
+ *     Author: Alex Williamson <alex.williamson@redhat.com>
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/err.h>
+#include <linux/vfio.h>
+#include <asm/iommu.h>
+
+#define DRIVER_VERSION  "0.1"
+#define DRIVER_AUTHOR   "aik@ozlabs.ru"
+#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group);
+
+/*
+ * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
+ */
+
+/*
+ * The container descriptor supports only a single group per container.
+ * Required by the API as the container is not supplied with the IOMMU group
+ * at the moment of initialization.
+ */
+struct tce_container {
+	struct mutex lock;
+	struct iommu_table *tbl;
+};
+
+static void *tce_iommu_open(unsigned long arg)
+{
+	struct tce_container *container;
+
+	if (arg != VFIO_SPAPR_TCE_IOMMU) {
+		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	container = kzalloc(sizeof(*container), GFP_KERNEL);
+	if (!container)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&container->lock);
+
+	return container;
+}
+
+static void tce_iommu_release(void *iommu_data)
+{
+	struct tce_container *container = iommu_data;
+
+	WARN_ON(container->tbl && !container->tbl->it_group);
+	if (container->tbl && container->tbl->it_group)
+		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
+
+	mutex_destroy(&container->lock);
+
+	kfree(container);
+}
+
+static long tce_iommu_ioctl(void *iommu_data,
+				 unsigned int cmd, unsigned long arg)
+{
+	struct tce_container *container = iommu_data;
+	unsigned long minsz;
+
+	switch (cmd) {
+	case VFIO_CHECK_EXTENSION: {
+		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
+	}
+	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
+		struct vfio_iommu_spapr_tce_info info;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
+				dma64_window_size);
+
+		if (copy_from_user(&info, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (info.argsz < minsz)
+			return -EINVAL;
+
+		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
+		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
+		info.dma64_window_start = 0;
+		info.dma64_window_size = 0;
+		info.flags = 0;
+
+		if (copy_to_user((void __user *)arg, &info, minsz))
+			return -EFAULT;
+
+		return 0;
+	}
+	case VFIO_IOMMU_MAP_DMA: {
+		vfio_iommu_spapr_tce_dma_map param;
+		struct iommu_table *tbl = container->tbl;
+		enum dma_data_direction direction = DMA_NONE;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
+				(param.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
+			direction = DMA_BIDIRECTIONAL;
+		} else if (param.flags & VFIO_DMA_MAP_FLAG_READ) {
+			direction = DMA_TO_DEVICE;
+		} else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) {
+			direction = DMA_FROM_DEVICE;
+		}
+
+		param.size += param.iova & ~IOMMU_PAGE_MASK;
+		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
+
+		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				param.vaddr & IOMMU_PAGE_MASK, direction,
+				param.size >> IOMMU_PAGE_SHIFT);
+	}
+	case VFIO_IOMMU_UNMAP_DMA: {
+		vfio_iommu_spapr_tce_dma_unmap param;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		param.size += param.iova & ~IOMMU_PAGE_MASK;
+		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
+
+		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				0, DMA_NONE, param.size >> IOMMU_PAGE_SHIFT);
+	}
+	default:
+		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
+	}
+
+	return -ENOTTY;
+}
+
+static int tce_iommu_attach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
+			iommu_group_id(iommu_group), iommu_group);
+	if (container->tbl) {
+		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
+				iommu_group_id(container->tbl->it_group),
+				iommu_group_id(iommu_group));
+		mutex_unlock(&container->lock);
+		return -EBUSY;
+	}
+
+	container->tbl = tbl;
+	mutex_unlock(&container->lock);
+
+	return 0;
+}
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	if (tbl != container->tbl) {
+		printk(KERN_WARNING "tce_vfio: detaching group #%u, expected group is #%u\n",
+				iommu_group_id(iommu_group),
+				iommu_group_id(tbl->it_group));
+	} else {
+
+		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
+				iommu_group_id(iommu_group), iommu_group);
+
+		iommu_put_tces(tbl, tbl->it_offset, 0, DMA_NONE, tbl->it_size);
+		container->tbl = NULL;
+	}
+	mutex_unlock(&container->lock);
+}
+
+const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
+	.name		= "iommu-vfio-powerpc",
+	.owner		= THIS_MODULE,
+	.open		= tce_iommu_open,
+	.release	= tce_iommu_release,
+	.ioctl		= tce_iommu_ioctl,
+	.attach_group	= tce_iommu_attach_group,
+	.detach_group	= tce_iommu_detach_group,
+};
+
+static int __init tce_iommu_init(void)
+{
+	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+
+MODULE_VERSION(DRIVER_VERSION);
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR(DRIVER_AUTHOR);
+MODULE_DESCRIPTION(DRIVER_DESC);
+
diff --git a/include/linux/vfio.h b/include/linux/vfio.h
index 0a4f180..3ecd65c 100644
--- a/include/linux/vfio.h
+++ b/include/linux/vfio.h
@@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
 /* Extensions */
 
 #define VFIO_TYPE1_IOMMU		1
+#define VFIO_SPAPR_TCE_IOMMU		2
 
 /*
  * The IOCTL interface is designed for extensibility by embedding the
@@ -442,4 +443,23 @@ struct vfio_iommu_type1_dma_unmap {
 
 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 
+/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
+
+struct vfio_iommu_spapr_tce_info {
+	__u32 argsz;
+	__u32 flags;
+	__u32 dma32_window_start;
+	__u32 dma32_window_size;
+	__u64 dma64_window_start;
+	__u64 dma64_window_size;
+};
+
+#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
+
+/* Reuse type1 map/unmap structs as they are the same at the moment */
+typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
+typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
+
+/* ***************************************************************** */
+
 #endif /* VFIO_H */
-- 
1.7.10.4

Re: [PATCH 1/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alex Williamson]

On Fri, 2012-11-23 at 20:03 +1100, Alexey Kardashevskiy wrote:
> VFIO implements platform independent stuff such as
> a PCI driver, BAR access (via read/write on a file descriptor
> or direct mapping when possible) and IRQ signaling.
> 
> The platform dependent part includes IOMMU initialization
> and handling. This patch implements an IOMMU driver for VFIO
> which does mapping/unmapping pages for the guest IO and
> provides information about DMA window (required by a POWERPC
> guest).
> 
> The counterpart in QEMU is required to support this functionality.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  drivers/vfio/Kconfig                |    6 +
>  drivers/vfio/Makefile               |    1 +
>  drivers/vfio/vfio_iommu_spapr_tce.c |  247 +++++++++++++++++++++++++++++++++++
>  include/linux/vfio.h                |   20 +++
>  4 files changed, 274 insertions(+)
>  create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> 
> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> index 7cd5dec..b464687 100644
> --- a/drivers/vfio/Kconfig
> +++ b/drivers/vfio/Kconfig
> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
>  	depends on VFIO
>  	default n
>  
> +config VFIO_IOMMU_SPAPR_TCE
> +	tristate
> +	depends on VFIO && SPAPR_TCE_IOMMU
> +	default n
> +
>  menuconfig VFIO
>  	tristate "VFIO Non-Privileged userspace driver framework"
>  	depends on IOMMU_API
>  	select VFIO_IOMMU_TYPE1 if X86
> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>  	help
>  	  VFIO provides a framework for secure userspace device drivers.
>  	  See Documentation/vfio.txt for more details.
> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> index 2398d4a..72bfabc 100644
> --- a/drivers/vfio/Makefile
> +++ b/drivers/vfio/Makefile
> @@ -1,3 +1,4 @@
>  obj-$(CONFIG_VFIO) += vfio.o
>  obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>  obj-$(CONFIG_VFIO_PCI) += pci/
> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> new file mode 100644
> index 0000000..46a6298
> --- /dev/null
> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> @@ -0,0 +1,247 @@
> +/*
> + * VFIO: IOMMU DMA mapping support for TCE on POWER
> + *
> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + *
> + * Derived from original vfio_iommu_type1.c:
> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> + *     Author: Alex Williamson <alex.williamson@redhat.com>
> + */
> +
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +#include <linux/slab.h>
> +#include <linux/uaccess.h>
> +#include <linux/err.h>
> +#include <linux/vfio.h>
> +#include <asm/iommu.h>
> +
> +#define DRIVER_VERSION  "0.1"
> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group);
> +
> +/*
> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> + */
> +
> +/*
> + * The container descriptor supports only a single group per container.
> + * Required by the API as the container is not supplied with the IOMMU group
> + * at the moment of initialization.
> + */
> +struct tce_container {
> +	struct mutex lock;
> +	struct iommu_table *tbl;
> +};
> +
> +static void *tce_iommu_open(unsigned long arg)
> +{
> +	struct tce_container *container;
> +
> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> +		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
> +	if (!container)
> +		return ERR_PTR(-ENOMEM);
> +
> +	mutex_init(&container->lock);
> +
> +	return container;
> +}
> +
> +static void tce_iommu_release(void *iommu_data)
> +{
> +	struct tce_container *container = iommu_data;
> +
> +	WARN_ON(container->tbl && !container->tbl->it_group);

I think your patch ordering is backwards here.  it_group isn't added
until 2/2.  I'd really like to see the arch/powerpc code approved and
merged by the powerpc maintainer before we add the code that makes use
of it into vfio.  Otherwise we just get lots of churn if interfaces
change or they disapprove of it altogether.

> +	if (container->tbl && container->tbl->it_group)
> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> +
> +	mutex_destroy(&container->lock);
> +
> +	kfree(container);
> +}
> +
> +static long tce_iommu_ioctl(void *iommu_data,
> +				 unsigned int cmd, unsigned long arg)
> +{
> +	struct tce_container *container = iommu_data;
> +	unsigned long minsz;
> +
> +	switch (cmd) {
> +	case VFIO_CHECK_EXTENSION: {
> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> +	}
> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> +		struct vfio_iommu_spapr_tce_info info;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> +				dma64_window_size);
> +
> +		if (copy_from_user(&info, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (info.argsz < minsz)
> +			return -EINVAL;
> +
> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> +		info.dma64_window_start = 0;
> +		info.dma64_window_size = 0;
> +		info.flags = 0;
> +
> +		if (copy_to_user((void __user *)arg, &info, minsz))
> +			return -EFAULT;
> +
> +		return 0;
> +	}
> +	case VFIO_IOMMU_MAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_map param;
> +		struct iommu_table *tbl = container->tbl;
> +		enum dma_data_direction direction = DMA_NONE;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
> +			direction = DMA_BIDIRECTIONAL;
> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_READ) {
> +			direction = DMA_TO_DEVICE;
> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) {
> +			direction = DMA_FROM_DEVICE;
> +		}
> +
> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);

On x86 we force iova, vaddr, and size to all be aligned to the smallest
page granularity of the iommu and return -EINVAL if it doesn't fit.
What does it imply to the user if they're always aligned to work here?
Won't this interface happily map overlapping entries with no indication
to the user that the previous mapping is no longer valid?

Maybe another reason why a combined unmap/map makes me nervous, we have
to assume the user knows what they're doing.

> +
> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				param.vaddr & IOMMU_PAGE_MASK, direction,
> +				param.size >> IOMMU_PAGE_SHIFT);
> +	}
> +	case VFIO_IOMMU_UNMAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_unmap param;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
> +
> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				0, DMA_NONE, param.size >> IOMMU_PAGE_SHIFT);
> +	}
> +	default:
> +		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);

pr_warn

> +	}
> +
> +	return -ENOTTY;
> +}
> +
> +static int tce_iommu_attach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> +			iommu_group_id(iommu_group), iommu_group);
> +	if (container->tbl) {
> +		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",

pr_warn

> +				iommu_group_id(container->tbl->it_group),
> +				iommu_group_id(iommu_group));
> +		mutex_unlock(&container->lock);
> +		return -EBUSY;
> +	}
> +
> +	container->tbl = tbl;

Would it be too much paranoia to clear all the tce here as you do below
on detach?  ie. is there any risk that there's leftover programming?
x86 allocates a new domain on open of the iommu, so we always start out
clean.

> +	mutex_unlock(&container->lock);
> +
> +	return 0;
> +}
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	if (tbl != container->tbl) {
> +		printk(KERN_WARNING "tce_vfio: detaching group #%u, expected group is #%u\n",

pr_warn

> +				iommu_group_id(iommu_group),
> +				iommu_group_id(tbl->it_group));
> +	} else {
> +
> +		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
> +				iommu_group_id(iommu_group), iommu_group);
> +
> +		iommu_put_tces(tbl, tbl->it_offset, 0, DMA_NONE, tbl->it_size);

So this cleans out any mappings when vfio is closed, good.

> +		container->tbl = NULL;
> +	}
> +	mutex_unlock(&container->lock);
> +}
> +
> +const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
> +	.name		= "iommu-vfio-powerpc",
> +	.owner		= THIS_MODULE,
> +	.open		= tce_iommu_open,
> +	.release	= tce_iommu_release,
> +	.ioctl		= tce_iommu_ioctl,
> +	.attach_group	= tce_iommu_attach_group,
> +	.detach_group	= tce_iommu_detach_group,
> +};
> +
> +static int __init tce_iommu_init(void)
> +{
> +	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +
> +MODULE_VERSION(DRIVER_VERSION);
> +MODULE_LICENSE("GPL v2");
> +MODULE_AUTHOR(DRIVER_AUTHOR);
> +MODULE_DESCRIPTION(DRIVER_DESC);
> +
> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
> index 0a4f180..3ecd65c 100644
> --- a/include/linux/vfio.h
> +++ b/include/linux/vfio.h
> @@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
>  /* Extensions */
>  
>  #define VFIO_TYPE1_IOMMU		1
> +#define VFIO_SPAPR_TCE_IOMMU		2
>  
>  /*
>   * The IOCTL interface is designed for extensibility by embedding the
> @@ -442,4 +443,23 @@ struct vfio_iommu_type1_dma_unmap {
>  
>  #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
>  
> +/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
> +
> +struct vfio_iommu_spapr_tce_info {
> +	__u32 argsz;
> +	__u32 flags;
> +	__u32 dma32_window_start;
> +	__u32 dma32_window_size;
> +	__u64 dma64_window_start;
> +	__u64 dma64_window_size;
> +};

Is there anything we can document about this?  It should probably list
that size is in bytes.  Is there any need to communicate the IOMMU page
size here?

> +
> +#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
> +
> +/* Reuse type1 map/unmap structs as they are the same at the moment */
> +typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
> +typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
> +
> +/* ***************************************************************** */
> +
>  #endif /* VFIO_H */

Thanks,

Alex

Re: [PATCH 1/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

On 27/11/12 05:20, Alex Williamson wrote:
> On Fri, 2012-11-23 at 20:03 +1100, Alexey Kardashevskiy wrote:
>> VFIO implements platform independent stuff such as
>> a PCI driver, BAR access (via read/write on a file descriptor
>> or direct mapping when possible) and IRQ signaling.
>>
>> The platform dependent part includes IOMMU initialization
>> and handling. This patch implements an IOMMU driver for VFIO
>> which does mapping/unmapping pages for the guest IO and
>> provides information about DMA window (required by a POWERPC
>> guest).
>>
>> The counterpart in QEMU is required to support this functionality.
>>
>> Cc: David Gibson <david@gibson.dropbear.id.au>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>> ---
>>   drivers/vfio/Kconfig                |    6 +
>>   drivers/vfio/Makefile               |    1 +
>>   drivers/vfio/vfio_iommu_spapr_tce.c |  247 +++++++++++++++++++++++++++++++++++
>>   include/linux/vfio.h                |   20 +++
>>   4 files changed, 274 insertions(+)
>>   create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
>>
>> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
>> index 7cd5dec..b464687 100644
>> --- a/drivers/vfio/Kconfig
>> +++ b/drivers/vfio/Kconfig
>> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
>>   	depends on VFIO
>>   	default n
>>
>> +config VFIO_IOMMU_SPAPR_TCE
>> +	tristate
>> +	depends on VFIO && SPAPR_TCE_IOMMU
>> +	default n
>> +
>>   menuconfig VFIO
>>   	tristate "VFIO Non-Privileged userspace driver framework"
>>   	depends on IOMMU_API
>>   	select VFIO_IOMMU_TYPE1 if X86
>> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>>   	help
>>   	  VFIO provides a framework for secure userspace device drivers.
>>   	  See Documentation/vfio.txt for more details.
>> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
>> index 2398d4a..72bfabc 100644
>> --- a/drivers/vfio/Makefile
>> +++ b/drivers/vfio/Makefile
>> @@ -1,3 +1,4 @@
>>   obj-$(CONFIG_VFIO) += vfio.o
>>   obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
>> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>>   obj-$(CONFIG_VFIO_PCI) += pci/
>> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
>> new file mode 100644
>> index 0000000..46a6298
>> --- /dev/null
>> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
>> @@ -0,0 +1,247 @@
>> +/*
>> + * VFIO: IOMMU DMA mapping support for TCE on POWER
>> + *
>> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
>> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
>> + *
>> + * This program is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License version 2 as
>> + * published by the Free Software Foundation.
>> + *
>> + * Derived from original vfio_iommu_type1.c:
>> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
>> + *     Author: Alex Williamson <alex.williamson@redhat.com>
>> + */
>> +
>> +#include <linux/module.h>
>> +#include <linux/pci.h>
>> +#include <linux/slab.h>
>> +#include <linux/uaccess.h>
>> +#include <linux/err.h>
>> +#include <linux/vfio.h>
>> +#include <asm/iommu.h>
>> +
>> +#define DRIVER_VERSION  "0.1"
>> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
>> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
>> +
>> +static void tce_iommu_detach_group(void *iommu_data,
>> +		struct iommu_group *iommu_group);
>> +
>> +/*
>> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
>> + */
>> +
>> +/*
>> + * The container descriptor supports only a single group per container.
>> + * Required by the API as the container is not supplied with the IOMMU group
>> + * at the moment of initialization.
>> + */
>> +struct tce_container {
>> +	struct mutex lock;
>> +	struct iommu_table *tbl;
>> +};
>> +
>> +static void *tce_iommu_open(unsigned long arg)
>> +{
>> +	struct tce_container *container;
>> +
>> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
>> +		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
>> +		return ERR_PTR(-EINVAL);
>> +	}
>> +
>> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
>> +	if (!container)
>> +		return ERR_PTR(-ENOMEM);
>> +
>> +	mutex_init(&container->lock);
>> +
>> +	return container;
>> +}
>> +
>> +static void tce_iommu_release(void *iommu_data)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +
>> +	WARN_ON(container->tbl && !container->tbl->it_group);
>
> I think your patch ordering is backwards here.  it_group isn't added
> until 2/2.  I'd really like to see the arch/powerpc code approved and
> merged by the powerpc maintainer before we add the code that makes use
> of it into vfio.  Otherwise we just get lots of churn if interfaces
> change or they disapprove of it altogether.


Makes sense, thanks.


>> +	if (container->tbl && container->tbl->it_group)
>> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
>> +
>> +	mutex_destroy(&container->lock);
>> +
>> +	kfree(container);
>> +}
>> +
>> +static long tce_iommu_ioctl(void *iommu_data,
>> +				 unsigned int cmd, unsigned long arg)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +	unsigned long minsz;
>> +
>> +	switch (cmd) {
>> +	case VFIO_CHECK_EXTENSION: {
>> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
>> +	}
>> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
>> +		struct vfio_iommu_spapr_tce_info info;
>> +		struct iommu_table *tbl = container->tbl;
>> +
>> +		if (WARN_ON(!tbl))
>> +			return -ENXIO;
>> +
>> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
>> +				dma64_window_size);
>> +
>> +		if (copy_from_user(&info, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (info.argsz < minsz)
>> +			return -EINVAL;
>> +
>> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
>> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
>> +		info.dma64_window_start = 0;
>> +		info.dma64_window_size = 0;
>> +		info.flags = 0;
>> +
>> +		if (copy_to_user((void __user *)arg, &info, minsz))
>> +			return -EFAULT;
>> +
>> +		return 0;
>> +	}
>> +	case VFIO_IOMMU_MAP_DMA: {
>> +		vfio_iommu_spapr_tce_dma_map param;
>> +		struct iommu_table *tbl = container->tbl;
>> +		enum dma_data_direction direction = DMA_NONE;
>> +
>> +		if (WARN_ON(!tbl))
>> +			return -ENXIO;
>> +
>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
>> +
>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (param.argsz < minsz)
>> +			return -EINVAL;
>> +
>> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
>> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
>> +			direction = DMA_BIDIRECTIONAL;
>> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_READ) {
>> +			direction = DMA_TO_DEVICE;
>> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) {
>> +			direction = DMA_FROM_DEVICE;
>> +		}
>> +
>> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
>> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
>
> On x86 we force iova, vaddr, and size to all be aligned to the smallest
> page granularity of the iommu and return -EINVAL if it doesn't fit.
> What does it imply to the user if they're always aligned to work here?
> Won't this interface happily map overlapping entries with no indication
> to the user that the previous mapping is no longer valid?
> Maybe another reason why a combined unmap/map makes me nervous, we have
> to assume the user knows what they're doing.


I got used to guests which do know what they are doing so I am pretty calm :)
but ok, I'll move alignment to the QEMU, it makes sense.


>> +
>> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
>> +				param.vaddr & IOMMU_PAGE_MASK, direction,
>> +				param.size >> IOMMU_PAGE_SHIFT);
>> +	}
>> +	case VFIO_IOMMU_UNMAP_DMA: {
>> +		vfio_iommu_spapr_tce_dma_unmap param;
>> +		struct iommu_table *tbl = container->tbl;
>> +
>> +		if (WARN_ON(!tbl))
>> +			return -ENXIO;
>> +
>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
>> +
>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (param.argsz < minsz)
>> +			return -EINVAL;
>> +
>> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
>> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
>> +
>> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
>> +				0, DMA_NONE, param.size >> IOMMU_PAGE_SHIFT);
>> +	}
>> +	default:
>> +		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
>
> pr_warn
>
>> +	}
>> +
>> +	return -ENOTTY;
>> +}
>> +
>> +static int tce_iommu_attach_group(void *iommu_data,
>> +		struct iommu_group *iommu_group)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
>> +
>> +	BUG_ON(!tbl);
>> +	mutex_lock(&container->lock);
>> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
>> +			iommu_group_id(iommu_group), iommu_group);
>> +	if (container->tbl) {
>> +		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
>
> pr_warn
>
>> +				iommu_group_id(container->tbl->it_group),
>> +				iommu_group_id(iommu_group));
>> +		mutex_unlock(&container->lock);
>> +		return -EBUSY;
>> +	}
>> +
>> +	container->tbl = tbl;
>
> Would it be too much paranoia to clear all the tce here as you do below
> on detach?

Guess so. I do unmap on detach() and the guest calls put_tce(0) (i.e. 
unmaps) the whole DMA window at the boot time.


> ie. is there any risk that there's leftover programming?
> x86 allocates a new domain on open of the iommu, so we always start out
> clean.


>> +	mutex_unlock(&container->lock);
>> +
>> +	return 0;
>> +}
>> +
>> +static void tce_iommu_detach_group(void *iommu_data,
>> +		struct iommu_group *iommu_group)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
>> +
>> +	BUG_ON(!tbl);
>> +	mutex_lock(&container->lock);
>> +	if (tbl != container->tbl) {
>> +		printk(KERN_WARNING "tce_vfio: detaching group #%u, expected group is #%u\n",
>
> pr_warn
>
>> +				iommu_group_id(iommu_group),
>> +				iommu_group_id(tbl->it_group));
>> +	} else {
>> +
>> +		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
>> +				iommu_group_id(iommu_group), iommu_group);
>> +
>> +		iommu_put_tces(tbl, tbl->it_offset, 0, DMA_NONE, tbl->it_size);
>
> So this cleans out any mappings when vfio is closed, good.
>
>> +		container->tbl = NULL;
>> +	}
>> +	mutex_unlock(&container->lock);
>> +}
>> +
>> +const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
>> +	.name		= "iommu-vfio-powerpc",
>> +	.owner		= THIS_MODULE,
>> +	.open		= tce_iommu_open,
>> +	.release	= tce_iommu_release,
>> +	.ioctl		= tce_iommu_ioctl,
>> +	.attach_group	= tce_iommu_attach_group,
>> +	.detach_group	= tce_iommu_detach_group,
>> +};
>> +
>> +static int __init tce_iommu_init(void)
>> +{
>> +	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
>> +}
>> +
>> +static void __exit tce_iommu_cleanup(void)
>> +{
>> +	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
>> +}
>> +
>> +module_init(tce_iommu_init);
>> +module_exit(tce_iommu_cleanup);
>> +
>> +MODULE_VERSION(DRIVER_VERSION);
>> +MODULE_LICENSE("GPL v2");
>> +MODULE_AUTHOR(DRIVER_AUTHOR);
>> +MODULE_DESCRIPTION(DRIVER_DESC);
>> +
>> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
>> index 0a4f180..3ecd65c 100644
>> --- a/include/linux/vfio.h
>> +++ b/include/linux/vfio.h
>> @@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
>>   /* Extensions */
>>
>>   #define VFIO_TYPE1_IOMMU		1
>> +#define VFIO_SPAPR_TCE_IOMMU		2
>>
>>   /*
>>    * The IOCTL interface is designed for extensibility by embedding the
>> @@ -442,4 +443,23 @@ struct vfio_iommu_type1_dma_unmap {
>>
>>   #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
>>
>> +/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
>> +
>> +struct vfio_iommu_spapr_tce_info {
>> +	__u32 argsz;
>> +	__u32 flags;
>> +	__u32 dma32_window_start;
>> +	__u32 dma32_window_size;
>> +	__u64 dma64_window_start;
>> +	__u64 dma64_window_size;
>> +};
>
> Is there anything we can document about this?

I'll put some.

> It should probably list that size is in bytes.  Is there any need to communicate the IOMMU page
> size here?

It is always 4k. I'll put it to comments.

>> +
>> +#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
>> +
>> +/* Reuse type1 map/unmap structs as they are the same at the moment */
>> +typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
>> +typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
>> +
>> +/* ***************************************************************** */
>> +
>>   #endif /* VFIO_H */
>
> Thanks,
>
> Alex
>
>
>


-- 
Alexey

Re: [PATCH 1/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alex Williamson]

On Tue, 2012-11-27 at 15:06 +1100, Alexey Kardashevskiy wrote:
> On 27/11/12 05:20, Alex Williamson wrote:
> > On Fri, 2012-11-23 at 20:03 +1100, Alexey Kardashevskiy wrote:
> >> VFIO implements platform independent stuff such as
> >> a PCI driver, BAR access (via read/write on a file descriptor
> >> or direct mapping when possible) and IRQ signaling.
> >>
> >> The platform dependent part includes IOMMU initialization
> >> and handling. This patch implements an IOMMU driver for VFIO
> >> which does mapping/unmapping pages for the guest IO and
> >> provides information about DMA window (required by a POWERPC
> >> guest).
> >>
> >> The counterpart in QEMU is required to support this functionality.
> >>
> >> Cc: David Gibson <david@gibson.dropbear.id.au>
> >> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >> ---
> >>   drivers/vfio/Kconfig                |    6 +
> >>   drivers/vfio/Makefile               |    1 +
> >>   drivers/vfio/vfio_iommu_spapr_tce.c |  247 +++++++++++++++++++++++++++++++++++
> >>   include/linux/vfio.h                |   20 +++
> >>   4 files changed, 274 insertions(+)
> >>   create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> >>
> >> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> >> index 7cd5dec..b464687 100644
> >> --- a/drivers/vfio/Kconfig
> >> +++ b/drivers/vfio/Kconfig
> >> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
> >>   	depends on VFIO
> >>   	default n
> >>
> >> +config VFIO_IOMMU_SPAPR_TCE
> >> +	tristate
> >> +	depends on VFIO && SPAPR_TCE_IOMMU
> >> +	default n
> >> +
> >>   menuconfig VFIO
> >>   	tristate "VFIO Non-Privileged userspace driver framework"
> >>   	depends on IOMMU_API
> >>   	select VFIO_IOMMU_TYPE1 if X86
> >> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
> >>   	help
> >>   	  VFIO provides a framework for secure userspace device drivers.
> >>   	  See Documentation/vfio.txt for more details.
> >> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> >> index 2398d4a..72bfabc 100644
> >> --- a/drivers/vfio/Makefile
> >> +++ b/drivers/vfio/Makefile
> >> @@ -1,3 +1,4 @@
> >>   obj-$(CONFIG_VFIO) += vfio.o
> >>   obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> >> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
> >>   obj-$(CONFIG_VFIO_PCI) += pci/
> >> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> >> new file mode 100644
> >> index 0000000..46a6298
> >> --- /dev/null
> >> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> >> @@ -0,0 +1,247 @@
> >> +/*
> >> + * VFIO: IOMMU DMA mapping support for TCE on POWER
> >> + *
> >> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
> >> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> >> + *
> >> + * This program is free software; you can redistribute it and/or modify
> >> + * it under the terms of the GNU General Public License version 2 as
> >> + * published by the Free Software Foundation.
> >> + *
> >> + * Derived from original vfio_iommu_type1.c:
> >> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> >> + *     Author: Alex Williamson <alex.williamson@redhat.com>
> >> + */
> >> +
> >> +#include <linux/module.h>
> >> +#include <linux/pci.h>
> >> +#include <linux/slab.h>
> >> +#include <linux/uaccess.h>
> >> +#include <linux/err.h>
> >> +#include <linux/vfio.h>
> >> +#include <asm/iommu.h>
> >> +
> >> +#define DRIVER_VERSION  "0.1"
> >> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> >> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> >> +
> >> +static void tce_iommu_detach_group(void *iommu_data,
> >> +		struct iommu_group *iommu_group);
> >> +
> >> +/*
> >> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> >> + */
> >> +
> >> +/*
> >> + * The container descriptor supports only a single group per container.
> >> + * Required by the API as the container is not supplied with the IOMMU group
> >> + * at the moment of initialization.
> >> + */
> >> +struct tce_container {
> >> +	struct mutex lock;
> >> +	struct iommu_table *tbl;
> >> +};
> >> +
> >> +static void *tce_iommu_open(unsigned long arg)
> >> +{
> >> +	struct tce_container *container;
> >> +
> >> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> >> +		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
> >> +		return ERR_PTR(-EINVAL);
> >> +	}
> >> +
> >> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
> >> +	if (!container)
> >> +		return ERR_PTR(-ENOMEM);
> >> +
> >> +	mutex_init(&container->lock);
> >> +
> >> +	return container;
> >> +}
> >> +
> >> +static void tce_iommu_release(void *iommu_data)
> >> +{
> >> +	struct tce_container *container = iommu_data;
> >> +
> >> +	WARN_ON(container->tbl && !container->tbl->it_group);
> >
> > I think your patch ordering is backwards here.  it_group isn't added
> > until 2/2.  I'd really like to see the arch/powerpc code approved and
> > merged by the powerpc maintainer before we add the code that makes use
> > of it into vfio.  Otherwise we just get lots of churn if interfaces
> > change or they disapprove of it altogether.
> 
> 
> Makes sense, thanks.
> 
> 
> >> +	if (container->tbl && container->tbl->it_group)
> >> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> >> +
> >> +	mutex_destroy(&container->lock);
> >> +
> >> +	kfree(container);
> >> +}
> >> +
> >> +static long tce_iommu_ioctl(void *iommu_data,
> >> +				 unsigned int cmd, unsigned long arg)
> >> +{
> >> +	struct tce_container *container = iommu_data;
> >> +	unsigned long minsz;
> >> +
> >> +	switch (cmd) {
> >> +	case VFIO_CHECK_EXTENSION: {
> >> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> >> +	}
> >> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> >> +		struct vfio_iommu_spapr_tce_info info;
> >> +		struct iommu_table *tbl = container->tbl;
> >> +
> >> +		if (WARN_ON(!tbl))
> >> +			return -ENXIO;
> >> +
> >> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> >> +				dma64_window_size);
> >> +
> >> +		if (copy_from_user(&info, (void __user *)arg, minsz))
> >> +			return -EFAULT;
> >> +
> >> +		if (info.argsz < minsz)
> >> +			return -EINVAL;
> >> +
> >> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> >> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> >> +		info.dma64_window_start = 0;
> >> +		info.dma64_window_size = 0;
> >> +		info.flags = 0;
> >> +
> >> +		if (copy_to_user((void __user *)arg, &info, minsz))
> >> +			return -EFAULT;
> >> +
> >> +		return 0;
> >> +	}
> >> +	case VFIO_IOMMU_MAP_DMA: {
> >> +		vfio_iommu_spapr_tce_dma_map param;
> >> +		struct iommu_table *tbl = container->tbl;
> >> +		enum dma_data_direction direction = DMA_NONE;
> >> +
> >> +		if (WARN_ON(!tbl))
> >> +			return -ENXIO;
> >> +
> >> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> >> +
> >> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> >> +			return -EFAULT;
> >> +
> >> +		if (param.argsz < minsz)
> >> +			return -EINVAL;
> >> +
> >> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> >> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
> >> +			direction = DMA_BIDIRECTIONAL;
> >> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_READ) {
> >> +			direction = DMA_TO_DEVICE;
> >> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) {
> >> +			direction = DMA_FROM_DEVICE;
> >> +		}
> >> +
> >> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
> >> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
> >
> > On x86 we force iova, vaddr, and size to all be aligned to the smallest
> > page granularity of the iommu and return -EINVAL if it doesn't fit.
> > What does it imply to the user if they're always aligned to work here?
> > Won't this interface happily map overlapping entries with no indication
> > to the user that the previous mapping is no longer valid?
> > Maybe another reason why a combined unmap/map makes me nervous, we have
> > to assume the user knows what they're doing.
> 
> 
> I got used to guests which do know what they are doing so I am pretty calm :)
> but ok, I'll move alignment to the QEMU, it makes sense.
> 
> 
> >> +
> >> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> >> +				param.vaddr & IOMMU_PAGE_MASK, direction,
> >> +				param.size >> IOMMU_PAGE_SHIFT);
> >> +	}
> >> +	case VFIO_IOMMU_UNMAP_DMA: {
> >> +		vfio_iommu_spapr_tce_dma_unmap param;
> >> +		struct iommu_table *tbl = container->tbl;
> >> +
> >> +		if (WARN_ON(!tbl))
> >> +			return -ENXIO;
> >> +
> >> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> >> +
> >> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> >> +			return -EFAULT;
> >> +
> >> +		if (param.argsz < minsz)
> >> +			return -EINVAL;
> >> +
> >> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
> >> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
> >> +
> >> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> >> +				0, DMA_NONE, param.size >> IOMMU_PAGE_SHIFT);
> >> +	}
> >> +	default:
> >> +		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
> >
> > pr_warn
> >
> >> +	}
> >> +
> >> +	return -ENOTTY;
> >> +}
> >> +
> >> +static int tce_iommu_attach_group(void *iommu_data,
> >> +		struct iommu_group *iommu_group)
> >> +{
> >> +	struct tce_container *container = iommu_data;
> >> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> >> +
> >> +	BUG_ON(!tbl);
> >> +	mutex_lock(&container->lock);
> >> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> >> +			iommu_group_id(iommu_group), iommu_group);
> >> +	if (container->tbl) {
> >> +		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
> >
> > pr_warn
> >
> >> +				iommu_group_id(container->tbl->it_group),
> >> +				iommu_group_id(iommu_group));
> >> +		mutex_unlock(&container->lock);
> >> +		return -EBUSY;
> >> +	}
> >> +
> >> +	container->tbl = tbl;
> >
> > Would it be too much paranoia to clear all the tce here as you do below
> > on detach?
> 
> Guess so. I do unmap on detach() and the guest calls put_tce(0) (i.e. 
> unmaps) the whole DMA window at the boot time.

But that's just one user of this interface, we can't assume they'll all
be so agreeable.  If any tces were enabled here, a malicious user would
have a window to host memory, right?  Thanks,

Alex

Re: [PATCH 1/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

On 27/11/12 15:29, Alex Williamson wrote:
> On Tue, 2012-11-27 at 15:06 +1100, Alexey Kardashevskiy wrote:
>> On 27/11/12 05:20, Alex Williamson wrote:
>>> On Fri, 2012-11-23 at 20:03 +1100, Alexey Kardashevskiy wrote:
>>>> VFIO implements platform independent stuff such as
>>>> a PCI driver, BAR access (via read/write on a file descriptor
>>>> or direct mapping when possible) and IRQ signaling.
>>>>
>>>> The platform dependent part includes IOMMU initialization
>>>> and handling. This patch implements an IOMMU driver for VFIO
>>>> which does mapping/unmapping pages for the guest IO and
>>>> provides information about DMA window (required by a POWERPC
>>>> guest).
>>>>
>>>> The counterpart in QEMU is required to support this functionality.
>>>>
>>>> Cc: David Gibson <david@gibson.dropbear.id.au>
>>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>>> ---
>>>>    drivers/vfio/Kconfig                |    6 +
>>>>    drivers/vfio/Makefile               |    1 +
>>>>    drivers/vfio/vfio_iommu_spapr_tce.c |  247 +++++++++++++++++++++++++++++++++++
>>>>    include/linux/vfio.h                |   20 +++
>>>>    4 files changed, 274 insertions(+)
>>>>    create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
>>>>
>>>> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
>>>> index 7cd5dec..b464687 100644
>>>> --- a/drivers/vfio/Kconfig
>>>> +++ b/drivers/vfio/Kconfig
>>>> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
>>>>    	depends on VFIO
>>>>    	default n
>>>>
>>>> +config VFIO_IOMMU_SPAPR_TCE
>>>> +	tristate
>>>> +	depends on VFIO && SPAPR_TCE_IOMMU
>>>> +	default n
>>>> +
>>>>    menuconfig VFIO
>>>>    	tristate "VFIO Non-Privileged userspace driver framework"
>>>>    	depends on IOMMU_API
>>>>    	select VFIO_IOMMU_TYPE1 if X86
>>>> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>>>>    	help
>>>>    	  VFIO provides a framework for secure userspace device drivers.
>>>>    	  See Documentation/vfio.txt for more details.
>>>> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
>>>> index 2398d4a..72bfabc 100644
>>>> --- a/drivers/vfio/Makefile
>>>> +++ b/drivers/vfio/Makefile
>>>> @@ -1,3 +1,4 @@
>>>>    obj-$(CONFIG_VFIO) += vfio.o
>>>>    obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
>>>> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>>>>    obj-$(CONFIG_VFIO_PCI) += pci/
>>>> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
>>>> new file mode 100644
>>>> index 0000000..46a6298
>>>> --- /dev/null
>>>> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
>>>> @@ -0,0 +1,247 @@
>>>> +/*
>>>> + * VFIO: IOMMU DMA mapping support for TCE on POWER
>>>> + *
>>>> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
>>>> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
>>>> + *
>>>> + * This program is free software; you can redistribute it and/or modify
>>>> + * it under the terms of the GNU General Public License version 2 as
>>>> + * published by the Free Software Foundation.
>>>> + *
>>>> + * Derived from original vfio_iommu_type1.c:
>>>> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
>>>> + *     Author: Alex Williamson <alex.williamson@redhat.com>
>>>> + */
>>>> +
>>>> +#include <linux/module.h>
>>>> +#include <linux/pci.h>
>>>> +#include <linux/slab.h>
>>>> +#include <linux/uaccess.h>
>>>> +#include <linux/err.h>
>>>> +#include <linux/vfio.h>
>>>> +#include <asm/iommu.h>
>>>> +
>>>> +#define DRIVER_VERSION  "0.1"
>>>> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
>>>> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
>>>> +
>>>> +static void tce_iommu_detach_group(void *iommu_data,
>>>> +		struct iommu_group *iommu_group);
>>>> +
>>>> +/*
>>>> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
>>>> + */
>>>> +
>>>> +/*
>>>> + * The container descriptor supports only a single group per container.
>>>> + * Required by the API as the container is not supplied with the IOMMU group
>>>> + * at the moment of initialization.
>>>> + */
>>>> +struct tce_container {
>>>> +	struct mutex lock;
>>>> +	struct iommu_table *tbl;
>>>> +};
>>>> +
>>>> +static void *tce_iommu_open(unsigned long arg)
>>>> +{
>>>> +	struct tce_container *container;
>>>> +
>>>> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
>>>> +		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
>>>> +		return ERR_PTR(-EINVAL);
>>>> +	}
>>>> +
>>>> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
>>>> +	if (!container)
>>>> +		return ERR_PTR(-ENOMEM);
>>>> +
>>>> +	mutex_init(&container->lock);
>>>> +
>>>> +	return container;
>>>> +}
>>>> +
>>>> +static void tce_iommu_release(void *iommu_data)
>>>> +{
>>>> +	struct tce_container *container = iommu_data;
>>>> +
>>>> +	WARN_ON(container->tbl && !container->tbl->it_group);
>>>
>>> I think your patch ordering is backwards here.  it_group isn't added
>>> until 2/2.  I'd really like to see the arch/powerpc code approved and
>>> merged by the powerpc maintainer before we add the code that makes use
>>> of it into vfio.  Otherwise we just get lots of churn if interfaces
>>> change or they disapprove of it altogether.
>>
>>
>> Makes sense, thanks.
>>
>>
>>>> +	if (container->tbl && container->tbl->it_group)
>>>> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
>>>> +
>>>> +	mutex_destroy(&container->lock);
>>>> +
>>>> +	kfree(container);
>>>> +}
>>>> +
>>>> +static long tce_iommu_ioctl(void *iommu_data,
>>>> +				 unsigned int cmd, unsigned long arg)
>>>> +{
>>>> +	struct tce_container *container = iommu_data;
>>>> +	unsigned long minsz;
>>>> +
>>>> +	switch (cmd) {
>>>> +	case VFIO_CHECK_EXTENSION: {
>>>> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
>>>> +	}
>>>> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
>>>> +		struct vfio_iommu_spapr_tce_info info;
>>>> +		struct iommu_table *tbl = container->tbl;
>>>> +
>>>> +		if (WARN_ON(!tbl))
>>>> +			return -ENXIO;
>>>> +
>>>> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
>>>> +				dma64_window_size);
>>>> +
>>>> +		if (copy_from_user(&info, (void __user *)arg, minsz))
>>>> +			return -EFAULT;
>>>> +
>>>> +		if (info.argsz < minsz)
>>>> +			return -EINVAL;
>>>> +
>>>> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
>>>> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
>>>> +		info.dma64_window_start = 0;
>>>> +		info.dma64_window_size = 0;
>>>> +		info.flags = 0;
>>>> +
>>>> +		if (copy_to_user((void __user *)arg, &info, minsz))
>>>> +			return -EFAULT;
>>>> +
>>>> +		return 0;
>>>> +	}
>>>> +	case VFIO_IOMMU_MAP_DMA: {
>>>> +		vfio_iommu_spapr_tce_dma_map param;
>>>> +		struct iommu_table *tbl = container->tbl;
>>>> +		enum dma_data_direction direction = DMA_NONE;
>>>> +
>>>> +		if (WARN_ON(!tbl))
>>>> +			return -ENXIO;
>>>> +
>>>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
>>>> +
>>>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>>>> +			return -EFAULT;
>>>> +
>>>> +		if (param.argsz < minsz)
>>>> +			return -EINVAL;
>>>> +
>>>> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
>>>> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
>>>> +			direction = DMA_BIDIRECTIONAL;
>>>> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_READ) {
>>>> +			direction = DMA_TO_DEVICE;
>>>> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) {
>>>> +			direction = DMA_FROM_DEVICE;
>>>> +		}
>>>> +
>>>> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
>>>> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
>>>
>>> On x86 we force iova, vaddr, and size to all be aligned to the smallest
>>> page granularity of the iommu and return -EINVAL if it doesn't fit.
>>> What does it imply to the user if they're always aligned to work here?
>>> Won't this interface happily map overlapping entries with no indication
>>> to the user that the previous mapping is no longer valid?
>>> Maybe another reason why a combined unmap/map makes me nervous, we have
>>> to assume the user knows what they're doing.
>>
>>
>> I got used to guests which do know what they are doing so I am pretty calm :)
>> but ok, I'll move alignment to the QEMU, it makes sense.
>>
>>
>>>> +
>>>> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
>>>> +				param.vaddr & IOMMU_PAGE_MASK, direction,
>>>> +				param.size >> IOMMU_PAGE_SHIFT);
>>>> +	}
>>>> +	case VFIO_IOMMU_UNMAP_DMA: {
>>>> +		vfio_iommu_spapr_tce_dma_unmap param;
>>>> +		struct iommu_table *tbl = container->tbl;
>>>> +
>>>> +		if (WARN_ON(!tbl))
>>>> +			return -ENXIO;
>>>> +
>>>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
>>>> +
>>>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>>>> +			return -EFAULT;
>>>> +
>>>> +		if (param.argsz < minsz)
>>>> +			return -EINVAL;
>>>> +
>>>> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
>>>> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
>>>> +
>>>> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
>>>> +				0, DMA_NONE, param.size >> IOMMU_PAGE_SHIFT);
>>>> +	}
>>>> +	default:
>>>> +		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
>>>
>>> pr_warn
>>>
>>>> +	}
>>>> +
>>>> +	return -ENOTTY;
>>>> +}
>>>> +
>>>> +static int tce_iommu_attach_group(void *iommu_data,
>>>> +		struct iommu_group *iommu_group)
>>>> +{
>>>> +	struct tce_container *container = iommu_data;
>>>> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
>>>> +
>>>> +	BUG_ON(!tbl);
>>>> +	mutex_lock(&container->lock);
>>>> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
>>>> +			iommu_group_id(iommu_group), iommu_group);
>>>> +	if (container->tbl) {
>>>> +		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
>>>
>>> pr_warn
>>>
>>>> +				iommu_group_id(container->tbl->it_group),
>>>> +				iommu_group_id(iommu_group));
>>>> +		mutex_unlock(&container->lock);
>>>> +		return -EBUSY;
>>>> +	}
>>>> +
>>>> +	container->tbl = tbl;
>>>
>>> Would it be too much paranoia to clear all the tce here as you do below
>>> on detach?
>>
>> Guess so. I do unmap on detach() and the guest calls put_tce(0) (i.e.
>> unmaps) the whole DMA window at the boot time.
>
> But that's just one user of this interface, we can't assume they'll all
> be so agreeable.  If any tces were enabled here, a malicious user would
> have a window to host memory, right?  Thanks,


But I still release pages on detach(), how can this code be not called on 
the guest exit (normal or crashed)?



>
> Alex
>


-- 
Alexey

Re: [PATCH 1/2] vfio powerpc: implemented IOMMU driver for VFIO

[David Gibson]

On Tue, Nov 27, 2012 at 03:58:14PM +1100, Alexey Kardashevskiy wrote:
> On 27/11/12 15:29, Alex Williamson wrote:
> >On Tue, 2012-11-27 at 15:06 +1100, Alexey Kardashevskiy wrote:
> >>On 27/11/12 05:20, Alex Williamson wrote:
> >>>On Fri, 2012-11-23 at 20:03 +1100, Alexey Kardashevskiy wrote:
> >>>>VFIO implements platform independent stuff such as
> >>>>a PCI driver, BAR access (via read/write on a file descriptor
> >>>>or direct mapping when possible) and IRQ signaling.
> >>>>
> >>>>The platform dependent part includes IOMMU initialization
> >>>>and handling. This patch implements an IOMMU driver for VFIO
> >>>>which does mapping/unmapping pages for the guest IO and
> >>>>provides information about DMA window (required by a POWERPC
> >>>>guest).
> >>>>
> >>>>The counterpart in QEMU is required to support this functionality.
> >>>>
> >>>>Cc: David Gibson <david@gibson.dropbear.id.au>
> >>>>Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >>>>---
> >>>>   drivers/vfio/Kconfig                |    6 +
> >>>>   drivers/vfio/Makefile               |    1 +
> >>>>   drivers/vfio/vfio_iommu_spapr_tce.c |  247 +++++++++++++++++++++++++++++++++++
> >>>>   include/linux/vfio.h                |   20 +++
> >>>>   4 files changed, 274 insertions(+)
> >>>>   create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> >>>>
> >>>>diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> >>>>index 7cd5dec..b464687 100644
> >>>>--- a/drivers/vfio/Kconfig
> >>>>+++ b/drivers/vfio/Kconfig
> >>>>@@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
> >>>>   	depends on VFIO
> >>>>   	default n
> >>>>
> >>>>+config VFIO_IOMMU_SPAPR_TCE
> >>>>+	tristate
> >>>>+	depends on VFIO && SPAPR_TCE_IOMMU
> >>>>+	default n
> >>>>+
> >>>>   menuconfig VFIO
> >>>>   	tristate "VFIO Non-Privileged userspace driver framework"
> >>>>   	depends on IOMMU_API
> >>>>   	select VFIO_IOMMU_TYPE1 if X86
> >>>>+	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
> >>>>   	help
> >>>>   	  VFIO provides a framework for secure userspace device drivers.
> >>>>   	  See Documentation/vfio.txt for more details.
> >>>>diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> >>>>index 2398d4a..72bfabc 100644
> >>>>--- a/drivers/vfio/Makefile
> >>>>+++ b/drivers/vfio/Makefile
> >>>>@@ -1,3 +1,4 @@
> >>>>   obj-$(CONFIG_VFIO) += vfio.o
> >>>>   obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> >>>>+obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
> >>>>   obj-$(CONFIG_VFIO_PCI) += pci/
> >>>>diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> >>>>new file mode 100644
> >>>>index 0000000..46a6298
> >>>>--- /dev/null
> >>>>+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> >>>>@@ -0,0 +1,247 @@
> >>>>+/*
> >>>>+ * VFIO: IOMMU DMA mapping support for TCE on POWER
> >>>>+ *
> >>>>+ * Copyright (C) 2012 IBM Corp.  All rights reserved.
> >>>>+ *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> >>>>+ *
> >>>>+ * This program is free software; you can redistribute it and/or modify
> >>>>+ * it under the terms of the GNU General Public License version 2 as
> >>>>+ * published by the Free Software Foundation.
> >>>>+ *
> >>>>+ * Derived from original vfio_iommu_type1.c:
> >>>>+ * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> >>>>+ *     Author: Alex Williamson <alex.williamson@redhat.com>
> >>>>+ */
> >>>>+
> >>>>+#include <linux/module.h>
> >>>>+#include <linux/pci.h>
> >>>>+#include <linux/slab.h>
> >>>>+#include <linux/uaccess.h>
> >>>>+#include <linux/err.h>
> >>>>+#include <linux/vfio.h>
> >>>>+#include <asm/iommu.h>
> >>>>+
> >>>>+#define DRIVER_VERSION  "0.1"
> >>>>+#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> >>>>+#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> >>>>+
> >>>>+static void tce_iommu_detach_group(void *iommu_data,
> >>>>+		struct iommu_group *iommu_group);
> >>>>+
> >>>>+/*
> >>>>+ * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> >>>>+ */
> >>>>+
> >>>>+/*
> >>>>+ * The container descriptor supports only a single group per container.
> >>>>+ * Required by the API as the container is not supplied with the IOMMU group
> >>>>+ * at the moment of initialization.
> >>>>+ */
> >>>>+struct tce_container {
> >>>>+	struct mutex lock;
> >>>>+	struct iommu_table *tbl;
> >>>>+};
> >>>>+
> >>>>+static void *tce_iommu_open(unsigned long arg)
> >>>>+{
> >>>>+	struct tce_container *container;
> >>>>+
> >>>>+	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> >>>>+		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
> >>>>+		return ERR_PTR(-EINVAL);
> >>>>+	}
> >>>>+
> >>>>+	container = kzalloc(sizeof(*container), GFP_KERNEL);
> >>>>+	if (!container)
> >>>>+		return ERR_PTR(-ENOMEM);
> >>>>+
> >>>>+	mutex_init(&container->lock);
> >>>>+
> >>>>+	return container;
> >>>>+}
> >>>>+
> >>>>+static void tce_iommu_release(void *iommu_data)
> >>>>+{
> >>>>+	struct tce_container *container = iommu_data;
> >>>>+
> >>>>+	WARN_ON(container->tbl && !container->tbl->it_group);
> >>>
> >>>I think your patch ordering is backwards here.  it_group isn't added
> >>>until 2/2.  I'd really like to see the arch/powerpc code approved and
> >>>merged by the powerpc maintainer before we add the code that makes use
> >>>of it into vfio.  Otherwise we just get lots of churn if interfaces
> >>>change or they disapprove of it altogether.
> >>
> >>
> >>Makes sense, thanks.
> >>
> >>
> >>>>+	if (container->tbl && container->tbl->it_group)
> >>>>+		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> >>>>+
> >>>>+	mutex_destroy(&container->lock);
> >>>>+
> >>>>+	kfree(container);
> >>>>+}
> >>>>+
> >>>>+static long tce_iommu_ioctl(void *iommu_data,
> >>>>+				 unsigned int cmd, unsigned long arg)
> >>>>+{
> >>>>+	struct tce_container *container = iommu_data;
> >>>>+	unsigned long minsz;
> >>>>+
> >>>>+	switch (cmd) {
> >>>>+	case VFIO_CHECK_EXTENSION: {
> >>>>+		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> >>>>+	}
> >>>>+	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> >>>>+		struct vfio_iommu_spapr_tce_info info;
> >>>>+		struct iommu_table *tbl = container->tbl;
> >>>>+
> >>>>+		if (WARN_ON(!tbl))
> >>>>+			return -ENXIO;
> >>>>+
> >>>>+		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> >>>>+				dma64_window_size);
> >>>>+
> >>>>+		if (copy_from_user(&info, (void __user *)arg, minsz))
> >>>>+			return -EFAULT;
> >>>>+
> >>>>+		if (info.argsz < minsz)
> >>>>+			return -EINVAL;
> >>>>+
> >>>>+		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> >>>>+		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> >>>>+		info.dma64_window_start = 0;
> >>>>+		info.dma64_window_size = 0;
> >>>>+		info.flags = 0;
> >>>>+
> >>>>+		if (copy_to_user((void __user *)arg, &info, minsz))
> >>>>+			return -EFAULT;
> >>>>+
> >>>>+		return 0;
> >>>>+	}
> >>>>+	case VFIO_IOMMU_MAP_DMA: {
> >>>>+		vfio_iommu_spapr_tce_dma_map param;
> >>>>+		struct iommu_table *tbl = container->tbl;
> >>>>+		enum dma_data_direction direction = DMA_NONE;
> >>>>+
> >>>>+		if (WARN_ON(!tbl))
> >>>>+			return -ENXIO;
> >>>>+
> >>>>+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> >>>>+
> >>>>+		if (copy_from_user(&param, (void __user *)arg, minsz))
> >>>>+			return -EFAULT;
> >>>>+
> >>>>+		if (param.argsz < minsz)
> >>>>+			return -EINVAL;
> >>>>+
> >>>>+		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> >>>>+				(param.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
> >>>>+			direction = DMA_BIDIRECTIONAL;
> >>>>+		} else if (param.flags & VFIO_DMA_MAP_FLAG_READ) {
> >>>>+			direction = DMA_TO_DEVICE;
> >>>>+		} else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) {
> >>>>+			direction = DMA_FROM_DEVICE;
> >>>>+		}
> >>>>+
> >>>>+		param.size += param.iova & ~IOMMU_PAGE_MASK;
> >>>>+		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
> >>>
> >>>On x86 we force iova, vaddr, and size to all be aligned to the smallest
> >>>page granularity of the iommu and return -EINVAL if it doesn't fit.
> >>>What does it imply to the user if they're always aligned to work here?
> >>>Won't this interface happily map overlapping entries with no indication
> >>>to the user that the previous mapping is no longer valid?
> >>>Maybe another reason why a combined unmap/map makes me nervous, we have
> >>>to assume the user knows what they're doing.
> >>
> >>
> >>I got used to guests which do know what they are doing so I am pretty calm :)
> >>but ok, I'll move alignment to the QEMU, it makes sense.
> >>
> >>
> >>>>+
> >>>>+		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> >>>>+				param.vaddr & IOMMU_PAGE_MASK, direction,
> >>>>+				param.size >> IOMMU_PAGE_SHIFT);
> >>>>+	}
> >>>>+	case VFIO_IOMMU_UNMAP_DMA: {
> >>>>+		vfio_iommu_spapr_tce_dma_unmap param;
> >>>>+		struct iommu_table *tbl = container->tbl;
> >>>>+
> >>>>+		if (WARN_ON(!tbl))
> >>>>+			return -ENXIO;
> >>>>+
> >>>>+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> >>>>+
> >>>>+		if (copy_from_user(&param, (void __user *)arg, minsz))
> >>>>+			return -EFAULT;
> >>>>+
> >>>>+		if (param.argsz < minsz)
> >>>>+			return -EINVAL;
> >>>>+
> >>>>+		param.size += param.iova & ~IOMMU_PAGE_MASK;
> >>>>+		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
> >>>>+
> >>>>+		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> >>>>+				0, DMA_NONE, param.size >> IOMMU_PAGE_SHIFT);
> >>>>+	}
> >>>>+	default:
> >>>>+		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
> >>>
> >>>pr_warn
> >>>
> >>>>+	}
> >>>>+
> >>>>+	return -ENOTTY;
> >>>>+}
> >>>>+
> >>>>+static int tce_iommu_attach_group(void *iommu_data,
> >>>>+		struct iommu_group *iommu_group)
> >>>>+{
> >>>>+	struct tce_container *container = iommu_data;
> >>>>+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> >>>>+
> >>>>+	BUG_ON(!tbl);
> >>>>+	mutex_lock(&container->lock);
> >>>>+	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> >>>>+			iommu_group_id(iommu_group), iommu_group);
> >>>>+	if (container->tbl) {
> >>>>+		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
> >>>
> >>>pr_warn
> >>>
> >>>>+				iommu_group_id(container->tbl->it_group),
> >>>>+				iommu_group_id(iommu_group));
> >>>>+		mutex_unlock(&container->lock);
> >>>>+		return -EBUSY;
> >>>>+	}
> >>>>+
> >>>>+	container->tbl = tbl;
> >>>
> >>>Would it be too much paranoia to clear all the tce here as you do below
> >>>on detach?
> >>
> >>Guess so. I do unmap on detach() and the guest calls put_tce(0) (i.e.
> >>unmaps) the whole DMA window at the boot time.
> >
> >But that's just one user of this interface, we can't assume they'll all
> >be so agreeable.  If any tces were enabled here, a malicious user would
> >have a window to host memory, right?  Thanks,
> 
> 
> But I still release pages on detach(), how can this code be not
> called on the guest exit (normal or crashed)?

I think the concern is about robustness if some bug elsewhere in the
kernel left some TCE entries in place before the table was handed over
to VFIO.

-- 
David Gibson			| I'll have my music baroque, and my code
david AT gibson.dropbear.id.au	| minimalist, thank you.  NOT _the_ _other_
				| _way_ _around_!
http://www.ozlabs.org/~dgibson

Re: [PATCH 1/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alex Williamson]

On Tue, 2012-11-27 at 15:58 +1100, Alexey Kardashevskiy wrote:
> On 27/11/12 15:29, Alex Williamson wrote:
> > On Tue, 2012-11-27 at 15:06 +1100, Alexey Kardashevskiy wrote:
> >> On 27/11/12 05:20, Alex Williamson wrote:
> >>> On Fri, 2012-11-23 at 20:03 +1100, Alexey Kardashevskiy wrote:
> >>>> VFIO implements platform independent stuff such as
> >>>> a PCI driver, BAR access (via read/write on a file descriptor
> >>>> or direct mapping when possible) and IRQ signaling.
> >>>>
> >>>> The platform dependent part includes IOMMU initialization
> >>>> and handling. This patch implements an IOMMU driver for VFIO
> >>>> which does mapping/unmapping pages for the guest IO and
> >>>> provides information about DMA window (required by a POWERPC
> >>>> guest).
> >>>>
> >>>> The counterpart in QEMU is required to support this functionality.
> >>>>
> >>>> Cc: David Gibson <david@gibson.dropbear.id.au>
> >>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >>>> ---
> >>>>    drivers/vfio/Kconfig                |    6 +
> >>>>    drivers/vfio/Makefile               |    1 +
> >>>>    drivers/vfio/vfio_iommu_spapr_tce.c |  247 +++++++++++++++++++++++++++++++++++
> >>>>    include/linux/vfio.h                |   20 +++
> >>>>    4 files changed, 274 insertions(+)
> >>>>    create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> >>>>
> >>>> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> >>>> index 7cd5dec..b464687 100644
> >>>> --- a/drivers/vfio/Kconfig
> >>>> +++ b/drivers/vfio/Kconfig
> >>>> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
> >>>>    	depends on VFIO
> >>>>    	default n
> >>>>
> >>>> +config VFIO_IOMMU_SPAPR_TCE
> >>>> +	tristate
> >>>> +	depends on VFIO && SPAPR_TCE_IOMMU
> >>>> +	default n
> >>>> +
> >>>>    menuconfig VFIO
> >>>>    	tristate "VFIO Non-Privileged userspace driver framework"
> >>>>    	depends on IOMMU_API
> >>>>    	select VFIO_IOMMU_TYPE1 if X86
> >>>> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
> >>>>    	help
> >>>>    	  VFIO provides a framework for secure userspace device drivers.
> >>>>    	  See Documentation/vfio.txt for more details.
> >>>> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> >>>> index 2398d4a..72bfabc 100644
> >>>> --- a/drivers/vfio/Makefile
> >>>> +++ b/drivers/vfio/Makefile
> >>>> @@ -1,3 +1,4 @@
> >>>>    obj-$(CONFIG_VFIO) += vfio.o
> >>>>    obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> >>>> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
> >>>>    obj-$(CONFIG_VFIO_PCI) += pci/
> >>>> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> >>>> new file mode 100644
> >>>> index 0000000..46a6298
> >>>> --- /dev/null
> >>>> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> >>>> @@ -0,0 +1,247 @@
> >>>> +/*
> >>>> + * VFIO: IOMMU DMA mapping support for TCE on POWER
> >>>> + *
> >>>> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
> >>>> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> >>>> + *
> >>>> + * This program is free software; you can redistribute it and/or modify
> >>>> + * it under the terms of the GNU General Public License version 2 as
> >>>> + * published by the Free Software Foundation.
> >>>> + *
> >>>> + * Derived from original vfio_iommu_type1.c:
> >>>> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> >>>> + *     Author: Alex Williamson <alex.williamson@redhat.com>
> >>>> + */
> >>>> +
> >>>> +#include <linux/module.h>
> >>>> +#include <linux/pci.h>
> >>>> +#include <linux/slab.h>
> >>>> +#include <linux/uaccess.h>
> >>>> +#include <linux/err.h>
> >>>> +#include <linux/vfio.h>
> >>>> +#include <asm/iommu.h>
> >>>> +
> >>>> +#define DRIVER_VERSION  "0.1"
> >>>> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> >>>> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> >>>> +
> >>>> +static void tce_iommu_detach_group(void *iommu_data,
> >>>> +		struct iommu_group *iommu_group);
> >>>> +
> >>>> +/*
> >>>> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> >>>> + */
> >>>> +
> >>>> +/*
> >>>> + * The container descriptor supports only a single group per container.
> >>>> + * Required by the API as the container is not supplied with the IOMMU group
> >>>> + * at the moment of initialization.
> >>>> + */
> >>>> +struct tce_container {
> >>>> +	struct mutex lock;
> >>>> +	struct iommu_table *tbl;
> >>>> +};
> >>>> +
> >>>> +static void *tce_iommu_open(unsigned long arg)
> >>>> +{
> >>>> +	struct tce_container *container;
> >>>> +
> >>>> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> >>>> +		printk(KERN_ERR "tce_vfio: Wrong IOMMU type\n");
> >>>> +		return ERR_PTR(-EINVAL);
> >>>> +	}
> >>>> +
> >>>> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
> >>>> +	if (!container)
> >>>> +		return ERR_PTR(-ENOMEM);
> >>>> +
> >>>> +	mutex_init(&container->lock);
> >>>> +
> >>>> +	return container;
> >>>> +}
> >>>> +
> >>>> +static void tce_iommu_release(void *iommu_data)
> >>>> +{
> >>>> +	struct tce_container *container = iommu_data;
> >>>> +
> >>>> +	WARN_ON(container->tbl && !container->tbl->it_group);
> >>>
> >>> I think your patch ordering is backwards here.  it_group isn't added
> >>> until 2/2.  I'd really like to see the arch/powerpc code approved and
> >>> merged by the powerpc maintainer before we add the code that makes use
> >>> of it into vfio.  Otherwise we just get lots of churn if interfaces
> >>> change or they disapprove of it altogether.
> >>
> >>
> >> Makes sense, thanks.
> >>
> >>
> >>>> +	if (container->tbl && container->tbl->it_group)
> >>>> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> >>>> +
> >>>> +	mutex_destroy(&container->lock);
> >>>> +
> >>>> +	kfree(container);
> >>>> +}
> >>>> +
> >>>> +static long tce_iommu_ioctl(void *iommu_data,
> >>>> +				 unsigned int cmd, unsigned long arg)
> >>>> +{
> >>>> +	struct tce_container *container = iommu_data;
> >>>> +	unsigned long minsz;
> >>>> +
> >>>> +	switch (cmd) {
> >>>> +	case VFIO_CHECK_EXTENSION: {
> >>>> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> >>>> +	}
> >>>> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> >>>> +		struct vfio_iommu_spapr_tce_info info;
> >>>> +		struct iommu_table *tbl = container->tbl;
> >>>> +
> >>>> +		if (WARN_ON(!tbl))
> >>>> +			return -ENXIO;
> >>>> +
> >>>> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> >>>> +				dma64_window_size);
> >>>> +
> >>>> +		if (copy_from_user(&info, (void __user *)arg, minsz))
> >>>> +			return -EFAULT;
> >>>> +
> >>>> +		if (info.argsz < minsz)
> >>>> +			return -EINVAL;
> >>>> +
> >>>> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> >>>> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> >>>> +		info.dma64_window_start = 0;
> >>>> +		info.dma64_window_size = 0;
> >>>> +		info.flags = 0;
> >>>> +
> >>>> +		if (copy_to_user((void __user *)arg, &info, minsz))
> >>>> +			return -EFAULT;
> >>>> +
> >>>> +		return 0;
> >>>> +	}
> >>>> +	case VFIO_IOMMU_MAP_DMA: {
> >>>> +		vfio_iommu_spapr_tce_dma_map param;
> >>>> +		struct iommu_table *tbl = container->tbl;
> >>>> +		enum dma_data_direction direction = DMA_NONE;
> >>>> +
> >>>> +		if (WARN_ON(!tbl))
> >>>> +			return -ENXIO;
> >>>> +
> >>>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> >>>> +
> >>>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> >>>> +			return -EFAULT;
> >>>> +
> >>>> +		if (param.argsz < minsz)
> >>>> +			return -EINVAL;
> >>>> +
> >>>> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> >>>> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE)) {
> >>>> +			direction = DMA_BIDIRECTIONAL;
> >>>> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_READ) {
> >>>> +			direction = DMA_TO_DEVICE;
> >>>> +		} else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE) {
> >>>> +			direction = DMA_FROM_DEVICE;
> >>>> +		}
> >>>> +
> >>>> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
> >>>> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
> >>>
> >>> On x86 we force iova, vaddr, and size to all be aligned to the smallest
> >>> page granularity of the iommu and return -EINVAL if it doesn't fit.
> >>> What does it imply to the user if they're always aligned to work here?
> >>> Won't this interface happily map overlapping entries with no indication
> >>> to the user that the previous mapping is no longer valid?
> >>> Maybe another reason why a combined unmap/map makes me nervous, we have
> >>> to assume the user knows what they're doing.
> >>
> >>
> >> I got used to guests which do know what they are doing so I am pretty calm :)
> >> but ok, I'll move alignment to the QEMU, it makes sense.
> >>
> >>
> >>>> +
> >>>> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> >>>> +				param.vaddr & IOMMU_PAGE_MASK, direction,
> >>>> +				param.size >> IOMMU_PAGE_SHIFT);
> >>>> +	}
> >>>> +	case VFIO_IOMMU_UNMAP_DMA: {
> >>>> +		vfio_iommu_spapr_tce_dma_unmap param;
> >>>> +		struct iommu_table *tbl = container->tbl;
> >>>> +
> >>>> +		if (WARN_ON(!tbl))
> >>>> +			return -ENXIO;
> >>>> +
> >>>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> >>>> +
> >>>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> >>>> +			return -EFAULT;
> >>>> +
> >>>> +		if (param.argsz < minsz)
> >>>> +			return -EINVAL;
> >>>> +
> >>>> +		param.size += param.iova & ~IOMMU_PAGE_MASK;
> >>>> +		param.size = _ALIGN_UP(param.size, IOMMU_PAGE_SIZE);
> >>>> +
> >>>> +		return iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> >>>> +				0, DMA_NONE, param.size >> IOMMU_PAGE_SHIFT);
> >>>> +	}
> >>>> +	default:
> >>>> +		printk(KERN_WARNING "tce_vfio: unexpected cmd %x\n", cmd);
> >>>
> >>> pr_warn
> >>>
> >>>> +	}
> >>>> +
> >>>> +	return -ENOTTY;
> >>>> +}
> >>>> +
> >>>> +static int tce_iommu_attach_group(void *iommu_data,
> >>>> +		struct iommu_group *iommu_group)
> >>>> +{
> >>>> +	struct tce_container *container = iommu_data;
> >>>> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> >>>> +
> >>>> +	BUG_ON(!tbl);
> >>>> +	mutex_lock(&container->lock);
> >>>> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> >>>> +			iommu_group_id(iommu_group), iommu_group);
> >>>> +	if (container->tbl) {
> >>>> +		printk(KERN_WARNING "tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
> >>>
> >>> pr_warn
> >>>
> >>>> +				iommu_group_id(container->tbl->it_group),
> >>>> +				iommu_group_id(iommu_group));
> >>>> +		mutex_unlock(&container->lock);
> >>>> +		return -EBUSY;
> >>>> +	}
> >>>> +
> >>>> +	container->tbl = tbl;
> >>>
> >>> Would it be too much paranoia to clear all the tce here as you do below
> >>> on detach?
> >>
> >> Guess so. I do unmap on detach() and the guest calls put_tce(0) (i.e.
> >> unmaps) the whole DMA window at the boot time.
> >
> > But that's just one user of this interface, we can't assume they'll all
> > be so agreeable.  If any tces were enabled here, a malicious user would
> > have a window to host memory, right?  Thanks,
> 
> 
> But I still release pages on detach(), how can this code be not called on 
> the guest exit (normal or crashed)?

What's the initial state?  You leave it clean, but who came before you?
Thanks,

Alex

[PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

VFIO implements platform independent stuff such as
a PCI driver, BAR access (via read/write on a file descriptor
or direct mapping when possible) and IRQ signaling.

The platform dependent part includes IOMMU initialization
and handling. This patch implements an IOMMU driver for VFIO
which does mapping/unmapping pages for the guest IO and
provides information about DMA window (required by a POWERPC
guest).

The counterpart in QEMU is required to support this functionality.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/vfio/Kconfig                |    6 +
 drivers/vfio/Makefile               |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c |  332 +++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                |   33 ++++
 4 files changed, 372 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
index 7cd5dec..b464687 100644
--- a/drivers/vfio/Kconfig
+++ b/drivers/vfio/Kconfig
@@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
 	depends on VFIO
 	default n
 
+config VFIO_IOMMU_SPAPR_TCE
+	tristate
+	depends on VFIO && SPAPR_TCE_IOMMU
+	default n
+
 menuconfig VFIO
 	tristate "VFIO Non-Privileged userspace driver framework"
 	depends on IOMMU_API
 	select VFIO_IOMMU_TYPE1 if X86
+	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
 	help
 	  VFIO provides a framework for secure userspace device drivers.
 	  See Documentation/vfio.txt for more details.
diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
index 2398d4a..72bfabc 100644
--- a/drivers/vfio/Makefile
+++ b/drivers/vfio/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_VFIO) += vfio.o
 obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
+obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
 obj-$(CONFIG_VFIO_PCI) += pci/
diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
new file mode 100644
index 0000000..b98770e
--- /dev/null
+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
@@ -0,0 +1,332 @@
+/*
+ * VFIO: IOMMU DMA mapping support for TCE on POWER
+ *
+ * Copyright (C) 2012 IBM Corp.  All rights reserved.
+ *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Derived from original vfio_iommu_type1.c:
+ * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
+ *     Author: Alex Williamson <alex.williamson@redhat.com>
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/err.h>
+#include <linux/vfio.h>
+#include <asm/iommu.h>
+
+#define DRIVER_VERSION  "0.1"
+#define DRIVER_AUTHOR   "aik@ozlabs.ru"
+#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group);
+
+/*
+ * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
+ */
+
+/*
+ * This code handles mapping and unmapping of user data buffers
+ * into DMA'ble space using the IOMMU
+ */
+
+#define NPAGE_TO_SIZE(npage)	((size_t)(npage) << PAGE_SHIFT)
+
+struct vwork {
+	struct mm_struct	*mm;
+	long			npage;
+	struct work_struct	work;
+};
+
+/* delayed decrement/increment for locked_vm */
+static void lock_acct_bg(struct work_struct *work)
+{
+	struct vwork *vwork = container_of(work, struct vwork, work);
+	struct mm_struct *mm;
+
+	mm = vwork->mm;
+	down_write(&mm->mmap_sem);
+	mm->locked_vm += vwork->npage;
+	up_write(&mm->mmap_sem);
+	mmput(mm);
+	kfree(vwork);
+}
+
+static void lock_acct(long npage)
+{
+	struct vwork *vwork;
+	struct mm_struct *mm;
+
+	if (!current->mm)
+		return; /* process exited */
+
+	if (down_write_trylock(&current->mm->mmap_sem)) {
+		current->mm->locked_vm += npage;
+		up_write(&current->mm->mmap_sem);
+		return;
+	}
+
+	/*
+	 * Couldn't get mmap_sem lock, so must setup to update
+	 * mm->locked_vm later. If locked_vm were atomic, we
+	 * wouldn't need this silliness
+	 */
+	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
+	if (!vwork)
+		return;
+	mm = get_task_mm(current);
+	if (!mm) {
+		kfree(vwork);
+		return;
+	}
+	INIT_WORK(&vwork->work, lock_acct_bg);
+	vwork->mm = mm;
+	vwork->npage = npage;
+	schedule_work(&vwork->work);
+}
+
+/*
+ * The container descriptor supports only a single group per container.
+ * Required by the API as the container is not supplied with the IOMMU group
+ * at the moment of initialization.
+ */
+struct tce_container {
+	struct mutex lock;
+	struct iommu_table *tbl;
+};
+
+static void *tce_iommu_open(unsigned long arg)
+{
+	struct tce_container *container;
+
+	if (arg != VFIO_SPAPR_TCE_IOMMU) {
+		pr_err("tce_vfio: Wrong IOMMU type\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	container = kzalloc(sizeof(*container), GFP_KERNEL);
+	if (!container)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&container->lock);
+
+	return container;
+}
+
+static void tce_iommu_release(void *iommu_data)
+{
+	struct tce_container *container = iommu_data;
+
+	WARN_ON(container->tbl && !container->tbl->it_group);
+	if (container->tbl && container->tbl->it_group)
+		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
+
+	mutex_destroy(&container->lock);
+
+	kfree(container);
+}
+
+static long tce_iommu_ioctl(void *iommu_data,
+				 unsigned int cmd, unsigned long arg)
+{
+	struct tce_container *container = iommu_data;
+	unsigned long minsz;
+	long ret;
+
+	switch (cmd) {
+	case VFIO_CHECK_EXTENSION: {
+		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
+	}
+	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
+		struct vfio_iommu_spapr_tce_info info;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
+				dma64_window_size);
+
+		if (copy_from_user(&info, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (info.argsz < minsz)
+			return -EINVAL;
+
+		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
+		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
+		info.dma64_window_start = 0;
+		info.dma64_window_size = 0;
+		info.flags = 0;
+
+		if (copy_to_user((void __user *)arg, &info, minsz))
+			return -EFAULT;
+
+		return 0;
+	}
+	case VFIO_IOMMU_MAP_DMA: {
+		vfio_iommu_spapr_tce_dma_map param;
+		struct iommu_table *tbl = container->tbl;
+		enum dma_data_direction direction;
+		unsigned long locked, lock_limit;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
+				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
+			direction = DMA_BIDIRECTIONAL;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
+			direction = DMA_TO_DEVICE;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
+			direction = DMA_FROM_DEVICE;
+		else
+			return -EINVAL;
+
+		if ((param.size & ~IOMMU_PAGE_MASK) ||
+				(param.iova & ~IOMMU_PAGE_MASK) ||
+				(param.vaddr & ~IOMMU_PAGE_MASK))
+			return -EINVAL;
+
+		/* Account for locked pages */
+		locked = current->mm->locked_vm +
+				(param.size >> IOMMU_PAGE_SHIFT);
+		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+		if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
+			pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
+					rlimit(RLIMIT_MEMLOCK));
+			return -ENOMEM;
+		}
+
+		ret = iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				param.vaddr, direction,
+				param.size >> IOMMU_PAGE_SHIFT);
+		if (ret > 0)
+			lock_acct(ret);
+
+		return ret;
+	}
+	case VFIO_IOMMU_UNMAP_DMA: {
+		vfio_iommu_spapr_tce_dma_unmap param;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.size & ~IOMMU_PAGE_MASK) ||
+				(param.iova & ~IOMMU_PAGE_MASK))
+			return -EINVAL;
+
+		ret = iommu_clear_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				param.size >> IOMMU_PAGE_SHIFT);
+		if (ret > 0)
+			lock_acct(-ret);
+
+		return ret;
+	}
+	default:
+		pr_warn("tce_vfio: unexpected cmd %x\n", cmd);
+	}
+
+	return -ENOTTY;
+}
+
+static int tce_iommu_attach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
+			iommu_group_id(iommu_group), iommu_group);
+	if (container->tbl) {
+		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
+				iommu_group_id(container->tbl->it_group),
+				iommu_group_id(iommu_group));
+		mutex_unlock(&container->lock);
+		return -EBUSY;
+	}
+
+	container->tbl = tbl;
+	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
+	mutex_unlock(&container->lock);
+
+	return 0;
+}
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	if (tbl != container->tbl) {
+		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
+				iommu_group_id(iommu_group),
+				iommu_group_id(tbl->it_group));
+	} else {
+
+		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
+				iommu_group_id(iommu_group), iommu_group);
+
+		iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
+		container->tbl = NULL;
+	}
+	mutex_unlock(&container->lock);
+}
+
+const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
+	.name		= "iommu-vfio-powerpc",
+	.owner		= THIS_MODULE,
+	.open		= tce_iommu_open,
+	.release	= tce_iommu_release,
+	.ioctl		= tce_iommu_ioctl,
+	.attach_group	= tce_iommu_attach_group,
+	.detach_group	= tce_iommu_detach_group,
+};
+
+static int __init tce_iommu_init(void)
+{
+	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+
+MODULE_VERSION(DRIVER_VERSION);
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR(DRIVER_AUTHOR);
+MODULE_DESCRIPTION(DRIVER_DESC);
+
diff --git a/include/linux/vfio.h b/include/linux/vfio.h
index 0a4f180..820af1e 100644
--- a/include/linux/vfio.h
+++ b/include/linux/vfio.h
@@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
 /* Extensions */
 
 #define VFIO_TYPE1_IOMMU		1
+#define VFIO_SPAPR_TCE_IOMMU		2
 
 /*
  * The IOCTL interface is designed for extensibility by embedding the
@@ -442,4 +443,36 @@ struct vfio_iommu_type1_dma_unmap {
 
 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 
+/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
+
+/*
+ * The SPAPR TCE info struct provides the information about the PCI bus
+ * address ranges available for DMA, these values are programmed into
+ * the hardware so the guest has to know that information.
+ *
+ * Pages within 32 bit window should be explicitely mapped/unmapped via ioctls.
+ * 64 bit window (not supported at the moment for the guest) is supposed to
+ * be mapped completely to the guest memory so the devices capable of 64bit
+ * DMA will not have to use map/unmap ioctls.
+ *
+ * The IOMMU page size is always 4K.
+ */
+
+struct vfio_iommu_spapr_tce_info {
+	__u32 argsz;
+	__u32 flags;			/* reserved for future use */
+	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
+	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
+	__u64 dma64_window_start;	/* 64 bit window start (bytes) */
+	__u64 dma64_window_size;	/* 64 bit window size (bytes) */
+};
+
+#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
+
+/* Reuse type1 map/unmap structs as they are the same at the moment */
+typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
+typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
+
+/* ***************************************************************** */
+
 #endif /* VFIO_H */
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alex Williamson]

On Wed, 2012-11-28 at 18:21 +1100, Alexey Kardashevskiy wrote:
> VFIO implements platform independent stuff such as
> a PCI driver, BAR access (via read/write on a file descriptor
> or direct mapping when possible) and IRQ signaling.
> 
> The platform dependent part includes IOMMU initialization
> and handling. This patch implements an IOMMU driver for VFIO
> which does mapping/unmapping pages for the guest IO and
> provides information about DMA window (required by a POWERPC
> guest).
> 
> The counterpart in QEMU is required to support this functionality.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  drivers/vfio/Kconfig                |    6 +
>  drivers/vfio/Makefile               |    1 +
>  drivers/vfio/vfio_iommu_spapr_tce.c |  332 +++++++++++++++++++++++++++++++++++
>  include/linux/vfio.h                |   33 ++++
>  4 files changed, 372 insertions(+)
>  create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> 
> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> index 7cd5dec..b464687 100644
> --- a/drivers/vfio/Kconfig
> +++ b/drivers/vfio/Kconfig
> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
>  	depends on VFIO
>  	default n
>  
> +config VFIO_IOMMU_SPAPR_TCE
> +	tristate
> +	depends on VFIO && SPAPR_TCE_IOMMU
> +	default n
> +
>  menuconfig VFIO
>  	tristate "VFIO Non-Privileged userspace driver framework"
>  	depends on IOMMU_API
>  	select VFIO_IOMMU_TYPE1 if X86
> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>  	help
>  	  VFIO provides a framework for secure userspace device drivers.
>  	  See Documentation/vfio.txt for more details.
> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> index 2398d4a..72bfabc 100644
> --- a/drivers/vfio/Makefile
> +++ b/drivers/vfio/Makefile
> @@ -1,3 +1,4 @@
>  obj-$(CONFIG_VFIO) += vfio.o
>  obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>  obj-$(CONFIG_VFIO_PCI) += pci/
> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> new file mode 100644
> index 0000000..b98770e
> --- /dev/null
> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> @@ -0,0 +1,332 @@
> +/*
> + * VFIO: IOMMU DMA mapping support for TCE on POWER
> + *
> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + *
> + * Derived from original vfio_iommu_type1.c:
> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> + *     Author: Alex Williamson <alex.williamson@redhat.com>
> + */
> +
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +#include <linux/slab.h>
> +#include <linux/uaccess.h>
> +#include <linux/err.h>
> +#include <linux/vfio.h>
> +#include <asm/iommu.h>
> +
> +#define DRIVER_VERSION  "0.1"
> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group);
> +
> +/*
> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> + */
> +
> +/*
> + * This code handles mapping and unmapping of user data buffers
> + * into DMA'ble space using the IOMMU
> + */
> +
> +#define NPAGE_TO_SIZE(npage)	((size_t)(npage) << PAGE_SHIFT)
> +
> +struct vwork {
> +	struct mm_struct	*mm;
> +	long			npage;
> +	struct work_struct	work;
> +};
> +
> +/* delayed decrement/increment for locked_vm */
> +static void lock_acct_bg(struct work_struct *work)
> +{
> +	struct vwork *vwork = container_of(work, struct vwork, work);
> +	struct mm_struct *mm;
> +
> +	mm = vwork->mm;
> +	down_write(&mm->mmap_sem);
> +	mm->locked_vm += vwork->npage;
> +	up_write(&mm->mmap_sem);
> +	mmput(mm);
> +	kfree(vwork);
> +}
> +
> +static void lock_acct(long npage)
> +{
> +	struct vwork *vwork;
> +	struct mm_struct *mm;
> +
> +	if (!current->mm)
> +		return; /* process exited */
> +
> +	if (down_write_trylock(&current->mm->mmap_sem)) {
> +		current->mm->locked_vm += npage;
> +		up_write(&current->mm->mmap_sem);
> +		return;
> +	}
> +
> +	/*
> +	 * Couldn't get mmap_sem lock, so must setup to update
> +	 * mm->locked_vm later. If locked_vm were atomic, we
> +	 * wouldn't need this silliness
> +	 */
> +	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
> +	if (!vwork)
> +		return;
> +	mm = get_task_mm(current);
> +	if (!mm) {
> +		kfree(vwork);
> +		return;
> +	}
> +	INIT_WORK(&vwork->work, lock_acct_bg);
> +	vwork->mm = mm;
> +	vwork->npage = npage;
> +	schedule_work(&vwork->work);
> +}

This looks familiar, should we split it out to a common file instead of
duplicating it?

> +
> +/*
> + * The container descriptor supports only a single group per container.
> + * Required by the API as the container is not supplied with the IOMMU group
> + * at the moment of initialization.
> + */
> +struct tce_container {
> +	struct mutex lock;
> +	struct iommu_table *tbl;
> +};
> +
> +static void *tce_iommu_open(unsigned long arg)
> +{
> +	struct tce_container *container;
> +
> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> +		pr_err("tce_vfio: Wrong IOMMU type\n");
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
> +	if (!container)
> +		return ERR_PTR(-ENOMEM);
> +
> +	mutex_init(&container->lock);
> +
> +	return container;
> +}
> +
> +static void tce_iommu_release(void *iommu_data)
> +{
> +	struct tce_container *container = iommu_data;
> +
> +	WARN_ON(container->tbl && !container->tbl->it_group);
> +	if (container->tbl && container->tbl->it_group)
> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> +
> +	mutex_destroy(&container->lock);
> +
> +	kfree(container);
> +}
> +
> +static long tce_iommu_ioctl(void *iommu_data,
> +				 unsigned int cmd, unsigned long arg)
> +{
> +	struct tce_container *container = iommu_data;
> +	unsigned long minsz;
> +	long ret;
> +
> +	switch (cmd) {
> +	case VFIO_CHECK_EXTENSION: {
> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> +	}
> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> +		struct vfio_iommu_spapr_tce_info info;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> +				dma64_window_size);
> +
> +		if (copy_from_user(&info, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (info.argsz < minsz)
> +			return -EINVAL;
> +
> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> +		info.dma64_window_start = 0;
> +		info.dma64_window_size = 0;
> +		info.flags = 0;
> +
> +		if (copy_to_user((void __user *)arg, &info, minsz))
> +			return -EFAULT;
> +
> +		return 0;
> +	}
> +	case VFIO_IOMMU_MAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_map param;
> +		struct iommu_table *tbl = container->tbl;
> +		enum dma_data_direction direction;
> +		unsigned long locked, lock_limit;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
> +			direction = DMA_BIDIRECTIONAL;
> +		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
> +			direction = DMA_TO_DEVICE;
> +		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
> +			direction = DMA_FROM_DEVICE;
> +		else
> +			return -EINVAL;
> +
> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
> +				(param.iova & ~IOMMU_PAGE_MASK) ||
> +				(param.vaddr & ~IOMMU_PAGE_MASK))
> +			return -EINVAL;
> +
> +		/* Account for locked pages */
> +		locked = current->mm->locked_vm +
> +				(param.size >> IOMMU_PAGE_SHIFT);
> +		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;

This page accounting doesn't look right.  PAGE_SIZE is several orders
bigger than IOMMU_PAGE_SIZE (right?), but we mix them here, which seems
like it will over penalize the user.  For example, if a user maps 4x4k
(assume aligned and contiguous) IOMMU pages, isn't that only a single
pinned system page (assuming >=16k pages).

> +		if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
> +			pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
> +					rlimit(RLIMIT_MEMLOCK));
> +			return -ENOMEM;
> +		}
> +
> +		ret = iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				param.vaddr, direction,
> +				param.size >> IOMMU_PAGE_SHIFT);
> +		if (ret > 0)
> +			lock_acct(ret);
> +
> +		return ret;
> +	}
> +	case VFIO_IOMMU_UNMAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_unmap param;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
> +				(param.iova & ~IOMMU_PAGE_MASK))
> +			return -EINVAL;
> +
> +		ret = iommu_clear_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				param.size >> IOMMU_PAGE_SHIFT);
> +		if (ret > 0)
> +			lock_acct(-ret);
> +
> +		return ret;
> +	}
> +	default:
> +		pr_warn("tce_vfio: unexpected cmd %x\n", cmd);
> +	}
> +
> +	return -ENOTTY;
> +}
> +
> +static int tce_iommu_attach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> +			iommu_group_id(iommu_group), iommu_group);
> +	if (container->tbl) {
> +		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
> +				iommu_group_id(container->tbl->it_group),
> +				iommu_group_id(iommu_group));
> +		mutex_unlock(&container->lock);
> +		return -EBUSY;
> +	}
> +
> +	container->tbl = tbl;
> +	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
> +	mutex_unlock(&container->lock);
> +
> +	return 0;
> +}
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	if (tbl != container->tbl) {
> +		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
> +				iommu_group_id(iommu_group),
> +				iommu_group_id(tbl->it_group));
> +	} else {
> +
> +		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
> +				iommu_group_id(iommu_group), iommu_group);
> +
> +		iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
> +		container->tbl = NULL;
> +	}
> +	mutex_unlock(&container->lock);
> +}
> +
> +const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
> +	.name		= "iommu-vfio-powerpc",
> +	.owner		= THIS_MODULE,
> +	.open		= tce_iommu_open,
> +	.release	= tce_iommu_release,
> +	.ioctl		= tce_iommu_ioctl,
> +	.attach_group	= tce_iommu_attach_group,
> +	.detach_group	= tce_iommu_detach_group,
> +};
> +
> +static int __init tce_iommu_init(void)
> +{
> +	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +
> +MODULE_VERSION(DRIVER_VERSION);
> +MODULE_LICENSE("GPL v2");
> +MODULE_AUTHOR(DRIVER_AUTHOR);
> +MODULE_DESCRIPTION(DRIVER_DESC);
> +
> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
> index 0a4f180..820af1e 100644
> --- a/include/linux/vfio.h
> +++ b/include/linux/vfio.h
> @@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
>  /* Extensions */
>  
>  #define VFIO_TYPE1_IOMMU		1
> +#define VFIO_SPAPR_TCE_IOMMU		2
>  
>  /*
>   * The IOCTL interface is designed for extensibility by embedding the
> @@ -442,4 +443,36 @@ struct vfio_iommu_type1_dma_unmap {
>  
>  #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
>  
> +/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
> +
> +/*
> + * The SPAPR TCE info struct provides the information about the PCI bus
> + * address ranges available for DMA, these values are programmed into
> + * the hardware so the guest has to know that information.
> + *
> + * Pages within 32 bit window should be explicitely mapped/unmapped via ioctls.
                                          ^^^^^^^^^^^
explicitly

> + * 64 bit window (not supported at the moment for the guest) is supposed to
> + * be mapped completely to the guest memory so the devices capable of 64bit
> + * DMA will not have to use map/unmap ioctls.
> + *
> + * The IOMMU page size is always 4K.
> + */

Thanks,

Alex

> +
> +struct vfio_iommu_spapr_tce_info {
> +	__u32 argsz;
> +	__u32 flags;			/* reserved for future use */
> +	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
> +	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
> +	__u64 dma64_window_start;	/* 64 bit window start (bytes) */
> +	__u64 dma64_window_size;	/* 64 bit window size (bytes) */
> +};
> +
> +#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
> +
> +/* Reuse type1 map/unmap structs as they are the same at the moment */
> +typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
> +typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
> +
> +/* ***************************************************************** */
> +
>  #endif /* VFIO_H */

Re: [PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

On 29/11/12 08:01, Alex Williamson wrote:
> On Wed, 2012-11-28 at 18:21 +1100, Alexey Kardashevskiy wrote:
>> VFIO implements platform independent stuff such as
>> a PCI driver, BAR access (via read/write on a file descriptor
>> or direct mapping when possible) and IRQ signaling.
>>
>> The platform dependent part includes IOMMU initialization
>> and handling. This patch implements an IOMMU driver for VFIO
>> which does mapping/unmapping pages for the guest IO and
>> provides information about DMA window (required by a POWERPC
>> guest).
>>
>> The counterpart in QEMU is required to support this functionality.
>>
>> Cc: David Gibson <david@gibson.dropbear.id.au>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>> ---
>>   drivers/vfio/Kconfig                |    6 +
>>   drivers/vfio/Makefile               |    1 +
>>   drivers/vfio/vfio_iommu_spapr_tce.c |  332 +++++++++++++++++++++++++++++++++++
>>   include/linux/vfio.h                |   33 ++++
>>   4 files changed, 372 insertions(+)
>>   create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
>>
>> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
>> index 7cd5dec..b464687 100644
>> --- a/drivers/vfio/Kconfig
>> +++ b/drivers/vfio/Kconfig
>> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
>>   	depends on VFIO
>>   	default n
>>
>> +config VFIO_IOMMU_SPAPR_TCE
>> +	tristate
>> +	depends on VFIO && SPAPR_TCE_IOMMU
>> +	default n
>> +
>>   menuconfig VFIO
>>   	tristate "VFIO Non-Privileged userspace driver framework"
>>   	depends on IOMMU_API
>>   	select VFIO_IOMMU_TYPE1 if X86
>> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>>   	help
>>   	  VFIO provides a framework for secure userspace device drivers.
>>   	  See Documentation/vfio.txt for more details.
>> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
>> index 2398d4a..72bfabc 100644
>> --- a/drivers/vfio/Makefile
>> +++ b/drivers/vfio/Makefile
>> @@ -1,3 +1,4 @@
>>   obj-$(CONFIG_VFIO) += vfio.o
>>   obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
>> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>>   obj-$(CONFIG_VFIO_PCI) += pci/
>> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
>> new file mode 100644
>> index 0000000..b98770e
>> --- /dev/null
>> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
>> @@ -0,0 +1,332 @@
>> +/*
>> + * VFIO: IOMMU DMA mapping support for TCE on POWER
>> + *
>> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
>> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
>> + *
>> + * This program is free software; you can redistribute it and/or modify
>> + * it under the terms of the GNU General Public License version 2 as
>> + * published by the Free Software Foundation.
>> + *
>> + * Derived from original vfio_iommu_type1.c:
>> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
>> + *     Author: Alex Williamson <alex.williamson@redhat.com>
>> + */
>> +
>> +#include <linux/module.h>
>> +#include <linux/pci.h>
>> +#include <linux/slab.h>
>> +#include <linux/uaccess.h>
>> +#include <linux/err.h>
>> +#include <linux/vfio.h>
>> +#include <asm/iommu.h>
>> +
>> +#define DRIVER_VERSION  "0.1"
>> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
>> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
>> +
>> +static void tce_iommu_detach_group(void *iommu_data,
>> +		struct iommu_group *iommu_group);
>> +
>> +/*
>> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
>> + */
>> +
>> +/*
>> + * This code handles mapping and unmapping of user data buffers
>> + * into DMA'ble space using the IOMMU
>> + */
>> +
>> +#define NPAGE_TO_SIZE(npage)	((size_t)(npage) << PAGE_SHIFT)
>> +
>> +struct vwork {
>> +	struct mm_struct	*mm;
>> +	long			npage;
>> +	struct work_struct	work;
>> +};
>> +
>> +/* delayed decrement/increment for locked_vm */
>> +static void lock_acct_bg(struct work_struct *work)
>> +{
>> +	struct vwork *vwork = container_of(work, struct vwork, work);
>> +	struct mm_struct *mm;
>> +
>> +	mm = vwork->mm;
>> +	down_write(&mm->mmap_sem);
>> +	mm->locked_vm += vwork->npage;
>> +	up_write(&mm->mmap_sem);
>> +	mmput(mm);
>> +	kfree(vwork);
>> +}
>> +
>> +static void lock_acct(long npage)
>> +{
>> +	struct vwork *vwork;
>> +	struct mm_struct *mm;
>> +
>> +	if (!current->mm)
>> +		return; /* process exited */
>> +
>> +	if (down_write_trylock(&current->mm->mmap_sem)) {
>> +		current->mm->locked_vm += npage;
>> +		up_write(&current->mm->mmap_sem);
>> +		return;
>> +	}
>> +
>> +	/*
>> +	 * Couldn't get mmap_sem lock, so must setup to update
>> +	 * mm->locked_vm later. If locked_vm were atomic, we
>> +	 * wouldn't need this silliness
>> +	 */
>> +	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
>> +	if (!vwork)
>> +		return;
>> +	mm = get_task_mm(current);
>> +	if (!mm) {
>> +		kfree(vwork);
>> +		return;
>> +	}
>> +	INIT_WORK(&vwork->work, lock_acct_bg);
>> +	vwork->mm = mm;
>> +	vwork->npage = npage;
>> +	schedule_work(&vwork->work);
>> +}
>
> This looks familiar, should we split it out to a common file instead of
> duplicating it?

It is simple cut-n-paste from type1 driver :)
Moving it to a separate file is up to you but it is quite small piece of 
code to move it somewhere, and I have not fixed rlimit handling yet, so 
wait a bit.


>> +
>> +/*
>> + * The container descriptor supports only a single group per container.
>> + * Required by the API as the container is not supplied with the IOMMU group
>> + * at the moment of initialization.
>> + */
>> +struct tce_container {
>> +	struct mutex lock;
>> +	struct iommu_table *tbl;
>> +};
>> +
>> +static void *tce_iommu_open(unsigned long arg)
>> +{
>> +	struct tce_container *container;
>> +
>> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
>> +		pr_err("tce_vfio: Wrong IOMMU type\n");
>> +		return ERR_PTR(-EINVAL);
>> +	}
>> +
>> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
>> +	if (!container)
>> +		return ERR_PTR(-ENOMEM);
>> +
>> +	mutex_init(&container->lock);
>> +
>> +	return container;
>> +}
>> +
>> +static void tce_iommu_release(void *iommu_data)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +
>> +	WARN_ON(container->tbl && !container->tbl->it_group);
>> +	if (container->tbl && container->tbl->it_group)
>> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
>> +
>> +	mutex_destroy(&container->lock);
>> +
>> +	kfree(container);
>> +}
>> +
>> +static long tce_iommu_ioctl(void *iommu_data,
>> +				 unsigned int cmd, unsigned long arg)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +	unsigned long minsz;
>> +	long ret;
>> +
>> +	switch (cmd) {
>> +	case VFIO_CHECK_EXTENSION: {
>> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
>> +	}
>> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
>> +		struct vfio_iommu_spapr_tce_info info;
>> +		struct iommu_table *tbl = container->tbl;
>> +
>> +		if (WARN_ON(!tbl))
>> +			return -ENXIO;
>> +
>> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
>> +				dma64_window_size);
>> +
>> +		if (copy_from_user(&info, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (info.argsz < minsz)
>> +			return -EINVAL;
>> +
>> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
>> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
>> +		info.dma64_window_start = 0;
>> +		info.dma64_window_size = 0;
>> +		info.flags = 0;
>> +
>> +		if (copy_to_user((void __user *)arg, &info, minsz))
>> +			return -EFAULT;
>> +
>> +		return 0;
>> +	}
>> +	case VFIO_IOMMU_MAP_DMA: {
>> +		vfio_iommu_spapr_tce_dma_map param;
>> +		struct iommu_table *tbl = container->tbl;
>> +		enum dma_data_direction direction;
>> +		unsigned long locked, lock_limit;
>> +
>> +		if (WARN_ON(!tbl))
>> +			return -ENXIO;
>> +
>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
>> +
>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (param.argsz < minsz)
>> +			return -EINVAL;
>> +
>> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
>> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
>> +			direction = DMA_BIDIRECTIONAL;
>> +		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
>> +			direction = DMA_TO_DEVICE;
>> +		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
>> +			direction = DMA_FROM_DEVICE;
>> +		else
>> +			return -EINVAL;
>> +
>> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
>> +				(param.iova & ~IOMMU_PAGE_MASK) ||
>> +				(param.vaddr & ~IOMMU_PAGE_MASK))
>> +			return -EINVAL;
>> +
>> +		/* Account for locked pages */
>> +		locked = current->mm->locked_vm +
>> +				(param.size >> IOMMU_PAGE_SHIFT);
>> +		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
>
> This page accounting doesn't look right.  PAGE_SIZE is several orders
> bigger than IOMMU_PAGE_SIZE (right?), but we mix them here, which seems
> like it will over penalize the user.  For example, if a user maps 4x4k
> (assume aligned and contiguous) IOMMU pages, isn't that only a single
> pinned system page (assuming >=16k pages).

Oops. My bad. IOMMU_PAGE_SHIFT should be PAGE_SHIFT and should return the 
number of system pages.

But we do not track 4K pages so I do not see any easy solution here. Except 
fixing iommu_put_tces/iommu_clear_tces (*) to return the number of the very 
first 4K IOMMU pages within system 64K pages.
This won't be too accurate but should work, no?

I'll post it as a patch in reply to "vfio powerpc: enabled on powernv 
platform".



>> +		if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
>> +			pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
>> +					rlimit(RLIMIT_MEMLOCK));
>> +			return -ENOMEM;
>> +		}
>> +
>> +		ret = iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
>> +				param.vaddr, direction,
>> +				param.size >> IOMMU_PAGE_SHIFT);
>> +		if (ret > 0)
>> +			lock_acct(ret);
>> +
>> +		return ret;
>> +	}
>> +	case VFIO_IOMMU_UNMAP_DMA: {
>> +		vfio_iommu_spapr_tce_dma_unmap param;
>> +		struct iommu_table *tbl = container->tbl;
>> +
>> +		if (WARN_ON(!tbl))
>> +			return -ENXIO;
>> +
>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
>> +
>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (param.argsz < minsz)
>> +			return -EINVAL;
>> +
>> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
>> +				(param.iova & ~IOMMU_PAGE_MASK))
>> +			return -EINVAL;
>> +
>> +		ret = iommu_clear_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
>> +				param.size >> IOMMU_PAGE_SHIFT);
>> +		if (ret > 0)
>> +			lock_acct(-ret);
>> +
>> +		return ret;
>> +	}
>> +	default:
>> +		pr_warn("tce_vfio: unexpected cmd %x\n", cmd);
>> +	}
>> +
>> +	return -ENOTTY;
>> +}
>> +
>> +static int tce_iommu_attach_group(void *iommu_data,
>> +		struct iommu_group *iommu_group)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
>> +
>> +	BUG_ON(!tbl);
>> +	mutex_lock(&container->lock);
>> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
>> +			iommu_group_id(iommu_group), iommu_group);
>> +	if (container->tbl) {
>> +		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
>> +				iommu_group_id(container->tbl->it_group),
>> +				iommu_group_id(iommu_group));
>> +		mutex_unlock(&container->lock);
>> +		return -EBUSY;
>> +	}
>> +
>> +	container->tbl = tbl;
>> +	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
>> +	mutex_unlock(&container->lock);
>> +
>> +	return 0;
>> +}
>> +
>> +static void tce_iommu_detach_group(void *iommu_data,
>> +		struct iommu_group *iommu_group)
>> +{
>> +	struct tce_container *container = iommu_data;
>> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
>> +
>> +	BUG_ON(!tbl);
>> +	mutex_lock(&container->lock);
>> +	if (tbl != container->tbl) {
>> +		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
>> +				iommu_group_id(iommu_group),
>> +				iommu_group_id(tbl->it_group));
>> +	} else {
>> +
>> +		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
>> +				iommu_group_id(iommu_group), iommu_group);
>> +
>> +		iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
>> +		container->tbl = NULL;
>> +	}
>> +	mutex_unlock(&container->lock);
>> +}
>> +
>> +const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
>> +	.name		= "iommu-vfio-powerpc",
>> +	.owner		= THIS_MODULE,
>> +	.open		= tce_iommu_open,
>> +	.release	= tce_iommu_release,
>> +	.ioctl		= tce_iommu_ioctl,
>> +	.attach_group	= tce_iommu_attach_group,
>> +	.detach_group	= tce_iommu_detach_group,
>> +};
>> +
>> +static int __init tce_iommu_init(void)
>> +{
>> +	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
>> +}
>> +
>> +static void __exit tce_iommu_cleanup(void)
>> +{
>> +	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
>> +}
>> +
>> +module_init(tce_iommu_init);
>> +module_exit(tce_iommu_cleanup);
>> +
>> +MODULE_VERSION(DRIVER_VERSION);
>> +MODULE_LICENSE("GPL v2");
>> +MODULE_AUTHOR(DRIVER_AUTHOR);
>> +MODULE_DESCRIPTION(DRIVER_DESC);
>> +
>> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
>> index 0a4f180..820af1e 100644
>> --- a/include/linux/vfio.h
>> +++ b/include/linux/vfio.h
>> @@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
>>   /* Extensions */
>>
>>   #define VFIO_TYPE1_IOMMU		1
>> +#define VFIO_SPAPR_TCE_IOMMU		2
>>
>>   /*
>>    * The IOCTL interface is designed for extensibility by embedding the
>> @@ -442,4 +443,36 @@ struct vfio_iommu_type1_dma_unmap {
>>
>>   #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
>>
>> +/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
>> +
>> +/*
>> + * The SPAPR TCE info struct provides the information about the PCI bus
>> + * address ranges available for DMA, these values are programmed into
>> + * the hardware so the guest has to know that information.
>> + *
>> + * Pages within 32 bit window should be explicitely mapped/unmapped via ioctls.
>                                            ^^^^^^^^^^^
> explicitly
>
>> + * 64 bit window (not supported at the moment for the guest) is supposed to
>> + * be mapped completely to the guest memory so the devices capable of 64bit
>> + * DMA will not have to use map/unmap ioctls.
>> + *
>> + * The IOMMU page size is always 4K.
>> + */
>
> Thanks,
>
> Alex
>
>> +
>> +struct vfio_iommu_spapr_tce_info {
>> +	__u32 argsz;
>> +	__u32 flags;			/* reserved for future use */
>> +	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
>> +	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
>> +	__u64 dma64_window_start;	/* 64 bit window start (bytes) */
>> +	__u64 dma64_window_size;	/* 64 bit window size (bytes) */
>> +};
>> +
>> +#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
>> +
>> +/* Reuse type1 map/unmap structs as they are the same at the moment */
>> +typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
>> +typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
>> +
>> +/* ***************************************************************** */
>> +
>>   #endif /* VFIO_H */
>
>
>


-- 
Alexey

[PATCH 2/2] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan on POWERNV
(POWER non virtualized) platform. The IOMMU groups are
to be used later by VFIO driver (PCI pass through).

It also implements an API for mapping/unmapping pages for
guest PCI drivers and providing DMA window properties.
This API is going to be used later by QEMU-VFIO to handle
h_put_tce hypercalls from the KVM guest.

Although this driver has been tested only on the POWERNV
platform, it should work on any platform which supports
TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option and configure VFIO as required.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |    6 ++
 arch/powerpc/kernel/iommu.c          |  141 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 4 files changed, 290 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..5ba66cb 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -76,6 +76,9 @@ struct iommu_table {
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
@@ -147,5 +150,8 @@ static inline void iommu_restore(void)
 }
 #endif
 
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
+		enum dma_data_direction direction, unsigned long pages);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..c8dad1f 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -44,6 +44,7 @@
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
@@ -856,3 +857,143 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+static struct page *free_tce(struct iommu_table *tbl, unsigned long entry)
+{
+	struct page *page;
+	unsigned long oldtce;
+
+	oldtce = ppc_md.tce_get(tbl, entry);
+
+	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
+		return NULL;
+
+	page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+	WARN_ON(!page);
+	if (page && (oldtce & TCE_PCI_WRITE))
+		SetPageDirty(page);
+	ppc_md.tce_free(tbl, entry, 1);
+
+	return page;
+}
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset;
+
+	/* Map new TCE */
+	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret < 1) {
+		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		if (!ret)
+			ret = -EFAULT;
+		return ret;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	entry += tbl->it_offset; /* Offset into real TCE table */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	return 0;
+}
+
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
+		enum dma_data_direction direction, unsigned long pages)
+{
+	int i, ret = 0, pages_to_put = 0;
+	struct page *page;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+	struct page **oldpages;
+	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+
+	/* Handle a single page request without allocation
+	   of pages-to-release array */
+	if (pages == 1) {
+		spin_lock(&(pool->lock));
+		page = free_tce(tbl, entry);
+
+		if (direction != DMA_NONE)
+			ret = put_tce(tbl, entry, tce, direction);
+
+		tce_flush(tbl);
+
+		if (page)
+			put_page(page);
+
+		spin_unlock(&(pool->lock));
+		return ret;
+	}
+
+	/* Releasing multiple pages */
+	/* Allocate an array for pages to be released after TCE table
+	   is updated */
+	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
+	if (!oldpages)
+		return -ENOMEM;
+
+	spin_lock(&(pool->lock));
+
+	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce += IOMMU_PAGE_SIZE) {
+		page = free_tce(tbl, entry);
+		if (page) {
+			oldpages[pages_to_put] = page;
+			++pages_to_put;
+		}
+
+		if (direction != DMA_NONE)
+			ret = put_tce(tbl, entry, tce, direction);
+
+		/* Release old pages if we reached the end of oldpages[] or
+		   it is the last page or we are about to exit the loop */
+		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret) {
+			tce_flush(tbl);
+
+			/* Release pages after removing them from TCE table */
+			while (pages_to_put) {
+				--pages_to_put;
+				put_page(oldpages[pages_to_put]);
+			}
+		}
+	}
+
+	spin_unlock(&(pool->lock));
+	kfree(oldpages);
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..660dcc6 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -20,6 +20,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
@@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev_name(dev),
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev_name(dev));
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev_name(dev), iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
+				dev_name(dev), ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			printk(KERN_INFO "tce_vfio: cannot create "
+					"new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
-- 
1.7.10.4

Re: [PATCH 2/2] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Fri, 2012-11-23 at 20:03 +1100, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
> 
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
> 
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |    6 ++
>  arch/powerpc/kernel/iommu.c          |  141 ++++++++++++++++++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++++++++++
>  drivers/iommu/Kconfig                |    8 ++
>  4 files changed, 290 insertions(+)
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..5ba66cb 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -76,6 +76,9 @@ struct iommu_table {
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
> @@ -147,5 +150,8 @@ static inline void iommu_restore(void)
>  }
>  #endif
>  
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
> +		enum dma_data_direction direction, unsigned long pages);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..c8dad1f 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -44,6 +44,7 @@
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
> @@ -856,3 +857,143 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +static struct page *free_tce(struct iommu_table *tbl, unsigned long entry)
> +{
> +	struct page *page;
> +	unsigned long oldtce;
> +
> +	oldtce = ppc_md.tce_get(tbl, entry);
> +
> +	if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> +		return NULL;
> +
> +	page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +	WARN_ON(!page);
> +	if (page && (oldtce & TCE_PCI_WRITE))
> +		SetPageDirty(page);
> +	ppc_md.tce_free(tbl, entry, 1);
> +
> +	return page;
> +}
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset;
> +
> +	/* Map new TCE */
> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);

We're locking memory here on behalf of the user, but I don't see where
rlimit gets checked to verify the user has privileges to lock the pages.
I know you're locking a much smaller set of memory than x86 does, but
are we just foregoing that added security?

> +	if (ret < 1) {
> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> +		if (!ret)
> +			ret = -EFAULT;
> +		return ret;
> +	}
> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	entry += tbl->it_offset; /* Offset into real TCE table */
> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	return 0;
> +}
> +
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry, uint64_t tce,
> +		enum dma_data_direction direction, unsigned long pages)
> +{
> +	int i, ret = 0, pages_to_put = 0;
> +	struct page *page;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +	struct page **oldpages;
> +	const int oldpagesnum = PAGE_SIZE/sizeof(*oldpages);
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +
> +	/* Handle a single page request without allocation
> +	   of pages-to-release array */

nit, this comment style doesn't seem to match anything existing in this
file.  I'd also be tempted to use pr_err/warn in this file, but I'll
leave that for the maintainers.  Thanks,

Alex

> +	if (pages == 1) {
> +		spin_lock(&(pool->lock));
> +		page = free_tce(tbl, entry);
> +
> +		if (direction != DMA_NONE)
> +			ret = put_tce(tbl, entry, tce, direction);
> +
> +		tce_flush(tbl);
> +
> +		if (page)
> +			put_page(page);
> +
> +		spin_unlock(&(pool->lock));
> +		return ret;
> +	}
> +
> +	/* Releasing multiple pages */
> +	/* Allocate an array for pages to be released after TCE table
> +	   is updated */
> +	oldpages = kmalloc(PAGE_SIZE, GFP_KERNEL);
> +	if (!oldpages)
> +		return -ENOMEM;
> +
> +	spin_lock(&(pool->lock));
> +
> +	for (i = 0; (i < pages) && !ret; ++i, ++entry, tce += IOMMU_PAGE_SIZE) {
> +		page = free_tce(tbl, entry);
> +		if (page) {
> +			oldpages[pages_to_put] = page;
> +			++pages_to_put;
> +		}
> +
> +		if (direction != DMA_NONE)
> +			ret = put_tce(tbl, entry, tce, direction);
> +
> +		/* Release old pages if we reached the end of oldpages[] or
> +		   it is the last page or we are about to exit the loop */
> +		if ((pages_to_put == oldpagesnum) || (i == pages - 1) || ret) {
> +			tce_flush(tbl);
> +
> +			/* Release pages after removing them from TCE table */
> +			while (pages_to_put) {
> +				--pages_to_put;
> +				put_page(oldpages[pages_to_put]);
> +			}
> +		}
> +	}
> +
> +	spin_unlock(&(pool->lock));
> +	kfree(oldpages);
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..660dcc6 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -20,6 +20,7 @@
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
> @@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev_name(dev),
> +				iommu_group_id(dev->iommu_group));
> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev_name(dev));
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev_name(dev), iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> +				dev_name(dev), ret);
> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			printk(KERN_INFO "tce_vfio: cannot create "
> +					"new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));
> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);
> +	}
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

[PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan on POWERNV
(POWER non virtualized) platform. The IOMMU groups are
to be used later by VFIO driver (PCI pass through).

It also implements an API for mapping/unmapping pages for
guest PCI drivers and providing DMA window properties.
This API is going to be used later by QEMU-VFIO to handle
h_put_tce hypercalls from the KVM guest.

Although this driver has been tested only on the POWERNV
platform, it should work on any platform which supports
TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option and configure VFIO as required.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |    9 +++
 arch/powerpc/kernel/iommu.c          |  147 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 4 files changed, 299 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..5c7087a 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -76,6 +76,9 @@ struct iommu_table {
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
@@ -147,5 +150,11 @@ static inline void iommu_restore(void)
 }
 #endif
 
+extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages);
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..1456b6e 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -44,6 +44,7 @@
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
@@ -856,3 +857,149 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of pages
+ * which it called put_page() on.
+ */
+static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int i, pages_put = 0;
+	unsigned long oldtce;
+	struct page *page;
+
+	for (i = 0; i < pages; ++i) {
+		oldtce = ppc_md.tce_get(tbl, entry + i);
+		ppc_md.tce_free(tbl, entry + i, 1);
+
+		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
+			continue;
+
+		page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+		WARN_ON(!page);
+		if (!page)
+			continue;
+
+		if (oldtce & TCE_PCI_WRITE)
+			SetPageDirty(page);
+
+		++pages_put;
+		put_page(page);
+	}
+
+	return pages_put;
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of released pages
+ */
+long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int ret;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	spin_lock(&(pool->lock));
+	ret = clear_tces_nolock(tbl, entry, pages);
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_clear_tces);
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset;
+
+	/* Map new TCE */
+	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
+
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret < 1) {
+		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		if (!ret)
+			ret = -EFAULT;
+		return ret;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	entry += tbl->it_offset; /* Offset into real TCE table */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	return 0;
+}
+
+/*
+ * iommu_put_tces builds tces and returned the number of actually locked pages
+ */
+long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages)
+{
+	int i, ret = 0;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+	BUG_ON(direction == DMA_NONE);
+
+	spin_lock(&(pool->lock));
+
+	/* Check if any is in use */
+	for (i = 0; i < pages; ++i) {
+		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
+		if (oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) {
+			spin_unlock(&(pool->lock));
+			return -EBUSY;
+		}
+	}
+
+	/* Put tces to the table */
+	for (i = 0; (i < pages) && !ret; ++i, tce += IOMMU_PAGE_SIZE)
+		ret = put_tce(tbl, entry + i, tce, direction);
+
+	/* If failed, release locked pages, otherwise return the number of pages */
+	if (ret)
+		clear_tces_nolock(tbl, entry, i);
+	else
+		ret = pages;
+
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..21250ef 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -20,6 +20,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
@@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev_name(dev),
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev_name(dev));
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev_name(dev), iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
+				dev_name(dev), ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			printk(KERN_INFO "tce_vfio: cannot create "
+					"new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Wed, 2012-11-28 at 18:18 +1100, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
> 
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
> 
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |    9 +++
>  arch/powerpc/kernel/iommu.c          |  147 ++++++++++++++++++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++++++++++++++
>  drivers/iommu/Kconfig                |    8 ++
>  4 files changed, 299 insertions(+)
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..5c7087a 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -76,6 +76,9 @@ struct iommu_table {
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
> @@ -147,5 +150,11 @@ static inline void iommu_restore(void)
>  }
>  #endif
>  
> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages);
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..1456b6e 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -44,6 +44,7 @@
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
> @@ -856,3 +857,149 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of pages
> + * which it called put_page() on.
> + */
> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int i, pages_put = 0;
> +	unsigned long oldtce;
> +	struct page *page;
> +
> +	for (i = 0; i < pages; ++i) {
> +		oldtce = ppc_md.tce_get(tbl, entry + i);
> +		ppc_md.tce_free(tbl, entry + i, 1);
> +
> +		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> +			continue;
> +
> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +		WARN_ON(!page);
> +		if (!page)
> +			continue;
> +
> +		if (oldtce & TCE_PCI_WRITE)
> +			SetPageDirty(page);
> +
> +		++pages_put;
> +		put_page(page);
> +	}
> +
> +	return pages_put;
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of released pages
> + */
> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int ret;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	spin_lock(&(pool->lock));
> +	ret = clear_tces_nolock(tbl, entry, pages);
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset;
> +
> +	/* Map new TCE */
> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> +
> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);
> +	if (ret < 1) {
> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> +		if (!ret)
> +			ret = -EFAULT;
> +		return ret;
> +	}
> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	entry += tbl->it_offset; /* Offset into real TCE table */
> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	return 0;
> +}
> +
> +/*
> + * iommu_put_tces builds tces and returned the number of actually locked pages
> + */
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages)
> +{
> +	int i, ret = 0;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +	BUG_ON(direction == DMA_NONE);
> +
> +	spin_lock(&(pool->lock));
> +
> +	/* Check if any is in use */
> +	for (i = 0; i < pages; ++i) {
> +		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
> +		if (oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) {
> +			spin_unlock(&(pool->lock));
> +			return -EBUSY;
> +		}
> +	}
> +
> +	/* Put tces to the table */
> +	for (i = 0; (i < pages) && !ret; ++i, tce += IOMMU_PAGE_SIZE)
> +		ret = put_tce(tbl, entry + i, tce, direction);
> +
> +	/* If failed, release locked pages, otherwise return the number of pages */
> +	if (ret)
> +		clear_tces_nolock(tbl, entry, i);
> +	else
> +		ret = pages;
> +
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}

Nice, no more kmalloc!  I'm still concerned about the IOMMU page size
mismatch here.  If nothing else, the comment is misleading since we're
locking system pages, but returning tce pages.  The user would therefore
need to multiply their locked memory limit by (PAGE_SHIFT -
IOMMU_PAGE_SHIFT).  Thanks,

Alex

> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..21250ef 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -20,6 +20,7 @@
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
> @@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev_name(dev),
> +				iommu_group_id(dev->iommu_group));
> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev_name(dev));
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev_name(dev), iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> +				dev_name(dev), ret);
> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			printk(KERN_INFO "tce_vfio: cannot create "
> +					"new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));
> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);
> +	}
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

[PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan on POWERNV
(POWER non virtualized) platform. The IOMMU groups are
to be used later by VFIO driver (PCI pass through).

It also implements an API for mapping/unmapping pages for
guest PCI drivers and providing DMA window properties.
This API is going to be used later by QEMU-VFIO to handle
h_put_tce hypercalls from the KVM guest.

Although this driver has been tested only on the POWERNV
platform, it should work on any platform which supports
TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option and configure VFIO as required.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |    9 ++
 arch/powerpc/kernel/iommu.c          |  159 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 4 files changed, 311 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..5c7087a 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -76,6 +76,9 @@ struct iommu_table {
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
@@ -147,5 +150,11 @@ static inline void iommu_restore(void)
 }
 #endif
 
+extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages);
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..1225fbb 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -44,6 +44,7 @@
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
@@ -856,3 +857,161 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of pages
+ * which it called put_page() on.
+ */
+static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int i, retpages = 0;
+	unsigned long oldtce;
+	struct page *page;
+
+	for (i = 0; i < pages; ++i) {
+		oldtce = ppc_md.tce_get(tbl, entry + i);
+		ppc_md.tce_free(tbl, entry + i, 1);
+
+		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
+			continue;
+
+		page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+		WARN_ON(!page);
+		if (!page)
+			continue;
+
+		if (oldtce & TCE_PCI_WRITE)
+			SetPageDirty(page);
+
+		if (!(oldtce & ~PAGE_MASK))
+			++retpages;
+
+		put_page(page);
+	}
+
+	return retpages;
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of released pages
+ */
+long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int ret;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	spin_lock(&(pool->lock));
+	ret = clear_tces_nolock(tbl, entry, pages);
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_clear_tces);
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset;
+
+	/* Map new TCE */
+	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
+
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret < 1) {
+		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		if (!ret)
+			ret = -EFAULT;
+		return ret;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	entry += tbl->it_offset; /* Offset into real TCE table */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	return 0;
+}
+
+/*
+ * iommu_put_tces builds tces and returned the number of actually
+ * locked system pages
+ */
+long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages)
+{
+	int i, ret = 0, retpages = 0;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+	BUG_ON(direction == DMA_NONE);
+
+	spin_lock(&(pool->lock));
+
+	/* Check if any is in use */
+	for (i = 0; i < pages; ++i) {
+		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
+		if (oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) {
+			spin_unlock(&(pool->lock));
+			return -EBUSY;
+		}
+	}
+
+	/* Put tces to the table */
+	for (i = 0; (i < pages) && !ret; ++i, tce += IOMMU_PAGE_SIZE) {
+		ret = put_tce(tbl, entry + i, tce, direction);
+		/*
+		 * As IOMMU page size is always 4K, the system page size
+		 * can be 64K and there is no special tracking for IOMMU pages,
+		 * we only do rlimit check/update for the very first
+		 * 4K IOMMUpage within 64K system page.
+		 */
+		if (!(tce & ~PAGE_MASK))
+			++retpages;
+	}
+
+	/* If failed, release locked pages, otherwise return the number of pages */
+	if (ret)
+		clear_tces_nolock(tbl, entry, i);
+	else
+		ret = retpages;
+
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..21250ef 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -20,6 +20,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
@@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev_name(dev),
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev_name(dev));
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev_name(dev), iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
+				dev_name(dev), ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			printk(KERN_INFO "tce_vfio: cannot create "
+					"new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Thu, 2012-11-29 at 14:53 +1100, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
> 
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
> 
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |    9 ++
>  arch/powerpc/kernel/iommu.c          |  159 ++++++++++++++++++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++++++++++++++++++
>  drivers/iommu/Kconfig                |    8 ++
>  4 files changed, 311 insertions(+)
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..5c7087a 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -76,6 +76,9 @@ struct iommu_table {
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
> @@ -147,5 +150,11 @@ static inline void iommu_restore(void)
>  }
>  #endif
>  
> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages);
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..1225fbb 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -44,6 +44,7 @@
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
> @@ -856,3 +857,161 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of pages
> + * which it called put_page() on.
> + */
> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int i, retpages = 0;
> +	unsigned long oldtce;
> +	struct page *page;
> +
> +	for (i = 0; i < pages; ++i) {
> +		oldtce = ppc_md.tce_get(tbl, entry + i);
> +		ppc_md.tce_free(tbl, entry + i, 1);
> +
> +		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> +			continue;
> +
> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +		WARN_ON(!page);
> +		if (!page)
> +			continue;
> +
> +		if (oldtce & TCE_PCI_WRITE)
> +			SetPageDirty(page);
> +
> +		if (!(oldtce & ~PAGE_MASK))
> +			++retpages;

I'm confused, it looks like you're trying to only increment the counter
for tce pages aligned at the start of a page, but don't we need to mask
out the read/write and valid bits?  Trickiness like this demands a
comment.

> +
> +		put_page(page);
> +	}
> +
> +	return retpages;
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of released pages
> + */
> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int ret;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	spin_lock(&(pool->lock));
> +	ret = clear_tces_nolock(tbl, entry, pages);
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset;
> +
> +	/* Map new TCE */
> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> +
> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);
> +	if (ret < 1) {
> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> +		if (!ret)
> +			ret = -EFAULT;
> +		return ret;
> +	}
> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	entry += tbl->it_offset; /* Offset into real TCE table */
> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	return 0;
> +}
> +
> +/*
> + * iommu_put_tces builds tces and returned the number of actually
> + * locked system pages
> + */
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages)
> +{
> +	int i, ret = 0, retpages = 0;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +	BUG_ON(direction == DMA_NONE);
> +
> +	spin_lock(&(pool->lock));
> +
> +	/* Check if any is in use */
> +	for (i = 0; i < pages; ++i) {
> +		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
> +		if (oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) {
> +			spin_unlock(&(pool->lock));
> +			return -EBUSY;
> +		}
> +	}
> +
> +	/* Put tces to the table */
> +	for (i = 0; (i < pages) && !ret; ++i, tce += IOMMU_PAGE_SIZE) {
> +		ret = put_tce(tbl, entry + i, tce, direction);
> +		/*
> +		 * As IOMMU page size is always 4K, the system page size
> +		 * can be 64K and there is no special tracking for IOMMU pages,
> +		 * we only do rlimit check/update for the very first
> +		 * 4K IOMMUpage within 64K system page.
> +		 */
> +		if (!(tce & ~PAGE_MASK))
> +			++retpages;

Ah, here's the comment I was looking for, though I'm still not sure
about the read/write bits.

Isn't there an exploit here that a user can lock pages beyond their
limits if they just skip mapping the first 4k of each page?  Thanks,

Alex

> +	}
> +
> +	/* If failed, release locked pages, otherwise return the number of pages */
> +	if (ret)
> +		clear_tces_nolock(tbl, entry, i);
> +	else
> +		ret = retpages;
> +
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..21250ef 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -20,6 +20,7 @@
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
> @@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev_name(dev),
> +				iommu_group_id(dev->iommu_group));
> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev_name(dev));
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev_name(dev), iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> +				dev_name(dev), ret);
> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			printk(KERN_INFO "tce_vfio: cannot create "
> +					"new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));
> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);
> +	}
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

[PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan on POWERNV
(POWER non virtualized) platform. The IOMMU groups are
to be used later by VFIO driver (PCI pass through).

It also implements an API for mapping/unmapping pages for
guest PCI drivers and providing DMA window properties.
This API is going to be used later by QEMU-VFIO to handle
h_put_tce hypercalls from the KVM guest.

Although this driver has been tested only on the POWERNV
platform, it should work on any platform which supports
TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option and configure VFIO as required.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |    9 ++
 arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 4 files changed, 338 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..5c7087a 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -76,6 +76,9 @@ struct iommu_table {
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
@@ -147,5 +150,11 @@ static inline void iommu_restore(void)
 }
 #endif
 
+extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages);
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..0646c50 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -44,6 +44,7 @@
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
@@ -856,3 +857,188 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+
+/*
+ * Returns the number of used IOMMU pages (4K) within
+ * the same system page (4K or 64K).
+ * bitmap_weight is not used as it does not support bigendian maps.
+ */
+static int syspage_weight(unsigned long *map, unsigned long entry)
+{
+	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
+
+	/* Aligns TCE entry number to system page boundary */
+	entry &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
+
+	/* Count used 4K pages */
+	while (nbits--)
+		ret += (test_bit(entry++, map) == 0) ? 0 : 1;
+
+	return ret;
+}
+
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of system pages
+ * which it called put_page() on
+ */
+static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int i, retpages = 0;
+	unsigned long oldtce, oldweight;
+	struct page *page;
+
+	for (i = 0; i < pages; ++i) {
+		oldtce = ppc_md.tce_get(tbl, entry + i);
+		ppc_md.tce_free(tbl, entry + i, 1);
+
+		oldweight = syspage_weight(tbl->it_map, entry);
+		__clear_bit(entry, tbl->it_map);
+
+		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
+			continue;
+
+		page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+		WARN_ON(!page);
+		if (!page)
+			continue;
+
+		if (oldtce & TCE_PCI_WRITE)
+			SetPageDirty(page);
+
+		put_page(page);
+
+		/* That was the last IOMMU page within the system page */
+		if ((oldweight == 1) && !syspage_weight(tbl->it_map, entry))
+			++retpages;
+	}
+
+	return retpages;
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number
+ / of released system pages
+ */
+long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int ret;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	spin_lock(&(pool->lock));
+	ret = clear_tces_nolock(tbl, entry, pages);
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_clear_tces);
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset, oldweight;
+
+	/* Map new TCE */
+	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret < 1) {
+		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		if (!ret || (ret > 1))
+			ret = -EFAULT;
+		return ret;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	entry += tbl->it_offset; /* Offset into real TCE table */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	/* Calculate if new system page has been locked */
+	oldweight = syspage_weight(tbl->it_map, entry);
+	__set_bit(entry, tbl->it_map);
+
+	return (oldweight == 0) ? 1 : 0;
+}
+
+/*
+ * iommu_put_tces builds tces and returned the number of actually
+ * locked system pages
+ */
+long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages)
+{
+	int i, ret = 0, retpages = 0;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+	BUG_ON(direction == DMA_NONE);
+
+	spin_lock(&(pool->lock));
+
+	/* Check if any is in use */
+	for (i = 0; i < pages; ++i) {
+		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
+		if ((oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) ||
+				test_bit(entry + i, tbl->it_map)) {
+			WARN_ON(test_bit(entry + i, tbl->it_map));
+			spin_unlock(&(pool->lock));
+			return -EBUSY;
+		}
+	}
+
+	/* Put tces to the table */
+	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
+		ret = put_tce(tbl, entry + i, tce, direction);
+		if (ret == 1)
+			++retpages;
+	}
+
+	/*
+	 * If failed, release locked pages, otherwise return the number
+	 * of locked system pages
+	 */
+	if (ret < 0)
+		clear_tces_nolock(tbl, entry, i);
+	else
+		ret = retpages;
+
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..21250ef 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -20,6 +20,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
@@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev_name(dev),
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev_name(dev));
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev_name(dev), iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
+				dev_name(dev), ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			printk(KERN_INFO "tce_vfio: cannot create "
+					"new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Fri, 2012-11-30 at 17:14 +1100, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
> 
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
> 
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |    9 ++
>  arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++
>  drivers/iommu/Kconfig                |    8 ++
>  4 files changed, 338 insertions(+)
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..5c7087a 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -76,6 +76,9 @@ struct iommu_table {
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
> @@ -147,5 +150,11 @@ static inline void iommu_restore(void)
>  }
>  #endif
>  
> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages);
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..0646c50 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -44,6 +44,7 @@
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
> @@ -856,3 +857,188 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +
> +/*
> + * Returns the number of used IOMMU pages (4K) within
> + * the same system page (4K or 64K).
> + * bitmap_weight is not used as it does not support bigendian maps.
> + */
> +static int syspage_weight(unsigned long *map, unsigned long entry)
> +{
> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
> +
> +	/* Aligns TCE entry number to system page boundary */
> +	entry &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> +
> +	/* Count used 4K pages */
> +	while (nbits--)
> +		ret += (test_bit(entry++, map) == 0) ? 0 : 1;

Ok, entry is the iova page number.  So presumably it's relative to the
start of dma32_window_start since you're unlikely to have a bitmap that
covers all of memory.  I hadn't realized that previously.  Doesn't that
mean that it's actually impossible to create an ioctl based interface to
the dma64_window since we're not going to know which window is the
target?  I know you're not planning on one, but it seems limiting.  We
at least need some documentation here, but I'm wondering if iova
shouldn't be zero based so we can determine which window it hits.  Also,
now that I look at it, I can't find any range checking on the iova.
Thanks,

Alex

> +
> +	return ret;
> +}
> +
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of system pages
> + * which it called put_page() on
> + */
> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int i, retpages = 0;
> +	unsigned long oldtce, oldweight;
> +	struct page *page;
> +
> +	for (i = 0; i < pages; ++i) {
> +		oldtce = ppc_md.tce_get(tbl, entry + i);
> +		ppc_md.tce_free(tbl, entry + i, 1);
> +
> +		oldweight = syspage_weight(tbl->it_map, entry);
> +		__clear_bit(entry, tbl->it_map);
> +
> +		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> +			continue;
> +
> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +		WARN_ON(!page);
> +		if (!page)
> +			continue;
> +
> +		if (oldtce & TCE_PCI_WRITE)
> +			SetPageDirty(page);
> +
> +		put_page(page);
> +
> +		/* That was the last IOMMU page within the system page */
> +		if ((oldweight == 1) && !syspage_weight(tbl->it_map, entry))
> +			++retpages;
> +	}
> +
> +	return retpages;
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number
> + / of released system pages
> + */
> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int ret;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	spin_lock(&(pool->lock));
> +	ret = clear_tces_nolock(tbl, entry, pages);
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset, oldweight;
> +
> +	/* Map new TCE */
> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);
> +	if (ret < 1) {
> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> +		if (!ret || (ret > 1))
> +			ret = -EFAULT;
> +		return ret;
> +	}
> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	entry += tbl->it_offset; /* Offset into real TCE table */
> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	/* Calculate if new system page has been locked */
> +	oldweight = syspage_weight(tbl->it_map, entry);
> +	__set_bit(entry, tbl->it_map);
> +
> +	return (oldweight == 0) ? 1 : 0;
> +}
> +
> +/*
> + * iommu_put_tces builds tces and returned the number of actually
> + * locked system pages
> + */
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages)
> +{
> +	int i, ret = 0, retpages = 0;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +	BUG_ON(direction == DMA_NONE);
> +
> +	spin_lock(&(pool->lock));
> +
> +	/* Check if any is in use */
> +	for (i = 0; i < pages; ++i) {
> +		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
> +		if ((oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) ||
> +				test_bit(entry + i, tbl->it_map)) {
> +			WARN_ON(test_bit(entry + i, tbl->it_map));
> +			spin_unlock(&(pool->lock));
> +			return -EBUSY;
> +		}
> +	}
> +
> +	/* Put tces to the table */
> +	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
> +		ret = put_tce(tbl, entry + i, tce, direction);
> +		if (ret == 1)
> +			++retpages;
> +	}
> +
> +	/*
> +	 * If failed, release locked pages, otherwise return the number
> +	 * of locked system pages
> +	 */
> +	if (ret < 0)
> +		clear_tces_nolock(tbl, entry, i);
> +	else
> +		ret = retpages;
> +
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..21250ef 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -20,6 +20,7 @@
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
> @@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev_name(dev),
> +				iommu_group_id(dev->iommu_group));
> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev_name(dev));
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev_name(dev), iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> +				dev_name(dev), ret);
> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			printk(KERN_INFO "tce_vfio: cannot create "
> +					"new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));
> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);
> +	}
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

On 01/12/12 03:48, Alex Williamson wrote:
> On Fri, 2012-11-30 at 17:14 +1100, Alexey Kardashevskiy wrote:
>> This patch initializes IOMMU groups based on the IOMMU
>> configuration discovered during the PCI scan on POWERNV
>> (POWER non virtualized) platform. The IOMMU groups are
>> to be used later by VFIO driver (PCI pass through).
>>
>> It also implements an API for mapping/unmapping pages for
>> guest PCI drivers and providing DMA window properties.
>> This API is going to be used later by QEMU-VFIO to handle
>> h_put_tce hypercalls from the KVM guest.
>>
>> Although this driver has been tested only on the POWERNV
>> platform, it should work on any platform which supports
>> TCE tables.
>>
>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
>> option and configure VFIO as required.
>>
>> Cc: David Gibson <david@gibson.dropbear.id.au>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>> ---
>>   arch/powerpc/include/asm/iommu.h     |    9 ++
>>   arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++++++++++++++++++
>>   arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++
>>   drivers/iommu/Kconfig                |    8 ++
>>   4 files changed, 338 insertions(+)
>>
>> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
>> index cbfe678..5c7087a 100644
>> --- a/arch/powerpc/include/asm/iommu.h
>> +++ b/arch/powerpc/include/asm/iommu.h
>> @@ -76,6 +76,9 @@ struct iommu_table {
>>   	struct iommu_pool large_pool;
>>   	struct iommu_pool pools[IOMMU_NR_POOLS];
>>   	unsigned long *it_map;       /* A simple allocation bitmap for now */
>> +#ifdef CONFIG_IOMMU_API
>> +	struct iommu_group *it_group;
>> +#endif
>>   };
>>
>>   struct scatterlist;
>> @@ -147,5 +150,11 @@ static inline void iommu_restore(void)
>>   }
>>   #endif
>>
>> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
>> +		unsigned long pages);
>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
>> +		uint64_t tce, enum dma_data_direction direction,
>> +		unsigned long pages);
>> +
>>   #endif /* __KERNEL__ */
>>   #endif /* _ASM_IOMMU_H */
>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
>> index ff5a6ce..0646c50 100644
>> --- a/arch/powerpc/kernel/iommu.c
>> +++ b/arch/powerpc/kernel/iommu.c
>> @@ -44,6 +44,7 @@
>>   #include <asm/kdump.h>
>>   #include <asm/fadump.h>
>>   #include <asm/vio.h>
>> +#include <asm/tce.h>
>>
>>   #define DBG(...)
>>
>> @@ -856,3 +857,188 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>>   		free_pages((unsigned long)vaddr, get_order(size));
>>   	}
>>   }
>> +
>> +#ifdef CONFIG_IOMMU_API
>> +/*
>> + * SPAPR TCE API
>> + */
>> +
>> +/*
>> + * Returns the number of used IOMMU pages (4K) within
>> + * the same system page (4K or 64K).
>> + * bitmap_weight is not used as it does not support bigendian maps.
>> + */
>> +static int syspage_weight(unsigned long *map, unsigned long entry)
>> +{
>> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
>> +
>> +	/* Aligns TCE entry number to system page boundary */
>> +	entry &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
>> +
>> +	/* Count used 4K pages */
>> +	while (nbits--)
>> +		ret += (test_bit(entry++, map) == 0) ? 0 : 1;
>
> Ok, entry is the iova page number.  So presumably it's relative to the
> start of dma32_window_start since you're unlikely to have a bitmap that
> covers all of memory.  I hadn't realized that previously.

No, it is zero based. The DMA window is a filter but not offset. But you 
are right, the it_map does not cover the whole global table (one per PHB, 
roughly), will fix it, thanks for pointing. On my test system IOMMU group 
is a whole PHB and DMA window always starts from 0 so tests do not show 
everything :)

> Doesn't that
> mean that it's actually impossible to create an ioctl based interface to
> the dma64_window since we're not going to know which window is the
> target?  I know you're not planning on one, but it seems limiting.

No ,it is not limiting as iova is zero based. Even if it was, there are 
flags in map/unmap ioctls which we could use, no?

> We
> at least need some documentation here, but I'm wondering if iova
> shouldn't be zero based so we can determine which window it hits.  Also,
> now that I look at it, I can't find any range checking on the iova.

True... Have not hit this problem yet :) Good point, will fix, thanks.



-- 
Alexey

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

On 29/11/12 15:20, Alex Williamson wrote:

>> +	/* Put tces to the table */
>> +	for (i = 0; (i < pages) && !ret; ++i, tce += IOMMU_PAGE_SIZE) {
>> +		ret = put_tce(tbl, entry + i, tce, direction);
>> +		/*
>> +		 * As IOMMU page size is always 4K, the system page size
>> +		 * can be 64K and there is no special tracking for IOMMU pages,
>> +		 * we only do rlimit check/update for the very first
>> +		 * 4K IOMMUpage within 64K system page.
>> +		 */
>> +		if (!(tce & ~PAGE_MASK))
>> +			++retpages;
>
> Ah, here's the comment I was looking for, though I'm still not sure
> about the read/write bits.
>
> Isn't there an exploit here that a user can lock pages beyond their
> limits if they just skip mapping the first 4k of each page?  Thanks,


Heh. True. Posted another patch with 4K pages per system page usage tracking.



-- 
Alexey

[PATCH 0/2] vfio on power: yet another try

[Alexey Kardashevskiy]

The set includes 2 patches.

The first one adds necessary support for VFIO IOMMU support,
the second one adds a SPAPR TCE IOMMU driver to VFIO.

At the moment we have decided to get rid of DMA64 window
properties because we need more API than just 2 properties
(such as dynamic window allocation) but have not decided
about its actual design yet.

Alexey Kardashevskiy (2):
  vfio powerpc: enabled on powernv platform
  vfio powerpc: implemented IOMMU driver for VFIO

 arch/powerpc/include/asm/iommu.h     |    9 +
 arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 +++++++++++++
 drivers/iommu/Kconfig                |    8 +
 drivers/vfio/Kconfig                 |    6 +
 drivers/vfio/Makefile                |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c  |  350 ++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                 |   26 +++
 8 files changed, 721 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

-- 
1.7.10.4

[PATCH 1/2] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan on POWERNV
(POWER non virtualized) platform. The IOMMU groups are
to be used later by VFIO driver (PCI pass through).

It also implements an API for mapping/unmapping pages for
guest PCI drivers and providing DMA window properties.
This API is going to be used later by QEMU-VFIO to handle
h_put_tce hypercalls from the KVM guest.

Although this driver has been tested only on the POWERNV
platform, it should work on any platform which supports
TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option and configure VFIO as required.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |    9 ++
 arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 4 files changed, 338 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..5c7087a 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -76,6 +76,9 @@ struct iommu_table {
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
@@ -147,5 +150,11 @@ static inline void iommu_restore(void)
 }
 #endif
 
+extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages);
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..2738aa4 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -44,6 +44,7 @@
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
@@ -856,3 +857,188 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+
+/*
+ * Returns the number of used IOMMU pages (4K) within
+ * the same system page (4K or 64K).
+ * bitmap_weight is not used as it does not support bigendian maps.
+ */
+static int syspage_weight(unsigned long *map, unsigned long entry)
+{
+	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
+
+	/* Aligns TCE entry number to system page boundary */
+	entry &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
+
+	/* Count used 4K pages */
+	while (nbits--)
+		ret += (test_bit(entry++, map) == 0) ? 0 : 1;
+
+	return ret;
+}
+
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of system pages
+ * which it called put_page() on
+ */
+static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int i, retpages = 0;
+	unsigned long oldtce, oldweight;
+	struct page *page;
+
+	for (i = 0; i < pages; ++i) {
+		oldtce = ppc_md.tce_get(tbl, entry + i);
+		ppc_md.tce_free(tbl, entry + i, 1);
+
+		oldweight = syspage_weight(tbl->it_map, entry);
+		__clear_bit(entry - tbl->it_offset, tbl->it_map);
+
+		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
+			continue;
+
+		page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+		WARN_ON(!page);
+		if (!page)
+			continue;
+
+		if (oldtce & TCE_PCI_WRITE)
+			SetPageDirty(page);
+
+		put_page(page);
+
+		/* That was the last IOMMU page within the system page */
+		if ((oldweight == 1) && !syspage_weight(tbl->it_map, entry))
+			++retpages;
+	}
+
+	return retpages;
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number
+ / of released system pages
+ */
+long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int ret;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	spin_lock(&(pool->lock));
+	ret = clear_tces_nolock(tbl, entry, pages);
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_clear_tces);
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset, oldweight;
+
+	/* Map new TCE */
+	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret < 1) {
+		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		if (!ret || (ret > 1))
+			ret = -EFAULT;
+		return ret;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	entry += tbl->it_offset; /* Offset into real TCE table */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	/* Calculate if new system page has been locked */
+	oldweight = syspage_weight(tbl->it_map, entry);
+	__set_bit(entry - tbl->it_offset, tbl->it_map);
+
+	return (oldweight == 0) ? 1 : 0;
+}
+
+/*
+ * iommu_put_tces builds tces and returned the number of actually
+ * locked system pages
+ */
+long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages)
+{
+	int i, ret = 0, retpages = 0;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+	BUG_ON(direction == DMA_NONE);
+
+	spin_lock(&(pool->lock));
+
+	/* Check if any is in use */
+	for (i = 0; i < pages; ++i) {
+		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
+		if ((oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) ||
+				test_bit(entry + i, tbl->it_map)) {
+			WARN_ON(test_bit(entry + i, tbl->it_map));
+			spin_unlock(&(pool->lock));
+			return -EBUSY;
+		}
+	}
+
+	/* Put tces to the table */
+	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
+		ret = put_tce(tbl, entry + i, tce, direction);
+		if (ret == 1)
+			++retpages;
+	}
+
+	/*
+	 * If failed, release locked pages, otherwise return the number
+	 * of locked system pages
+	 */
+	if (ret < 0)
+		clear_tces_nolock(tbl, entry, i);
+	else
+		ret = retpages;
+
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..21250ef 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -20,6 +20,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
@@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev_name(dev),
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev_name(dev));
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev_name(dev), iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
+				dev_name(dev), ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			printk(KERN_INFO "tce_vfio: cannot create "
+					"new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
-- 
1.7.10.4

Re: [PATCH 1/2] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Mon, 2012-12-03 at 13:52 +1100, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
> 
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
> 
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |    9 ++
>  arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++
>  drivers/iommu/Kconfig                |    8 ++
>  4 files changed, 338 insertions(+)
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..5c7087a 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -76,6 +76,9 @@ struct iommu_table {
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
> @@ -147,5 +150,11 @@ static inline void iommu_restore(void)
>  }
>  #endif
>  
> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages);
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..2738aa4 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -44,6 +44,7 @@
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
> @@ -856,3 +857,188 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +
> +/*
> + * Returns the number of used IOMMU pages (4K) within
> + * the same system page (4K or 64K).
> + * bitmap_weight is not used as it does not support bigendian maps.
> + */
> +static int syspage_weight(unsigned long *map, unsigned long entry)
> +{
> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
> +
> +	/* Aligns TCE entry number to system page boundary */
> +	entry &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> +
> +	/* Count used 4K pages */
> +	while (nbits--)
> +		ret += (test_bit(entry++, map) == 0) ? 0 : 1;
> +
> +	return ret;
> +}
> +
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of system pages
> + * which it called put_page() on
> + */
> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int i, retpages = 0;
> +	unsigned long oldtce, oldweight;
> +	struct page *page;
> +
> +	for (i = 0; i < pages; ++i) {
> +		oldtce = ppc_md.tce_get(tbl, entry + i);
> +		ppc_md.tce_free(tbl, entry + i, 1);
> +
> +		oldweight = syspage_weight(tbl->it_map, entry);
> +		__clear_bit(entry - tbl->it_offset, tbl->it_map);
> +
> +		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> +			continue;

Could this happen earlier, above syspage_weight() and __clear_bit()?

> +
> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +		WARN_ON(!page);
> +		if (!page)
> +			continue;
> +
> +		if (oldtce & TCE_PCI_WRITE)
> +			SetPageDirty(page);
> +
> +		put_page(page);
> +
> +		/* That was the last IOMMU page within the system page */
> +		if ((oldweight == 1) && !syspage_weight(tbl->it_map, entry))
> +			++retpages;

If you used __test_and_clear_bit() above I think you could avoid this
2nd call to syspage_weight.  A minor optimization though.

> +	}
> +
> +	return retpages;
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number
> + / of released system pages
> + */

Something bad happened to your comments here.

> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int ret;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	spin_lock(&(pool->lock));
> +	ret = clear_tces_nolock(tbl, entry, pages);
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset, oldweight;
> +
> +	/* Map new TCE */
> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);

Maybe the compiler will figure this out, but isn't this the same as tce
& (IOMMU_PAGE_MASK & PAGE_MASK)?

> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);
> +	if (ret < 1) {

Probably (ret != 1) here or else we never get to your >1 case below.

> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);

Use pr_err

> +		if (!ret || (ret > 1))

Then (ret >= 0) here.  Or return (ret >= 0) ? -EFAULT : ret

> +			ret = -EFAULT;
> +		return ret;
> +	}

You're missing the code from x86 that handles mapping mmap'd ranges.
This is intended to allow peer-to-peer DMA between devices.  Is that
intentional?

> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	entry += tbl->it_offset; /* Offset into real TCE table */

Here's what makes me call the entry "relative" rather than zero-based.
The iova is relative to the start of dma32_window_start, ie. if the
window starts at bus address 512MB and I want to create a translation at
bus address 512MB, I pass in an iova of 0, right?  The above adds the
window offset.  So you've removed dma64 window, but we really need to
define iova better.

> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);

Use pr_err

> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	/* Calculate if new system page has been locked */
> +	oldweight = syspage_weight(tbl->it_map, entry);
> +	__set_bit(entry - tbl->it_offset, tbl->it_map);
> +
> +	return (oldweight == 0) ? 1 : 0;
> +}
> +
> +/*
> + * iommu_put_tces builds tces and returned the number of actually
> + * locked system pages
> + */
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages)
> +{
> +	int i, ret = 0, retpages = 0;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +	BUG_ON(direction == DMA_NONE);

This doesn't seem BUG worthy, -EINVAL?  We can't assume tce_iommu_ioctl
will always be the only caller of this function.

> +
> +	spin_lock(&(pool->lock));
> +
> +	/* Check if any is in use */
> +	for (i = 0; i < pages; ++i) {
> +		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
> +		if ((oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) ||
> +				test_bit(entry + i, tbl->it_map)) {
> +			WARN_ON(test_bit(entry + i, tbl->it_map));

The WARN_ON seems to confirm that these are redundant tests, does that
imply we don't trust it_map?  It would be a lot faster if we could rely
on it_map exclusively here.

> +			spin_unlock(&(pool->lock));
> +			return -EBUSY;
> +		}
> +	}
> +
> +	/* Put tces to the table */
> +	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
> +		ret = put_tce(tbl, entry + i, tce, direction);
> +		if (ret == 1)
> +			++retpages;
> +	}
> +
> +	/*
> +	 * If failed, release locked pages, otherwise return the number
> +	 * of locked system pages
> +	 */
> +	if (ret < 0)
> +		clear_tces_nolock(tbl, entry, i);
> +	else
> +		ret = retpages;
> +
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..21250ef 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -20,6 +20,7 @@
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
> @@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev_name(dev),
> +				iommu_group_id(dev->iommu_group));

Use pr_warn

> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev_name(dev));
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev_name(dev), iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> +				dev_name(dev), ret);

Use pr_err

> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			printk(KERN_INFO "tce_vfio: cannot create "
> +					"new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));

Use pr_info

> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);
> +	}
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

Thanks,

Alex

Re: [PATCH 1/2] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

On 04/12/12 04:35, Alex Williamson wrote:
> On Mon, 2012-12-03 at 13:52 +1100, Alexey Kardashevskiy wrote:
>> This patch initializes IOMMU groups based on the IOMMU
>> configuration discovered during the PCI scan on POWERNV
>> (POWER non virtualized) platform. The IOMMU groups are
>> to be used later by VFIO driver (PCI pass through).
>>
>> It also implements an API for mapping/unmapping pages for
>> guest PCI drivers and providing DMA window properties.
>> This API is going to be used later by QEMU-VFIO to handle
>> h_put_tce hypercalls from the KVM guest.
>>
>> Although this driver has been tested only on the POWERNV
>> platform, it should work on any platform which supports
>> TCE tables.
>>
>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
>> option and configure VFIO as required.
>>
>> Cc: David Gibson <david@gibson.dropbear.id.au>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>> ---
>>   arch/powerpc/include/asm/iommu.h     |    9 ++
>>   arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++++++++++++++++++
>>   arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++
>>   drivers/iommu/Kconfig                |    8 ++
>>   4 files changed, 338 insertions(+)
>>
>> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
>> index cbfe678..5c7087a 100644
>> --- a/arch/powerpc/include/asm/iommu.h
>> +++ b/arch/powerpc/include/asm/iommu.h
>> @@ -76,6 +76,9 @@ struct iommu_table {
>>   	struct iommu_pool large_pool;
>>   	struct iommu_pool pools[IOMMU_NR_POOLS];
>>   	unsigned long *it_map;       /* A simple allocation bitmap for now */
>> +#ifdef CONFIG_IOMMU_API
>> +	struct iommu_group *it_group;
>> +#endif
>>   };
>>
>>   struct scatterlist;
>> @@ -147,5 +150,11 @@ static inline void iommu_restore(void)
>>   }
>>   #endif
>>
>> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
>> +		unsigned long pages);
>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
>> +		uint64_t tce, enum dma_data_direction direction,
>> +		unsigned long pages);
>> +
>>   #endif /* __KERNEL__ */
>>   #endif /* _ASM_IOMMU_H */
>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
>> index ff5a6ce..2738aa4 100644
>> --- a/arch/powerpc/kernel/iommu.c
>> +++ b/arch/powerpc/kernel/iommu.c
>> @@ -44,6 +44,7 @@
>>   #include <asm/kdump.h>
>>   #include <asm/fadump.h>
>>   #include <asm/vio.h>
>> +#include <asm/tce.h>
>>
>>   #define DBG(...)
>>
>> @@ -856,3 +857,188 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>>   		free_pages((unsigned long)vaddr, get_order(size));
>>   	}
>>   }
>> +
>> +#ifdef CONFIG_IOMMU_API
>> +/*
>> + * SPAPR TCE API
>> + */
>> +
>> +/*
>> + * Returns the number of used IOMMU pages (4K) within
>> + * the same system page (4K or 64K).
>> + * bitmap_weight is not used as it does not support bigendian maps.
>> + */
>> +static int syspage_weight(unsigned long *map, unsigned long entry)
>> +{
>> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
>> +
>> +	/* Aligns TCE entry number to system page boundary */
>> +	entry &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
>> +
>> +	/* Count used 4K pages */
>> +	while (nbits--)
>> +		ret += (test_bit(entry++, map) == 0) ? 0 : 1;
>> +
>> +	return ret;
>> +}
>> +
>> +static void tce_flush(struct iommu_table *tbl)
>> +{
>> +	/* Flush/invalidate TLB caches if necessary */
>> +	if (ppc_md.tce_flush)
>> +		ppc_md.tce_flush(tbl);
>> +
>> +	/* Make sure updates are seen by hardware */
>> +	mb();
>> +}
>> +
>> +/*
>> + * iommu_clear_tces clears tces and returned the number of system pages
>> + * which it called put_page() on
>> + */
>> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
>> +		unsigned long pages)
>> +{
>> +	int i, retpages = 0;
>> +	unsigned long oldtce, oldweight;
>> +	struct page *page;
>> +
>> +	for (i = 0; i < pages; ++i) {
>> +		oldtce = ppc_md.tce_get(tbl, entry + i);
>> +		ppc_md.tce_free(tbl, entry + i, 1);
>> +
>> +		oldweight = syspage_weight(tbl->it_map, entry);
>> +		__clear_bit(entry - tbl->it_offset, tbl->it_map);
>> +
>> +		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
>> +			continue;
>
> Could this happen earlier, above syspage_weight() and __clear_bit()?


Want to clear it anyway if it is not cleared by some reason. Added WARN_ON.


>> +
>> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
>> +
>> +		WARN_ON(!page);
>> +		if (!page)
>> +			continue;
>> +
>> +		if (oldtce & TCE_PCI_WRITE)
>> +			SetPageDirty(page);
>> +
>> +		put_page(page);
>> +
>> +		/* That was the last IOMMU page within the system page */
>> +		if ((oldweight == 1) && !syspage_weight(tbl->it_map, entry))
>> +			++retpages;
>
> If you used __test_and_clear_bit() above I think you could avoid this
> 2nd call to syspage_weight.  A minor optimization though.
>
>> +	}
>> +
>> +	return retpages;
>> +}
>> +
>> +/*
>> + * iommu_clear_tces clears tces and returned the number
>> + / of released system pages
>> + */
>
> Something bad happened to your comments here.
>
>> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
>> +		unsigned long pages)
>> +{
>> +	int ret;
>> +	struct iommu_pool *pool = get_pool(tbl, entry);
>> +
>> +	spin_lock(&(pool->lock));
>> +	ret = clear_tces_nolock(tbl, entry, pages);
>> +	tce_flush(tbl);
>> +	spin_unlock(&(pool->lock));
>> +
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
>> +
>> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
>> +		uint64_t tce, enum dma_data_direction direction)
>> +{
>> +	int ret;
>> +	struct page *page = NULL;
>> +	unsigned long kva, offset, oldweight;
>> +
>> +	/* Map new TCE */
>> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
>
> Maybe the compiler will figure this out, but isn't this the same as tce
> & (IOMMU_PAGE_MASK & PAGE_MASK)?


it is rather (tce & (IOMMU_PAGE_MASK & ~PAGE_MASK)) but I cannot see how it 
is simpler and I doubt that it is faster enough to notice it anyhow :)


>> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
>> +			direction != DMA_TO_DEVICE, &page);
>> +	if (ret < 1) {
>
> Probably (ret != 1) here or else we never get to your >1 case below.
>
>> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
>> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
>
> Use pr_err
 >
>> +		if (!ret || (ret > 1))
>
> Then (ret >= 0) here.  Or return (ret >= 0) ? -EFAULT : ret
>
>> +			ret = -EFAULT;
>> +		return ret;
>> +	}
>
> You're missing the code from x86 that handles mapping mmap'd ranges.
> This is intended to allow peer-to-peer DMA between devices.  Is that
> intentional?

I am not following you here. What code exactly are talking about? We do not 
track ranges at all and I do not see how it helps with p2p dma.


>> +
>> +	kva = (unsigned long) page_address(page);
>> +	kva += offset;
>> +
>> +	/* tce_build receives a virtual address */
>> +	entry += tbl->it_offset; /* Offset into real TCE table */
>
> Here's what makes me call the entry "relative" rather than zero-based.

This is the bug actually, I overlooked it and I removed it now. Thanks for 
being so picky :)


> The iova is relative to the start of dma32_window_start, ie. if the
> window starts at bus address 512MB and I want to create a translation at
> bus address 512MB, I pass in an iova of 0, right?  The above adds the
> window offset.  So you've removed dma64 window, but we really need to
> define iova better.




>> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
>> +
>> +	/* tce_build() only returns non-zero for transient errors */
>> +	if (unlikely(ret)) {
>> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
>> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
>
> Use pr_err
>
>> +		put_page(page);
>> +		return -EIO;
>> +	}
>> +
>> +	/* Calculate if new system page has been locked */
>> +	oldweight = syspage_weight(tbl->it_map, entry);
>> +	__set_bit(entry - tbl->it_offset, tbl->it_map);
>> +
>> +	return (oldweight == 0) ? 1 : 0;
>> +}
>> +
>> +/*
>> + * iommu_put_tces builds tces and returned the number of actually
>> + * locked system pages
>> + */
>> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
>> +		uint64_t tce, enum dma_data_direction direction,
>> +		unsigned long pages)
>> +{
>> +	int i, ret = 0, retpages = 0;
>> +	struct iommu_pool *pool = get_pool(tbl, entry);
>> +
>> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
>> +	BUG_ON(direction == DMA_NONE);
>
> This doesn't seem BUG worthy, -EINVAL?  We can't assume tce_iommu_ioctl
> will always be the only caller of this function.


This is what other function does in this file.


>> +
>> +	spin_lock(&(pool->lock));
>> +
>> +	/* Check if any is in use */
>> +	for (i = 0; i < pages; ++i) {
>> +		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
>> +		if ((oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) ||
>> +				test_bit(entry + i, tbl->it_map)) {
>> +			WARN_ON(test_bit(entry + i, tbl->it_map));
>
> The WARN_ON seems to confirm that these are redundant tests, does that
> imply we don't trust it_map?  It would be a lot faster if we could rely
> on it_map exclusively here.


As for me, pretty minor optimization. I'm testing it now to see if I do not 
miss bits.



>> +			spin_unlock(&(pool->lock));
>> +			return -EBUSY;
>> +		}
>> +	}
>> +
>> +	/* Put tces to the table */
>> +	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
>> +		ret = put_tce(tbl, entry + i, tce, direction);
>> +		if (ret == 1)
>> +			++retpages;
>> +	}
>> +
>> +	/*
>> +	 * If failed, release locked pages, otherwise return the number
>> +	 * of locked system pages
>> +	 */
>> +	if (ret < 0)
>> +		clear_tces_nolock(tbl, entry, i);
>> +	else
>> +		ret = retpages;
>> +
>> +	tce_flush(tbl);
>> +	spin_unlock(&(pool->lock));
>> +
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_put_tces);
>> +#endif /* CONFIG_IOMMU_API */
>> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
>> index 05205cf..21250ef 100644
>> --- a/arch/powerpc/platforms/powernv/pci.c
>> +++ b/arch/powerpc/platforms/powernv/pci.c
>> @@ -20,6 +20,7 @@
>>   #include <linux/irq.h>
>>   #include <linux/io.h>
>>   #include <linux/msi.h>
>> +#include <linux/iommu.h>
>>
>>   #include <asm/sections.h>
>>   #include <asm/io.h>
>> @@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
>>   	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>>   #endif
>>   }
>> +
>> +#ifdef CONFIG_IOMMU_API
>> +/*
>> + * IOMMU groups support required by VFIO
>> + */
>> +static int add_device(struct device *dev)
>> +{
>> +	struct iommu_table *tbl;
>> +	int ret = 0;
>> +
>> +	if (WARN_ON(dev->iommu_group)) {
>> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
>> +				dev_name(dev),
>> +				iommu_group_id(dev->iommu_group));
>
> Use pr_warn
>
>> +		return -EBUSY;
>> +	}
>> +
>> +	tbl = get_iommu_table_base(dev);
>> +	if (!tbl) {
>> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
>> +				dev_name(dev));
>> +		return 0;
>> +	}
>> +
>> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
>> +			dev_name(dev), iommu_group_id(tbl->it_group));
>> +
>> +	ret = iommu_group_add_device(tbl->it_group, dev);
>> +	if (ret < 0)
>> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
>> +				dev_name(dev), ret);
>
> Use pr_err
>
>> +
>> +	return ret;
>> +}
>> +
>> +static void del_device(struct device *dev)
>> +{
>> +	iommu_group_remove_device(dev);
>> +}
>> +
>> +static int iommu_bus_notifier(struct notifier_block *nb,
>> +			      unsigned long action, void *data)
>> +{
>> +	struct device *dev = data;
>> +
>> +	switch (action) {
>> +	case BUS_NOTIFY_ADD_DEVICE:
>> +		return add_device(dev);
>> +	case BUS_NOTIFY_DEL_DEVICE:
>> +		del_device(dev);
>> +		return 0;
>> +	default:
>> +		return 0;
>> +	}
>> +}
>> +
>> +static struct notifier_block tce_iommu_bus_nb = {
>> +	.notifier_call = iommu_bus_notifier,
>> +};
>> +
>> +static void group_release(void *iommu_data)
>> +{
>> +	struct iommu_table *tbl = iommu_data;
>> +	tbl->it_group = NULL;
>> +}
>> +
>> +static int __init tce_iommu_init(void)
>> +{
>> +	struct pci_dev *pdev = NULL;
>> +	struct iommu_table *tbl;
>> +	struct iommu_group *grp;
>> +
>> +	/* Allocate and initialize IOMMU groups */
>> +	for_each_pci_dev(pdev) {
>> +		tbl = get_iommu_table_base(&pdev->dev);
>> +		if (!tbl)
>> +			continue;
>> +
>> +		/* Skip already initialized */
>> +		if (tbl->it_group)
>> +			continue;
>> +
>> +		grp = iommu_group_alloc();
>> +		if (IS_ERR(grp)) {
>> +			printk(KERN_INFO "tce_vfio: cannot create "
>> +					"new IOMMU group, ret=%ld\n",
>> +					PTR_ERR(grp));
>
> Use pr_info
>
>> +			return PTR_ERR(grp);
>> +		}
>> +		tbl->it_group = grp;
>> +		iommu_group_set_iommudata(grp, tbl, group_release);
>> +	}
>> +
>> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
>> +
>> +	/* Add PCI devices to VFIO groups */
>> +	for_each_pci_dev(pdev)
>> +		add_device(&pdev->dev);
>> +
>> +	return 0;
>> +}
>> +
>> +static void __exit tce_iommu_cleanup(void)
>> +{
>> +	struct pci_dev *pdev = NULL;
>> +	struct iommu_table *tbl;
>> +	struct iommu_group *grp = NULL;
>> +
>> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
>> +
>> +	/* Delete PCI devices from VFIO groups */
>> +	for_each_pci_dev(pdev)
>> +		del_device(&pdev->dev);
>> +
>> +	/* Release VFIO groups */
>> +	for_each_pci_dev(pdev) {
>> +		tbl = get_iommu_table_base(&pdev->dev);
>> +		if (!tbl)
>> +			continue;
>> +		grp = tbl->it_group;
>> +
>> +		/* Skip (already) uninitialized */
>> +		if (!grp)
>> +			continue;
>> +
>> +		/* Do actual release, group_release() is expected to work */
>> +		iommu_group_put(grp);
>> +		BUG_ON(tbl->it_group);
>> +	}
>> +}
>> +
>> +module_init(tce_iommu_init);
>> +module_exit(tce_iommu_cleanup);
>> +#endif /* CONFIG_IOMMU_API */
>> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
>> index 9f69b56..29d11dc 100644
>> --- a/drivers/iommu/Kconfig
>> +++ b/drivers/iommu/Kconfig
>> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>>
>>   	  Say N unless you need kernel log message for IOMMU debugging
>>
>> +config SPAPR_TCE_IOMMU
>> +	bool "sPAPR TCE IOMMU Support"
>> +	depends on PPC_POWERNV
>> +	select IOMMU_API
>> +	help
>> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
>> +	  still not implemented.
>> +
>>   endif # IOMMU_SUPPORT
>
> Thanks,
>
> Alex
>


-- 
Alexey

Re: [PATCH 1/2] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Tue, 2012-12-04 at 19:12 +1100, Alexey Kardashevskiy wrote:
> On 04/12/12 04:35, Alex Williamson wrote:
> > On Mon, 2012-12-03 at 13:52 +1100, Alexey Kardashevskiy wrote:
> >> This patch initializes IOMMU groups based on the IOMMU
> >> configuration discovered during the PCI scan on POWERNV
> >> (POWER non virtualized) platform. The IOMMU groups are
> >> to be used later by VFIO driver (PCI pass through).
> >>
> >> It also implements an API for mapping/unmapping pages for
> >> guest PCI drivers and providing DMA window properties.
> >> This API is going to be used later by QEMU-VFIO to handle
> >> h_put_tce hypercalls from the KVM guest.
> >>
> >> Although this driver has been tested only on the POWERNV
> >> platform, it should work on any platform which supports
> >> TCE tables.
> >>
> >> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> >> option and configure VFIO as required.
> >>
> >> Cc: David Gibson <david@gibson.dropbear.id.au>
> >> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> >> ---
> >>   arch/powerpc/include/asm/iommu.h     |    9 ++
> >>   arch/powerpc/kernel/iommu.c          |  186 ++++++++++++++++++++++++++++++++++
> >>   arch/powerpc/platforms/powernv/pci.c |  135 ++++++++++++++++++++++++
> >>   drivers/iommu/Kconfig                |    8 ++
> >>   4 files changed, 338 insertions(+)
> >>
> >> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> >> index cbfe678..5c7087a 100644
> >> --- a/arch/powerpc/include/asm/iommu.h
> >> +++ b/arch/powerpc/include/asm/iommu.h
> >> @@ -76,6 +76,9 @@ struct iommu_table {
> >>   	struct iommu_pool large_pool;
> >>   	struct iommu_pool pools[IOMMU_NR_POOLS];
> >>   	unsigned long *it_map;       /* A simple allocation bitmap for now */
> >> +#ifdef CONFIG_IOMMU_API
> >> +	struct iommu_group *it_group;
> >> +#endif
> >>   };
> >>
> >>   struct scatterlist;
> >> @@ -147,5 +150,11 @@ static inline void iommu_restore(void)
> >>   }
> >>   #endif
> >>
> >> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> >> +		unsigned long pages);
> >> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> >> +		uint64_t tce, enum dma_data_direction direction,
> >> +		unsigned long pages);
> >> +
> >>   #endif /* __KERNEL__ */
> >>   #endif /* _ASM_IOMMU_H */
> >> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> >> index ff5a6ce..2738aa4 100644
> >> --- a/arch/powerpc/kernel/iommu.c
> >> +++ b/arch/powerpc/kernel/iommu.c
> >> @@ -44,6 +44,7 @@
> >>   #include <asm/kdump.h>
> >>   #include <asm/fadump.h>
> >>   #include <asm/vio.h>
> >> +#include <asm/tce.h>
> >>
> >>   #define DBG(...)
> >>
> >> @@ -856,3 +857,188 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
> >>   		free_pages((unsigned long)vaddr, get_order(size));
> >>   	}
> >>   }
> >> +
> >> +#ifdef CONFIG_IOMMU_API
> >> +/*
> >> + * SPAPR TCE API
> >> + */
> >> +
> >> +/*
> >> + * Returns the number of used IOMMU pages (4K) within
> >> + * the same system page (4K or 64K).
> >> + * bitmap_weight is not used as it does not support bigendian maps.
> >> + */
> >> +static int syspage_weight(unsigned long *map, unsigned long entry)
> >> +{
> >> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
> >> +
> >> +	/* Aligns TCE entry number to system page boundary */
> >> +	entry &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> >> +
> >> +	/* Count used 4K pages */
> >> +	while (nbits--)
> >> +		ret += (test_bit(entry++, map) == 0) ? 0 : 1;
> >> +
> >> +	return ret;
> >> +}
> >> +
> >> +static void tce_flush(struct iommu_table *tbl)
> >> +{
> >> +	/* Flush/invalidate TLB caches if necessary */
> >> +	if (ppc_md.tce_flush)
> >> +		ppc_md.tce_flush(tbl);
> >> +
> >> +	/* Make sure updates are seen by hardware */
> >> +	mb();
> >> +}
> >> +
> >> +/*
> >> + * iommu_clear_tces clears tces and returned the number of system pages
> >> + * which it called put_page() on
> >> + */
> >> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> >> +		unsigned long pages)
> >> +{
> >> +	int i, retpages = 0;
> >> +	unsigned long oldtce, oldweight;
> >> +	struct page *page;
> >> +
> >> +	for (i = 0; i < pages; ++i) {
> >> +		oldtce = ppc_md.tce_get(tbl, entry + i);
> >> +		ppc_md.tce_free(tbl, entry + i, 1);
> >> +
> >> +		oldweight = syspage_weight(tbl->it_map, entry);
> >> +		__clear_bit(entry - tbl->it_offset, tbl->it_map);
> >> +
> >> +		if (!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)))
> >> +			continue;
> >
> > Could this happen earlier, above syspage_weight() and __clear_bit()?
> 
> 
> Want to clear it anyway if it is not cleared by some reason. Added WARN_ON.

The map shouldn't be set unless read/write is enabled, right?  It seems
like we don't have a lot of trust in this bitmap.

> >> +
> >> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> >> +
> >> +		WARN_ON(!page);
> >> +		if (!page)
> >> +			continue;
> >> +
> >> +		if (oldtce & TCE_PCI_WRITE)
> >> +			SetPageDirty(page);
> >> +
> >> +		put_page(page);
> >> +
> >> +		/* That was the last IOMMU page within the system page */
> >> +		if ((oldweight == 1) && !syspage_weight(tbl->it_map, entry))
> >> +			++retpages;
> >
> > If you used __test_and_clear_bit() above I think you could avoid this
> > 2nd call to syspage_weight.  A minor optimization though.
> >
> >> +	}
> >> +
> >> +	return retpages;
> >> +}
> >> +
> >> +/*
> >> + * iommu_clear_tces clears tces and returned the number
> >> + / of released system pages
> >> + */
> >
> > Something bad happened to your comments here.
> >
> >> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> >> +		unsigned long pages)
> >> +{
> >> +	int ret;
> >> +	struct iommu_pool *pool = get_pool(tbl, entry);
> >> +
> >> +	spin_lock(&(pool->lock));
> >> +	ret = clear_tces_nolock(tbl, entry, pages);
> >> +	tce_flush(tbl);
> >> +	spin_unlock(&(pool->lock));
> >> +
> >> +	return ret;
> >> +}
> >> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> >> +
> >> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> >> +		uint64_t tce, enum dma_data_direction direction)
> >> +{
> >> +	int ret;
> >> +	struct page *page = NULL;
> >> +	unsigned long kva, offset, oldweight;
> >> +
> >> +	/* Map new TCE */
> >> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> >
> > Maybe the compiler will figure this out, but isn't this the same as tce
> > & (IOMMU_PAGE_MASK & PAGE_MASK)?
> 
> 
> it is rather (tce & (IOMMU_PAGE_MASK & ~PAGE_MASK)) but I cannot see how it 
> is simpler and I doubt that it is faster enough to notice it anyhow :)

Yes, ~PAGE_MASK.  IMHO, it's more intuitive.

> >> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> >> +			direction != DMA_TO_DEVICE, &page);
> >> +	if (ret < 1) {
> >
> > Probably (ret != 1) here or else we never get to your >1 case below.
> >
> >> +		printk(KERN_ERR "tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> >> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> >
> > Use pr_err
>  >
> >> +		if (!ret || (ret > 1))
> >
> > Then (ret >= 0) here.  Or return (ret >= 0) ? -EFAULT : ret
> >
> >> +			ret = -EFAULT;
> >> +		return ret;
> >> +	}
> >
> > You're missing the code from x86 that handles mapping mmap'd ranges.
> > This is intended to allow peer-to-peer DMA between devices.  Is that
> > intentional?
> 
> I am not following you here. What code exactly are talking about? We do not 
> track ranges at all and I do not see how it helps with p2p dma.

The code in type1 that checks PFNMAP and reserved pages that I only
sometimes remember how it works ;)  The idea there is to allow p2p dma
by inserting iommu translations for non-page backed memory, ie. the
mmap'd BARs of other devices.  It may be that the POWER topology is not
amenable to this since you have a whole PCI bus in your group and
intra-group p2p isn't iommu translated.  I'm not sure how useful it is
even on x86, but KVM device assignment does it, so I added it to type1.

> >> +
> >> +	kva = (unsigned long) page_address(page);
> >> +	kva += offset;
> >> +
> >> +	/* tce_build receives a virtual address */
> >> +	entry += tbl->it_offset; /* Offset into real TCE table */
> >
> > Here's what makes me call the entry "relative" rather than zero-based.
> 
> This is the bug actually, I overlooked it and I removed it now. Thanks for 
> being so picky :)

Ah, ok.  I'll look for it on the next round and make sure I understand
it.

> > The iova is relative to the start of dma32_window_start, ie. if the
> > window starts at bus address 512MB and I want to create a translation at
> > bus address 512MB, I pass in an iova of 0, right?  The above adds the
> > window offset.  So you've removed dma64 window, but we really need to
> > define iova better.
> 
> 
> 
> 
> >> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> >> +
> >> +	/* tce_build() only returns non-zero for transient errors */
> >> +	if (unlikely(ret)) {
> >> +		printk(KERN_ERR "tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> >> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> >
> > Use pr_err
> >
> >> +		put_page(page);
> >> +		return -EIO;
> >> +	}
> >> +
> >> +	/* Calculate if new system page has been locked */
> >> +	oldweight = syspage_weight(tbl->it_map, entry);
> >> +	__set_bit(entry - tbl->it_offset, tbl->it_map);
> >> +
> >> +	return (oldweight == 0) ? 1 : 0;
> >> +}
> >> +
> >> +/*
> >> + * iommu_put_tces builds tces and returned the number of actually
> >> + * locked system pages
> >> + */
> >> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> >> +		uint64_t tce, enum dma_data_direction direction,
> >> +		unsigned long pages)
> >> +{
> >> +	int i, ret = 0, retpages = 0;
> >> +	struct iommu_pool *pool = get_pool(tbl, entry);
> >> +
> >> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> >> +	BUG_ON(direction == DMA_NONE);
> >
> > This doesn't seem BUG worthy, -EINVAL?  We can't assume tce_iommu_ioctl
> > will always be the only caller of this function.
> 
> 
> This is what other function does in this file.

Blech, ok.

> >> +
> >> +	spin_lock(&(pool->lock));
> >> +
> >> +	/* Check if any is in use */
> >> +	for (i = 0; i < pages; ++i) {
> >> +		unsigned long oldtce = ppc_md.tce_get(tbl, entry + i);
> >> +		if ((oldtce & (TCE_PCI_WRITE | TCE_PCI_READ)) ||
> >> +				test_bit(entry + i, tbl->it_map)) {
> >> +			WARN_ON(test_bit(entry + i, tbl->it_map));
> >
> > The WARN_ON seems to confirm that these are redundant tests, does that
> > imply we don't trust it_map?  It would be a lot faster if we could rely
> > on it_map exclusively here.
> 
> 
> As for me, pretty minor optimization. I'm testing it now to see if I do not 
> miss bits.

It would be a lot more re-assuring if we didn't need it ;)  Thanks,

Alex

> >> +			spin_unlock(&(pool->lock));
> >> +			return -EBUSY;
> >> +		}
> >> +	}
> >> +
> >> +	/* Put tces to the table */
> >> +	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
> >> +		ret = put_tce(tbl, entry + i, tce, direction);
> >> +		if (ret == 1)
> >> +			++retpages;
> >> +	}
> >> +
> >> +	/*
> >> +	 * If failed, release locked pages, otherwise return the number
> >> +	 * of locked system pages
> >> +	 */
> >> +	if (ret < 0)
> >> +		clear_tces_nolock(tbl, entry, i);
> >> +	else
> >> +		ret = retpages;
> >> +
> >> +	tce_flush(tbl);
> >> +	spin_unlock(&(pool->lock));
> >> +
> >> +	return ret;
> >> +}
> >> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> >> +#endif /* CONFIG_IOMMU_API */
> >> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> >> index 05205cf..21250ef 100644
> >> --- a/arch/powerpc/platforms/powernv/pci.c
> >> +++ b/arch/powerpc/platforms/powernv/pci.c
> >> @@ -20,6 +20,7 @@
> >>   #include <linux/irq.h>
> >>   #include <linux/io.h>
> >>   #include <linux/msi.h>
> >> +#include <linux/iommu.h>
> >>
> >>   #include <asm/sections.h>
> >>   #include <asm/io.h>
> >> @@ -613,3 +614,137 @@ void __init pnv_pci_init(void)
> >>   	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
> >>   #endif
> >>   }
> >> +
> >> +#ifdef CONFIG_IOMMU_API
> >> +/*
> >> + * IOMMU groups support required by VFIO
> >> + */
> >> +static int add_device(struct device *dev)
> >> +{
> >> +	struct iommu_table *tbl;
> >> +	int ret = 0;
> >> +
> >> +	if (WARN_ON(dev->iommu_group)) {
> >> +		printk(KERN_WARNING "tce_vfio: device %s is already in iommu group %d, skipping\n",
> >> +				dev_name(dev),
> >> +				iommu_group_id(dev->iommu_group));
> >
> > Use pr_warn
> >
> >> +		return -EBUSY;
> >> +	}
> >> +
> >> +	tbl = get_iommu_table_base(dev);
> >> +	if (!tbl) {
> >> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> >> +				dev_name(dev));
> >> +		return 0;
> >> +	}
> >> +
> >> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> >> +			dev_name(dev), iommu_group_id(tbl->it_group));
> >> +
> >> +	ret = iommu_group_add_device(tbl->it_group, dev);
> >> +	if (ret < 0)
> >> +		printk(KERN_ERR "tce_vfio: %s has not been added, ret=%d\n",
> >> +				dev_name(dev), ret);
> >
> > Use pr_err
> >
> >> +
> >> +	return ret;
> >> +}
> >> +
> >> +static void del_device(struct device *dev)
> >> +{
> >> +	iommu_group_remove_device(dev);
> >> +}
> >> +
> >> +static int iommu_bus_notifier(struct notifier_block *nb,
> >> +			      unsigned long action, void *data)
> >> +{
> >> +	struct device *dev = data;
> >> +
> >> +	switch (action) {
> >> +	case BUS_NOTIFY_ADD_DEVICE:
> >> +		return add_device(dev);
> >> +	case BUS_NOTIFY_DEL_DEVICE:
> >> +		del_device(dev);
> >> +		return 0;
> >> +	default:
> >> +		return 0;
> >> +	}
> >> +}
> >> +
> >> +static struct notifier_block tce_iommu_bus_nb = {
> >> +	.notifier_call = iommu_bus_notifier,
> >> +};
> >> +
> >> +static void group_release(void *iommu_data)
> >> +{
> >> +	struct iommu_table *tbl = iommu_data;
> >> +	tbl->it_group = NULL;
> >> +}
> >> +
> >> +static int __init tce_iommu_init(void)
> >> +{
> >> +	struct pci_dev *pdev = NULL;
> >> +	struct iommu_table *tbl;
> >> +	struct iommu_group *grp;
> >> +
> >> +	/* Allocate and initialize IOMMU groups */
> >> +	for_each_pci_dev(pdev) {
> >> +		tbl = get_iommu_table_base(&pdev->dev);
> >> +		if (!tbl)
> >> +			continue;
> >> +
> >> +		/* Skip already initialized */
> >> +		if (tbl->it_group)
> >> +			continue;
> >> +
> >> +		grp = iommu_group_alloc();
> >> +		if (IS_ERR(grp)) {
> >> +			printk(KERN_INFO "tce_vfio: cannot create "
> >> +					"new IOMMU group, ret=%ld\n",
> >> +					PTR_ERR(grp));
> >
> > Use pr_info
> >
> >> +			return PTR_ERR(grp);
> >> +		}
> >> +		tbl->it_group = grp;
> >> +		iommu_group_set_iommudata(grp, tbl, group_release);
> >> +	}
> >> +
> >> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> >> +
> >> +	/* Add PCI devices to VFIO groups */
> >> +	for_each_pci_dev(pdev)
> >> +		add_device(&pdev->dev);
> >> +
> >> +	return 0;
> >> +}
> >> +
> >> +static void __exit tce_iommu_cleanup(void)
> >> +{
> >> +	struct pci_dev *pdev = NULL;
> >> +	struct iommu_table *tbl;
> >> +	struct iommu_group *grp = NULL;
> >> +
> >> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> >> +
> >> +	/* Delete PCI devices from VFIO groups */
> >> +	for_each_pci_dev(pdev)
> >> +		del_device(&pdev->dev);
> >> +
> >> +	/* Release VFIO groups */
> >> +	for_each_pci_dev(pdev) {
> >> +		tbl = get_iommu_table_base(&pdev->dev);
> >> +		if (!tbl)
> >> +			continue;
> >> +		grp = tbl->it_group;
> >> +
> >> +		/* Skip (already) uninitialized */
> >> +		if (!grp)
> >> +			continue;
> >> +
> >> +		/* Do actual release, group_release() is expected to work */
> >> +		iommu_group_put(grp);
> >> +		BUG_ON(tbl->it_group);
> >> +	}
> >> +}
> >> +
> >> +module_init(tce_iommu_init);
> >> +module_exit(tce_iommu_cleanup);
> >> +#endif /* CONFIG_IOMMU_API */
> >> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> >> index 9f69b56..29d11dc 100644
> >> --- a/drivers/iommu/Kconfig
> >> +++ b/drivers/iommu/Kconfig
> >> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
> >>
> >>   	  Say N unless you need kernel log message for IOMMU debugging
> >>
> >> +config SPAPR_TCE_IOMMU
> >> +	bool "sPAPR TCE IOMMU Support"
> >> +	depends on PPC_POWERNV
> >> +	select IOMMU_API
> >> +	help
> >> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> >> +	  still not implemented.
> >> +
> >>   endif # IOMMU_SUPPORT
> >
> > Thanks,
> >
> > Alex
> >
> 
> 

[PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan on POWERNV
(POWER non virtualized) platform. The IOMMU groups are
to be used later by VFIO driver (PCI pass through).

It also implements an API for mapping/unmapping pages for
guest PCI drivers and providing DMA window properties.
This API is going to be used later by QEMU-VFIO to handle
h_put_tce hypercalls from the KVM guest.

Although this driver has been tested only on the POWERNV
platform, it should work on any platform which supports
TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option and configure VFIO as required.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |   10 ++
 arch/powerpc/kernel/iommu.c          |  214 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  134 +++++++++++++++++++++
 drivers/iommu/Kconfig                |    8 ++
 4 files changed, 366 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..be3b11b 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -76,6 +76,9 @@ struct iommu_table {
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
@@ -147,5 +150,12 @@ static inline void iommu_restore(void)
 }
 #endif
 
+extern void iommu_reset_table(struct iommu_table *tbl, bool release);
+extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages);
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..123431a 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -44,6 +44,7 @@
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
@@ -856,3 +857,216 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+
+/*
+ * iommu_reset_table is called when it started/stopped being used
+ */
+void iommu_reset_table(struct iommu_table *tbl, bool release)
+{
+	/*
+	 * Page at 0 is marked as used in iommu_init_table,
+	 * so here we clear it when called with release=false...
+	 */
+	if (!release && (tbl->it_offset == 0))
+		clear_bit(0, tbl->it_map);
+
+	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
+
+	memset(tbl->it_map, 0, (tbl->it_size + 7) >> 3);
+
+	/*
+	 * ... or restore when release=true
+	 */
+	if (release && (tbl->it_offset == 0))
+		set_bit(0, tbl->it_map);
+}
+EXPORT_SYMBOL_GPL(iommu_reset_table);
+
+/*
+ * Returns the number of used IOMMU pages (4K) within
+ * the same system page (4K or 64K).
+ * bitmap_weight is not used as it does not support bigendian maps.
+ * "offset" is an IOMMU page number relative to DMA window start.
+ */
+static int syspage_weight(unsigned long *map, unsigned long offset)
+{
+	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
+
+	/* Aligns TCE entry number to system page boundary */
+	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
+
+	/* Count used 4K pages */
+	while (nbits) {
+		if (test_bit(offset, map))
+			++ret;
+		--nbits;
+		++offset;
+	}
+
+	return ret;
+}
+
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of system pages
+ * which it called put_page() on
+ */
+static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int i, retpages = 0, clr;
+	unsigned long oldtce, oldweight;
+	struct page *page;
+
+	for (i = 0; i < pages; ++i) {
+		if (!test_bit(entry + i - tbl->it_offset, tbl->it_map))
+			continue;
+
+		oldtce = ppc_md.tce_get(tbl, entry + i);
+		ppc_md.tce_free(tbl, entry + i, 1);
+
+		oldweight = syspage_weight(tbl->it_map,
+				entry + i - tbl->it_offset);
+		clr = __test_and_clear_bit(entry + i - tbl->it_offset,
+				tbl->it_map);
+
+		if (WARN_ON(!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ))))
+			continue;
+
+		page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+		if (WARN_ON(!page))
+			continue;
+
+		if (oldtce & TCE_PCI_WRITE)
+			SetPageDirty(page);
+
+		put_page(page);
+
+		/* That was the last IOMMU page within the system page */
+		if ((oldweight == 1) && clr)
+			++retpages;
+	}
+
+	return retpages;
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number
+ * of released system pages
+ */
+long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int ret;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	spin_lock(&(pool->lock));
+	ret = clear_tces_nolock(tbl, entry, pages);
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_clear_tces);
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset, oldweight;
+
+	/* Map new TCE */
+	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret != 1) {
+		pr_err("tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		return -EFAULT;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		pr_err("tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	/* Calculate if new system page has been locked */
+	oldweight = syspage_weight(tbl->it_map, entry - tbl->it_offset);
+	__set_bit(entry - tbl->it_offset, tbl->it_map);
+
+	return (oldweight == 0) ? 1 : 0;
+}
+
+/*
+ * iommu_put_tces builds tces and returned the number of actually
+ * locked system pages
+ */
+long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long pages)
+{
+	int i, ret = 0, retpages = 0;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+	BUG_ON(direction == DMA_NONE);
+
+	spin_lock(&(pool->lock));
+
+	/* Check if any is in use */
+	for (i = 0; i < pages; ++i) {
+		if (test_bit(entry + i - tbl->it_offset, tbl->it_map)) {
+			spin_unlock(&(pool->lock));
+			return -EBUSY;
+		}
+	}
+
+	/* Put tces to the table */
+	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
+		ret = put_tce(tbl, entry + i, tce, direction);
+		if (ret == 1)
+			++retpages;
+	}
+
+	/*
+	 * If failed, release locked pages, otherwise return the number
+	 * of locked system pages
+	 */
+	if (ret < 0)
+		clear_tces_nolock(tbl, entry, i);
+	else
+		ret = retpages;
+
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..1b970bf 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -20,6 +20,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
@@ -613,3 +614,136 @@ void __init pnv_pci_init(void)
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		pr_warn("tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev_name(dev),
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev_name(dev));
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev_name(dev), iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		pr_err("tce_vfio: %s has not been added, ret=%d\n",
+				dev_name(dev), ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			pr_info("tce_vfio: cannot create new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Fri, 2012-12-07 at 18:35 +1100, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
> 
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
> 
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |   10 ++
>  arch/powerpc/kernel/iommu.c          |  214 ++++++++++++++++++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  134 +++++++++++++++++++++
>  drivers/iommu/Kconfig                |    8 ++
>  4 files changed, 366 insertions(+)
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..be3b11b 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -76,6 +76,9 @@ struct iommu_table {
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
> @@ -147,5 +150,12 @@ static inline void iommu_restore(void)
>  }
>  #endif
>  
> +extern void iommu_reset_table(struct iommu_table *tbl, bool release);
> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages);
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..123431a 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -44,6 +44,7 @@
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
> @@ -856,3 +857,216 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +
> +/*
> + * iommu_reset_table is called when it started/stopped being used
> + */
> +void iommu_reset_table(struct iommu_table *tbl, bool release)
> +{
> +	/*
> +	 * Page at 0 is marked as used in iommu_init_table,
> +	 * so here we clear it when called with release=false...
> +	 */
> +	if (!release && (tbl->it_offset == 0))
> +		clear_bit(0, tbl->it_map);

Isn't this redundant to the memset below?

> +
> +	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
> +
> +	memset(tbl->it_map, 0, (tbl->it_size + 7) >> 3);
> +
> +	/*
> +	 * ... or restore when release=true
> +	 */
> +	if (release && (tbl->it_offset == 0))
> +		set_bit(0, tbl->it_map);

"release" to me implies something is freed, maybe this should just be
called "restore".

> +}
> +EXPORT_SYMBOL_GPL(iommu_reset_table);
> +
> +/*
> + * Returns the number of used IOMMU pages (4K) within
> + * the same system page (4K or 64K).
> + * bitmap_weight is not used as it does not support bigendian maps.
> + * "offset" is an IOMMU page number relative to DMA window start.
> + */
> +static int syspage_weight(unsigned long *map, unsigned long offset)
> +{
> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
> +
> +	/* Aligns TCE entry number to system page boundary */
> +	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> +
> +	/* Count used 4K pages */
> +	while (nbits) {
> +		if (test_bit(offset, map))
> +			++ret;
> +		--nbits;
> +		++offset;
> +	}
> +
> +	return ret;
> +}
> +
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of system pages
> + * which it called put_page() on
> + */
> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int i, retpages = 0, clr;
> +	unsigned long oldtce, oldweight;
> +	struct page *page;
> +
> +	for (i = 0; i < pages; ++i) {

Any reason not to increment "entry" and avoid the 5 cases of "entry + i"
below?

> +		if (!test_bit(entry + i - tbl->it_offset, tbl->it_map))
> +			continue;
> +
> +		oldtce = ppc_md.tce_get(tbl, entry + i);
> +		ppc_md.tce_free(tbl, entry + i, 1);
> +
> +		oldweight = syspage_weight(tbl->it_map,
> +				entry + i - tbl->it_offset);
> +		clr = __test_and_clear_bit(entry + i - tbl->it_offset,
> +				tbl->it_map);
> +
> +		if (WARN_ON(!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ))))
> +			continue;
> +
> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +		if (WARN_ON(!page))
> +			continue;
> +
> +		if (oldtce & TCE_PCI_WRITE)
> +			SetPageDirty(page);
> +
> +		put_page(page);
> +
> +		/* That was the last IOMMU page within the system page */
> +		if ((oldweight == 1) && clr)
> +			++retpages;
> +	}
> +
> +	return retpages;
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number
> + * of released system pages
> + */
> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int ret;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	spin_lock(&(pool->lock));
> +	ret = clear_tces_nolock(tbl, entry, pages);
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset, oldweight;
> +
> +	/* Map new TCE */
> +	offset = (tce & IOMMU_PAGE_MASK) - (tce & PAGE_MASK);
> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);
> +	if (ret != 1) {
> +		pr_err("tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> +		return -EFAULT;
> +	}
> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		pr_err("tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	/* Calculate if new system page has been locked */
> +	oldweight = syspage_weight(tbl->it_map, entry - tbl->it_offset);
> +	__set_bit(entry - tbl->it_offset, tbl->it_map);
> +
> +	return (oldweight == 0) ? 1 : 0;

It seems like there's an optimization for syspage_weight since you only
care about two cases, ie. syspage_weight_one and syspage_weight_zero.
The zero test is easy, just mask and return !! the value.  Testing
weight 1 means you don't have to find more than 2 bits set.  I won't
hold you to that optimization, just fyi.

> +}
> +
> +/*
> + * iommu_put_tces builds tces and returned the number of actually
> + * locked system pages
> + */
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long pages)
> +{
> +	int i, ret = 0, retpages = 0;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +	BUG_ON(direction == DMA_NONE);
> +
> +	spin_lock(&(pool->lock));
> +
> +	/* Check if any is in use */
> +	for (i = 0; i < pages; ++i) {
> +		if (test_bit(entry + i - tbl->it_offset, tbl->it_map)) {

Again, seems like there's an optimization here that avoids individually
testing bits since you only care about zero or non-zero for a sequential
run.

> +			spin_unlock(&(pool->lock));
> +			return -EBUSY;
> +		}
> +	}
> +
> +	/* Put tces to the table */
> +	for (i = 0; (i < pages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
> +		ret = put_tce(tbl, entry + i, tce, direction);
> +		if (ret == 1)
> +			++retpages;
> +	}
> +
> +	/*
> +	 * If failed, release locked pages, otherwise return the number
> +	 * of locked system pages
> +	 */
> +	if (ret < 0)
> +		clear_tces_nolock(tbl, entry, i);
> +	else
> +		ret = retpages;
> +
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..1b970bf 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -20,6 +20,7 @@
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
> @@ -613,3 +614,136 @@ void __init pnv_pci_init(void)
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		pr_warn("tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev_name(dev),
> +				iommu_group_id(dev->iommu_group));
> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev_name(dev));
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev_name(dev), iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		pr_err("tce_vfio: %s has not been added, ret=%d\n",
> +				dev_name(dev), ret);
> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			pr_info("tce_vfio: cannot create new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));
> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);

BTW, groups have a name property that shows up in sysfs that can be set
with iommu_group_set_name().  IIRC, this was a feature David requested
for PEs.  It'd be nice if it was used for PEs...  Thanks,

Alex

> +	}
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

On 08/12/12 04:38, Alex Williamson wrote:
>> +static int __init tce_iommu_init(void)
>> +{
>> +	struct pci_dev *pdev = NULL;
>> +	struct iommu_table *tbl;
>> +	struct iommu_group *grp;
>> +
>> +	/* Allocate and initialize IOMMU groups */
>> +	for_each_pci_dev(pdev) {
>> +		tbl = get_iommu_table_base(&pdev->dev);
>> +		if (!tbl)
>> +			continue;
>> +
>> +		/* Skip already initialized */
>> +		if (tbl->it_group)
>> +			continue;
>> +
>> +		grp = iommu_group_alloc();
>> +		if (IS_ERR(grp)) {
>> +			pr_info("tce_vfio: cannot create new IOMMU group, ret=%ld\n",
>> +					PTR_ERR(grp));
>> +			return PTR_ERR(grp);
>> +		}
>> +		tbl->it_group = grp;
>> +		iommu_group_set_iommudata(grp, tbl, group_release);
>
> BTW, groups have a name property that shows up in sysfs that can be set
> with iommu_group_set_name().  IIRC, this was a feature David requested
> for PEs.  It'd be nice if it was used for PEs...  Thanks,



But what would I put there?... IOMMU ID is more than enough at the moment 
and struct iommu_table does not have anything what would have made sense to 
show in the sysfs...


-- 
Alexey

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Wed, 2012-12-12 at 17:14 +1100, Alexey Kardashevskiy wrote:
> On 08/12/12 04:38, Alex Williamson wrote:
> >> +static int __init tce_iommu_init(void)
> >> +{
> >> +	struct pci_dev *pdev = NULL;
> >> +	struct iommu_table *tbl;
> >> +	struct iommu_group *grp;
> >> +
> >> +	/* Allocate and initialize IOMMU groups */
> >> +	for_each_pci_dev(pdev) {
> >> +		tbl = get_iommu_table_base(&pdev->dev);
> >> +		if (!tbl)
> >> +			continue;
> >> +
> >> +		/* Skip already initialized */
> >> +		if (tbl->it_group)
> >> +			continue;
> >> +
> >> +		grp = iommu_group_alloc();
> >> +		if (IS_ERR(grp)) {
> >> +			pr_info("tce_vfio: cannot create new IOMMU group, ret=%ld\n",
> >> +					PTR_ERR(grp));
> >> +			return PTR_ERR(grp);
> >> +		}
> >> +		tbl->it_group = grp;
> >> +		iommu_group_set_iommudata(grp, tbl, group_release);
> >
> > BTW, groups have a name property that shows up in sysfs that can be set
> > with iommu_group_set_name().  IIRC, this was a feature David requested
> > for PEs.  It'd be nice if it was used for PEs...  Thanks,
> 
> 
> 
> But what would I put there?... IOMMU ID is more than enough at the moment 
> and struct iommu_table does not have anything what would have made sense to 
> show in the sysfs...

I believe David mentioned that PEs had user visible names.  Perhaps they
match an enclosure location or something.  Group numbers are rather
arbitrary and really have no guarantee of persistence.  Thanks,

Alex

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Benjamin Herrenschmidt]

On Wed, 2012-12-12 at 07:34 -0700, Alex Williamson wrote:
> > But what would I put there?... IOMMU ID is more than enough at the moment 
> > and struct iommu_table does not have anything what would have made sense to 
> > show in the sysfs...
> 
> I believe David mentioned that PEs had user visible names.  Perhaps they
> match an enclosure location or something.  Group numbers are rather
> arbitrary and really have no guarantee of persistence.  Thanks, 

I agree. Make up something, for example domain[PE] or something like
that.

Cheers,
Ben.

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

On 13/12/12 13:29, Benjamin Herrenschmidt wrote:
> On Wed, 2012-12-12 at 07:34 -0700, Alex Williamson wrote:
>>> But what would I put there?... IOMMU ID is more than enough at the moment
>>> and struct iommu_table does not have anything what would have made sense to
>>> show in the sysfs...
>>
>> I believe David mentioned that PEs had user visible names.  Perhaps they
>> match an enclosure location or something.  Group numbers are rather
>> arbitrary and really have no guarantee of persistence.  Thanks,
>
> I agree. Make up something, for example domain[PE] or something like
> that.

To be able to add a PE number, I need to call iommu_group_alloc() in the 
correct place where I know this number OR I have to carry it in iommu_table 
till the moment the iommu_group_alloc() is called (acceptable but not cool).

I will post a patch which would help as a response to this mail.


-- 
Alexey

[PATCH] iommu: moving initialization earlier

[Alexey Kardashevskiy]

The iommu_init() call initializes IOMMU internal structures and data
required for the API to function such as iommu_group_alloc().
It is registered as a subsys_initcall.

One of the IOMMU users is a PCI subsystem on POWER which discovers new
IOMMU tables during the PCI scan so the most logical place to call
iommu_group_alloc() is when a new group is just discovered. However
PCI scan is done from subsys_initcall hook as well what makes
using of the IOMMU API impossible.

The patch moves IOMMU subsystem initialization one step earlier.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/iommu/iommu.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index de857bb..b0afd3d 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -865,7 +865,7 @@ printk("%s %u\n", __func__, __LINE__);
 
 	return 0;
 }
-subsys_initcall(iommu_init);
+arch_initcall(iommu_init);
 
 int iommu_domain_get_attr(struct iommu_domain *domain,
 			  enum iommu_attr attr, void *data)
-- 
1.7.10.4

Re: [PATCH] iommu: moving initialization earlier

[Alex Williamson]

Probably a good idea to CC the iommu list and maintainer...

On Thu, 2012-12-13 at 17:28 +1100, Alexey Kardashevskiy wrote:
> The iommu_init() call initializes IOMMU internal structures and data
> required for the API to function such as iommu_group_alloc().
> It is registered as a subsys_initcall.
> 
> One of the IOMMU users is a PCI subsystem on POWER which discovers new
> IOMMU tables during the PCI scan so the most logical place to call
> iommu_group_alloc() is when a new group is just discovered. However
> PCI scan is done from subsys_initcall hook as well what makes
> using of the IOMMU API impossible.
> 
> The patch moves IOMMU subsystem initialization one step earlier.
> 
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  drivers/iommu/iommu.c |    2 +-
>  1 file changed, 1 insertion(+), 1 deletion(-)
> 
> diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
> index de857bb..b0afd3d 100644
> --- a/drivers/iommu/iommu.c
> +++ b/drivers/iommu/iommu.c
> @@ -865,7 +865,7 @@ printk("%s %u\n", __func__, __LINE__);
>  
>  	return 0;
>  }
> -subsys_initcall(iommu_init);
> +arch_initcall(iommu_init);
>  
>  int iommu_domain_get_attr(struct iommu_domain *domain,
>  			  enum iommu_attr attr, void *data)

Re: [PATCH] iommu: moving initialization earlier

[Joerg Roedel]

Alexey,

On Thu, Dec 13, 2012 at 08:48:55AM -0700, Alex Williamson wrote:
> Probably a good idea to CC the iommu list and maintainer...
> 
> On Thu, 2012-12-13 at 17:28 +1100, Alexey Kardashevskiy wrote:
> > Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>

Please resend the patch when the merge-window is closed.

Thanks,

	Joerg

Re: [PATCH] iommu: moving initialization earlier

[Alexey Kardashevskiy]

On 16/12/12 22:20, Joerg Roedel wrote:
> Alexey,
>
> On Thu, Dec 13, 2012 at 08:48:55AM -0700, Alex Williamson wrote:
>> Probably a good idea to CC the iommu list and maintainer...
>>
>> On Thu, 2012-12-13 at 17:28 +1100, Alexey Kardashevskiy wrote:
>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>
> Please resend the patch when the merge-window is closed.

is it closed now? not sure I entirely understand what window you kept in 
mind :)


-- 
Alexey

Re: [PATCH] iommu: moving initialization earlier

[Joerg Roedel]

On Fri, Jan 04, 2013 at 07:21:34PM +1100, Alexey Kardashevskiy wrote:
> On 16/12/12 22:20, Joerg Roedel wrote:

> >Please resend the patch when the merge-window is closed.
> 
> is it closed now? not sure I entirely understand what window you
> kept in mind :)

Yes, it is closed now :-) The merge-window is the time between a Linux
kernel release (like v3.7) and the next release candiate (like
v3.8-rc1).


	Joerg

[PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

This patch initializes IOMMU groups based on the IOMMU
configuration discovered during the PCI scan on POWERNV
(POWER non virtualized) platform. The IOMMU groups are
to be used later by VFIO driver (PCI pass through).

It also implements an API for mapping/unmapping pages for
guest PCI drivers and providing DMA window properties.
This API is going to be used later by QEMU-VFIO to handle
h_put_tce hypercalls from the KVM guest.

Although this driver has been tested only on the POWERNV
platform, it should work on any platform which supports
TCE tables.

To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
option and configure VFIO as required.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 arch/powerpc/include/asm/iommu.h     |   10 ++
 arch/powerpc/kernel/iommu.c          |  329 ++++++++++++++++++++++++++++++++++
 arch/powerpc/platforms/powernv/pci.c |  134 ++++++++++++++
 drivers/iommu/Kconfig                |    8 +
 4 files changed, 481 insertions(+)

diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
index cbfe678..3c861ae 100644
--- a/arch/powerpc/include/asm/iommu.h
+++ b/arch/powerpc/include/asm/iommu.h
@@ -76,6 +76,9 @@ struct iommu_table {
 	struct iommu_pool large_pool;
 	struct iommu_pool pools[IOMMU_NR_POOLS];
 	unsigned long *it_map;       /* A simple allocation bitmap for now */
+#ifdef CONFIG_IOMMU_API
+	struct iommu_group *it_group;
+#endif
 };
 
 struct scatterlist;
@@ -147,5 +150,12 @@ static inline void iommu_restore(void)
 }
 #endif
 
+extern void iommu_reset_table(struct iommu_table *tbl, bool restore);
+extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long ioba,
+		unsigned long size);
+extern long iommu_put_tces(struct iommu_table *tbl, unsigned long ioba,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long size);
+
 #endif /* __KERNEL__ */
 #endif /* _ASM_IOMMU_H */
diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
index ff5a6ce..f3bb2e7 100644
--- a/arch/powerpc/kernel/iommu.c
+++ b/arch/powerpc/kernel/iommu.c
@@ -36,6 +36,7 @@
 #include <linux/hash.h>
 #include <linux/fault-inject.h>
 #include <linux/pci.h>
+#include <linux/uaccess.h>
 #include <asm/io.h>
 #include <asm/prom.h>
 #include <asm/iommu.h>
@@ -44,6 +45,7 @@
 #include <asm/kdump.h>
 #include <asm/fadump.h>
 #include <asm/vio.h>
+#include <asm/tce.h>
 
 #define DBG(...)
 
@@ -856,3 +858,330 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
 		free_pages((unsigned long)vaddr, get_order(size));
 	}
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * SPAPR TCE API
+ */
+
+struct vwork {
+	struct mm_struct	*mm;
+	long			npage;
+	struct work_struct	work;
+};
+
+/* delayed decrement/increment for locked_vm */
+static void lock_acct_bg(struct work_struct *work)
+{
+	struct vwork *vwork = container_of(work, struct vwork, work);
+	struct mm_struct *mm;
+
+	mm = vwork->mm;
+	down_write(&mm->mmap_sem);
+	mm->locked_vm += vwork->npage;
+	up_write(&mm->mmap_sem);
+	mmput(mm);
+	kfree(vwork);
+}
+
+static void lock_acct(long npage)
+{
+	struct vwork *vwork;
+	struct mm_struct *mm;
+
+	if (!current->mm)
+		return; /* process exited */
+
+	if (down_write_trylock(&current->mm->mmap_sem)) {
+		current->mm->locked_vm += npage;
+		up_write(&current->mm->mmap_sem);
+		return;
+	}
+
+	/*
+	 * Couldn't get mmap_sem lock, so must setup to update
+	 * mm->locked_vm later. If locked_vm were atomic, we
+	 * wouldn't need this silliness
+	 */
+	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
+	if (!vwork)
+		return;
+	mm = get_task_mm(current);
+	if (!mm) {
+		kfree(vwork);
+		return;
+	}
+	INIT_WORK(&vwork->work, lock_acct_bg);
+	vwork->mm = mm;
+	vwork->npage = npage;
+	schedule_work(&vwork->work);
+}
+
+/*
+ * iommu_reset_table is called when it started/stopped being used.
+ *
+ * restore==true says to bring the iommu_table into the state as it was
+ * before being used by VFIO.
+ */
+void iommu_reset_table(struct iommu_table *tbl, bool restore)
+{
+	/* Page#0 is marked as used in iommu_init_table, so we clear it... */
+	if (!restore && (tbl->it_offset == 0))
+		clear_bit(0, tbl->it_map);
+
+	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
+
+	/* ... or restore  */
+	if (restore && (tbl->it_offset == 0))
+		set_bit(0, tbl->it_map);
+}
+EXPORT_SYMBOL_GPL(iommu_reset_table);
+
+/*
+ * Returns the number of used IOMMU pages (4K) within
+ * the same system page (4K or 64K).
+ *
+ * syspage_weight_zero is optimized for expected case == 0
+ * syspage_weight_one is optimized for expected case > 1
+ * Other case are not used in this file.
+ */
+#if PAGE_SIZE == IOMMU_PAGE_SIZE
+
+#define syspage_weight_zero(map, offset)	test_bit((map), (offset))
+#define syspage_weight_one(map, offset)		test_bit((map), (offset))
+
+#elif PAGE_SIZE/IOMMU_PAGE_SIZE == 16
+
+static int syspage_weight_zero(unsigned long *map, unsigned long offset)
+{
+	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
+	return 0xffffUL & (map[BIT_WORD(offset)] >>
+			(offset & (BITS_PER_LONG-1)));
+}
+
+static int syspage_weight_one(unsigned long *map, unsigned long offset)
+{
+	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
+
+	/* Aligns TCE entry number to system page boundary */
+	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
+
+	/* Count used 4K pages */
+	while (nbits && (ret < 2)) {
+		if (test_bit(offset, map))
+			++ret;
+
+		--nbits;
+		++offset;
+	}
+
+	return ret;
+}
+#else
+#error TODO: support other page size
+#endif
+
+static void tce_flush(struct iommu_table *tbl)
+{
+	/* Flush/invalidate TLB caches if necessary */
+	if (ppc_md.tce_flush)
+		ppc_md.tce_flush(tbl);
+
+	/* Make sure updates are seen by hardware */
+	mb();
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number of system pages
+ * which it called put_page() on
+ */
+static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
+		unsigned long pages)
+{
+	int i, retpages = 0, clr;
+	unsigned long oldtce, oldweight;
+	struct page *page;
+
+	for (i = 0; i < pages; ++i, ++entry) {
+		if (!test_bit(entry - tbl->it_offset, tbl->it_map))
+			continue;
+
+		oldtce = ppc_md.tce_get(tbl, entry);
+		ppc_md.tce_free(tbl, entry, 1);
+
+		oldweight = syspage_weight_one(tbl->it_map,
+				entry - tbl->it_offset);
+		clr = __test_and_clear_bit(entry - tbl->it_offset,
+				tbl->it_map);
+
+		if (WARN_ON(!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ))))
+			continue;
+
+		page = pfn_to_page(oldtce >> PAGE_SHIFT);
+
+		if (WARN_ON(!page))
+			continue;
+
+		if (oldtce & TCE_PCI_WRITE)
+			SetPageDirty(page);
+
+		put_page(page);
+
+		/* That was the last IOMMU page within the system page */
+		if ((oldweight == 1) && clr)
+			++retpages;
+	}
+
+	return retpages;
+}
+
+/*
+ * iommu_clear_tces clears tces and returned the number
+ * of released system pages
+ */
+long iommu_clear_tces(struct iommu_table *tbl, unsigned long ioba,
+		unsigned long size)
+{
+	int ret;
+	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
+	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+
+	if ((size & ~IOMMU_PAGE_MASK) || (ioba & ~IOMMU_PAGE_MASK))
+		return -EINVAL;
+
+	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
+			<< IOMMU_PAGE_SHIFT))
+		return -EINVAL;
+
+	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
+		return -EINVAL;
+
+	spin_lock(&(pool->lock));
+	ret = clear_tces_nolock(tbl, entry, npages);
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	if (ret > 0) {
+		lock_acct(-ret);
+		return 0;
+	}
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_clear_tces);
+
+static int put_tce(struct iommu_table *tbl, unsigned long entry,
+		uint64_t tce, enum dma_data_direction direction)
+{
+	int ret;
+	struct page *page = NULL;
+	unsigned long kva, offset, oldweight;
+
+	/* Map new TCE */
+	offset = tce & IOMMU_PAGE_MASK & ~PAGE_MASK;
+	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
+			direction != DMA_TO_DEVICE, &page);
+	if (ret != 1) {
+		pr_err("tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, ret);
+		return -EFAULT;
+	}
+
+	kva = (unsigned long) page_address(page);
+	kva += offset;
+
+	/* tce_build receives a virtual address */
+	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
+
+	/* tce_build() only returns non-zero for transient errors */
+	if (unlikely(ret)) {
+		pr_err("tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
+				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
+		put_page(page);
+		return -EIO;
+	}
+
+	/* Calculate if new system page has been locked */
+	oldweight = syspage_weight_zero(tbl->it_map, entry - tbl->it_offset);
+	__set_bit(entry - tbl->it_offset, tbl->it_map);
+
+	return (oldweight == 0) ? 1 : 0;
+}
+
+/*
+ * iommu_put_tces builds tces and returned the number of actually
+ * locked system pages
+ */
+long iommu_put_tces(struct iommu_table *tbl, unsigned long ioba,
+		uint64_t tce, enum dma_data_direction direction,
+		unsigned long size)
+{
+	int i, ret = 0, retpages = 0;
+	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
+	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
+	struct iommu_pool *pool = get_pool(tbl, entry);
+	unsigned long locked, lock_limit;
+
+	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
+	BUG_ON(direction == DMA_NONE);
+
+	if ((size & ~IOMMU_PAGE_MASK) ||
+			(ioba & ~IOMMU_PAGE_MASK) ||
+			(tce & ~IOMMU_PAGE_MASK))
+		return -EINVAL;
+
+	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
+			 << IOMMU_PAGE_SHIFT))
+		return -EINVAL;
+
+	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
+		return -EINVAL;
+
+	/* Account for locked pages */
+	locked = current->mm->locked_vm +
+		(_ALIGN_UP(size, PAGE_SIZE) >> PAGE_SHIFT);
+	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+	if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
+		pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
+				rlimit(RLIMIT_MEMLOCK));
+		return -ENOMEM;
+	}
+
+	spin_lock(&(pool->lock));
+
+	/* Check if any is in use */
+	for (i = 0; i < npages; ++i) {
+		if (test_bit(entry + i - tbl->it_offset, tbl->it_map)) {
+			spin_unlock(&(pool->lock));
+			return -EBUSY;
+		}
+	}
+
+	/* Put tces to the table */
+	for (i = 0; (i < npages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
+		ret = put_tce(tbl, entry + i, tce, direction);
+		if (ret == 1)
+			++retpages;
+	}
+
+	/*
+	 * If failed, release locked pages, otherwise return the number
+	 * of locked system pages
+	 */
+	if (ret < 0) {
+		clear_tces_nolock(tbl, entry, i);
+	} else {
+		if (retpages)
+			lock_acct(retpages);
+		ret = 0;
+	}
+
+	tce_flush(tbl);
+	spin_unlock(&(pool->lock));
+
+	return ret;
+}
+EXPORT_SYMBOL_GPL(iommu_put_tces);
+
+#endif /* CONFIG_IOMMU_API */
diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
index 05205cf..1b970bf 100644
--- a/arch/powerpc/platforms/powernv/pci.c
+++ b/arch/powerpc/platforms/powernv/pci.c
@@ -20,6 +20,7 @@
 #include <linux/irq.h>
 #include <linux/io.h>
 #include <linux/msi.h>
+#include <linux/iommu.h>
 
 #include <asm/sections.h>
 #include <asm/io.h>
@@ -613,3 +614,136 @@ void __init pnv_pci_init(void)
 	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
 #endif
 }
+
+#ifdef CONFIG_IOMMU_API
+/*
+ * IOMMU groups support required by VFIO
+ */
+static int add_device(struct device *dev)
+{
+	struct iommu_table *tbl;
+	int ret = 0;
+
+	if (WARN_ON(dev->iommu_group)) {
+		pr_warn("tce_vfio: device %s is already in iommu group %d, skipping\n",
+				dev_name(dev),
+				iommu_group_id(dev->iommu_group));
+		return -EBUSY;
+	}
+
+	tbl = get_iommu_table_base(dev);
+	if (!tbl) {
+		pr_debug("tce_vfio: skipping device %s with no tbl\n",
+				dev_name(dev));
+		return 0;
+	}
+
+	pr_debug("tce_vfio: adding %s to iommu group %d\n",
+			dev_name(dev), iommu_group_id(tbl->it_group));
+
+	ret = iommu_group_add_device(tbl->it_group, dev);
+	if (ret < 0)
+		pr_err("tce_vfio: %s has not been added, ret=%d\n",
+				dev_name(dev), ret);
+
+	return ret;
+}
+
+static void del_device(struct device *dev)
+{
+	iommu_group_remove_device(dev);
+}
+
+static int iommu_bus_notifier(struct notifier_block *nb,
+			      unsigned long action, void *data)
+{
+	struct device *dev = data;
+
+	switch (action) {
+	case BUS_NOTIFY_ADD_DEVICE:
+		return add_device(dev);
+	case BUS_NOTIFY_DEL_DEVICE:
+		del_device(dev);
+		return 0;
+	default:
+		return 0;
+	}
+}
+
+static struct notifier_block tce_iommu_bus_nb = {
+	.notifier_call = iommu_bus_notifier,
+};
+
+static void group_release(void *iommu_data)
+{
+	struct iommu_table *tbl = iommu_data;
+	tbl->it_group = NULL;
+}
+
+static int __init tce_iommu_init(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp;
+
+	/* Allocate and initialize IOMMU groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+
+		/* Skip already initialized */
+		if (tbl->it_group)
+			continue;
+
+		grp = iommu_group_alloc();
+		if (IS_ERR(grp)) {
+			pr_info("tce_vfio: cannot create new IOMMU group, ret=%ld\n",
+					PTR_ERR(grp));
+			return PTR_ERR(grp);
+		}
+		tbl->it_group = grp;
+		iommu_group_set_iommudata(grp, tbl, group_release);
+	}
+
+	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Add PCI devices to VFIO groups */
+	for_each_pci_dev(pdev)
+		add_device(&pdev->dev);
+
+	return 0;
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	struct pci_dev *pdev = NULL;
+	struct iommu_table *tbl;
+	struct iommu_group *grp = NULL;
+
+	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
+
+	/* Delete PCI devices from VFIO groups */
+	for_each_pci_dev(pdev)
+		del_device(&pdev->dev);
+
+	/* Release VFIO groups */
+	for_each_pci_dev(pdev) {
+		tbl = get_iommu_table_base(&pdev->dev);
+		if (!tbl)
+			continue;
+		grp = tbl->it_group;
+
+		/* Skip (already) uninitialized */
+		if (!grp)
+			continue;
+
+		/* Do actual release, group_release() is expected to work */
+		iommu_group_put(grp);
+		BUG_ON(tbl->it_group);
+	}
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+#endif /* CONFIG_IOMMU_API */
diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
index 9f69b56..29d11dc 100644
--- a/drivers/iommu/Kconfig
+++ b/drivers/iommu/Kconfig
@@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
 
 	  Say N unless you need kernel log message for IOMMU debugging
 
+config SPAPR_TCE_IOMMU
+	bool "sPAPR TCE IOMMU Support"
+	depends on PPC_POWERNV
+	select IOMMU_API
+	help
+	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
+	  still not implemented.
+
 endif # IOMMU_SUPPORT
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

Hi Alex,

I posted other pair of patches. While debugging and testing my stuff I 
implemented some rough hack to support IOMMU mappings without passing those 
hypercalls to the QEMU, this is why I moved pieces of code around - want to 
support both QEMU-VFIO and kernel optimized H_PUT_TCE handler.



On 12/12/12 23:34, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
>
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
>
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
>
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
>
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---


-- 
Alexey

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Wed, 2012-12-12 at 23:34 +1100, Alexey Kardashevskiy wrote:
> This patch initializes IOMMU groups based on the IOMMU
> configuration discovered during the PCI scan on POWERNV
> (POWER non virtualized) platform. The IOMMU groups are
> to be used later by VFIO driver (PCI pass through).
> 
> It also implements an API for mapping/unmapping pages for
> guest PCI drivers and providing DMA window properties.
> This API is going to be used later by QEMU-VFIO to handle
> h_put_tce hypercalls from the KVM guest.
> 
> Although this driver has been tested only on the POWERNV
> platform, it should work on any platform which supports
> TCE tables.
> 
> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
> option and configure VFIO as required.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  arch/powerpc/include/asm/iommu.h     |   10 ++
>  arch/powerpc/kernel/iommu.c          |  329 ++++++++++++++++++++++++++++++++++
>  arch/powerpc/platforms/powernv/pci.c |  134 ++++++++++++++
>  drivers/iommu/Kconfig                |    8 +
>  4 files changed, 481 insertions(+)
> 
> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
> index cbfe678..3c861ae 100644
> --- a/arch/powerpc/include/asm/iommu.h
> +++ b/arch/powerpc/include/asm/iommu.h
> @@ -76,6 +76,9 @@ struct iommu_table {
>  	struct iommu_pool large_pool;
>  	struct iommu_pool pools[IOMMU_NR_POOLS];
>  	unsigned long *it_map;       /* A simple allocation bitmap for now */
> +#ifdef CONFIG_IOMMU_API
> +	struct iommu_group *it_group;
> +#endif
>  };
>  
>  struct scatterlist;
> @@ -147,5 +150,12 @@ static inline void iommu_restore(void)
>  }
>  #endif
>  
> +extern void iommu_reset_table(struct iommu_table *tbl, bool restore);
> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long ioba,
> +		unsigned long size);
> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long ioba,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long size);
> +
>  #endif /* __KERNEL__ */
>  #endif /* _ASM_IOMMU_H */
> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
> index ff5a6ce..f3bb2e7 100644
> --- a/arch/powerpc/kernel/iommu.c
> +++ b/arch/powerpc/kernel/iommu.c
> @@ -36,6 +36,7 @@
>  #include <linux/hash.h>
>  #include <linux/fault-inject.h>
>  #include <linux/pci.h>
> +#include <linux/uaccess.h>
>  #include <asm/io.h>
>  #include <asm/prom.h>
>  #include <asm/iommu.h>
> @@ -44,6 +45,7 @@
>  #include <asm/kdump.h>
>  #include <asm/fadump.h>
>  #include <asm/vio.h>
> +#include <asm/tce.h>
>  
>  #define DBG(...)
>  
> @@ -856,3 +858,330 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>  		free_pages((unsigned long)vaddr, get_order(size));
>  	}
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * SPAPR TCE API
> + */
> +
> +struct vwork {
> +	struct mm_struct	*mm;
> +	long			npage;
> +	struct work_struct	work;
> +};
> +
> +/* delayed decrement/increment for locked_vm */
> +static void lock_acct_bg(struct work_struct *work)
> +{
> +	struct vwork *vwork = container_of(work, struct vwork, work);
> +	struct mm_struct *mm;
> +
> +	mm = vwork->mm;
> +	down_write(&mm->mmap_sem);
> +	mm->locked_vm += vwork->npage;
> +	up_write(&mm->mmap_sem);
> +	mmput(mm);
> +	kfree(vwork);
> +}
> +
> +static void lock_acct(long npage)
> +{
> +	struct vwork *vwork;
> +	struct mm_struct *mm;
> +
> +	if (!current->mm)
> +		return; /* process exited */
> +
> +	if (down_write_trylock(&current->mm->mmap_sem)) {
> +		current->mm->locked_vm += npage;
> +		up_write(&current->mm->mmap_sem);
> +		return;
> +	}
> +
> +	/*
> +	 * Couldn't get mmap_sem lock, so must setup to update
> +	 * mm->locked_vm later. If locked_vm were atomic, we
> +	 * wouldn't need this silliness
> +	 */
> +	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
> +	if (!vwork)
> +		return;
> +	mm = get_task_mm(current);
> +	if (!mm) {
> +		kfree(vwork);
> +		return;
> +	}
> +	INIT_WORK(&vwork->work, lock_acct_bg);
> +	vwork->mm = mm;
> +	vwork->npage = npage;
> +	schedule_work(&vwork->work);
> +}

Locked page accounting in this version is very, very broken.  How do
powerpc folks feel about seemingly generic kernel iommu interfaces
messing with the current task mm?  Besides that, more problems below...

> +
> +/*
> + * iommu_reset_table is called when it started/stopped being used.
> + *
> + * restore==true says to bring the iommu_table into the state as it was
> + * before being used by VFIO.
> + */
> +void iommu_reset_table(struct iommu_table *tbl, bool restore)
> +{
> +	/* Page#0 is marked as used in iommu_init_table, so we clear it... */
> +	if (!restore && (tbl->it_offset == 0))
> +		clear_bit(0, tbl->it_map);
> +
> +	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);

This does locked page accounting and unpins pages, even on startup when
the pages aren't necessarily pinned or accounted against the current
process.

> +
> +	/* ... or restore  */
> +	if (restore && (tbl->it_offset == 0))
> +		set_bit(0, tbl->it_map);
> +}
> +EXPORT_SYMBOL_GPL(iommu_reset_table);
> +
> +/*
> + * Returns the number of used IOMMU pages (4K) within
> + * the same system page (4K or 64K).
> + *
> + * syspage_weight_zero is optimized for expected case == 0
> + * syspage_weight_one is optimized for expected case > 1
> + * Other case are not used in this file.
> + */
> +#if PAGE_SIZE == IOMMU_PAGE_SIZE
> +
> +#define syspage_weight_zero(map, offset)	test_bit((map), (offset))
> +#define syspage_weight_one(map, offset)		test_bit((map), (offset))
> +
> +#elif PAGE_SIZE/IOMMU_PAGE_SIZE == 16
> +
> +static int syspage_weight_zero(unsigned long *map, unsigned long offset)
> +{
> +	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> +	return 0xffffUL & (map[BIT_WORD(offset)] >>
> +			(offset & (BITS_PER_LONG-1)));
> +}

I would have expected these to be bools and return true if the weight
matches the value.

If you replaced 0xffff above w/ this, would you need the #error below?

(1UL << (PAGE_SIZE/IOMMU_PAGE_SIZE)) - 1)

> +
> +static int syspage_weight_one(unsigned long *map, unsigned long offset)
> +{
> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
> +
> +	/* Aligns TCE entry number to system page boundary */
> +	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> +
> +	/* Count used 4K pages */
> +	while (nbits && (ret < 2)) {

Don't you have a ffs()?  Could also be used for _zero.  Surely there are
some bitops helpers that could help here even on big endian.  hweight
really doesn't work?

> +		if (test_bit(offset, map))
> +			++ret;
> +
> +		--nbits;
> +		++offset;
> +	}
> +
> +	return ret;
> +}
> +#else
> +#error TODO: support other page size
> +#endif
> +
> +static void tce_flush(struct iommu_table *tbl)
> +{
> +	/* Flush/invalidate TLB caches if necessary */
> +	if (ppc_md.tce_flush)
> +		ppc_md.tce_flush(tbl);
> +
> +	/* Make sure updates are seen by hardware */
> +	mb();
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number of system pages
> + * which it called put_page() on
> + */
> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> +		unsigned long pages)
> +{
> +	int i, retpages = 0, clr;
> +	unsigned long oldtce, oldweight;
> +	struct page *page;
> +
> +	for (i = 0; i < pages; ++i, ++entry) {
> +		if (!test_bit(entry - tbl->it_offset, tbl->it_map))
> +			continue;
> +
> +		oldtce = ppc_md.tce_get(tbl, entry);
> +		ppc_md.tce_free(tbl, entry, 1);
> +
> +		oldweight = syspage_weight_one(tbl->it_map,
> +				entry - tbl->it_offset);
> +		clr = __test_and_clear_bit(entry - tbl->it_offset,
> +				tbl->it_map);
> +
> +		if (WARN_ON(!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ))))
> +			continue;
> +
> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> +
> +		if (WARN_ON(!page))
> +			continue;
> +
> +		if (oldtce & TCE_PCI_WRITE)
> +			SetPageDirty(page);
> +
> +		put_page(page);
> +
> +		/* That was the last IOMMU page within the system page */
> +		if ((oldweight == 1) && clr)
> +			++retpages;
> +	}
> +
> +	return retpages;
> +}
> +
> +/*
> + * iommu_clear_tces clears tces and returned the number
> + * of released system pages
> + */
> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long ioba,
> +		unsigned long size)
> +{
> +	int ret;
> +	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
> +	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +
> +	if ((size & ~IOMMU_PAGE_MASK) || (ioba & ~IOMMU_PAGE_MASK))
> +		return -EINVAL;
> +
> +	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
> +			<< IOMMU_PAGE_SHIFT))
> +		return -EINVAL;
> +
> +	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> +		return -EINVAL;
> +
> +	spin_lock(&(pool->lock));
> +	ret = clear_tces_nolock(tbl, entry, npages);
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	if (ret > 0) {
> +		lock_acct(-ret);
> +		return 0;
> +	}
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> +
> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> +		uint64_t tce, enum dma_data_direction direction)
> +{
> +	int ret;
> +	struct page *page = NULL;
> +	unsigned long kva, offset, oldweight;
> +
> +	/* Map new TCE */
> +	offset = tce & IOMMU_PAGE_MASK & ~PAGE_MASK;
> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> +			direction != DMA_TO_DEVICE, &page);
> +	if (ret != 1) {
> +		pr_err("tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> +		return -EFAULT;
> +	}
> +
> +	kva = (unsigned long) page_address(page);
> +	kva += offset;
> +
> +	/* tce_build receives a virtual address */
> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> +
> +	/* tce_build() only returns non-zero for transient errors */
> +	if (unlikely(ret)) {
> +		pr_err("tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> +		put_page(page);
> +		return -EIO;
> +	}
> +
> +	/* Calculate if new system page has been locked */
> +	oldweight = syspage_weight_zero(tbl->it_map, entry - tbl->it_offset);
> +	__set_bit(entry - tbl->it_offset, tbl->it_map);
> +
> +	return (oldweight == 0) ? 1 : 0;
> +}
> +
> +/*
> + * iommu_put_tces builds tces and returned the number of actually
> + * locked system pages
> + */
> +long iommu_put_tces(struct iommu_table *tbl, unsigned long ioba,
> +		uint64_t tce, enum dma_data_direction direction,
> +		unsigned long size)
> +{
> +	int i, ret = 0, retpages = 0;
> +	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
> +	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
> +	struct iommu_pool *pool = get_pool(tbl, entry);
> +	unsigned long locked, lock_limit;
> +
> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> +	BUG_ON(direction == DMA_NONE);
> +
> +	if ((size & ~IOMMU_PAGE_MASK) ||
> +			(ioba & ~IOMMU_PAGE_MASK) ||
> +			(tce & ~IOMMU_PAGE_MASK))
> +		return -EINVAL;
> +
> +	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
> +			 << IOMMU_PAGE_SHIFT))
> +		return -EINVAL;
> +
> +	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> +		return -EINVAL;
> +
> +	/* Account for locked pages */
> +	locked = current->mm->locked_vm +
> +		(_ALIGN_UP(size, PAGE_SIZE) >> PAGE_SHIFT);

Looks like we just over penalize upfront and correct when mapped, that's
better, but not great.

> +	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
> +	if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
> +		pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
> +				rlimit(RLIMIT_MEMLOCK));
> +		return -ENOMEM;
> +	}
> +
> +	spin_lock(&(pool->lock));
> +
> +	/* Check if any is in use */
> +	for (i = 0; i < npages; ++i) {
> +		if (test_bit(entry + i - tbl->it_offset, tbl->it_map)) {
> +			spin_unlock(&(pool->lock));
> +			return -EBUSY;
> +		}
> +	}
> +
> +	/* Put tces to the table */
> +	for (i = 0; (i < npages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
> +		ret = put_tce(tbl, entry + i, tce, direction);
> +		if (ret == 1)
> +			++retpages;
> +	}
> +
> +	/*
> +	 * If failed, release locked pages, otherwise return the number
> +	 * of locked system pages
> +	 */
> +	if (ret < 0) {
> +		clear_tces_nolock(tbl, entry, i);
> +	} else {
> +		if (retpages)
> +			lock_acct(retpages);
> +		ret = 0;
> +	}

Bug, if it fails we clear, which decrements our locked pages, but we
haven't incremented them yet.  Thanks,

Alex

> +
> +	tce_flush(tbl);
> +	spin_unlock(&(pool->lock));
> +
> +	return ret;
> +}
> +EXPORT_SYMBOL_GPL(iommu_put_tces);
> +
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> index 05205cf..1b970bf 100644
> --- a/arch/powerpc/platforms/powernv/pci.c
> +++ b/arch/powerpc/platforms/powernv/pci.c
> @@ -20,6 +20,7 @@
>  #include <linux/irq.h>
>  #include <linux/io.h>
>  #include <linux/msi.h>
> +#include <linux/iommu.h>
>  
>  #include <asm/sections.h>
>  #include <asm/io.h>
> @@ -613,3 +614,136 @@ void __init pnv_pci_init(void)
>  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
>  #endif
>  }
> +
> +#ifdef CONFIG_IOMMU_API
> +/*
> + * IOMMU groups support required by VFIO
> + */
> +static int add_device(struct device *dev)
> +{
> +	struct iommu_table *tbl;
> +	int ret = 0;
> +
> +	if (WARN_ON(dev->iommu_group)) {
> +		pr_warn("tce_vfio: device %s is already in iommu group %d, skipping\n",
> +				dev_name(dev),
> +				iommu_group_id(dev->iommu_group));
> +		return -EBUSY;
> +	}
> +
> +	tbl = get_iommu_table_base(dev);
> +	if (!tbl) {
> +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> +				dev_name(dev));
> +		return 0;
> +	}
> +
> +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> +			dev_name(dev), iommu_group_id(tbl->it_group));
> +
> +	ret = iommu_group_add_device(tbl->it_group, dev);
> +	if (ret < 0)
> +		pr_err("tce_vfio: %s has not been added, ret=%d\n",
> +				dev_name(dev), ret);
> +
> +	return ret;
> +}
> +
> +static void del_device(struct device *dev)
> +{
> +	iommu_group_remove_device(dev);
> +}
> +
> +static int iommu_bus_notifier(struct notifier_block *nb,
> +			      unsigned long action, void *data)
> +{
> +	struct device *dev = data;
> +
> +	switch (action) {
> +	case BUS_NOTIFY_ADD_DEVICE:
> +		return add_device(dev);
> +	case BUS_NOTIFY_DEL_DEVICE:
> +		del_device(dev);
> +		return 0;
> +	default:
> +		return 0;
> +	}
> +}
> +
> +static struct notifier_block tce_iommu_bus_nb = {
> +	.notifier_call = iommu_bus_notifier,
> +};
> +
> +static void group_release(void *iommu_data)
> +{
> +	struct iommu_table *tbl = iommu_data;
> +	tbl->it_group = NULL;
> +}
> +
> +static int __init tce_iommu_init(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp;
> +
> +	/* Allocate and initialize IOMMU groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +
> +		/* Skip already initialized */
> +		if (tbl->it_group)
> +			continue;
> +
> +		grp = iommu_group_alloc();
> +		if (IS_ERR(grp)) {
> +			pr_info("tce_vfio: cannot create new IOMMU group, ret=%ld\n",
> +					PTR_ERR(grp));
> +			return PTR_ERR(grp);
> +		}
> +		tbl->it_group = grp;
> +		iommu_group_set_iommudata(grp, tbl, group_release);
> +	}
> +
> +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Add PCI devices to VFIO groups */
> +	for_each_pci_dev(pdev)
> +		add_device(&pdev->dev);
> +
> +	return 0;
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	struct pci_dev *pdev = NULL;
> +	struct iommu_table *tbl;
> +	struct iommu_group *grp = NULL;
> +
> +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> +
> +	/* Delete PCI devices from VFIO groups */
> +	for_each_pci_dev(pdev)
> +		del_device(&pdev->dev);
> +
> +	/* Release VFIO groups */
> +	for_each_pci_dev(pdev) {
> +		tbl = get_iommu_table_base(&pdev->dev);
> +		if (!tbl)
> +			continue;
> +		grp = tbl->it_group;
> +
> +		/* Skip (already) uninitialized */
> +		if (!grp)
> +			continue;
> +
> +		/* Do actual release, group_release() is expected to work */
> +		iommu_group_put(grp);
> +		BUG_ON(tbl->it_group);
> +	}
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +#endif /* CONFIG_IOMMU_API */
> diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> index 9f69b56..29d11dc 100644
> --- a/drivers/iommu/Kconfig
> +++ b/drivers/iommu/Kconfig
> @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
>  
>  	  Say N unless you need kernel log message for IOMMU debugging
>  
> +config SPAPR_TCE_IOMMU
> +	bool "sPAPR TCE IOMMU Support"
> +	depends on PPC_POWERNV
> +	select IOMMU_API
> +	help
> +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> +	  still not implemented.
> +
>  endif # IOMMU_SUPPORT

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alexey Kardashevskiy]

On 13/12/12 10:30, Alex Williamson wrote:
> On Wed, 2012-12-12 at 23:34 +1100, Alexey Kardashevskiy wrote:
>> This patch initializes IOMMU groups based on the IOMMU
>> configuration discovered during the PCI scan on POWERNV
>> (POWER non virtualized) platform. The IOMMU groups are
>> to be used later by VFIO driver (PCI pass through).
>>
>> It also implements an API for mapping/unmapping pages for
>> guest PCI drivers and providing DMA window properties.
>> This API is going to be used later by QEMU-VFIO to handle
>> h_put_tce hypercalls from the KVM guest.
>>
>> Although this driver has been tested only on the POWERNV
>> platform, it should work on any platform which supports
>> TCE tables.
>>
>> To enable VFIO on POWER, enable SPAPR_TCE_IOMMU config
>> option and configure VFIO as required.
>>
>> Cc: David Gibson <david@gibson.dropbear.id.au>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>> ---
>>   arch/powerpc/include/asm/iommu.h     |   10 ++
>>   arch/powerpc/kernel/iommu.c          |  329 ++++++++++++++++++++++++++++++++++
>>   arch/powerpc/platforms/powernv/pci.c |  134 ++++++++++++++
>>   drivers/iommu/Kconfig                |    8 +
>>   4 files changed, 481 insertions(+)
>>
>> diff --git a/arch/powerpc/include/asm/iommu.h b/arch/powerpc/include/asm/iommu.h
>> index cbfe678..3c861ae 100644
>> --- a/arch/powerpc/include/asm/iommu.h
>> +++ b/arch/powerpc/include/asm/iommu.h
>> @@ -76,6 +76,9 @@ struct iommu_table {
>>   	struct iommu_pool large_pool;
>>   	struct iommu_pool pools[IOMMU_NR_POOLS];
>>   	unsigned long *it_map;       /* A simple allocation bitmap for now */
>> +#ifdef CONFIG_IOMMU_API
>> +	struct iommu_group *it_group;
>> +#endif
>>   };
>>
>>   struct scatterlist;
>> @@ -147,5 +150,12 @@ static inline void iommu_restore(void)
>>   }
>>   #endif
>>
>> +extern void iommu_reset_table(struct iommu_table *tbl, bool restore);
>> +extern long iommu_clear_tces(struct iommu_table *tbl, unsigned long ioba,
>> +		unsigned long size);
>> +extern long iommu_put_tces(struct iommu_table *tbl, unsigned long ioba,
>> +		uint64_t tce, enum dma_data_direction direction,
>> +		unsigned long size);
>> +
>>   #endif /* __KERNEL__ */
>>   #endif /* _ASM_IOMMU_H */
>> diff --git a/arch/powerpc/kernel/iommu.c b/arch/powerpc/kernel/iommu.c
>> index ff5a6ce..f3bb2e7 100644
>> --- a/arch/powerpc/kernel/iommu.c
>> +++ b/arch/powerpc/kernel/iommu.c
>> @@ -36,6 +36,7 @@
>>   #include <linux/hash.h>
>>   #include <linux/fault-inject.h>
>>   #include <linux/pci.h>
>> +#include <linux/uaccess.h>
>>   #include <asm/io.h>
>>   #include <asm/prom.h>
>>   #include <asm/iommu.h>
>> @@ -44,6 +45,7 @@
>>   #include <asm/kdump.h>
>>   #include <asm/fadump.h>
>>   #include <asm/vio.h>
>> +#include <asm/tce.h>
>>
>>   #define DBG(...)
>>
>> @@ -856,3 +858,330 @@ void iommu_free_coherent(struct iommu_table *tbl, size_t size,
>>   		free_pages((unsigned long)vaddr, get_order(size));
>>   	}
>>   }
>> +
>> +#ifdef CONFIG_IOMMU_API
>> +/*
>> + * SPAPR TCE API
>> + */
>> +
>> +struct vwork {
>> +	struct mm_struct	*mm;
>> +	long			npage;
>> +	struct work_struct	work;
>> +};
>> +
>> +/* delayed decrement/increment for locked_vm */
>> +static void lock_acct_bg(struct work_struct *work)
>> +{
>> +	struct vwork *vwork = container_of(work, struct vwork, work);
>> +	struct mm_struct *mm;
>> +
>> +	mm = vwork->mm;
>> +	down_write(&mm->mmap_sem);
>> +	mm->locked_vm += vwork->npage;
>> +	up_write(&mm->mmap_sem);
>> +	mmput(mm);
>> +	kfree(vwork);
>> +}
>> +
>> +static void lock_acct(long npage)
>> +{
>> +	struct vwork *vwork;
>> +	struct mm_struct *mm;
>> +
>> +	if (!current->mm)
>> +		return; /* process exited */
>> +
>> +	if (down_write_trylock(&current->mm->mmap_sem)) {
>> +		current->mm->locked_vm += npage;
>> +		up_write(&current->mm->mmap_sem);
>> +		return;
>> +	}
>> +
>> +	/*
>> +	 * Couldn't get mmap_sem lock, so must setup to update
>> +	 * mm->locked_vm later. If locked_vm were atomic, we
>> +	 * wouldn't need this silliness
>> +	 */
>> +	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
>> +	if (!vwork)
>> +		return;
>> +	mm = get_task_mm(current);
>> +	if (!mm) {
>> +		kfree(vwork);
>> +		return;
>> +	}
>> +	INIT_WORK(&vwork->work, lock_acct_bg);
>> +	vwork->mm = mm;
>> +	vwork->npage = npage;
>> +	schedule_work(&vwork->work);
>> +}
>
> Locked page accounting in this version is very, very broken.  How do
> powerpc folks feel about seemingly generic kernel iommu interfaces
> messing with the current task mm?  Besides that, more problems below...
>
>> +
>> +/*
>> + * iommu_reset_table is called when it started/stopped being used.
>> + *
>> + * restore==true says to bring the iommu_table into the state as it was
>> + * before being used by VFIO.
>> + */
>> +void iommu_reset_table(struct iommu_table *tbl, bool restore)
>> +{
>> +	/* Page#0 is marked as used in iommu_init_table, so we clear it... */
>> +	if (!restore && (tbl->it_offset == 0))
>> +		clear_bit(0, tbl->it_map);
>> +
>> +	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
>
> This does locked page accounting and unpins pages, even on startup when
> the pages aren't necessarily pinned or accounted against the current
> process.
 >
>> +
>> +	/* ... or restore  */
>> +	if (restore && (tbl->it_offset == 0))
>> +		set_bit(0, tbl->it_map);
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_reset_table);
>> +
>> +/*
>> + * Returns the number of used IOMMU pages (4K) within
>> + * the same system page (4K or 64K).
>> + *
>> + * syspage_weight_zero is optimized for expected case == 0
>> + * syspage_weight_one is optimized for expected case > 1
>> + * Other case are not used in this file.
>> + */
>> +#if PAGE_SIZE == IOMMU_PAGE_SIZE
>> +
>> +#define syspage_weight_zero(map, offset)	test_bit((map), (offset))
>> +#define syspage_weight_one(map, offset)		test_bit((map), (offset))
>> +
>> +#elif PAGE_SIZE/IOMMU_PAGE_SIZE == 16
>> +
>> +static int syspage_weight_zero(unsigned long *map, unsigned long offset)
>> +{
>> +	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
>> +	return 0xffffUL & (map[BIT_WORD(offset)] >>
>> +			(offset & (BITS_PER_LONG-1)));
>> +}
>
> I would have expected these to be bools and return true if the weight
> matches the value.

My expectation was different but ok, I'll fix :)


> If you replaced 0xffff above w/ this, would you need the #error below?
> (1UL << (PAGE_SIZE/IOMMU_PAGE_SIZE)) - 1)


We have 3 pages size on POWER - 4K, 64K and 16MB. We already handle 4K and 
64K and the 16MB case will require much different approach and I am not 
sure how/when we will add this so I'd keep it as an error.


>> +
>> +static int syspage_weight_one(unsigned long *map, unsigned long offset)
>> +{
>> +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
>> +
>> +	/* Aligns TCE entry number to system page boundary */
>> +	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
>> +
>> +	/* Count used 4K pages */
>> +	while (nbits && (ret < 2)) {
>
> Don't you have a ffs()?  Could also be used for _zero.  Surely there are
> some bitops helpers that could help here even on big endian.  hweight
> really doesn't work?
>
>> +		if (test_bit(offset, map))
>> +			++ret;
>> +
>> +		--nbits;
>> +		++offset;
>> +	}
>> +
>> +	return ret;
>> +}
>> +#else
>> +#error TODO: support other page size
>> +#endif
>> +
>> +static void tce_flush(struct iommu_table *tbl)
>> +{
>> +	/* Flush/invalidate TLB caches if necessary */
>> +	if (ppc_md.tce_flush)
>> +		ppc_md.tce_flush(tbl);
>> +
>> +	/* Make sure updates are seen by hardware */
>> +	mb();
>> +}
>> +
>> +/*
>> + * iommu_clear_tces clears tces and returned the number of system pages
>> + * which it called put_page() on
>> + */
>> +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
>> +		unsigned long pages)
>> +{
>> +	int i, retpages = 0, clr;
>> +	unsigned long oldtce, oldweight;
>> +	struct page *page;
>> +
>> +	for (i = 0; i < pages; ++i, ++entry) {
>> +		if (!test_bit(entry - tbl->it_offset, tbl->it_map))
>> +			continue;
>> +
>> +		oldtce = ppc_md.tce_get(tbl, entry);
>> +		ppc_md.tce_free(tbl, entry, 1);
>> +
>> +		oldweight = syspage_weight_one(tbl->it_map,
>> +				entry - tbl->it_offset);
>> +		clr = __test_and_clear_bit(entry - tbl->it_offset,
>> +				tbl->it_map);
>> +
>> +		if (WARN_ON(!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ))))
>> +			continue;
>> +
>> +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
>> +
>> +		if (WARN_ON(!page))
>> +			continue;
>> +
>> +		if (oldtce & TCE_PCI_WRITE)
>> +			SetPageDirty(page);
>> +
>> +		put_page(page);
>> +
>> +		/* That was the last IOMMU page within the system page */
>> +		if ((oldweight == 1) && clr)
>> +			++retpages;
>> +	}
>> +
>> +	return retpages;
>> +}
>> +
>> +/*
>> + * iommu_clear_tces clears tces and returned the number
>> + * of released system pages
>> + */
>> +long iommu_clear_tces(struct iommu_table *tbl, unsigned long ioba,
>> +		unsigned long size)
>> +{
>> +	int ret;
>> +	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
>> +	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
>> +	struct iommu_pool *pool = get_pool(tbl, entry);
>> +
>> +	if ((size & ~IOMMU_PAGE_MASK) || (ioba & ~IOMMU_PAGE_MASK))
>> +		return -EINVAL;
>> +
>> +	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
>> +			<< IOMMU_PAGE_SHIFT))
>> +		return -EINVAL;
>> +
>> +	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
>> +		return -EINVAL;
>> +
>> +	spin_lock(&(pool->lock));
>> +	ret = clear_tces_nolock(tbl, entry, npages);
>> +	tce_flush(tbl);
>> +	spin_unlock(&(pool->lock));
>> +
>> +	if (ret > 0) {
>> +		lock_acct(-ret);
>> +		return 0;
>> +	}
>> +
>> +	return ret;
>> +}
>> +EXPORT_SYMBOL_GPL(iommu_clear_tces);
>> +
>> +static int put_tce(struct iommu_table *tbl, unsigned long entry,
>> +		uint64_t tce, enum dma_data_direction direction)
>> +{
>> +	int ret;
>> +	struct page *page = NULL;
>> +	unsigned long kva, offset, oldweight;
>> +
>> +	/* Map new TCE */
>> +	offset = tce & IOMMU_PAGE_MASK & ~PAGE_MASK;
>> +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
>> +			direction != DMA_TO_DEVICE, &page);
>> +	if (ret != 1) {
>> +		pr_err("tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
>> +				tce, entry << IOMMU_PAGE_SHIFT, ret);
>> +		return -EFAULT;
>> +	}
>> +
>> +	kva = (unsigned long) page_address(page);
>> +	kva += offset;
>> +
>> +	/* tce_build receives a virtual address */
>> +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
>> +
>> +	/* tce_build() only returns non-zero for transient errors */
>> +	if (unlikely(ret)) {
>> +		pr_err("tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
>> +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
>> +		put_page(page);
>> +		return -EIO;
>> +	}
>> +
>> +	/* Calculate if new system page has been locked */
>> +	oldweight = syspage_weight_zero(tbl->it_map, entry - tbl->it_offset);
>> +	__set_bit(entry - tbl->it_offset, tbl->it_map);
>> +
>> +	return (oldweight == 0) ? 1 : 0;
>> +}
>> +
>> +/*
>> + * iommu_put_tces builds tces and returned the number of actually
>> + * locked system pages
>> + */
>> +long iommu_put_tces(struct iommu_table *tbl, unsigned long ioba,
>> +		uint64_t tce, enum dma_data_direction direction,
>> +		unsigned long size)
>> +{
>> +	int i, ret = 0, retpages = 0;
>> +	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
>> +	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
>> +	struct iommu_pool *pool = get_pool(tbl, entry);
>> +	unsigned long locked, lock_limit;
>> +
>> +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
>> +	BUG_ON(direction == DMA_NONE);
>> +
>> +	if ((size & ~IOMMU_PAGE_MASK) ||
>> +			(ioba & ~IOMMU_PAGE_MASK) ||
>> +			(tce & ~IOMMU_PAGE_MASK))
>> +		return -EINVAL;
>> +
>> +	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
>> +			 << IOMMU_PAGE_SHIFT))
>> +		return -EINVAL;
>> +
>> +	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
>> +		return -EINVAL;
>> +
>> +	/* Account for locked pages */
>> +	locked = current->mm->locked_vm +
>> +		(_ALIGN_UP(size, PAGE_SIZE) >> PAGE_SHIFT);
>
> Looks like we just over penalize upfront and correct when mapped, that's
> better, but not great.

What would be great? :)


>> +	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
>> +	if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
>> +		pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
>> +				rlimit(RLIMIT_MEMLOCK));
>> +		return -ENOMEM;
>> +	}
>> +
>> +	spin_lock(&(pool->lock));
>> +
>> +	/* Check if any is in use */
>> +	for (i = 0; i < npages; ++i) {
>> +		if (test_bit(entry + i - tbl->it_offset, tbl->it_map)) {
>> +			spin_unlock(&(pool->lock));
>> +			return -EBUSY;
>> +		}
>> +	}
>> +
>> +	/* Put tces to the table */
>> +	for (i = 0; (i < npages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
>> +		ret = put_tce(tbl, entry + i, tce, direction);
>> +		if (ret == 1)
>> +			++retpages;
>> +	}
>> +
>> +	/*
>> +	 * If failed, release locked pages, otherwise return the number
>> +	 * of locked system pages
>> +	 */
>> +	if (ret < 0) {
>> +		clear_tces_nolock(tbl, entry, i);
>> +	} else {
>> +		if (retpages)
>> +			lock_acct(retpages);
>> +		ret = 0;
>> +	}
>
> Bug, if it fails we clear, which decrements our locked pages, but we
> haven't incremented them yet.  Thanks,


static clear_tces_nolock does not touch the counter, extern 
iommu_clear_tces does or I missed your point.


-- 
Alexey

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Benjamin Herrenschmidt]

On Wed, 2012-12-12 at 16:30 -0700, Alex Williamson wrote:

> Locked page accounting in this version is very, very broken.  How do
> powerpc folks feel about seemingly generic kernel iommu interfaces
> messing with the current task mm?  Besides that, more problems below...

Not good at all :-)

I don't understand tho ... H_PUT_TCE calls should be in the qemu context
(or the guest) as current at the point of the call, so everything should
be accounted fine on the *current* task when those calls occur, what's
the point of the work queue Alexey ?

This code looks horribly complicated ... where does it come from ?

> > +/*
> > + * iommu_reset_table is called when it started/stopped being used.
> > + *
> > + * restore==true says to bring the iommu_table into the state as it was
> > + * before being used by VFIO.
> > + */
> > +void iommu_reset_table(struct iommu_table *tbl, bool restore)
> > +{
> > +	/* Page#0 is marked as used in iommu_init_table, so we clear it... */
> > +	if (!restore && (tbl->it_offset == 0))
> > +		clear_bit(0, tbl->it_map);
> > +
> > +	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
> 
> This does locked page accounting and unpins pages, even on startup when
> the pages aren't necessarily pinned or accounted against the current
> process.

Not sure what you mean Alex, and not sure either what Alexey
implementation actually does but indeed, pages inside an iommu table
that was used by the host don't have their refcount elevated by the fact
that they are there.

So when taking ownership of an iommu for vfio, you probably need to FAIL
if any page is already mapped. Only once you know the iommu is clear for
use, then you can start populating it and account for anything you put
in it (and de-account anything you remove from it when cleaning things
up).

> > +
> > +	/* ... or restore  */
> > +	if (restore && (tbl->it_offset == 0))
> > +		set_bit(0, tbl->it_map);
> > +}
> > +EXPORT_SYMBOL_GPL(iommu_reset_table);
> > +
> > +/*
> > + * Returns the number of used IOMMU pages (4K) within
> > + * the same system page (4K or 64K).
> > + *
> > + * syspage_weight_zero is optimized for expected case == 0
> > + * syspage_weight_one is optimized for expected case > 1
> > + * Other case are not used in this file.
> > + */
> > +#if PAGE_SIZE == IOMMU_PAGE_SIZE
> > +
> > +#define syspage_weight_zero(map, offset)	test_bit((map), (offset))
> > +#define syspage_weight_one(map, offset)		test_bit((map), (offset))
> > +
> > +#elif PAGE_SIZE/IOMMU_PAGE_SIZE == 16
> > +
> > +static int syspage_weight_zero(unsigned long *map, unsigned long offset)
> > +{
> > +	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> > +	return 0xffffUL & (map[BIT_WORD(offset)] >>
> > +			(offset & (BITS_PER_LONG-1)));
> > +}
> 
> I would have expected these to be bools and return true if the weight
> matches the value.

What is that business anyway ? It's very obscure.

> If you replaced 0xffff above w/ this, would you need the #error below?
> 
> (1UL << (PAGE_SIZE/IOMMU_PAGE_SIZE)) - 1)
> 
> > +
> > +static int syspage_weight_one(unsigned long *map, unsigned long offset)
> > +{
> > +	int ret = 0, nbits = PAGE_SIZE/IOMMU_PAGE_SIZE;
> > +
> > +	/* Aligns TCE entry number to system page boundary */
> > +	offset &= PAGE_MASK >> IOMMU_PAGE_SHIFT;
> > +
> > +	/* Count used 4K pages */
> > +	while (nbits && (ret < 2)) {
> 
> Don't you have a ffs()?  Could also be used for _zero.  Surely there are
> some bitops helpers that could help here even on big endian.  hweight
> really doesn't work?
> 
> > +		if (test_bit(offset, map))
> > +			++ret;
> > +
> > +		--nbits;
> > +		++offset;
> > +	}
> > +
> > +	return ret;
> > +}
> > +#else
> > +#error TODO: support other page size
> > +#endif

What combinations do you support ?

> > +static void tce_flush(struct iommu_table *tbl)
> > +{
> > +	/* Flush/invalidate TLB caches if necessary */
> > +	if (ppc_md.tce_flush)
> > +		ppc_md.tce_flush(tbl);
> > +
> > +	/* Make sure updates are seen by hardware */
> > +	mb();
> > +}
>> +
> > +/*
> > + * iommu_clear_tces clears tces and returned the number of system pages
> > + * which it called put_page() on
> > + */
> > +static long clear_tces_nolock(struct iommu_table *tbl, unsigned long entry,
> > +		unsigned long pages)
> > +{
> > +	int i, retpages = 0, clr;
> > +	unsigned long oldtce, oldweight;
> > +	struct page *page;
> > +
> > +	for (i = 0; i < pages; ++i, ++entry) {
> > +		if (!test_bit(entry - tbl->it_offset, tbl->it_map))
> > +			continue;
> > +
> > +		oldtce = ppc_md.tce_get(tbl, entry);
> > +		ppc_md.tce_free(tbl, entry, 1);
> > +
> > +		oldweight = syspage_weight_one(tbl->it_map,
> > +				entry - tbl->it_offset);
> > +		clr = __test_and_clear_bit(entry - tbl->it_offset,
> > +				tbl->it_map);
> > +
> > +		if (WARN_ON(!(oldtce & (TCE_PCI_WRITE | TCE_PCI_READ))))
> > +			continue;
> > +
> > +		page = pfn_to_page(oldtce >> PAGE_SHIFT);
> > +
> > +		if (WARN_ON(!page))
> > +			continue;
> > +
> > +		if (oldtce & TCE_PCI_WRITE)
> > +			SetPageDirty(page);
> > +
> > +		put_page(page);
> > +
> > +		/* That was the last IOMMU page within the system page */
> > +		if ((oldweight == 1) && clr)
> > +			++retpages;
> > +	}
> > +
> > +	return retpages;
> > +}
> > +
> > +/*
> > + * iommu_clear_tces clears tces and returned the number
> > + * of released system pages
> > + */
> > +long iommu_clear_tces(struct iommu_table *tbl, unsigned long ioba,
> > +		unsigned long size)
> > +{
> > +	int ret;
> > +	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
> > +	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
> > +	struct iommu_pool *pool = get_pool(tbl, entry);
> > +
> > +	if ((size & ~IOMMU_PAGE_MASK) || (ioba & ~IOMMU_PAGE_MASK))
> > +		return -EINVAL;
> > +
> > +	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
> > +			<< IOMMU_PAGE_SHIFT))
> > +		return -EINVAL;
> > +
> > +	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> > +		return -EINVAL;
> > +
> > +	spin_lock(&(pool->lock));
> > +	ret = clear_tces_nolock(tbl, entry, npages);
> > +	tce_flush(tbl);
> > +	spin_unlock(&(pool->lock));

Why are you messing with the pools and their locks ? These are only
relevant for the in-kernel use of the table. The table should be locked
out of kernel use when given to vfio (we could add a flag to make any
kernel dma mapping attempt to fail).

> > +	if (ret > 0) {
> > +		lock_acct(-ret);
> > +		return 0;
> > +	}
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(iommu_clear_tces);
> > +
> > +static int put_tce(struct iommu_table *tbl, unsigned long entry,
> > +		uint64_t tce, enum dma_data_direction direction)
> > +{
> > +	int ret;
> > +	struct page *page = NULL;
> > +	unsigned long kva, offset, oldweight;
> > +
> > +	/* Map new TCE */
> > +	offset = tce & IOMMU_PAGE_MASK & ~PAGE_MASK;
> > +	ret = get_user_pages_fast(tce & PAGE_MASK, 1,
> > +			direction != DMA_TO_DEVICE, &page);
> > +	if (ret != 1) {
> > +		pr_err("tce_vfio: get_user_pages_fast failed tce=%llx ioba=%lx ret=%d\n",
> > +				tce, entry << IOMMU_PAGE_SHIFT, ret);
> > +		return -EFAULT;
> > +	}
> > +
> > +	kva = (unsigned long) page_address(page);
> > +	kva += offset;
> > +
> > +	/* tce_build receives a virtual address */
> > +	ret = ppc_md.tce_build(tbl, entry, 1, kva, direction, NULL);
> > +
> > +	/* tce_build() only returns non-zero for transient errors */
> > +	if (unlikely(ret)) {
> > +		pr_err("tce_vfio: tce_put failed on tce=%llx ioba=%lx kva=%lx ret=%d\n",
> > +				tce, entry << IOMMU_PAGE_SHIFT, kva, ret);
> > +		put_page(page);
> > +		return -EIO;
> > +	}
> > +
> > +	/* Calculate if new system page has been locked */
> > +	oldweight = syspage_weight_zero(tbl->it_map, entry - tbl->it_offset);
> > +	__set_bit(entry - tbl->it_offset, tbl->it_map);
> > +
> > +	return (oldweight == 0) ? 1 : 0;
> > +}
> > +
> > +/*
> > + * iommu_put_tces builds tces and returned the number of actually
> > + * locked system pages
> > + */
> > +long iommu_put_tces(struct iommu_table *tbl, unsigned long ioba,
> > +		uint64_t tce, enum dma_data_direction direction,
> > +		unsigned long size)
> > +{
> > +	int i, ret = 0, retpages = 0;
> > +	unsigned long entry = ioba >> IOMMU_PAGE_SHIFT;
> > +	unsigned long npages = size >> IOMMU_PAGE_SHIFT;
> > +	struct iommu_pool *pool = get_pool(tbl, entry);
> > +	unsigned long locked, lock_limit;
> > +
> > +	BUILD_BUG_ON(PAGE_SIZE < IOMMU_PAGE_SIZE);
> > +	BUG_ON(direction == DMA_NONE);
> > +
> > +	if ((size & ~IOMMU_PAGE_MASK) ||
> > +			(ioba & ~IOMMU_PAGE_MASK) ||
> > +			(tce & ~IOMMU_PAGE_MASK))
> > +		return -EINVAL;
> > +
> > +	if ((ioba + size) > ((tbl->it_offset + tbl->it_size)
> > +			 << IOMMU_PAGE_SHIFT))
> > +		return -EINVAL;
> > +
> > +	if (ioba < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> > +		return -EINVAL;
> > +
> > +	/* Account for locked pages */
> > +	locked = current->mm->locked_vm +
> > +		(_ALIGN_UP(size, PAGE_SIZE) >> PAGE_SHIFT);
> 
> Looks like we just over penalize upfront and correct when mapped, that's
> better, but not great.
> 
> > +	lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
> > +	if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
> > +		pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
> > +				rlimit(RLIMIT_MEMLOCK));
> > +		return -ENOMEM;
> > +	}
> > +
> > +	spin_lock(&(pool->lock));
> > +
> > +	/* Check if any is in use */
> > +	for (i = 0; i < npages; ++i) {
> > +		if (test_bit(entry + i - tbl->it_offset, tbl->it_map)) {
> > +			spin_unlock(&(pool->lock));
> > +			return -EBUSY;
> > +		}
> > +	}
> > +
> > +	/* Put tces to the table */
> > +	for (i = 0; (i < npages) && (ret >= 0); ++i, tce += IOMMU_PAGE_SIZE) {
> > +		ret = put_tce(tbl, entry + i, tce, direction);
> > +		if (ret == 1)
> > +			++retpages;
> > +	}
> > +
> > +	/*
> > +	 * If failed, release locked pages, otherwise return the number
> > +	 * of locked system pages
> > +	 */
> > +	if (ret < 0) {
> > +		clear_tces_nolock(tbl, entry, i);
> > +	} else {
> > +		if (retpages)
> > +			lock_acct(retpages);
> > +		ret = 0;
> > +	}
> 
> Bug, if it fails we clear, which decrements our locked pages, but we
> haven't incremented them yet.  Thanks,
> 
> Alex
> 
> > +
> > +	tce_flush(tbl);
> > +	spin_unlock(&(pool->lock));
> > +
> > +	return ret;
> > +}
> > +EXPORT_SYMBOL_GPL(iommu_put_tces);
> > +
> > +#endif /* CONFIG_IOMMU_API */
> > diff --git a/arch/powerpc/platforms/powernv/pci.c b/arch/powerpc/platforms/powernv/pci.c
> > index 05205cf..1b970bf 100644
> > --- a/arch/powerpc/platforms/powernv/pci.c
> > +++ b/arch/powerpc/platforms/powernv/pci.c
> > @@ -20,6 +20,7 @@
> >  #include <linux/irq.h>
> >  #include <linux/io.h>
> >  #include <linux/msi.h>
> > +#include <linux/iommu.h>
> >  
> >  #include <asm/sections.h>
> >  #include <asm/io.h>
> > @@ -613,3 +614,136 @@ void __init pnv_pci_init(void)
> >  	ppc_md.teardown_msi_irqs = pnv_teardown_msi_irqs;
> >  #endif
> >  }
> > +
> > +#ifdef CONFIG_IOMMU_API
> > +/*
> > + * IOMMU groups support required by VFIO
> > + */
> > +static int add_device(struct device *dev)
> > +{
> > +	struct iommu_table *tbl;
> > +	int ret = 0;
> > +
> > +	if (WARN_ON(dev->iommu_group)) {
> > +		pr_warn("tce_vfio: device %s is already in iommu group %d, skipping\n",
> > +				dev_name(dev),
> > +				iommu_group_id(dev->iommu_group));
> > +		return -EBUSY;
> > +	}
> > +
> > +	tbl = get_iommu_table_base(dev);
> > +	if (!tbl) {
> > +		pr_debug("tce_vfio: skipping device %s with no tbl\n",
> > +				dev_name(dev));
> > +		return 0;
> > +	}
> > +
> > +	pr_debug("tce_vfio: adding %s to iommu group %d\n",
> > +			dev_name(dev), iommu_group_id(tbl->it_group));
> > +
> > +	ret = iommu_group_add_device(tbl->it_group, dev);
> > +	if (ret < 0)
> > +		pr_err("tce_vfio: %s has not been added, ret=%d\n",
> > +				dev_name(dev), ret);
> > +
> > +	return ret;
> > +}
> > +
> > +static void del_device(struct device *dev)
> > +{
> > +	iommu_group_remove_device(dev);
> > +}
> > +
> > +static int iommu_bus_notifier(struct notifier_block *nb,
> > +			      unsigned long action, void *data)
> > +{
> > +	struct device *dev = data;
> > +
> > +	switch (action) {
> > +	case BUS_NOTIFY_ADD_DEVICE:
> > +		return add_device(dev);
> > +	case BUS_NOTIFY_DEL_DEVICE:
> > +		del_device(dev);
> > +		return 0;
> > +	default:
> > +		return 0;
> > +	}
> > +}
> > +
> > +static struct notifier_block tce_iommu_bus_nb = {
> > +	.notifier_call = iommu_bus_notifier,
> > +};
> > +
> > +static void group_release(void *iommu_data)
> > +{
> > +	struct iommu_table *tbl = iommu_data;
> > +	tbl->it_group = NULL;
> > +}
> > +
> > +static int __init tce_iommu_init(void)
> > +{
> > +	struct pci_dev *pdev = NULL;
> > +	struct iommu_table *tbl;
> > +	struct iommu_group *grp;
> > +
> > +	/* Allocate and initialize IOMMU groups */
> > +	for_each_pci_dev(pdev) {
> > +		tbl = get_iommu_table_base(&pdev->dev);
> > +		if (!tbl)
> > +			continue;
> > +
> > +		/* Skip already initialized */
> > +		if (tbl->it_group)
> > +			continue;
> > +
> > +		grp = iommu_group_alloc();
> > +		if (IS_ERR(grp)) {
> > +			pr_info("tce_vfio: cannot create new IOMMU group, ret=%ld\n",
> > +					PTR_ERR(grp));
> > +			return PTR_ERR(grp);
> > +		}
> > +		tbl->it_group = grp;
> > +		iommu_group_set_iommudata(grp, tbl, group_release);
> > +	}
> > +
> > +	bus_register_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> > +
> > +	/* Add PCI devices to VFIO groups */
> > +	for_each_pci_dev(pdev)
> > +		add_device(&pdev->dev);
> > +
> > +	return 0;
> > +}
> > +
> > +static void __exit tce_iommu_cleanup(void)
> > +{
> > +	struct pci_dev *pdev = NULL;
> > +	struct iommu_table *tbl;
> > +	struct iommu_group *grp = NULL;
> > +
> > +	bus_unregister_notifier(&pci_bus_type, &tce_iommu_bus_nb);
> > +
> > +	/* Delete PCI devices from VFIO groups */
> > +	for_each_pci_dev(pdev)
> > +		del_device(&pdev->dev);
> > +
> > +	/* Release VFIO groups */
> > +	for_each_pci_dev(pdev) {
> > +		tbl = get_iommu_table_base(&pdev->dev);
> > +		if (!tbl)
> > +			continue;
> > +		grp = tbl->it_group;
> > +
> > +		/* Skip (already) uninitialized */
> > +		if (!grp)
> > +			continue;
> > +
> > +		/* Do actual release, group_release() is expected to work */
> > +		iommu_group_put(grp);
> > +		BUG_ON(tbl->it_group);
> > +	}
> > +}
> > +
> > +module_init(tce_iommu_init);
> > +module_exit(tce_iommu_cleanup);
> > +#endif /* CONFIG_IOMMU_API */
> > diff --git a/drivers/iommu/Kconfig b/drivers/iommu/Kconfig
> > index 9f69b56..29d11dc 100644
> > --- a/drivers/iommu/Kconfig
> > +++ b/drivers/iommu/Kconfig
> > @@ -187,4 +187,12 @@ config EXYNOS_IOMMU_DEBUG
> >  
> >  	  Say N unless you need kernel log message for IOMMU debugging
> >  
> > +config SPAPR_TCE_IOMMU
> > +	bool "sPAPR TCE IOMMU Support"
> > +	depends on PPC_POWERNV
> > +	select IOMMU_API
> > +	help
> > +	  Enables bits of IOMMU API required by VFIO. The iommu_ops is
> > +	  still not implemented.
> > +
> >  endif # IOMMU_SUPPORT
> 
> 

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Benjamin Herrenschmidt]

On Wed, 2012-12-12 at 16:30 -0700, Alex Williamson wrote:
> Locked page accounting in this version is very, very broken.  How do
> powerpc folks feel about seemingly generic kernel iommu interfaces
> messing with the current task mm?  Besides that, more problems
> below...

After a second look & thought...

This whole accounting business is fucked. First, we simply can't just
randomly return errors from H_PUT_TCE because the process reached some
rlimit. This is not a proper failure mode. That means that the guest
will probably panic() ... possibly right in the middle of some disk
writeback or god knows what. Not good.

Also the overhead of doing all that crap on every TCE map/unmap is
ridiculous.

Finally, it's just not going to work for real mode which we really want,
since we can't take the mmap-sem in real mode anyway, so unless we
convert that counter to an atomic, we can't do it.

I'd suggest just not bothering, or if you want to bother, check once
when creating a TCE table that the rlimit is enough to bolt as many
pages as can be populated in that table and fail to create *that*. The
failure mode is much better, ie, qemu failing to create a PCI bus due to
insufficient rlimits.

Cheers,
Ben.

Re: [PATCH] vfio powerpc: enabled on powernv platform

[Alex Williamson]

On Thu, 2012-12-13 at 13:57 +1100, Benjamin Herrenschmidt wrote:
> On Wed, 2012-12-12 at 16:30 -0700, Alex Williamson wrote:
> > Locked page accounting in this version is very, very broken.  How do
> > powerpc folks feel about seemingly generic kernel iommu interfaces
> > messing with the current task mm?  Besides that, more problems
> > below...
> 
> After a second look & thought...
> 
> This whole accounting business is fucked. First, we simply can't just
> randomly return errors from H_PUT_TCE because the process reached some
> rlimit. This is not a proper failure mode. That means that the guest
> will probably panic() ... possibly right in the middle of some disk
> writeback or god knows what. Not good.
> 
> Also the overhead of doing all that crap on every TCE map/unmap is
> ridiculous.
> 
> Finally, it's just not going to work for real mode which we really want,
> since we can't take the mmap-sem in real mode anyway, so unless we
> convert that counter to an atomic, we can't do it.
> 
> I'd suggest just not bothering, or if you want to bother, check once
> when creating a TCE table that the rlimit is enough to bolt as many
> pages as can be populated in that table and fail to create *that*. The
> failure mode is much better, ie, qemu failing to create a PCI bus due to
> insufficient rlimits.

I agree, we don't seem to be headed in the right direction.  x86 needs
to track rlimits or else a user can exploit the interface to pin all the
memory in the system.  On power, only the iova window can be pinned, so
it's a fixed amount.  I could see it as granting access to a group
implicitly grants access to pinning the iova window.  We can still make
it more explicit by handling the rlimit accounting upfront.  Thanks,

Alex

[PATCH 2/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

VFIO implements platform independent stuff such as
a PCI driver, BAR access (via read/write on a file descriptor
or direct mapping when possible) and IRQ signaling.

The platform dependent part includes IOMMU initialization
and handling. This patch implements an IOMMU driver for VFIO
which does mapping/unmapping pages for the guest IO and
provides information about DMA window (required by a POWERPC
guest).

The counterpart in QEMU is required to support this functionality.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/vfio/Kconfig                |    6 +
 drivers/vfio/Makefile               |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c |  350 +++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                |   26 +++
 4 files changed, 383 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
index 7cd5dec..b464687 100644
--- a/drivers/vfio/Kconfig
+++ b/drivers/vfio/Kconfig
@@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
 	depends on VFIO
 	default n
 
+config VFIO_IOMMU_SPAPR_TCE
+	tristate
+	depends on VFIO && SPAPR_TCE_IOMMU
+	default n
+
 menuconfig VFIO
 	tristate "VFIO Non-Privileged userspace driver framework"
 	depends on IOMMU_API
 	select VFIO_IOMMU_TYPE1 if X86
+	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
 	help
 	  VFIO provides a framework for secure userspace device drivers.
 	  See Documentation/vfio.txt for more details.
diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
index 2398d4a..72bfabc 100644
--- a/drivers/vfio/Makefile
+++ b/drivers/vfio/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_VFIO) += vfio.o
 obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
+obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
 obj-$(CONFIG_VFIO_PCI) += pci/
diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
new file mode 100644
index 0000000..806ad9f
--- /dev/null
+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
@@ -0,0 +1,350 @@
+/*
+ * VFIO: IOMMU DMA mapping support for TCE on POWER
+ *
+ * Copyright (C) 2012 IBM Corp.  All rights reserved.
+ *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Derived from original vfio_iommu_type1.c:
+ * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
+ *     Author: Alex Williamson <alex.williamson@redhat.com>
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/err.h>
+#include <linux/vfio.h>
+#include <asm/iommu.h>
+
+#define DRIVER_VERSION  "0.1"
+#define DRIVER_AUTHOR   "aik@ozlabs.ru"
+#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group);
+
+/*
+ * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
+ */
+
+/*
+ * This code handles mapping and unmapping of user data buffers
+ * into DMA'ble space using the IOMMU
+ */
+
+#define NPAGE_TO_SIZE(npage)	((size_t)(npage) << PAGE_SHIFT)
+
+struct vwork {
+	struct mm_struct	*mm;
+	long			npage;
+	struct work_struct	work;
+};
+
+/* delayed decrement/increment for locked_vm */
+static void lock_acct_bg(struct work_struct *work)
+{
+	struct vwork *vwork = container_of(work, struct vwork, work);
+	struct mm_struct *mm;
+
+	mm = vwork->mm;
+	down_write(&mm->mmap_sem);
+	mm->locked_vm += vwork->npage;
+	up_write(&mm->mmap_sem);
+	mmput(mm);
+	kfree(vwork);
+}
+
+static void lock_acct(long npage)
+{
+	struct vwork *vwork;
+	struct mm_struct *mm;
+
+	if (!current->mm)
+		return; /* process exited */
+
+	if (down_write_trylock(&current->mm->mmap_sem)) {
+		current->mm->locked_vm += npage;
+		up_write(&current->mm->mmap_sem);
+		return;
+	}
+
+	/*
+	 * Couldn't get mmap_sem lock, so must setup to update
+	 * mm->locked_vm later. If locked_vm were atomic, we
+	 * wouldn't need this silliness
+	 */
+	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
+	if (!vwork)
+		return;
+	mm = get_task_mm(current);
+	if (!mm) {
+		kfree(vwork);
+		return;
+	}
+	INIT_WORK(&vwork->work, lock_acct_bg);
+	vwork->mm = mm;
+	vwork->npage = npage;
+	schedule_work(&vwork->work);
+}
+
+/*
+ * The container descriptor supports only a single group per container.
+ * Required by the API as the container is not supplied with the IOMMU group
+ * at the moment of initialization.
+ */
+struct tce_container {
+	struct mutex lock;
+	struct iommu_table *tbl;
+};
+
+static void *tce_iommu_open(unsigned long arg)
+{
+	struct tce_container *container;
+
+	if (arg != VFIO_SPAPR_TCE_IOMMU) {
+		pr_err("tce_vfio: Wrong IOMMU type\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	container = kzalloc(sizeof(*container), GFP_KERNEL);
+	if (!container)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&container->lock);
+
+	return container;
+}
+
+static void tce_iommu_release(void *iommu_data)
+{
+	struct tce_container *container = iommu_data;
+
+	WARN_ON(container->tbl && !container->tbl->it_group);
+	if (container->tbl && container->tbl->it_group)
+		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
+
+	mutex_destroy(&container->lock);
+
+	kfree(container);
+}
+
+static long tce_iommu_ioctl(void *iommu_data,
+				 unsigned int cmd, unsigned long arg)
+{
+	struct tce_container *container = iommu_data;
+	unsigned long minsz;
+	long ret;
+
+	switch (cmd) {
+	case VFIO_CHECK_EXTENSION: {
+		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
+	}
+	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
+		struct vfio_iommu_spapr_tce_info info;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
+				dma32_window_size);
+
+		if (copy_from_user(&info, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (info.argsz < minsz)
+			return -EINVAL;
+
+		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
+		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
+		info.flags = 0;
+
+		if (copy_to_user((void __user *)arg, &info, minsz))
+			return -EFAULT;
+
+		return 0;
+	}
+	case VFIO_IOMMU_MAP_DMA: {
+		vfio_iommu_spapr_tce_dma_map param;
+		struct iommu_table *tbl = container->tbl;
+		enum dma_data_direction direction;
+		unsigned long locked, lock_limit;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
+				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
+			direction = DMA_BIDIRECTIONAL;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
+			direction = DMA_TO_DEVICE;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
+			direction = DMA_FROM_DEVICE;
+		else
+			return -EINVAL;
+
+		if ((param.size & ~IOMMU_PAGE_MASK) ||
+				(param.iova & ~IOMMU_PAGE_MASK) ||
+				(param.vaddr & ~IOMMU_PAGE_MASK))
+			return -EINVAL;
+
+		if ((param.iova + param.size) >
+				(tbl->it_size << IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		/* Account for locked pages */
+		locked = current->mm->locked_vm +
+			(param.size >> PAGE_SHIFT);
+		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+		if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
+			pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
+					rlimit(RLIMIT_MEMLOCK));
+			return -ENOMEM;
+		}
+
+		ret = iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				param.vaddr, direction,
+				param.size >> IOMMU_PAGE_SHIFT);
+
+		if (ret > 0)
+			lock_acct(ret);
+
+		return ret;
+	}
+	case VFIO_IOMMU_UNMAP_DMA: {
+		vfio_iommu_spapr_tce_dma_unmap param;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.size & ~IOMMU_PAGE_MASK) ||
+				(param.iova & ~IOMMU_PAGE_MASK))
+			return -EINVAL;
+
+		if ((param.iova + param.size) >
+				(tbl->it_size << IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		ret = iommu_clear_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				param.size >> IOMMU_PAGE_SHIFT);
+
+		if (ret > 0)
+			lock_acct(-ret);
+
+		return ret;
+	}
+	default:
+		pr_warn("tce_vfio: unexpected cmd %x\n", cmd);
+	}
+
+	return -ENOTTY;
+}
+
+static int tce_iommu_attach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
+			iommu_group_id(iommu_group), iommu_group);
+	if (container->tbl) {
+		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
+				iommu_group_id(container->tbl->it_group),
+				iommu_group_id(iommu_group));
+		mutex_unlock(&container->lock);
+		return -EBUSY;
+	}
+
+	container->tbl = tbl;
+	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
+	mutex_unlock(&container->lock);
+
+	return 0;
+}
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	if (tbl != container->tbl) {
+		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
+				iommu_group_id(iommu_group),
+				iommu_group_id(tbl->it_group));
+	} else {
+
+		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
+				iommu_group_id(iommu_group), iommu_group);
+
+		iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
+		container->tbl = NULL;
+		/* Restore reserve for page 0 */
+		if (tbl->it_offset == 0)
+			set_bit(0, tbl->it_map);
+
+	}
+	mutex_unlock(&container->lock);
+}
+
+const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
+	.name		= "iommu-vfio-powerpc",
+	.owner		= THIS_MODULE,
+	.open		= tce_iommu_open,
+	.release	= tce_iommu_release,
+	.ioctl		= tce_iommu_ioctl,
+	.attach_group	= tce_iommu_attach_group,
+	.detach_group	= tce_iommu_detach_group,
+};
+
+static int __init tce_iommu_init(void)
+{
+	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+
+MODULE_VERSION(DRIVER_VERSION);
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR(DRIVER_AUTHOR);
+MODULE_DESCRIPTION(DRIVER_DESC);
+
diff --git a/include/linux/vfio.h b/include/linux/vfio.h
index 0a4f180..a12295c 100644
--- a/include/linux/vfio.h
+++ b/include/linux/vfio.h
@@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
 /* Extensions */
 
 #define VFIO_TYPE1_IOMMU		1
+#define VFIO_SPAPR_TCE_IOMMU		2
 
 /*
  * The IOCTL interface is designed for extensibility by embedding the
@@ -442,4 +443,29 @@ struct vfio_iommu_type1_dma_unmap {
 
 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 
+/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
+
+/*
+ * The SPAPR TCE info struct provides the information about the PCI bus
+ * address ranges available for DMA, these values are programmed into
+ * the hardware so the guest has to know that information.
+ *
+ * The IOMMU page size is always 4K.
+ */
+
+struct vfio_iommu_spapr_tce_info {
+	__u32 argsz;
+	__u32 flags;			/* reserved for future use */
+	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
+	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
+};
+
+#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
+
+/* Reuse type1 map/unmap structs as they are the same at the moment */
+typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
+typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
+
+/* ***************************************************************** */
+
 #endif /* VFIO_H */
-- 
1.7.10.4

Re: [PATCH 2/2] vfio powerpc: implemented IOMMU driver for VFIO

[Alex Williamson]

On Mon, 2012-12-03 at 13:52 +1100, Alexey Kardashevskiy wrote:
> VFIO implements platform independent stuff such as
> a PCI driver, BAR access (via read/write on a file descriptor
> or direct mapping when possible) and IRQ signaling.
> 
> The platform dependent part includes IOMMU initialization
> and handling. This patch implements an IOMMU driver for VFIO
> which does mapping/unmapping pages for the guest IO and
> provides information about DMA window (required by a POWERPC
> guest).
> 
> The counterpart in QEMU is required to support this functionality.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  drivers/vfio/Kconfig                |    6 +
>  drivers/vfio/Makefile               |    1 +
>  drivers/vfio/vfio_iommu_spapr_tce.c |  350 +++++++++++++++++++++++++++++++++++
>  include/linux/vfio.h                |   26 +++
>  4 files changed, 383 insertions(+)
>  create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> 
> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> index 7cd5dec..b464687 100644
> --- a/drivers/vfio/Kconfig
> +++ b/drivers/vfio/Kconfig
> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
>  	depends on VFIO
>  	default n
>  
> +config VFIO_IOMMU_SPAPR_TCE
> +	tristate
> +	depends on VFIO && SPAPR_TCE_IOMMU
> +	default n
> +
>  menuconfig VFIO
>  	tristate "VFIO Non-Privileged userspace driver framework"
>  	depends on IOMMU_API
>  	select VFIO_IOMMU_TYPE1 if X86
> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>  	help
>  	  VFIO provides a framework for secure userspace device drivers.
>  	  See Documentation/vfio.txt for more details.
> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> index 2398d4a..72bfabc 100644
> --- a/drivers/vfio/Makefile
> +++ b/drivers/vfio/Makefile
> @@ -1,3 +1,4 @@
>  obj-$(CONFIG_VFIO) += vfio.o
>  obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>  obj-$(CONFIG_VFIO_PCI) += pci/
> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> new file mode 100644
> index 0000000..806ad9f
> --- /dev/null
> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> @@ -0,0 +1,350 @@
> +/*
> + * VFIO: IOMMU DMA mapping support for TCE on POWER
> + *
> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + *
> + * Derived from original vfio_iommu_type1.c:
> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> + *     Author: Alex Williamson <alex.williamson@redhat.com>
> + */
> +
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +#include <linux/slab.h>
> +#include <linux/uaccess.h>
> +#include <linux/err.h>
> +#include <linux/vfio.h>
> +#include <asm/iommu.h>
> +
> +#define DRIVER_VERSION  "0.1"
> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group);
> +
> +/*
> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> + */
> +
> +/*
> + * This code handles mapping and unmapping of user data buffers
> + * into DMA'ble space using the IOMMU
> + */
> +
> +#define NPAGE_TO_SIZE(npage)	((size_t)(npage) << PAGE_SHIFT)
> +
> +struct vwork {
> +	struct mm_struct	*mm;
> +	long			npage;
> +	struct work_struct	work;
> +};
> +
> +/* delayed decrement/increment for locked_vm */
> +static void lock_acct_bg(struct work_struct *work)
> +{
> +	struct vwork *vwork = container_of(work, struct vwork, work);
> +	struct mm_struct *mm;
> +
> +	mm = vwork->mm;
> +	down_write(&mm->mmap_sem);
> +	mm->locked_vm += vwork->npage;
> +	up_write(&mm->mmap_sem);
> +	mmput(mm);
> +	kfree(vwork);
> +}
> +
> +static void lock_acct(long npage)
> +{
> +	struct vwork *vwork;
> +	struct mm_struct *mm;
> +
> +	if (!current->mm)
> +		return; /* process exited */
> +
> +	if (down_write_trylock(&current->mm->mmap_sem)) {
> +		current->mm->locked_vm += npage;
> +		up_write(&current->mm->mmap_sem);
> +		return;
> +	}
> +
> +	/*
> +	 * Couldn't get mmap_sem lock, so must setup to update
> +	 * mm->locked_vm later. If locked_vm were atomic, we
> +	 * wouldn't need this silliness
> +	 */
> +	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
> +	if (!vwork)
> +		return;
> +	mm = get_task_mm(current);
> +	if (!mm) {
> +		kfree(vwork);
> +		return;
> +	}
> +	INIT_WORK(&vwork->work, lock_acct_bg);
> +	vwork->mm = mm;
> +	vwork->npage = npage;
> +	schedule_work(&vwork->work);
> +}
> +
> +/*
> + * The container descriptor supports only a single group per container.
> + * Required by the API as the container is not supplied with the IOMMU group
> + * at the moment of initialization.
> + */
> +struct tce_container {
> +	struct mutex lock;
> +	struct iommu_table *tbl;
> +};
> +
> +static void *tce_iommu_open(unsigned long arg)
> +{
> +	struct tce_container *container;
> +
> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> +		pr_err("tce_vfio: Wrong IOMMU type\n");
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
> +	if (!container)
> +		return ERR_PTR(-ENOMEM);
> +
> +	mutex_init(&container->lock);
> +
> +	return container;
> +}
> +
> +static void tce_iommu_release(void *iommu_data)
> +{
> +	struct tce_container *container = iommu_data;
> +
> +	WARN_ON(container->tbl && !container->tbl->it_group);
> +	if (container->tbl && container->tbl->it_group)
> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> +
> +	mutex_destroy(&container->lock);
> +
> +	kfree(container);
> +}
> +
> +static long tce_iommu_ioctl(void *iommu_data,
> +				 unsigned int cmd, unsigned long arg)
> +{
> +	struct tce_container *container = iommu_data;
> +	unsigned long minsz;
> +	long ret;
> +
> +	switch (cmd) {
> +	case VFIO_CHECK_EXTENSION: {
> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> +	}

nit, {}s are unnecessary for this case

> +	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> +		struct vfio_iommu_spapr_tce_info info;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> +				dma32_window_size);
> +
> +		if (copy_from_user(&info, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (info.argsz < minsz)
> +			return -EINVAL;
> +
> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> +		info.flags = 0;
> +
> +		if (copy_to_user((void __user *)arg, &info, minsz))
> +			return -EFAULT;
> +
> +		return 0;
> +	}
> +	case VFIO_IOMMU_MAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_map param;
> +		struct iommu_table *tbl = container->tbl;
> +		enum dma_data_direction direction;
> +		unsigned long locked, lock_limit;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
> +			direction = DMA_BIDIRECTIONAL;
> +		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
> +			direction = DMA_TO_DEVICE;
> +		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
> +			direction = DMA_FROM_DEVICE;
> +		else
> +			return -EINVAL;
> +
> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
> +				(param.iova & ~IOMMU_PAGE_MASK) ||
> +				(param.vaddr & ~IOMMU_PAGE_MASK))
> +			return -EINVAL;
> +
> +		if ((param.iova + param.size) >
> +				(tbl->it_size << IOMMU_PAGE_SHIFT))
> +			return -EINVAL;
> +
> +		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> +			return -EINVAL;

This is confusing me, in 1/2 we had:

entry += tbl->it_offset; /* Offset into real TCE table */

Which implies to me that entry is relative to it_offset.  So I'm not
sure how iova can be less than it_offset (it's u64, so it can't be
negative).  If iova is not relative then the iova + size > it_size
doesn't make any sense.  Looks like you have a few extra ()s in these
too.  Same for unmap case below.

> +
> +		/* Account for locked pages */
> +		locked = current->mm->locked_vm +
> +			(param.size >> PAGE_SHIFT);
> +		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
> +		if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
> +			pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
> +					rlimit(RLIMIT_MEMLOCK));
> +			return -ENOMEM;
> +		}
> +
> +		ret = iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				param.vaddr, direction,
> +				param.size >> IOMMU_PAGE_SHIFT);
> +
> +		if (ret > 0)
> +			lock_acct(ret);
> +
> +		return ret;
> +	}
> +	case VFIO_IOMMU_UNMAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_unmap param;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
> +				(param.iova & ~IOMMU_PAGE_MASK))
> +			return -EINVAL;
> +
> +		if ((param.iova + param.size) >
> +				(tbl->it_size << IOMMU_PAGE_SHIFT))
> +			return -EINVAL;
> +
> +		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> +			return -EINVAL;
> +
> +		ret = iommu_clear_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				param.size >> IOMMU_PAGE_SHIFT);
> +
> +		if (ret > 0)
> +			lock_acct(-ret);
> +
> +		return ret;
> +	}
> +	default:
> +		pr_warn("tce_vfio: unexpected cmd %x\n", cmd);
> +	}
> +
> +	return -ENOTTY;
> +}
> +
> +static int tce_iommu_attach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> +			iommu_group_id(iommu_group), iommu_group);
> +	if (container->tbl) {
> +		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
> +				iommu_group_id(container->tbl->it_group),
> +				iommu_group_id(iommu_group));
> +		mutex_unlock(&container->lock);
> +		return -EBUSY;
> +	}
> +
> +	container->tbl = tbl;
> +	iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
> +	mutex_unlock(&container->lock);
> +
> +	return 0;
> +}
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	if (tbl != container->tbl) {
> +		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
> +				iommu_group_id(iommu_group),
> +				iommu_group_id(tbl->it_group));
> +	} else {
> +
> +		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
> +				iommu_group_id(iommu_group), iommu_group);
> +
> +		iommu_clear_tces(tbl, tbl->it_offset, tbl->it_size);
> +		container->tbl = NULL;
> +		/* Restore reserve for page 0 */
> +		if (tbl->it_offset == 0)
> +			set_bit(0, tbl->it_map);
> +
> +	}
> +	mutex_unlock(&container->lock);
> +}
> +
> +const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
> +	.name		= "iommu-vfio-powerpc",
> +	.owner		= THIS_MODULE,
> +	.open		= tce_iommu_open,
> +	.release	= tce_iommu_release,
> +	.ioctl		= tce_iommu_ioctl,
> +	.attach_group	= tce_iommu_attach_group,
> +	.detach_group	= tce_iommu_detach_group,
> +};
> +
> +static int __init tce_iommu_init(void)
> +{
> +	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +
> +MODULE_VERSION(DRIVER_VERSION);
> +MODULE_LICENSE("GPL v2");
> +MODULE_AUTHOR(DRIVER_AUTHOR);
> +MODULE_DESCRIPTION(DRIVER_DESC);
> +
> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
> index 0a4f180..a12295c 100644
> --- a/include/linux/vfio.h
> +++ b/include/linux/vfio.h
> @@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
>  /* Extensions */
>  
>  #define VFIO_TYPE1_IOMMU		1
> +#define VFIO_SPAPR_TCE_IOMMU		2
>  
>  /*
>   * The IOCTL interface is designed for extensibility by embedding the
> @@ -442,4 +443,29 @@ struct vfio_iommu_type1_dma_unmap {
>  
>  #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
>  
> +/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
> +
> +/*
> + * The SPAPR TCE info struct provides the information about the PCI bus
> + * address ranges available for DMA, these values are programmed into
> + * the hardware so the guest has to know that information.
> + *
> + * The IOMMU page size is always 4K.
> + */
> +
> +struct vfio_iommu_spapr_tce_info {
> +	__u32 argsz;
> +	__u32 flags;			/* reserved for future use */
> +	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
> +	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
> +};
> +
> +#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
> +
> +/* Reuse type1 map/unmap structs as they are the same at the moment */
> +typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
> +typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;

One or both of us are still confused whether iova passed via these
structures are absolute or relative, so let's add some documentation for
that.  Thanks,

Alex

> +
> +/* ***************************************************************** */
> +
>  #endif /* VFIO_H */

[PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

VFIO implements platform independent stuff such as
a PCI driver, BAR access (via read/write on a file descriptor
or direct mapping when possible) and IRQ signaling.

The platform dependent part includes IOMMU initialization
and handling. This patch implements an IOMMU driver for VFIO
which does mapping/unmapping pages for the guest IO and
provides information about DMA window (required by a POWERPC
guest).

The counterpart in QEMU is required to support this functionality.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/vfio/Kconfig                |    6 +
 drivers/vfio/Makefile               |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c |  348 +++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                |   30 +++
 4 files changed, 385 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
index 7cd5dec..b464687 100644
--- a/drivers/vfio/Kconfig
+++ b/drivers/vfio/Kconfig
@@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
 	depends on VFIO
 	default n
 
+config VFIO_IOMMU_SPAPR_TCE
+	tristate
+	depends on VFIO && SPAPR_TCE_IOMMU
+	default n
+
 menuconfig VFIO
 	tristate "VFIO Non-Privileged userspace driver framework"
 	depends on IOMMU_API
 	select VFIO_IOMMU_TYPE1 if X86
+	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
 	help
 	  VFIO provides a framework for secure userspace device drivers.
 	  See Documentation/vfio.txt for more details.
diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
index 2398d4a..72bfabc 100644
--- a/drivers/vfio/Makefile
+++ b/drivers/vfio/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_VFIO) += vfio.o
 obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
+obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
 obj-$(CONFIG_VFIO_PCI) += pci/
diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
new file mode 100644
index 0000000..b0f81fe
--- /dev/null
+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
@@ -0,0 +1,348 @@
+/*
+ * VFIO: IOMMU DMA mapping support for TCE on POWER
+ *
+ * Copyright (C) 2012 IBM Corp.  All rights reserved.
+ *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Derived from original vfio_iommu_type1.c:
+ * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
+ *     Author: Alex Williamson <alex.williamson@redhat.com>
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/err.h>
+#include <linux/vfio.h>
+#include <asm/iommu.h>
+
+#define DRIVER_VERSION  "0.1"
+#define DRIVER_AUTHOR   "aik@ozlabs.ru"
+#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group);
+
+/*
+ * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
+ */
+
+/*
+ * This code handles mapping and unmapping of user data buffers
+ * into DMA'ble space using the IOMMU
+ */
+
+#define NPAGE_TO_SIZE(npage)	((size_t)(npage) << PAGE_SHIFT)
+
+struct vwork {
+	struct mm_struct	*mm;
+	long			npage;
+	struct work_struct	work;
+};
+
+/* delayed decrement/increment for locked_vm */
+static void lock_acct_bg(struct work_struct *work)
+{
+	struct vwork *vwork = container_of(work, struct vwork, work);
+	struct mm_struct *mm;
+
+	mm = vwork->mm;
+	down_write(&mm->mmap_sem);
+	mm->locked_vm += vwork->npage;
+	up_write(&mm->mmap_sem);
+	mmput(mm);
+	kfree(vwork);
+}
+
+static void lock_acct(long npage)
+{
+	struct vwork *vwork;
+	struct mm_struct *mm;
+
+	if (!current->mm)
+		return; /* process exited */
+
+	if (down_write_trylock(&current->mm->mmap_sem)) {
+		current->mm->locked_vm += npage;
+		up_write(&current->mm->mmap_sem);
+		return;
+	}
+
+	/*
+	 * Couldn't get mmap_sem lock, so must setup to update
+	 * mm->locked_vm later. If locked_vm were atomic, we
+	 * wouldn't need this silliness
+	 */
+	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
+	if (!vwork)
+		return;
+	mm = get_task_mm(current);
+	if (!mm) {
+		kfree(vwork);
+		return;
+	}
+	INIT_WORK(&vwork->work, lock_acct_bg);
+	vwork->mm = mm;
+	vwork->npage = npage;
+	schedule_work(&vwork->work);
+}
+
+/*
+ * The container descriptor supports only a single group per container.
+ * Required by the API as the container is not supplied with the IOMMU group
+ * at the moment of initialization.
+ */
+struct tce_container {
+	struct mutex lock;
+	struct iommu_table *tbl;
+};
+
+static void *tce_iommu_open(unsigned long arg)
+{
+	struct tce_container *container;
+
+	if (arg != VFIO_SPAPR_TCE_IOMMU) {
+		pr_err("tce_vfio: Wrong IOMMU type\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	container = kzalloc(sizeof(*container), GFP_KERNEL);
+	if (!container)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&container->lock);
+
+	return container;
+}
+
+static void tce_iommu_release(void *iommu_data)
+{
+	struct tce_container *container = iommu_data;
+
+	WARN_ON(container->tbl && !container->tbl->it_group);
+	if (container->tbl && container->tbl->it_group)
+		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
+
+	mutex_destroy(&container->lock);
+
+	kfree(container);
+}
+
+static long tce_iommu_ioctl(void *iommu_data,
+				 unsigned int cmd, unsigned long arg)
+{
+	struct tce_container *container = iommu_data;
+	unsigned long minsz;
+	long ret;
+
+	switch (cmd) {
+	case VFIO_CHECK_EXTENSION:
+		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
+
+ 	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
+		struct vfio_iommu_spapr_tce_info info;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
+				dma32_window_size);
+
+		if (copy_from_user(&info, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (info.argsz < minsz)
+			return -EINVAL;
+
+		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
+		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
+		info.flags = 0;
+
+		if (copy_to_user((void __user *)arg, &info, minsz))
+			return -EFAULT;
+
+		return 0;
+	}
+	case VFIO_IOMMU_MAP_DMA: {
+		vfio_iommu_spapr_tce_dma_map param;
+		struct iommu_table *tbl = container->tbl;
+		enum dma_data_direction direction;
+		unsigned long locked, lock_limit;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
+				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
+			direction = DMA_BIDIRECTIONAL;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
+			direction = DMA_TO_DEVICE;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
+			direction = DMA_FROM_DEVICE;
+		else
+			return -EINVAL;
+
+		if ((param.size & ~IOMMU_PAGE_MASK) ||
+				(param.iova & ~IOMMU_PAGE_MASK) ||
+				(param.vaddr & ~IOMMU_PAGE_MASK))
+			return -EINVAL;
+
+		if ((param.iova + param.size) >
+				((tbl->it_offset + tbl->it_size)
+				<< IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		/* Account for locked pages */
+		locked = current->mm->locked_vm +
+			(param.size >> PAGE_SHIFT);
+		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
+		if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
+			pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
+					rlimit(RLIMIT_MEMLOCK));
+			return -ENOMEM;
+		}
+
+		ret = iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				param.vaddr, direction,
+				param.size >> IOMMU_PAGE_SHIFT);
+
+		if (ret > 0)
+			lock_acct(ret);
+
+		return ret;
+	}
+	case VFIO_IOMMU_UNMAP_DMA: {
+		vfio_iommu_spapr_tce_dma_unmap param;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.size & ~IOMMU_PAGE_MASK) ||
+				(param.iova & ~IOMMU_PAGE_MASK))
+			return -EINVAL;
+
+		if ((param.iova + param.size) >
+				((tbl->it_offset + tbl->it_size)
+				<< IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
+			return -EINVAL;
+
+		ret = iommu_clear_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
+				param.size >> IOMMU_PAGE_SHIFT);
+
+		if (ret > 0)
+			lock_acct(-ret);
+
+		return ret;
+	}
+	default:
+		pr_warn("tce_vfio: unexpected cmd %x\n", cmd);
+	}
+
+	return -ENOTTY;
+}
+
+static int tce_iommu_attach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
+			iommu_group_id(iommu_group), iommu_group);
+	if (container->tbl) {
+		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
+				iommu_group_id(container->tbl->it_group),
+				iommu_group_id(iommu_group));
+		mutex_unlock(&container->lock);
+		return -EBUSY;
+	}
+
+	container->tbl = tbl;
+	iommu_reset_table(tbl, false);
+	mutex_unlock(&container->lock);
+
+	return 0;
+}
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	if (tbl != container->tbl) {
+		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
+				iommu_group_id(iommu_group),
+				iommu_group_id(tbl->it_group));
+	} else {
+
+		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
+				iommu_group_id(iommu_group), iommu_group);
+
+		container->tbl = NULL;
+		iommu_reset_table(tbl, true);
+	}
+	mutex_unlock(&container->lock);
+}
+
+const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
+	.name		= "iommu-vfio-powerpc",
+	.owner		= THIS_MODULE,
+	.open		= tce_iommu_open,
+	.release	= tce_iommu_release,
+	.ioctl		= tce_iommu_ioctl,
+	.attach_group	= tce_iommu_attach_group,
+	.detach_group	= tce_iommu_detach_group,
+};
+
+static int __init tce_iommu_init(void)
+{
+	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+
+MODULE_VERSION(DRIVER_VERSION);
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR(DRIVER_AUTHOR);
+MODULE_DESCRIPTION(DRIVER_DESC);
+
diff --git a/include/linux/vfio.h b/include/linux/vfio.h
index 0a4f180..23cfc79 100644
--- a/include/linux/vfio.h
+++ b/include/linux/vfio.h
@@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
 /* Extensions */
 
 #define VFIO_TYPE1_IOMMU		1
+#define VFIO_SPAPR_TCE_IOMMU		2
 
 /*
  * The IOCTL interface is designed for extensibility by embedding the
@@ -442,4 +443,33 @@ struct vfio_iommu_type1_dma_unmap {
 
 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 
+/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
+
+/*
+ * The SPAPR TCE info struct provides the information about the PCI bus
+ * address ranges available for DMA, these values are programmed into
+ * the hardware so the guest has to know that information.
+ *
+ * The DMA 32 bit window start is an absolute PCI bus address.
+ * The IOVA address passed via map/unmap ioctls are absolute PCI bus
+ * addresses too so the window works as a filter rather than an offset
+ * for IOVA addresses.
+ *
+ * At the momen the IOMMU page size is 4K.
+ */
+struct vfio_iommu_spapr_tce_info {
+	__u32 argsz;
+	__u32 flags;			/* reserved for future use */
+	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
+	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
+};
+
+#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
+
+/* Reuse type1 map/unmap structs as they are the same at the moment */
+typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
+typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
+
+/* ***************************************************************** */
+
 #endif /* VFIO_H */
-- 
1.7.10.4

Re: [PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alex Williamson]

On Fri, 2012-12-07 at 18:34 +1100, Alexey Kardashevskiy wrote:
> VFIO implements platform independent stuff such as
> a PCI driver, BAR access (via read/write on a file descriptor
> or direct mapping when possible) and IRQ signaling.
> 
> The platform dependent part includes IOMMU initialization
> and handling. This patch implements an IOMMU driver for VFIO
> which does mapping/unmapping pages for the guest IO and
> provides information about DMA window (required by a POWERPC
> guest).
> 
> The counterpart in QEMU is required to support this functionality.
> 
> Cc: David Gibson <david@gibson.dropbear.id.au>
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
> ---
>  drivers/vfio/Kconfig                |    6 +
>  drivers/vfio/Makefile               |    1 +
>  drivers/vfio/vfio_iommu_spapr_tce.c |  348 +++++++++++++++++++++++++++++++++++
>  include/linux/vfio.h                |   30 +++
>  4 files changed, 385 insertions(+)
>  create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c
> 
> diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
> index 7cd5dec..b464687 100644
> --- a/drivers/vfio/Kconfig
> +++ b/drivers/vfio/Kconfig
> @@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
>  	depends on VFIO
>  	default n
>  
> +config VFIO_IOMMU_SPAPR_TCE
> +	tristate
> +	depends on VFIO && SPAPR_TCE_IOMMU
> +	default n
> +
>  menuconfig VFIO
>  	tristate "VFIO Non-Privileged userspace driver framework"
>  	depends on IOMMU_API
>  	select VFIO_IOMMU_TYPE1 if X86
> +	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
>  	help
>  	  VFIO provides a framework for secure userspace device drivers.
>  	  See Documentation/vfio.txt for more details.
> diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
> index 2398d4a..72bfabc 100644
> --- a/drivers/vfio/Makefile
> +++ b/drivers/vfio/Makefile
> @@ -1,3 +1,4 @@
>  obj-$(CONFIG_VFIO) += vfio.o
>  obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
> +obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
>  obj-$(CONFIG_VFIO_PCI) += pci/
> diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
> new file mode 100644
> index 0000000..b0f81fe
> --- /dev/null
> +++ b/drivers/vfio/vfio_iommu_spapr_tce.c
> @@ -0,0 +1,348 @@
> +/*
> + * VFIO: IOMMU DMA mapping support for TCE on POWER
> + *
> + * Copyright (C) 2012 IBM Corp.  All rights reserved.
> + *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
> + *
> + * This program is free software; you can redistribute it and/or modify
> + * it under the terms of the GNU General Public License version 2 as
> + * published by the Free Software Foundation.
> + *
> + * Derived from original vfio_iommu_type1.c:
> + * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
> + *     Author: Alex Williamson <alex.williamson@redhat.com>
> + */
> +
> +#include <linux/module.h>
> +#include <linux/pci.h>
> +#include <linux/slab.h>
> +#include <linux/uaccess.h>
> +#include <linux/err.h>
> +#include <linux/vfio.h>
> +#include <asm/iommu.h>
> +
> +#define DRIVER_VERSION  "0.1"
> +#define DRIVER_AUTHOR   "aik@ozlabs.ru"
> +#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group);
> +
> +/*
> + * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
> + */
> +
> +/*
> + * This code handles mapping and unmapping of user data buffers
> + * into DMA'ble space using the IOMMU
> + */
> +
> +#define NPAGE_TO_SIZE(npage)	((size_t)(npage) << PAGE_SHIFT)
> +
> +struct vwork {
> +	struct mm_struct	*mm;
> +	long			npage;
> +	struct work_struct	work;
> +};
> +
> +/* delayed decrement/increment for locked_vm */
> +static void lock_acct_bg(struct work_struct *work)
> +{
> +	struct vwork *vwork = container_of(work, struct vwork, work);
> +	struct mm_struct *mm;
> +
> +	mm = vwork->mm;
> +	down_write(&mm->mmap_sem);
> +	mm->locked_vm += vwork->npage;
> +	up_write(&mm->mmap_sem);
> +	mmput(mm);
> +	kfree(vwork);
> +}
> +
> +static void lock_acct(long npage)
> +{
> +	struct vwork *vwork;
> +	struct mm_struct *mm;
> +
> +	if (!current->mm)
> +		return; /* process exited */
> +
> +	if (down_write_trylock(&current->mm->mmap_sem)) {
> +		current->mm->locked_vm += npage;
> +		up_write(&current->mm->mmap_sem);
> +		return;
> +	}
> +
> +	/*
> +	 * Couldn't get mmap_sem lock, so must setup to update
> +	 * mm->locked_vm later. If locked_vm were atomic, we
> +	 * wouldn't need this silliness
> +	 */
> +	vwork = kmalloc(sizeof(struct vwork), GFP_KERNEL);
> +	if (!vwork)
> +		return;
> +	mm = get_task_mm(current);
> +	if (!mm) {
> +		kfree(vwork);
> +		return;
> +	}
> +	INIT_WORK(&vwork->work, lock_acct_bg);
> +	vwork->mm = mm;
> +	vwork->npage = npage;
> +	schedule_work(&vwork->work);
> +}
> +
> +/*
> + * The container descriptor supports only a single group per container.
> + * Required by the API as the container is not supplied with the IOMMU group
> + * at the moment of initialization.
> + */
> +struct tce_container {
> +	struct mutex lock;
> +	struct iommu_table *tbl;
> +};
> +
> +static void *tce_iommu_open(unsigned long arg)
> +{
> +	struct tce_container *container;
> +
> +	if (arg != VFIO_SPAPR_TCE_IOMMU) {
> +		pr_err("tce_vfio: Wrong IOMMU type\n");
> +		return ERR_PTR(-EINVAL);
> +	}
> +
> +	container = kzalloc(sizeof(*container), GFP_KERNEL);
> +	if (!container)
> +		return ERR_PTR(-ENOMEM);
> +
> +	mutex_init(&container->lock);
> +
> +	return container;
> +}
> +
> +static void tce_iommu_release(void *iommu_data)
> +{
> +	struct tce_container *container = iommu_data;
> +
> +	WARN_ON(container->tbl && !container->tbl->it_group);
> +	if (container->tbl && container->tbl->it_group)
> +		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
> +
> +	mutex_destroy(&container->lock);
> +
> +	kfree(container);
> +}
> +
> +static long tce_iommu_ioctl(void *iommu_data,
> +				 unsigned int cmd, unsigned long arg)
> +{
> +	struct tce_container *container = iommu_data;
> +	unsigned long minsz;
> +	long ret;
> +
> +	switch (cmd) {
> +	case VFIO_CHECK_EXTENSION:
> +		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
> +
> + 	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
> +		struct vfio_iommu_spapr_tce_info info;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
> +				dma32_window_size);
> +
> +		if (copy_from_user(&info, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (info.argsz < minsz)
> +			return -EINVAL;
> +
> +		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
> +		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
> +		info.flags = 0;
> +
> +		if (copy_to_user((void __user *)arg, &info, minsz))
> +			return -EFAULT;
> +
> +		return 0;
> +	}
> +	case VFIO_IOMMU_MAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_map param;
> +		struct iommu_table *tbl = container->tbl;
> +		enum dma_data_direction direction;
> +		unsigned long locked, lock_limit;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
> +			direction = DMA_BIDIRECTIONAL;
> +		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
> +			direction = DMA_TO_DEVICE;
> +		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
> +			direction = DMA_FROM_DEVICE;
> +		else
> +			return -EINVAL;

flags needs to be sanitized too.  Return EINVAL if any unknown bit is
set or else sloppy users may make it very difficult to make use of those
flag bits later.

> +
> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
> +				(param.iova & ~IOMMU_PAGE_MASK) ||
> +				(param.vaddr & ~IOMMU_PAGE_MASK))
> +			return -EINVAL;
> +
> +		if ((param.iova + param.size) >
> +				((tbl->it_offset + tbl->it_size)
> +				<< IOMMU_PAGE_SHIFT))
> +			return -EINVAL;
> +
> +		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> +			return -EINVAL;
> +
> +		/* Account for locked pages */
> +		locked = current->mm->locked_vm +
> +			(param.size >> PAGE_SHIFT);

If a user were to map only IOMMU_PAGE_SIZE entries on a 64k PAGE_SIZE
system, they'd circumvent this test.

> +		lock_limit = rlimit(RLIMIT_MEMLOCK) >> PAGE_SHIFT;
> +		if (locked > lock_limit && !capable(CAP_IPC_LOCK)) {
> +			pr_warn("RLIMIT_MEMLOCK (%ld) exceeded\n",
> +					rlimit(RLIMIT_MEMLOCK));
> +			return -ENOMEM;
> +		}
> +
> +		ret = iommu_put_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				param.vaddr, direction,
> +				param.size >> IOMMU_PAGE_SHIFT);
> +
> +		if (ret > 0)
> +			lock_acct(ret);
> +
> +		return ret;
> +	}
> +	case VFIO_IOMMU_UNMAP_DMA: {
> +		vfio_iommu_spapr_tce_dma_unmap param;
> +		struct iommu_table *tbl = container->tbl;
> +
> +		if (WARN_ON(!tbl))
> +			return -ENXIO;
> +
> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
> +
> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> +			return -EFAULT;
> +
> +		if (param.argsz < minsz)
> +			return -EINVAL;
> +
> +		if ((param.size & ~IOMMU_PAGE_MASK) ||
> +				(param.iova & ~IOMMU_PAGE_MASK))
> +			return -EINVAL;
> +
> +		if ((param.iova + param.size) >
> +				((tbl->it_offset + tbl->it_size)
> +				<< IOMMU_PAGE_SHIFT))
> +			return -EINVAL;
> +
> +		if (param.iova < (tbl->it_offset << IOMMU_PAGE_SHIFT))
> +			return -EINVAL;

Sanitize flags here too.

> +
> +		ret = iommu_clear_tces(tbl, param.iova >> IOMMU_PAGE_SHIFT,
> +				param.size >> IOMMU_PAGE_SHIFT);
> +
> +		if (ret > 0)
> +			lock_acct(-ret);
> +
> +		return ret;
> +	}
> +	default:
> +		pr_warn("tce_vfio: unexpected cmd %x\n", cmd);

This is unnecessary additional future maintenance.  You only support
EXTENSION, INFO, MAP, and UNMAP.  It doesn't matter what else gets added
later.

> +	}
> +
> +	return -ENOTTY;
> +}
> +
> +static int tce_iommu_attach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
> +			iommu_group_id(iommu_group), iommu_group);
> +	if (container->tbl) {
> +		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
> +				iommu_group_id(container->tbl->it_group),
> +				iommu_group_id(iommu_group));
> +		mutex_unlock(&container->lock);
> +		return -EBUSY;
> +	}
> +
> +	container->tbl = tbl;
> +	iommu_reset_table(tbl, false);
> +	mutex_unlock(&container->lock);
> +
> +	return 0;
> +}
> +
> +static void tce_iommu_detach_group(void *iommu_data,
> +		struct iommu_group *iommu_group)
> +{
> +	struct tce_container *container = iommu_data;
> +	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
> +
> +	BUG_ON(!tbl);
> +	mutex_lock(&container->lock);
> +	if (tbl != container->tbl) {
> +		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
> +				iommu_group_id(iommu_group),
> +				iommu_group_id(tbl->it_group));
> +	} else {
> +
> +		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
> +				iommu_group_id(iommu_group), iommu_group);
> +
> +		container->tbl = NULL;
> +		iommu_reset_table(tbl, true);
> +	}
> +	mutex_unlock(&container->lock);
> +}
> +
> +const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
> +	.name		= "iommu-vfio-powerpc",
> +	.owner		= THIS_MODULE,
> +	.open		= tce_iommu_open,
> +	.release	= tce_iommu_release,
> +	.ioctl		= tce_iommu_ioctl,
> +	.attach_group	= tce_iommu_attach_group,
> +	.detach_group	= tce_iommu_detach_group,
> +};
> +
> +static int __init tce_iommu_init(void)
> +{
> +	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +static void __exit tce_iommu_cleanup(void)
> +{
> +	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
> +}
> +
> +module_init(tce_iommu_init);
> +module_exit(tce_iommu_cleanup);
> +
> +MODULE_VERSION(DRIVER_VERSION);
> +MODULE_LICENSE("GPL v2");
> +MODULE_AUTHOR(DRIVER_AUTHOR);
> +MODULE_DESCRIPTION(DRIVER_DESC);
> +
> diff --git a/include/linux/vfio.h b/include/linux/vfio.h
> index 0a4f180..23cfc79 100644
> --- a/include/linux/vfio.h
> +++ b/include/linux/vfio.h
> @@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
>  /* Extensions */
>  
>  #define VFIO_TYPE1_IOMMU		1
> +#define VFIO_SPAPR_TCE_IOMMU		2
>  
>  /*
>   * The IOCTL interface is designed for extensibility by embedding the
> @@ -442,4 +443,33 @@ struct vfio_iommu_type1_dma_unmap {
>  
>  #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
>  
> +/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
> +
> +/*
> + * The SPAPR TCE info struct provides the information about the PCI bus
> + * address ranges available for DMA, these values are programmed into
> + * the hardware so the guest has to know that information.
> + *
> + * The DMA 32 bit window start is an absolute PCI bus address.
> + * The IOVA address passed via map/unmap ioctls are absolute PCI bus
> + * addresses too so the window works as a filter rather than an offset
> + * for IOVA addresses.
> + *
> + * At the momen the IOMMU page size is 4K.

s/momen/moment/

A flag will need to be added if other page sizes are supported, so as
defined here, it's always 4k.  Thanks,

Alex

> + */
> +struct vfio_iommu_spapr_tce_info {
> +	__u32 argsz;
> +	__u32 flags;			/* reserved for future use */
> +	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
> +	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
> +};
> +
> +#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
> +
> +/* Reuse type1 map/unmap structs as they are the same at the moment */
> +typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
> +typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
> +
> +/* ***************************************************************** */
> +
>  #endif /* VFIO_H */

Re: [PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

On 08/12/12 04:01, Alex Williamson wrote:
>> +	case VFIO_IOMMU_MAP_DMA: {
>> +		vfio_iommu_spapr_tce_dma_map param;
>> +		struct iommu_table *tbl = container->tbl;
>> +		enum dma_data_direction direction;
>> +		unsigned long locked, lock_limit;
>> +
>> +		if (WARN_ON(!tbl))
>> +			return -ENXIO;
>> +
>> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
>> +
>> +		if (copy_from_user(&param, (void __user *)arg, minsz))
>> +			return -EFAULT;
>> +
>> +		if (param.argsz < minsz)
>> +			return -EINVAL;
>> +
>> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
>> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
>> +			direction = DMA_BIDIRECTIONAL;
>> +		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
>> +			direction = DMA_TO_DEVICE;
>> +		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
>> +			direction = DMA_FROM_DEVICE;
>> +		else
>> +			return -EINVAL;
>
> flags needs to be sanitized too.  Return EINVAL if any unknown bit is
> set or else sloppy users may make it very difficult to make use of those
> flag bits later.


It already returns -EINVAL on any bit set except READ/WRITE, no?


-- 
Alexey

Re: [PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alex Williamson]

On Wed, 2012-12-12 at 17:59 +1100, Alexey Kardashevskiy wrote:
> On 08/12/12 04:01, Alex Williamson wrote:
> >> +	case VFIO_IOMMU_MAP_DMA: {
> >> +		vfio_iommu_spapr_tce_dma_map param;
> >> +		struct iommu_table *tbl = container->tbl;
> >> +		enum dma_data_direction direction;
> >> +		unsigned long locked, lock_limit;
> >> +
> >> +		if (WARN_ON(!tbl))
> >> +			return -ENXIO;
> >> +
> >> +		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
> >> +
> >> +		if (copy_from_user(&param, (void __user *)arg, minsz))
> >> +			return -EFAULT;
> >> +
> >> +		if (param.argsz < minsz)
> >> +			return -EINVAL;
> >> +
> >> +		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
> >> +				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
> >> +			direction = DMA_BIDIRECTIONAL;
> >> +		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
> >> +			direction = DMA_TO_DEVICE;
> >> +		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
> >> +			direction = DMA_FROM_DEVICE;
> >> +		else
> >> +			return -EINVAL;
> >
> > flags needs to be sanitized too.  Return EINVAL if any unknown bit is
> > set or else sloppy users may make it very difficult to make use of those
> > flag bits later.
> 
> 
> It already returns -EINVAL on any bit set except READ/WRITE, no?

No.  I could pass flags ~0 through there to get a read/write mapping and
cause you problems if you later want to define another bit.  Thanks,

Alex

[PATCH] vfio powerpc: implemented IOMMU driver for VFIO

[Alexey Kardashevskiy]

VFIO implements platform independent stuff such as
a PCI driver, BAR access (via read/write on a file descriptor
or direct mapping when possible) and IRQ signaling.

The platform dependent part includes IOMMU initialization
and handling. This patch implements an IOMMU driver for VFIO
which does mapping/unmapping pages for the guest IO and
provides information about DMA window (required by a POWERPC
guest).

The counterpart in QEMU is required to support this functionality.

Cc: David Gibson <david@gibson.dropbear.id.au>
Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/vfio/Kconfig                |    6 +
 drivers/vfio/Makefile               |    1 +
 drivers/vfio/vfio_iommu_spapr_tce.c |  249 +++++++++++++++++++++++++++++++++++
 include/linux/vfio.h                |   31 +++++
 4 files changed, 287 insertions(+)
 create mode 100644 drivers/vfio/vfio_iommu_spapr_tce.c

diff --git a/drivers/vfio/Kconfig b/drivers/vfio/Kconfig
index 7cd5dec..b464687 100644
--- a/drivers/vfio/Kconfig
+++ b/drivers/vfio/Kconfig
@@ -3,10 +3,16 @@ config VFIO_IOMMU_TYPE1
 	depends on VFIO
 	default n
 
+config VFIO_IOMMU_SPAPR_TCE
+	tristate
+	depends on VFIO && SPAPR_TCE_IOMMU
+	default n
+
 menuconfig VFIO
 	tristate "VFIO Non-Privileged userspace driver framework"
 	depends on IOMMU_API
 	select VFIO_IOMMU_TYPE1 if X86
+	select VFIO_IOMMU_SPAPR_TCE if PPC_POWERNV
 	help
 	  VFIO provides a framework for secure userspace device drivers.
 	  See Documentation/vfio.txt for more details.
diff --git a/drivers/vfio/Makefile b/drivers/vfio/Makefile
index 2398d4a..72bfabc 100644
--- a/drivers/vfio/Makefile
+++ b/drivers/vfio/Makefile
@@ -1,3 +1,4 @@
 obj-$(CONFIG_VFIO) += vfio.o
 obj-$(CONFIG_VFIO_IOMMU_TYPE1) += vfio_iommu_type1.o
+obj-$(CONFIG_VFIO_IOMMU_SPAPR_TCE) += vfio_iommu_spapr_tce.o
 obj-$(CONFIG_VFIO_PCI) += pci/
diff --git a/drivers/vfio/vfio_iommu_spapr_tce.c b/drivers/vfio/vfio_iommu_spapr_tce.c
new file mode 100644
index 0000000..714bf57
--- /dev/null
+++ b/drivers/vfio/vfio_iommu_spapr_tce.c
@@ -0,0 +1,249 @@
+/*
+ * VFIO: IOMMU DMA mapping support for TCE on POWER
+ *
+ * Copyright (C) 2012 IBM Corp.  All rights reserved.
+ *     Author: Alexey Kardashevskiy <aik@ozlabs.ru>
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ *
+ * Derived from original vfio_iommu_type1.c:
+ * Copyright (C) 2012 Red Hat, Inc.  All rights reserved.
+ *     Author: Alex Williamson <alex.williamson@redhat.com>
+ */
+
+#include <linux/module.h>
+#include <linux/pci.h>
+#include <linux/slab.h>
+#include <linux/uaccess.h>
+#include <linux/err.h>
+#include <linux/vfio.h>
+#include <asm/iommu.h>
+
+#define DRIVER_VERSION  "0.1"
+#define DRIVER_AUTHOR   "aik@ozlabs.ru"
+#define DRIVER_DESC     "VFIO IOMMU SPAPR TCE"
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group);
+
+/*
+ * VFIO IOMMU fd for SPAPR_TCE IOMMU implementation
+ *
+ * This code handles mapping and unmapping of user data buffers
+ * into DMA'ble space using the IOMMU
+ */
+
+/*
+ * The container descriptor supports only a single group per container.
+ * Required by the API as the container is not supplied with the IOMMU group
+ * at the moment of initialization.
+ */
+struct tce_container {
+	struct mutex lock;
+	struct iommu_table *tbl;
+};
+
+static void *tce_iommu_open(unsigned long arg)
+{
+	struct tce_container *container;
+
+	if (arg != VFIO_SPAPR_TCE_IOMMU) {
+		pr_err("tce_vfio: Wrong IOMMU type\n");
+		return ERR_PTR(-EINVAL);
+	}
+
+	container = kzalloc(sizeof(*container), GFP_KERNEL);
+	if (!container)
+		return ERR_PTR(-ENOMEM);
+
+	mutex_init(&container->lock);
+
+	return container;
+}
+
+static void tce_iommu_release(void *iommu_data)
+{
+	struct tce_container *container = iommu_data;
+
+	WARN_ON(container->tbl && !container->tbl->it_group);
+	if (container->tbl && container->tbl->it_group)
+		tce_iommu_detach_group(iommu_data, container->tbl->it_group);
+
+	mutex_destroy(&container->lock);
+
+	kfree(container);
+}
+
+static long tce_iommu_ioctl(void *iommu_data,
+				 unsigned int cmd, unsigned long arg)
+{
+	struct tce_container *container = iommu_data;
+	unsigned long minsz;
+	long ret;
+
+	switch (cmd) {
+	case VFIO_CHECK_EXTENSION:
+		return (arg == VFIO_SPAPR_TCE_IOMMU) ? 1 : 0;
+
+ 	case VFIO_IOMMU_SPAPR_TCE_GET_INFO: {
+		struct vfio_iommu_spapr_tce_info info;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(struct vfio_iommu_spapr_tce_info,
+				dma32_window_size);
+
+		if (copy_from_user(&info, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (info.argsz < minsz)
+			return -EINVAL;
+
+		info.dma32_window_start = tbl->it_offset << IOMMU_PAGE_SHIFT;
+		info.dma32_window_size = tbl->it_size << IOMMU_PAGE_SHIFT;
+		info.flags = 0;
+
+		if (copy_to_user((void __user *)arg, &info, minsz))
+			return -EFAULT;
+
+		return 0;
+	}
+	case VFIO_IOMMU_MAP_DMA: {
+		vfio_iommu_spapr_tce_dma_map param;
+		struct iommu_table *tbl = container->tbl;
+		enum dma_data_direction direction;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_map, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		if ((param.flags & VFIO_DMA_MAP_FLAG_READ) &&
+				(param.flags & VFIO_DMA_MAP_FLAG_WRITE))
+			direction = DMA_BIDIRECTIONAL;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_READ)
+			direction = DMA_TO_DEVICE;
+		else if (param.flags & VFIO_DMA_MAP_FLAG_WRITE)
+			direction = DMA_FROM_DEVICE;
+		else
+			return -EINVAL;
+
+		ret = iommu_put_tces(tbl, param.iova, param.vaddr, direction,
+				param.size);
+
+		return ret;
+	}
+	case VFIO_IOMMU_UNMAP_DMA: {
+		vfio_iommu_spapr_tce_dma_unmap param;
+		struct iommu_table *tbl = container->tbl;
+
+		if (WARN_ON(!tbl))
+			return -ENXIO;
+
+		minsz = offsetofend(vfio_iommu_spapr_tce_dma_unmap, size);
+
+		if (copy_from_user(&param, (void __user *)arg, minsz))
+			return -EFAULT;
+
+		if (param.argsz < minsz)
+			return -EINVAL;
+
+		/* No flag is supported now */
+		if (param.flags)
+			return -EINVAL;
+
+		ret = iommu_clear_tces(tbl, param.iova, param.size);
+
+		return ret;
+	}
+	}
+
+	return -ENOTTY;
+}
+
+static int tce_iommu_attach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	pr_debug("tce_vfio: Attaching group #%u to iommu %p\n",
+			iommu_group_id(iommu_group), iommu_group);
+	if (container->tbl) {
+		pr_warn("tce_vfio: Only one group per IOMMU container is allowed, existing id=%d, attaching id=%d\n",
+				iommu_group_id(container->tbl->it_group),
+				iommu_group_id(iommu_group));
+		mutex_unlock(&container->lock);
+		return -EBUSY;
+	}
+
+	container->tbl = tbl;
+	iommu_reset_table(tbl, false);
+	mutex_unlock(&container->lock);
+
+	return 0;
+}
+
+static void tce_iommu_detach_group(void *iommu_data,
+		struct iommu_group *iommu_group)
+{
+	struct tce_container *container = iommu_data;
+	struct iommu_table *tbl = iommu_group_get_iommudata(iommu_group);
+
+	BUG_ON(!tbl);
+	mutex_lock(&container->lock);
+	if (tbl != container->tbl) {
+		pr_warn("tce_vfio: detaching group #%u, expected group is #%u\n",
+				iommu_group_id(iommu_group),
+				iommu_group_id(tbl->it_group));
+	} else {
+
+		pr_debug("tce_vfio: detaching group #%u from iommu %p\n",
+				iommu_group_id(iommu_group), iommu_group);
+
+		container->tbl = NULL;
+		iommu_reset_table(tbl, true);
+	}
+	mutex_unlock(&container->lock);
+}
+
+const struct vfio_iommu_driver_ops tce_iommu_driver_ops = {
+	.name		= "iommu-vfio-powerpc",
+	.owner		= THIS_MODULE,
+	.open		= tce_iommu_open,
+	.release	= tce_iommu_release,
+	.ioctl		= tce_iommu_ioctl,
+	.attach_group	= tce_iommu_attach_group,
+	.detach_group	= tce_iommu_detach_group,
+};
+
+static int __init tce_iommu_init(void)
+{
+	return vfio_register_iommu_driver(&tce_iommu_driver_ops);
+}
+
+static void __exit tce_iommu_cleanup(void)
+{
+	vfio_unregister_iommu_driver(&tce_iommu_driver_ops);
+}
+
+module_init(tce_iommu_init);
+module_exit(tce_iommu_cleanup);
+
+MODULE_VERSION(DRIVER_VERSION);
+MODULE_LICENSE("GPL v2");
+MODULE_AUTHOR(DRIVER_AUTHOR);
+MODULE_DESCRIPTION(DRIVER_DESC);
+
diff --git a/include/linux/vfio.h b/include/linux/vfio.h
index 0a4f180..b97697d 100644
--- a/include/linux/vfio.h
+++ b/include/linux/vfio.h
@@ -99,6 +99,7 @@ extern void vfio_unregister_iommu_driver(
 /* Extensions */
 
 #define VFIO_TYPE1_IOMMU		1
+#define VFIO_SPAPR_TCE_IOMMU		2
 
 /*
  * The IOCTL interface is designed for extensibility by embedding the
@@ -442,4 +443,34 @@ struct vfio_iommu_type1_dma_unmap {
 
 #define VFIO_IOMMU_UNMAP_DMA _IO(VFIO_TYPE, VFIO_BASE + 14)
 
+/* -------- Additional API for SPAPR TCE (Server POWERPC) IOMMU -------- */
+
+/*
+ * The SPAPR TCE info struct provides the information about the PCI bus
+ * address ranges available for DMA, these values are programmed into
+ * the hardware so the guest has to know that information.
+ *
+ * The DMA 32 bit window start is an absolute PCI bus address.
+ * The IOVA address passed via map/unmap ioctls are absolute PCI bus
+ * addresses too so the window works as a filter rather than an offset
+ * for IOVA addresses.
+ *
+ * A flag will need to be added if other page sizes are supported,
+ * so as defined here, it is always 4k.
+ */
+struct vfio_iommu_spapr_tce_info {
+	__u32 argsz;
+	__u32 flags;			/* reserved for future use */
+	__u32 dma32_window_start;	/* 32 bit window start (bytes) */
+	__u32 dma32_window_size;	/* 32 bit window size (bytes) */
+};
+
+#define VFIO_IOMMU_SPAPR_TCE_GET_INFO	_IO(VFIO_TYPE, VFIO_BASE + 12)
+
+/* Reuse type1 map/unmap structs as they are the same at the moment */
+typedef struct vfio_iommu_type1_dma_map vfio_iommu_spapr_tce_dma_map;
+typedef struct vfio_iommu_type1_dma_unmap vfio_iommu_spapr_tce_dma_unmap;
+
+/* ***************************************************************** */
+
 #endif /* VFIO_H */
-- 
1.7.10.4

[PATCH] iommu: moving initialization earlier

[Alexey Kardashevskiy]

The iommu_init() initializes IOMMU internal structures and data
required for the IOMMU API as iommu_group_alloc().
It is registered as a subsys_initcall now.

One of the IOMMU users is going to be a PCI subsystem on POWER.
It discovers new IOMMU tables during the PCI scan so the logical
place to call iommu_group_alloc() is the moment when a new group
is discovered. However PCI scan is done from subsys_initcall hook
as IOMMU does so PCI hook can be (and is) called before the IOMMU one.

The patch moves IOMMU subsystem initialization one step earlier
to make sure that IOMMU is initialized before PCI scan begins.

Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
---
 drivers/iommu/iommu.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/iommu/iommu.c b/drivers/iommu/iommu.c
index ddbdaca..1065a1a 100644
--- a/drivers/iommu/iommu.c
+++ b/drivers/iommu/iommu.c
@@ -861,7 +861,7 @@ static int __init iommu_init(void)
 
 	return 0;
 }
-subsys_initcall(iommu_init);
+arch_initcall(iommu_init);
 
 int iommu_domain_get_attr(struct iommu_domain *domain,
 			  enum iommu_attr attr, void *data)
-- 
1.7.10.4

Re: [PATCH] iommu: moving initialization earlier

[Joerg Roedel]

On Mon, Jan 07, 2013 at 06:51:52PM +1100, Alexey Kardashevskiy wrote:
> The iommu_init() initializes IOMMU internal structures and data
> required for the IOMMU API as iommu_group_alloc().
> It is registered as a subsys_initcall now.
> 
> One of the IOMMU users is going to be a PCI subsystem on POWER.
> It discovers new IOMMU tables during the PCI scan so the logical
> place to call iommu_group_alloc() is the moment when a new group
> is discovered. However PCI scan is done from subsys_initcall hook
> as IOMMU does so PCI hook can be (and is) called before the IOMMU one.
> 
> The patch moves IOMMU subsystem initialization one step earlier
> to make sure that IOMMU is initialized before PCI scan begins.
> 
> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>

Applied, thanks.

Re: [PATCH] iommu: moving initialization earlier

[Shuah Khan]

On Thu, Jan 10, 2013 at 10:09 AM, Joerg Roedel <joro@8bytes.org> wrote:
> On Mon, Jan 07, 2013 at 06:51:52PM +1100, Alexey Kardashevskiy wrote:
>> The iommu_init() initializes IOMMU internal structures and data
>> required for the IOMMU API as iommu_group_alloc().
>> It is registered as a subsys_initcall now.
>>
>> One of the IOMMU users is going to be a PCI subsystem on POWER.
>> It discovers new IOMMU tables during the PCI scan so the logical
>> place to call iommu_group_alloc() is the moment when a new group
>> is discovered. However PCI scan is done from subsys_initcall hook
>> as IOMMU does so PCI hook can be (and is) called before the IOMMU one.
>>
>> The patch moves IOMMU subsystem initialization one step earlier
>> to make sure that IOMMU is initialized before PCI scan begins.
>>
>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>
> Applied, thanks.

Joerg,

Could you please consider this patch for stable releases.

I am currently debugging IO_PAGE_FAULTS on 3.6.11 (happens on all
pre-3.7 releases). I root-caused the reason 3.7 works is because in
3.7 amd iommu driver moving up the early iommu initialization from
irq_remap_ops with the irq remapping feature.

I looked into back-porting a small sub-set of the changes from 3.7. In
the process, I back-ported the following change that fixes
IO_PAGE_FAULTS the problem to some extent.

33f28c59e18d83fd2aeef258d211be66b9b80eb3
[PATCH] iommu/amd: Split device table initialization into irq and
 dma part

My back-port reduced the IO_PAGE_FAULTS, but still happen. I applied
Alexey's patch on top of my back-ported change, and I no longer see
any IO_PAGE_FAULTS. So my second question/request is would you
consider my back-ported patch for stables which I will send out.

Here is the snippet from dmesg from 3.7 and 3.6.11 that illustrates
the change in early initialization during kernel boot process:

Snippet from 3.7:
[    0.009980] Freeing SMP alternatives: 24k freed
[    0.011486] ACPI: Core revision 20120913
[    0.017291] ftrace: allocating 24968 entries in 98 pages
[    0.025792] SK: before iommu_init_flags()
[    1.015392] SK: before iommu_set_device_table()
[    2.004989] SK: before iommu_enable_command_buffer()
[    2.994600] SK: before iommu_enable_event_buffer()
[    3.984469] SK: before iommu_set_exclusion_range()
[    4.974066] SK: before iommu_enable()
[    5.963662] SK: after iommu_flush_all_caches()
[    8.024776] ..TIMER: vector=0x30 apic1=0 pin1=2 apic2=-1 pin2=-1
[    8.064424] smpboot: CPU0: AMD Opteron(tm) Processor 6328 (fam: 15,
model: 02, stepping: 00)

Snippet from 3.6.11:
[    3.305383] PCI: CLS 64 bytes, default 64
[    3.305424] Trying to unpack rootfs image as initramfs...
[    5.137815] Freeing initrd memory: 118756k freed
[    5.169817] SK: before iommu_init_flags()
[    6.159717] SK: before iommu_set_device_table()
[    7.149435] SK: before iommu_enable_command_buffer()
[    8.139149] SK: before iommu_enable_event_buffer()
[    9.128868] SK: before iommu_set_exclusion_range()
[   10.118581] SK: before iommu_enable()
[   11.108460] SK: before iommu_flush_all_caches()
[   13.141141] SK: after iommu_flush_all_caches()
[   15.120755] pci 0000:00:00.2: can't derive routing for PCI INT A
[   15.120819] pci 0000:00:00.2: PCI INT A: no GSI
[   15.120880]

In 3.6, early iommu initialization occurs way later.

-- Shuah

Re: [PATCH] iommu: moving initialization earlier

[Shuah Khan]

On Thu, Jan 10, 2013 at 1:40 PM, Shuah Khan <shuahkhan@gmail.com> wrote:
> On Thu, Jan 10, 2013 at 10:09 AM, Joerg Roedel <joro@8bytes.org> wrote:
>> On Mon, Jan 07, 2013 at 06:51:52PM +1100, Alexey Kardashevskiy wrote:
>>> The iommu_init() initializes IOMMU internal structures and data
>>> required for the IOMMU API as iommu_group_alloc().
>>> It is registered as a subsys_initcall now.
>>>
>>> One of the IOMMU users is going to be a PCI subsystem on POWER.
>>> It discovers new IOMMU tables during the PCI scan so the logical
>>> place to call iommu_group_alloc() is the moment when a new group
>>> is discovered. However PCI scan is done from subsys_initcall hook
>>> as IOMMU does so PCI hook can be (and is) called before the IOMMU one.
>>>
>>> The patch moves IOMMU subsystem initialization one step earlier
>>> to make sure that IOMMU is initialized before PCI scan begins.
>>>
>>> Signed-off-by: Alexey Kardashevskiy <aik@ozlabs.ru>
>>
>> Applied, thanks.
>
> Joerg,
>
> Could you please consider this patch for stable releases.

The patch applies cleanly only on 3.6.y which is EOL. Guess I am out
of luck with this patch.

-- Shuah

-- Shuah

Re: [PATCH] iommu: moving initialization earlier

[Joerg Roedel]

On Thu, Jan 10, 2013 at 01:40:17PM -0700, Shuah Khan wrote:
> I am currently debugging IO_PAGE_FAULTS on 3.6.11 (happens on all
> pre-3.7 releases). I root-caused the reason 3.7 works is because in
> 3.7 amd iommu driver moving up the early iommu initialization from
> irq_remap_ops with the irq remapping feature.

Have you investigated the reason for those IO_PAGE_FAULTS? I guess they
come from the USB controlers and happen between the time the IOMMU is
enabled and the USB controlers are taken over by the Linux kernel from
the BIOS.

But I don't see why this patch can have any impact on the IO_PAGE_FAULTS
you are seeing.


	Joerg

Re: [PATCH] iommu: moving initialization earlier

[Shuah Khan]

On Fri, Jan 11, 2013 at 7:50 AM, Joerg Roedel <joro@8bytes.org> wrote:
> On Thu, Jan 10, 2013 at 01:40:17PM -0700, Shuah Khan wrote:
>> I am currently debugging IO_PAGE_FAULTS on 3.6.11 (happens on all
>> pre-3.7 releases). I root-caused the reason 3.7 works is because in
>> 3.7 amd iommu driver moving up the early iommu initialization from
>> irq_remap_ops with the irq remapping feature.
>
> Have you investigated the reason for those IO_PAGE_FAULTS? I guess they
> come from the USB controlers and happen between the time the IOMMU is
> enabled and the USB controlers are taken over by the Linux kernel from
> the BIOS.
>
> But I don't see why this patch can have any impact on the IO_PAGE_FAULTS
> you are seeing.

USB is one of them and SATA is another. I know with my back-port to
split dma ops initialization and this patch Alexey sent, the problem
goes away on 3.6.11 and I don't see the problem on 3.7. Same system,
and two different boots. I will investigate do more investigation into
the nature of these faults I am seeing and get back to you.

Thanks,
-- Shuah