[PATCH v2 0/3] xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

[PATCH v2 0/3] xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

[Stefano Stabellini]

Hi all,
Xen support in Linux for ARM and ARM64 suffers from lack of support for
multiple mfn to pfn mappings: whenever a frontend grants the same page
multiple times to the backend, the mfn to pfn accounting in
arch/arm/xen/p2m.c fails. The issue has become critical since v3.15,
when xen-netfront/xen-netback switched from grant copies to grant
mappings, therefore causing the issue to happen much more often.

Fixing the mfn to pfn accounting in p2m.c is difficult and expensive,
therefore we are looking for alternative solutions. One idea is avoiding
mfn to pfn conversions altogether. The only code path that needs them is
swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

To avoid mfn to pfn conversions we rely on a second p2m mapping done by
Xen (a separate patch series will be sent for Xen). In Linux we use it
to perform the cache maintenance operations without mfns conversions.


Changes in v2:
- introduce XENFEAT_grant_map_11;
- remeber the ptep corresponding to scratch pages so that we don't need
to calculate it again every time;
- do not acutally unmap the page on xen_mm32_unmap;
- properly account preempt_enable/disable;
- do not check for mfn in xen_add_phys_to_mach_entry.


Stefano Stabellini (3):
      xen/arm: introduce XENFEAT_grant_map_11
      xen/arm: reimplement xen_dma_unmap_page & friends
      xen/arm: remove mach_to_phys rbtree

 arch/arm/include/asm/xen/page-coherent.h |   25 ++--
 arch/arm/include/asm/xen/page.h          |    9 --
 arch/arm/xen/Makefile                    |    2 +-
 arch/arm/xen/enlighten.c                 |    6 +
 arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
 arch/arm/xen/p2m.c                       |   66 +---------
 include/xen/interface/features.h         |    3 +
 7 files changed, 220 insertions(+), 93 deletions(-)
 create mode 100644 arch/arm/xen/mm32.c

[PATCH v2 1/3] xen/arm: introduce XENFEAT_grant_map_11

[Stefano Stabellini]

The flag tells us that the hypervisor maps a grant page to guest
physical address == machine address of the page in addition to the
normal grant mapping address. It is needed to properly issue cache
maintenance operation at the completion of a DMA operation involving a
foreign grant.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
---
 arch/arm/xen/enlighten.c         |    6 ++++++
 include/xen/interface/features.h |    3 +++
 2 files changed, 9 insertions(+)

diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
index b96723e..ee3135a 100644
--- a/arch/arm/xen/enlighten.c
+++ b/arch/arm/xen/enlighten.c
@@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
 	xen_domain_type = XEN_HVM_DOMAIN;
 
 	xen_setup_features();
+
+	if (!xen_feature(XENFEAT_grant_map_11)) {
+		pr_warn("Please upgrade your Xen.\n"
+				"If your platform has any non-coherent DMA devices, they won't work properly.\n");
+	}
+
 	if (xen_feature(XENFEAT_dom0))
 		xen_start_info->flags |= SIF_INITDOMAIN|SIF_PRIVILEGED;
 	else
diff --git a/include/xen/interface/features.h b/include/xen/interface/features.h
index 131a6cc..6517dd5 100644
--- a/include/xen/interface/features.h
+++ b/include/xen/interface/features.h
@@ -53,6 +53,9 @@
 /* operation as Dom0 is supported */
 #define XENFEAT_dom0                      11
 
+/* Xen also maps grant references at pfn = mfn */
+#define XENFEAT_grant_map_11              12
+
 #define XENFEAT_NR_SUBMAPS 1
 
 #endif /* __XEN_PUBLIC_FEATURES_H__ */
-- 
1.7.10.4

[PATCH v2 2/3] xen/arm: reimplement xen_dma_unmap_page & friends

[Stefano Stabellini]

xen_dma_unmap_page, xen_dma_sync_single_for_cpu and
xen_dma_sync_single_for_device are currently implemented by calling into
the corresponding generic ARM implementation of these functions. In
order to do this, firstly the dma_addr_t handle, that on Xen is a
machine address, needs to be translated into a physical address.  The
operation is expensive and inaccurate, given that a single machine
address can correspond to multiple physical addresses in one domain,
because the same page can be granted multiple times by the frontend.

To avoid this problem, we introduce a Xen specific implementation of
xen_dma_unmap_page, xen_dma_sync_single_for_cpu and
xen_dma_sync_single_for_device, that can operate on machine addresses
directly.

The new implementation relies on the fact that the hypervisor creates a
second p2m mapping of any grant pages at physical address == machine
address of the page for dom0. Therefore we can access memory at physical
address == dma_addr_r handle and perform the cache flushing there. Some
cache maintenance operations require a virtual address. Instead of using
ioremap_cache, that is not safe in interrupt context, we allocate a
per-cpu PAGE_KERNEL scratch page and we manually update the pte for it.

arm64 doesn't need cache maintenance operations on unmap for now.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>

---

Changes in v2:
- check for XENFEAT_grant_map_11;
- remeber the ptep corresponding to scratch pages so that we don't need
to calculate it again every time;
- do not acutally unmap the page on xen_mm32_unmap;
- properly account preempt_enable/disable.
---
 arch/arm/include/asm/xen/page-coherent.h |   25 ++--
 arch/arm/xen/Makefile                    |    2 +-
 arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
 3 files changed, 210 insertions(+), 19 deletions(-)
 create mode 100644 arch/arm/xen/mm32.c

diff --git a/arch/arm/include/asm/xen/page-coherent.h b/arch/arm/include/asm/xen/page-coherent.h
index 1109017..e8275ea 100644
--- a/arch/arm/include/asm/xen/page-coherent.h
+++ b/arch/arm/include/asm/xen/page-coherent.h
@@ -26,25 +26,14 @@ static inline void xen_dma_map_page(struct device *hwdev, struct page *page,
 	__generic_dma_ops(hwdev)->map_page(hwdev, page, offset, size, dir, attrs);
 }
 
-static inline void xen_dma_unmap_page(struct device *hwdev, dma_addr_t handle,
+void xen_dma_unmap_page(struct device *hwdev, dma_addr_t handle,
 		size_t size, enum dma_data_direction dir,
-		struct dma_attrs *attrs)
-{
-	if (__generic_dma_ops(hwdev)->unmap_page)
-		__generic_dma_ops(hwdev)->unmap_page(hwdev, handle, size, dir, attrs);
-}
+		struct dma_attrs *attrs);
 
-static inline void xen_dma_sync_single_for_cpu(struct device *hwdev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
-{
-	if (__generic_dma_ops(hwdev)->sync_single_for_cpu)
-		__generic_dma_ops(hwdev)->sync_single_for_cpu(hwdev, handle, size, dir);
-}
+void xen_dma_sync_single_for_cpu(struct device *hwdev,
+		dma_addr_t handle, size_t size, enum dma_data_direction dir);
+
+void xen_dma_sync_single_for_device(struct device *hwdev,
+		dma_addr_t handle, size_t size, enum dma_data_direction dir);
 
-static inline void xen_dma_sync_single_for_device(struct device *hwdev,
-		dma_addr_t handle, size_t size, enum dma_data_direction dir)
-{
-	if (__generic_dma_ops(hwdev)->sync_single_for_device)
-		__generic_dma_ops(hwdev)->sync_single_for_device(hwdev, handle, size, dir);
-}
 #endif /* _ASM_ARM_XEN_PAGE_COHERENT_H */
diff --git a/arch/arm/xen/Makefile b/arch/arm/xen/Makefile
index 1296952..1f85bfe 100644
--- a/arch/arm/xen/Makefile
+++ b/arch/arm/xen/Makefile
@@ -1 +1 @@
-obj-y		:= enlighten.o hypercall.o grant-table.o p2m.o mm.o
+obj-y		:= enlighten.o hypercall.o grant-table.o p2m.o mm.o mm32.o
diff --git a/arch/arm/xen/mm32.c b/arch/arm/xen/mm32.c
new file mode 100644
index 0000000..964d0af
--- /dev/null
+++ b/arch/arm/xen/mm32.c
@@ -0,0 +1,202 @@
+#include <linux/cpu.h>
+#include <linux/dma-mapping.h>
+#include <linux/gfp.h>
+#include <linux/highmem.h>
+
+#include <xen/features.h>
+
+static DEFINE_PER_CPU(unsigned long, xen_mm32_scratch_virt);
+static DEFINE_PER_CPU(pte_t *, xen_mm32_scratch_ptep);
+
+static int alloc_xen_mm32_scratch_page(int cpu)
+{
+	struct page *page;
+	unsigned long virt;
+	pmd_t *pmdp;
+	pte_t *ptep;
+
+	if (per_cpu(xen_mm32_scratch_ptep, cpu) != NULL)
+		return 0;
+
+	page = alloc_page(GFP_KERNEL);
+	if (page == NULL) {
+		pr_warn("Failed to allocate xen_mm32_scratch_page for cpu %d\n", cpu);
+		return -ENOMEM;
+	}
+
+	virt = (unsigned long)__va(page_to_phys(page));
+	pmdp = pmd_offset(pud_offset(pgd_offset_k(virt), virt), virt);
+	ptep = pte_offset_kernel(pmdp, virt);
+
+	per_cpu(xen_mm32_scratch_virt, cpu) = virt;
+	per_cpu(xen_mm32_scratch_ptep, cpu) = ptep;
+
+	return 0;
+}
+
+static int xen_mm32_cpu_notify(struct notifier_block *self,
+				    unsigned long action, void *hcpu)
+{
+	int cpu = (long)hcpu;
+	switch (action) {
+	case CPU_UP_PREPARE:
+		if (alloc_xen_mm32_scratch_page(cpu))
+			return NOTIFY_BAD;
+		break;
+	default:
+		break;
+	}
+	return NOTIFY_OK;
+}
+
+static struct notifier_block xen_mm32_cpu_notifier = {
+	.notifier_call	= xen_mm32_cpu_notify,
+};
+
+static void* xen_mm32_remap_page(dma_addr_t handle)
+{
+	unsigned long virt = get_cpu_var(xen_mm32_scratch_virt);
+	pte_t *ptep = __get_cpu_var(xen_mm32_scratch_ptep);
+
+	*ptep = pfn_pte(handle >> PAGE_SHIFT, PAGE_KERNEL);
+	local_flush_tlb_kernel_page(virt);
+
+	return (void*)virt;
+}
+
+static void xen_mm32_unmap(void *vaddr)
+{
+	put_cpu_var(xen_mm32_scratch_virt);
+}
+
+
+/* functions called by SWIOTLB */
+
+static void dma_cache_maint(dma_addr_t handle, unsigned long offset,
+	size_t size, enum dma_data_direction dir,
+	void (*op)(const void *, size_t, int))
+{
+	unsigned long pfn;
+	size_t left = size;
+
+	pfn = (handle >> PAGE_SHIFT) + offset / PAGE_SIZE;
+	offset %= PAGE_SIZE;
+
+	do {
+		size_t len = left;
+		void *vaddr;
+	
+		if (!pfn_valid(pfn))
+		{
+			/* Cannot map the page, we don't know its physical address.
+			 * Return and hope for the best */
+			if (!xen_feature(XENFEAT_grant_map_11))
+				return;
+			vaddr = xen_mm32_remap_page(handle) + offset;
+			op(vaddr, len, dir);
+			xen_mm32_unmap(vaddr - offset);
+		} else {
+			struct page *page = pfn_to_page(pfn);
+
+			if (PageHighMem(page)) {
+				if (len + offset > PAGE_SIZE)
+					len = PAGE_SIZE - offset;
+
+				if (cache_is_vipt_nonaliasing()) {
+					vaddr = kmap_atomic(page);
+					op(vaddr + offset, len, dir);
+					kunmap_atomic(vaddr);
+				} else {
+					vaddr = kmap_high_get(page);
+					if (vaddr) {
+						op(vaddr + offset, len, dir);
+						kunmap_high(page);
+					}
+				}
+			} else {
+				vaddr = page_address(page) + offset;
+				op(vaddr, len, dir);
+			}
+		}
+
+		offset = 0;
+		pfn++;
+		left -= len;
+	} while (left);
+}
+
+static void __xen_dma_page_dev_to_cpu(struct device *hwdev, dma_addr_t handle,
+		size_t size, enum dma_data_direction dir)
+{
+	/* Cannot use __dma_page_dev_to_cpu because we don't have a
+	 * struct page for handle */
+
+	if (dir == DMA_TO_DEVICE)
+		outer_inv_range(handle, handle + size);
+
+	dma_cache_maint(handle & PAGE_MASK, handle & ~PAGE_MASK, size, dir, dmac_unmap_area);
+}
+
+static void __xen_dma_page_cpu_to_dev(struct device *hwdev, dma_addr_t handle,
+		size_t size, enum dma_data_direction dir)
+{
+
+	dma_cache_maint(handle & PAGE_MASK, handle & ~PAGE_MASK, size, dir, dmac_map_area);
+
+	if (dir == DMA_FROM_DEVICE) {
+		outer_inv_range(handle, handle + size);
+	} else {
+		outer_clean_range(handle, handle + size);
+	}
+}
+
+void xen_dma_unmap_page(struct device *hwdev, dma_addr_t handle,
+		size_t size, enum dma_data_direction dir,
+		struct dma_attrs *attrs)
+
+{
+	if (!__generic_dma_ops(hwdev)->unmap_page)
+		return;
+	if (dma_get_attr(DMA_ATTR_SKIP_CPU_SYNC, attrs))
+		return;
+
+	__xen_dma_page_dev_to_cpu(hwdev, handle, size, dir);
+}
+
+void xen_dma_sync_single_for_cpu(struct device *hwdev,
+		dma_addr_t handle, size_t size, enum dma_data_direction dir)
+{
+	if (!__generic_dma_ops(hwdev)->sync_single_for_cpu)
+		return;
+	__xen_dma_page_dev_to_cpu(hwdev, handle, size, dir);
+}
+
+void xen_dma_sync_single_for_device(struct device *hwdev,
+		dma_addr_t handle, size_t size, enum dma_data_direction dir)
+{
+	if (!__generic_dma_ops(hwdev)->sync_single_for_device)
+		return;
+	__xen_dma_page_cpu_to_dev(hwdev, handle, size, dir);
+}
+
+int __init xen_mm32_init(void)
+{
+	int cpu;
+
+	if (!xen_initial_domain())
+		return 0;
+
+	register_cpu_notifier(&xen_mm32_cpu_notifier);
+	get_online_cpus();
+	for_each_online_cpu(cpu) {
+		if (alloc_xen_mm32_scratch_page(cpu)) {
+			put_online_cpus();
+			unregister_cpu_notifier(&xen_mm32_cpu_notifier);
+			return -ENOMEM;
+		}
+	}
+	put_online_cpus();
+
+	return 0;
+}
+arch_initcall(xen_mm32_init);
-- 
1.7.10.4

[PATCH v2 3/3] xen/arm: remove mach_to_phys rbtree

[Stefano Stabellini]

Remove the rbtree used to keep track of machine to physical mappings:
the frontend can grant the same page multiple times, leading to errors
inserting or removing entries from the mach_to_phys tree.

Linux only needed to know the physical address corresponding to a given
machine address in swiotlb-xen. Now that swiotlb-xen can call the
xen_dma_* functions passing the machine address directly, we can remove
it.

Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>

---

Changes in v2:
- do not check for mfn in xen_add_phys_to_mach_entry.
---
 arch/arm/include/asm/xen/page.h |    9 ------
 arch/arm/xen/p2m.c              |   66 +--------------------------------------
 2 files changed, 1 insertion(+), 74 deletions(-)

diff --git a/arch/arm/include/asm/xen/page.h b/arch/arm/include/asm/xen/page.h
index ded062f..135c24a 100644
--- a/arch/arm/include/asm/xen/page.h
+++ b/arch/arm/include/asm/xen/page.h
@@ -33,7 +33,6 @@ typedef struct xpaddr {
 #define INVALID_P2M_ENTRY      (~0UL)
 
 unsigned long __pfn_to_mfn(unsigned long pfn);
-unsigned long __mfn_to_pfn(unsigned long mfn);
 extern struct rb_root phys_to_mach;
 
 static inline unsigned long pfn_to_mfn(unsigned long pfn)
@@ -51,14 +50,6 @@ static inline unsigned long pfn_to_mfn(unsigned long pfn)
 
 static inline unsigned long mfn_to_pfn(unsigned long mfn)
 {
-	unsigned long pfn;
-
-	if (phys_to_mach.rb_node != NULL) {
-		pfn = __mfn_to_pfn(mfn);
-		if (pfn != INVALID_P2M_ENTRY)
-			return pfn;
-	}
-
 	return mfn;
 }
 
diff --git a/arch/arm/xen/p2m.c b/arch/arm/xen/p2m.c
index 97baf44..0548577 100644
--- a/arch/arm/xen/p2m.c
+++ b/arch/arm/xen/p2m.c
@@ -21,14 +21,12 @@ struct xen_p2m_entry {
 	unsigned long pfn;
 	unsigned long mfn;
 	unsigned long nr_pages;
-	struct rb_node rbnode_mach;
 	struct rb_node rbnode_phys;
 };
 
 static rwlock_t p2m_lock;
 struct rb_root phys_to_mach = RB_ROOT;
 EXPORT_SYMBOL_GPL(phys_to_mach);
-static struct rb_root mach_to_phys = RB_ROOT;
 
 static int xen_add_phys_to_mach_entry(struct xen_p2m_entry *new)
 {
@@ -41,8 +39,6 @@ static int xen_add_phys_to_mach_entry(struct xen_p2m_entry *new)
 		parent = *link;
 		entry = rb_entry(parent, struct xen_p2m_entry, rbnode_phys);
 
-		if (new->mfn == entry->mfn)
-			goto err_out;
 		if (new->pfn == entry->pfn)
 			goto err_out;
 
@@ -88,64 +84,6 @@ unsigned long __pfn_to_mfn(unsigned long pfn)
 }
 EXPORT_SYMBOL_GPL(__pfn_to_mfn);
 
-static int xen_add_mach_to_phys_entry(struct xen_p2m_entry *new)
-{
-	struct rb_node **link = &mach_to_phys.rb_node;
-	struct rb_node *parent = NULL;
-	struct xen_p2m_entry *entry;
-	int rc = 0;
-
-	while (*link) {
-		parent = *link;
-		entry = rb_entry(parent, struct xen_p2m_entry, rbnode_mach);
-
-		if (new->mfn == entry->mfn)
-			goto err_out;
-		if (new->pfn == entry->pfn)
-			goto err_out;
-
-		if (new->mfn < entry->mfn)
-			link = &(*link)->rb_left;
-		else
-			link = &(*link)->rb_right;
-	}
-	rb_link_node(&new->rbnode_mach, parent, link);
-	rb_insert_color(&new->rbnode_mach, &mach_to_phys);
-	goto out;
-
-err_out:
-	rc = -EINVAL;
-	pr_warn("%s: cannot add pfn=%pa -> mfn=%pa: pfn=%pa -> mfn=%pa already exists\n",
-			__func__, &new->pfn, &new->mfn, &entry->pfn, &entry->mfn);
-out:
-	return rc;
-}
-
-unsigned long __mfn_to_pfn(unsigned long mfn)
-{
-	struct rb_node *n = mach_to_phys.rb_node;
-	struct xen_p2m_entry *entry;
-	unsigned long irqflags;
-
-	read_lock_irqsave(&p2m_lock, irqflags);
-	while (n) {
-		entry = rb_entry(n, struct xen_p2m_entry, rbnode_mach);
-		if (entry->mfn <= mfn &&
-				entry->mfn + entry->nr_pages > mfn) {
-			read_unlock_irqrestore(&p2m_lock, irqflags);
-			return entry->pfn + (mfn - entry->mfn);
-		}
-		if (mfn < entry->mfn)
-			n = n->rb_left;
-		else
-			n = n->rb_right;
-	}
-	read_unlock_irqrestore(&p2m_lock, irqflags);
-
-	return INVALID_P2M_ENTRY;
-}
-EXPORT_SYMBOL_GPL(__mfn_to_pfn);
-
 int set_foreign_p2m_mapping(struct gnttab_map_grant_ref *map_ops,
 			    struct gnttab_map_grant_ref *kmap_ops,
 			    struct page **pages, unsigned int count)
@@ -192,7 +130,6 @@ bool __set_phys_to_machine_multi(unsigned long pfn,
 			p2m_entry = rb_entry(n, struct xen_p2m_entry, rbnode_phys);
 			if (p2m_entry->pfn <= pfn &&
 					p2m_entry->pfn + p2m_entry->nr_pages > pfn) {
-				rb_erase(&p2m_entry->rbnode_mach, &mach_to_phys);
 				rb_erase(&p2m_entry->rbnode_phys, &phys_to_mach);
 				write_unlock_irqrestore(&p2m_lock, irqflags);
 				kfree(p2m_entry);
@@ -217,8 +154,7 @@ bool __set_phys_to_machine_multi(unsigned long pfn,
 	p2m_entry->mfn = mfn;
 
 	write_lock_irqsave(&p2m_lock, irqflags);
-	if ((rc = xen_add_phys_to_mach_entry(p2m_entry) < 0) ||
-		(rc = xen_add_mach_to_phys_entry(p2m_entry) < 0)) {
+	if ((rc = xen_add_phys_to_mach_entry(p2m_entry)) < 0) {
 		write_unlock_irqrestore(&p2m_lock, irqflags);
 		return false;
 	}
-- 
1.7.10.4

[PATCH v2 1/3] xen/arm: introduce XENFEAT_grant_map_11

[Ian Campbell]

On Tue, 2014-07-08 at 16:42 +0100, Stefano Stabellini wrote:
> The flag tells us that the hypervisor maps a grant page to guest
> physical address == machine address of the page in addition to the
> normal grant mapping address. It is needed to properly issue cache
> maintenance operation at the completion of a DMA operation involving a
> foreign grant.
> 
> Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> ---
>  arch/arm/xen/enlighten.c         |    6 ++++++
>  include/xen/interface/features.h |    3 +++
>  2 files changed, 9 insertions(+)
> 
> diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
> index b96723e..ee3135a 100644
> --- a/arch/arm/xen/enlighten.c
> +++ b/arch/arm/xen/enlighten.c
> @@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
>  	xen_domain_type = XEN_HVM_DOMAIN;
>  
>  	xen_setup_features();
> +
> +	if (!xen_feature(XENFEAT_grant_map_11)) {
> +		pr_warn("Please upgrade your Xen.\n"
> +				"If your platform has any non-coherent DMA devices, they won't work properly.\n");
> +	}

Unfortunately this isn't quite complete. On a system where all devices
are behind an SMMU then we would want to be able to disable the 1:1
workaround, which in turn would imply disabling this feature flag too
(since it is no longer necessary and also impossible to implement in
that case).

Ian.

[PATCH v2 1/3] xen/arm: introduce XENFEAT_grant_map_11

[Stefano Stabellini]

On Tue, 8 Jul 2014, Ian Campbell wrote:
> On Tue, 2014-07-08 at 16:42 +0100, Stefano Stabellini wrote:
> > The flag tells us that the hypervisor maps a grant page to guest
> > physical address == machine address of the page in addition to the
> > normal grant mapping address. It is needed to properly issue cache
> > maintenance operation at the completion of a DMA operation involving a
> > foreign grant.
> > 
> > Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> > ---
> >  arch/arm/xen/enlighten.c         |    6 ++++++
> >  include/xen/interface/features.h |    3 +++
> >  2 files changed, 9 insertions(+)
> > 
> > diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
> > index b96723e..ee3135a 100644
> > --- a/arch/arm/xen/enlighten.c
> > +++ b/arch/arm/xen/enlighten.c
> > @@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
> >  	xen_domain_type = XEN_HVM_DOMAIN;
> >  
> >  	xen_setup_features();
> > +
> > +	if (!xen_feature(XENFEAT_grant_map_11)) {
> > +		pr_warn("Please upgrade your Xen.\n"
> > +				"If your platform has any non-coherent DMA devices, they won't work properly.\n");
> > +	}
> 
> Unfortunately this isn't quite complete. On a system where all devices
> are behind an SMMU then we would want to be able to disable the 1:1
> workaround, which in turn would imply disabling this feature flag too
> (since it is no longer necessary and also impossible to implement in
> that case).

That is true, but in such a system we would have to tell the kernel that
DMAing is safe, so this will turn into:

if (!xen_feature(XENFEAT_grant_map_11) &&
    !xen_feature(XENFEAT_safe_dma)) {
	pr_warn("Please upgrade your Xen.\n"
			"If your platform has any non-coherent DMA devices, they won't work properly.\n");
}

[PATCH v2 1/3] xen/arm: introduce XENFEAT_grant_map_11

[Ian Campbell]

On Tue, 2014-07-08 at 16:54 +0100, Stefano Stabellini wrote:
> On Tue, 8 Jul 2014, Ian Campbell wrote:
> > On Tue, 2014-07-08 at 16:42 +0100, Stefano Stabellini wrote:
> > > The flag tells us that the hypervisor maps a grant page to guest
> > > physical address == machine address of the page in addition to the
> > > normal grant mapping address. It is needed to properly issue cache
> > > maintenance operation at the completion of a DMA operation involving a
> > > foreign grant.
> > > 
> > > Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> > > ---
> > >  arch/arm/xen/enlighten.c         |    6 ++++++
> > >  include/xen/interface/features.h |    3 +++
> > >  2 files changed, 9 insertions(+)
> > > 
> > > diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
> > > index b96723e..ee3135a 100644
> > > --- a/arch/arm/xen/enlighten.c
> > > +++ b/arch/arm/xen/enlighten.c
> > > @@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
> > >  	xen_domain_type = XEN_HVM_DOMAIN;
> > >  
> > >  	xen_setup_features();
> > > +
> > > +	if (!xen_feature(XENFEAT_grant_map_11)) {
> > > +		pr_warn("Please upgrade your Xen.\n"
> > > +				"If your platform has any non-coherent DMA devices, they won't work properly.\n");
> > > +	}
> > 
> > Unfortunately this isn't quite complete. On a system where all devices
> > are behind an SMMU then we would want to be able to disable the 1:1
> > workaround, which in turn would imply disabling this feature flag too
> > (since it is no longer necessary and also impossible to implement in
> > that case).
> 
> That is true, but in such a system we would have to tell the kernel that
> DMAing is safe, so this will turn into:

Oh right, yes. Good then ;-)

> 
> if (!xen_feature(XENFEAT_grant_map_11) &&
>     !xen_feature(XENFEAT_safe_dma)) {
> 	pr_warn("Please upgrade your Xen.\n"
> 			"If your platform has any non-coherent DMA devices, they won't work properly.\n");
> }

[Xen-devel] [PATCH v2 0/3] xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

[Konrad Rzeszutek Wilk]

On Tue, Jul 08, 2014 at 04:40:46PM +0100, Stefano Stabellini wrote:
> Hi all,
> Xen support in Linux for ARM and ARM64 suffers from lack of support for
> multiple mfn to pfn mappings: whenever a frontend grants the same page
> multiple times to the backend, the mfn to pfn accounting in
> arch/arm/xen/p2m.c fails. The issue has become critical since v3.15,
> when xen-netfront/xen-netback switched from grant copies to grant
> mappings, therefore causing the issue to happen much more often.
> 
> Fixing the mfn to pfn accounting in p2m.c is difficult and expensive,
> therefore we are looking for alternative solutions. One idea is avoiding
> mfn to pfn conversions altogether. The only code path that needs them is
> swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).

I seem to have lost track of that patch? Or is it in now in the kernel?

Could you include the git commit id or URL for it in the cover letter please?

Thanks
> 
> To avoid mfn to pfn conversions we rely on a second p2m mapping done by
> Xen (a separate patch series will be sent for Xen). In Linux we use it
> to perform the cache maintenance operations without mfns conversions.
> 
> 
> Changes in v2:
> - introduce XENFEAT_grant_map_11;
> - remeber the ptep corresponding to scratch pages so that we don't need
> to calculate it again every time;
> - do not acutally unmap the page on xen_mm32_unmap;
> - properly account preempt_enable/disable;
> - do not check for mfn in xen_add_phys_to_mach_entry.
> 
> 
> Stefano Stabellini (3):
>       xen/arm: introduce XENFEAT_grant_map_11
>       xen/arm: reimplement xen_dma_unmap_page & friends
>       xen/arm: remove mach_to_phys rbtree
> 
>  arch/arm/include/asm/xen/page-coherent.h |   25 ++--
>  arch/arm/include/asm/xen/page.h          |    9 --
>  arch/arm/xen/Makefile                    |    2 +-
>  arch/arm/xen/enlighten.c                 |    6 +
>  arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
>  arch/arm/xen/p2m.c                       |   66 +---------
>  include/xen/interface/features.h         |    3 +
>  7 files changed, 220 insertions(+), 93 deletions(-)
>  create mode 100644 arch/arm/xen/mm32.c
> 
> _______________________________________________
> Xen-devel mailing list
> Xen-devel at lists.xen.org
> http://lists.xen.org/xen-devel

[PATCH v2 1/3] xen/arm: introduce XENFEAT_grant_map_11

[David Vrabel]

On 08/07/14 16:42, Stefano Stabellini wrote:
> The flag tells us that the hypervisor maps a grant page to guest
> physical address == machine address of the page in addition to the
> normal grant mapping address. It is needed to properly issue cache
> maintenance operation at the completion of a DMA operation involving a
> foreign grant.
[...]
> +/* Xen also maps grant references at pfn = mfn */
> +#define XENFEAT_grant_map_11              12

I keep reading this as "grant map eleven".  I think you've used this
abbreviation else where so you should probably keep it as-is.

I might have picked XENFEAT_grant_map_identity.

David

[Xen-devel] [PATCH v2 1/3] xen/arm: introduce XENFEAT_grant_map_11

[Julien Grall]

Hi Ian and Stefano,

On 07/08/2014 04:59 PM, Ian Campbell wrote:
> On Tue, 2014-07-08 at 16:54 +0100, Stefano Stabellini wrote:
>> On Tue, 8 Jul 2014, Ian Campbell wrote:
>>> On Tue, 2014-07-08 at 16:42 +0100, Stefano Stabellini wrote:
>>>> The flag tells us that the hypervisor maps a grant page to guest
>>>> physical address == machine address of the page in addition to the
>>>> normal grant mapping address. It is needed to properly issue cache
>>>> maintenance operation at the completion of a DMA operation involving a
>>>> foreign grant.
>>>>
>>>> Signed-off-by: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
>>>> ---
>>>>  arch/arm/xen/enlighten.c         |    6 ++++++
>>>>  include/xen/interface/features.h |    3 +++
>>>>  2 files changed, 9 insertions(+)
>>>>
>>>> diff --git a/arch/arm/xen/enlighten.c b/arch/arm/xen/enlighten.c
>>>> index b96723e..ee3135a 100644
>>>> --- a/arch/arm/xen/enlighten.c
>>>> +++ b/arch/arm/xen/enlighten.c
>>>> @@ -262,6 +262,12 @@ static int __init xen_guest_init(void)
>>>>  	xen_domain_type = XEN_HVM_DOMAIN;
>>>>  
>>>>  	xen_setup_features();
>>>> +
>>>> +	if (!xen_feature(XENFEAT_grant_map_11)) {
>>>> +		pr_warn("Please upgrade your Xen.\n"
>>>> +				"If your platform has any non-coherent DMA devices, they won't work properly.\n");
>>>> +	}
>>>
>>> Unfortunately this isn't quite complete. On a system where all devices
>>> are behind an SMMU then we would want to be able to disable the 1:1
>>> workaround, which in turn would imply disabling this feature flag too
>>> (since it is no longer necessary and also impossible to implement in
>>> that case).
>>
>> That is true, but in such a system we would have to tell the kernel that
>> DMAing is safe, so this will turn into:
> 
> Oh right, yes. Good then ;-)

FWIW, I've sent a patch series a couple ago to avoid using swiotlb when
the device is protected (see https://patches.linaro.org/25070/).

I should take time to rework properly and send a new version.

Regards,

-- 
Julien Grall

[Xen-devel] [PATCH v2 0/3] xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

[Stefano Stabellini]

On Tue, 8 Jul 2014, Konrad Rzeszutek Wilk wrote:
> On Tue, Jul 08, 2014 at 04:40:46PM +0100, Stefano Stabellini wrote:
> > Hi all,
> > Xen support in Linux for ARM and ARM64 suffers from lack of support for
> > multiple mfn to pfn mappings: whenever a frontend grants the same page
> > multiple times to the backend, the mfn to pfn accounting in
> > arch/arm/xen/p2m.c fails. The issue has become critical since v3.15,
> > when xen-netfront/xen-netback switched from grant copies to grant
> > mappings, therefore causing the issue to happen much more often.
> > 
> > Fixing the mfn to pfn accounting in p2m.c is difficult and expensive,
> > therefore we are looking for alternative solutions. One idea is avoiding
> > mfn to pfn conversions altogether. The only code path that needs them is
> > swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).
> 
> I seem to have lost track of that patch? Or is it in now in the kernel?
> 
> Could you include the git commit id or URL for it in the cover letter please?

I take that you are asking the id of the commit that introduced the
function calls that need mfn to pfn conversions in swiotlb-xen, right?
The commit that introduced them is:

commit 6cf054636261ca5c88f3c2984058d51f927b8a2e
Author: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
Date:   Fri Oct 25 10:33:25 2013 +0000

    swiotlb-xen: use xen_dma_map/unmap_page, xen_dma_sync_single_for_cpu/device
     
xen_dma_unmap_page on x86 doesn't do anything but on ARM performs
important cache invalidate operations for non-dma-coherent devices.



> Thanks
> > 
> > To avoid mfn to pfn conversions we rely on a second p2m mapping done by
> > Xen (a separate patch series will be sent for Xen). In Linux we use it
> > to perform the cache maintenance operations without mfns conversions.
> > 
> > 
> > Changes in v2:
> > - introduce XENFEAT_grant_map_11;
> > - remeber the ptep corresponding to scratch pages so that we don't need
> > to calculate it again every time;
> > - do not acutally unmap the page on xen_mm32_unmap;
> > - properly account preempt_enable/disable;
> > - do not check for mfn in xen_add_phys_to_mach_entry.
> > 
> > 
> > Stefano Stabellini (3):
> >       xen/arm: introduce XENFEAT_grant_map_11
> >       xen/arm: reimplement xen_dma_unmap_page & friends
> >       xen/arm: remove mach_to_phys rbtree
> > 
> >  arch/arm/include/asm/xen/page-coherent.h |   25 ++--
> >  arch/arm/include/asm/xen/page.h          |    9 --
> >  arch/arm/xen/Makefile                    |    2 +-
> >  arch/arm/xen/enlighten.c                 |    6 +
> >  arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
> >  arch/arm/xen/p2m.c                       |   66 +---------
> >  include/xen/interface/features.h         |    3 +
> >  7 files changed, 220 insertions(+), 93 deletions(-)
> >  create mode 100644 arch/arm/xen/mm32.c
> > 
> > _______________________________________________
> > Xen-devel mailing list
> > Xen-devel at lists.xen.org
> > http://lists.xen.org/xen-devel
> 

[Xen-devel] [PATCH v2 0/3] xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

[Konrad Rzeszutek Wilk]

On Wed, Jul 09, 2014 at 11:30:49AM +0100, Stefano Stabellini wrote:
> On Tue, 8 Jul 2014, Konrad Rzeszutek Wilk wrote:
> > On Tue, Jul 08, 2014 at 04:40:46PM +0100, Stefano Stabellini wrote:
> > > Hi all,
> > > Xen support in Linux for ARM and ARM64 suffers from lack of support for
> > > multiple mfn to pfn mappings: whenever a frontend grants the same page
> > > multiple times to the backend, the mfn to pfn accounting in
> > > arch/arm/xen/p2m.c fails. The issue has become critical since v3.15,
> > > when xen-netfront/xen-netback switched from grant copies to grant
> > > mappings, therefore causing the issue to happen much more often.
> > > 
> > > Fixing the mfn to pfn accounting in p2m.c is difficult and expensive,
> > > therefore we are looking for alternative solutions. One idea is avoiding
> > > mfn to pfn conversions altogether. The only code path that needs them is
> > > swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).
> > 
> > I seem to have lost track of that patch? Or is it in now in the kernel?
> > 
> > Could you include the git commit id or URL for it in the cover letter please?
> 
> I take that you are asking the id of the commit that introduced the
> function calls that need mfn to pfn conversions in swiotlb-xen, right?

Earlier you said: "avoiding mfn to pfn conversions. The only code path
that needs them is .." So I suspected (it seems wrongly) that there was
some SWIOTLB patch that would do this and I had missed it.

But it seems that this soo fresh you hadn't yet posted the SWIOTLB changes
to use this new API - so I hadn't missed them :-)

> The commit that introduced them is:
> 
> commit 6cf054636261ca5c88f3c2984058d51f927b8a2e
> Author: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> Date:   Fri Oct 25 10:33:25 2013 +0000
> 
>     swiotlb-xen: use xen_dma_map/unmap_page, xen_dma_sync_single_for_cpu/device
>      
> xen_dma_unmap_page on x86 doesn't do anything but on ARM performs
> important cache invalidate operations for non-dma-coherent devices.
> 
> 
> 
> > Thanks
> > > 
> > > To avoid mfn to pfn conversions we rely on a second p2m mapping done by
> > > Xen (a separate patch series will be sent for Xen). In Linux we use it
> > > to perform the cache maintenance operations without mfns conversions.
> > > 
> > > 
> > > Changes in v2:
> > > - introduce XENFEAT_grant_map_11;
> > > - remeber the ptep corresponding to scratch pages so that we don't need
> > > to calculate it again every time;
> > > - do not acutally unmap the page on xen_mm32_unmap;
> > > - properly account preempt_enable/disable;
> > > - do not check for mfn in xen_add_phys_to_mach_entry.
> > > 
> > > 
> > > Stefano Stabellini (3):
> > >       xen/arm: introduce XENFEAT_grant_map_11
> > >       xen/arm: reimplement xen_dma_unmap_page & friends
> > >       xen/arm: remove mach_to_phys rbtree
> > > 
> > >  arch/arm/include/asm/xen/page-coherent.h |   25 ++--
> > >  arch/arm/include/asm/xen/page.h          |    9 --
> > >  arch/arm/xen/Makefile                    |    2 +-
> > >  arch/arm/xen/enlighten.c                 |    6 +
> > >  arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
> > >  arch/arm/xen/p2m.c                       |   66 +---------
> > >  include/xen/interface/features.h         |    3 +
> > >  7 files changed, 220 insertions(+), 93 deletions(-)
> > >  create mode 100644 arch/arm/xen/mm32.c
> > > 
> > > _______________________________________________
> > > Xen-devel mailing list
> > > Xen-devel at lists.xen.org
> > > http://lists.xen.org/xen-devel
> > 

[Xen-devel] [PATCH v2 0/3] xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

[Stefano Stabellini]

On Wed, 9 Jul 2014, Konrad Rzeszutek Wilk wrote:
> On Wed, Jul 09, 2014 at 11:30:49AM +0100, Stefano Stabellini wrote:
> > On Tue, 8 Jul 2014, Konrad Rzeszutek Wilk wrote:
> > > On Tue, Jul 08, 2014 at 04:40:46PM +0100, Stefano Stabellini wrote:
> > > > Hi all,
> > > > Xen support in Linux for ARM and ARM64 suffers from lack of support for
> > > > multiple mfn to pfn mappings: whenever a frontend grants the same page
> > > > multiple times to the backend, the mfn to pfn accounting in
> > > > arch/arm/xen/p2m.c fails. The issue has become critical since v3.15,
> > > > when xen-netfront/xen-netback switched from grant copies to grant
> > > > mappings, therefore causing the issue to happen much more often.
> > > > 
> > > > Fixing the mfn to pfn accounting in p2m.c is difficult and expensive,
> > > > therefore we are looking for alternative solutions. One idea is avoiding
> > > > mfn to pfn conversions altogether. The only code path that needs them is
> > > > swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).
> > > 
> > > I seem to have lost track of that patch? Or is it in now in the kernel?
> > > 
> > > Could you include the git commit id or URL for it in the cover letter please?
> > 
> > I take that you are asking the id of the commit that introduced the
> > function calls that need mfn to pfn conversions in swiotlb-xen, right?
> 
> Earlier you said: "avoiding mfn to pfn conversions. The only code path
> that needs them is .." So I suspected (it seems wrongly) that there was
> some SWIOTLB patch that would do this and I had missed it.
> 
> But it seems that this soo fresh you hadn't yet posted the SWIOTLB changes
> to use this new API - so I hadn't missed them :-)

I think you didn't notice them because they don't do anything at all on
x86 :-)

If you give a look at swiotlb-xen, you'll find that xen_bus_to_phys is
called in xen_unmap_single and xen_swiotlb_sync_single. The returned
physical address is passed to xen_dma_unmap_page and
xen_dma_sync_single_for_cpu.

xen_dma_unmap_page and xen_dma_sync_single_for_cpu have an empty
implementation on x86 (see arch/x86/include/asm/xen/page-coherent.h)
but on ARM they perform important cache maintenance operations.

These are the only place where we need mfn to pfn conversions.


> > The commit that introduced them is:
> > 
> > commit 6cf054636261ca5c88f3c2984058d51f927b8a2e
> > Author: Stefano Stabellini <stefano.stabellini@eu.citrix.com>
> > Date:   Fri Oct 25 10:33:25 2013 +0000
> > 
> >     swiotlb-xen: use xen_dma_map/unmap_page, xen_dma_sync_single_for_cpu/device
> >      
> > xen_dma_unmap_page on x86 doesn't do anything but on ARM performs
> > important cache invalidate operations for non-dma-coherent devices.
> > 
> > 
> > 
> > > Thanks
> > > > 
> > > > To avoid mfn to pfn conversions we rely on a second p2m mapping done by
> > > > Xen (a separate patch series will be sent for Xen). In Linux we use it
> > > > to perform the cache maintenance operations without mfns conversions.
> > > > 
> > > > 
> > > > Changes in v2:
> > > > - introduce XENFEAT_grant_map_11;
> > > > - remeber the ptep corresponding to scratch pages so that we don't need
> > > > to calculate it again every time;
> > > > - do not acutally unmap the page on xen_mm32_unmap;
> > > > - properly account preempt_enable/disable;
> > > > - do not check for mfn in xen_add_phys_to_mach_entry.
> > > > 
> > > > 
> > > > Stefano Stabellini (3):
> > > >       xen/arm: introduce XENFEAT_grant_map_11
> > > >       xen/arm: reimplement xen_dma_unmap_page & friends
> > > >       xen/arm: remove mach_to_phys rbtree
> > > > 
> > > >  arch/arm/include/asm/xen/page-coherent.h |   25 ++--
> > > >  arch/arm/include/asm/xen/page.h          |    9 --
> > > >  arch/arm/xen/Makefile                    |    2 +-
> > > >  arch/arm/xen/enlighten.c                 |    6 +
> > > >  arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
> > > >  arch/arm/xen/p2m.c                       |   66 +---------
> > > >  include/xen/interface/features.h         |    3 +
> > > >  7 files changed, 220 insertions(+), 93 deletions(-)
> > > >  create mode 100644 arch/arm/xen/mm32.c
> > > > 
> > > > _______________________________________________
> > > > Xen-devel mailing list
> > > > Xen-devel at lists.xen.org
> > > > http://lists.xen.org/xen-devel
> > > 
> 

[PATCH v2 0/3] xen/arm: fix "xen_add_mach_to_phys_entry: cannot add" problem

[Denis Schneider]

2014-07-08 17:40 GMT+02:00 Stefano Stabellini
<stefano.stabellini@eu.citrix.com>:
> Hi all,
> Xen support in Linux for ARM and ARM64 suffers from lack of support for
> multiple mfn to pfn mappings: whenever a frontend grants the same page
> multiple times to the backend, the mfn to pfn accounting in
> arch/arm/xen/p2m.c fails. The issue has become critical since v3.15,
> when xen-netfront/xen-netback switched from grant copies to grant
> mappings, therefore causing the issue to happen much more often.
>
> Fixing the mfn to pfn accounting in p2m.c is difficult and expensive,
> therefore we are looking for alternative solutions. One idea is avoiding
> mfn to pfn conversions altogether. The only code path that needs them is
> swiotlb-xen:unmap_page (and single_for_cpu and single_for_device).
>
> To avoid mfn to pfn conversions we rely on a second p2m mapping done by
> Xen (a separate patch series will be sent for Xen). In Linux we use it
> to perform the cache maintenance operations without mfns conversions.
>
>
> Changes in v2:
> - introduce XENFEAT_grant_map_11;
> - remeber the ptep corresponding to scratch pages so that we don't need
> to calculate it again every time;
> - do not acutally unmap the page on xen_mm32_unmap;
> - properly account preempt_enable/disable;
> - do not check for mfn in xen_add_phys_to_mach_entry.
>
>
> Stefano Stabellini (3):
>       xen/arm: introduce XENFEAT_grant_map_11
>       xen/arm: reimplement xen_dma_unmap_page & friends
>       xen/arm: remove mach_to_phys rbtree
>
>  arch/arm/include/asm/xen/page-coherent.h |   25 ++--
>  arch/arm/include/asm/xen/page.h          |    9 --
>  arch/arm/xen/Makefile                    |    2 +-
>  arch/arm/xen/enlighten.c                 |    6 +
>  arch/arm/xen/mm32.c                      |  202 ++++++++++++++++++++++++++++++
>  arch/arm/xen/p2m.c                       |   66 +---------
>  include/xen/interface/features.h         |    3 +
>  7 files changed, 220 insertions(+), 93 deletions(-)
>  create mode 100644 arch/arm/xen/mm32.c

Tested-by: Denis Schneider <v1ne2go@gmail.com>

Stress-tested with network I/O, disk I/O for hours without any problem. Thanks.