[PATCH v5 0/7] dax: I/O path enhancements

[PATCH v5 0/7] dax: I/O path enhancements

[Ross Zwisler]

The goal of this series is to enhance the DAX I/O path so that all operations
that store data (I/O writes, zeroing blocks, punching holes, etc.) properly
synchronize the stores to media using the PMEM API.  This ensures that the
data DAX is writing is durable on media before the operation completes.

Patches 1-4 are a few random cleanups.

Changes from v4:
 - rebased to libnvdimm-for-next branch:
https://git.kernel.org/cgit/linux/kernel/git/nvdimm/nvdimm.git/commit/?h=libnvdimm-for-next

The nvdimm repository doesn't have the DAX PMD changes that are in the -mm
tree.  I expect the merge will basically be these two hunks:

@@ -514,7 +528,7 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address,
        unsigned long pmd_addr = address & PMD_MASK;
        bool write = flags & FAULT_FLAG_WRITE;
        long length;
-       void *kaddr;
+       void __pmem *kaddr;
        pgoff_t size, pgoff;
        sector_t block, sector;
        unsigned long pfn;
@@ -608,7 +622,8 @@ int __dax_pmd_fault(struct vm_area_struct *vma, unsigned long address,
                if (buffer_unwritten(&bh) || buffer_new(&bh)) {
                        int i;
                        for (i = 0; i < PTRS_PER_PMD; i++)
-                               clear_page(kaddr + i * PAGE_SIZE);
+                               clear_pmem(kaddr + i * PAGE_SIZE, PAGE_SIZE);
+                       wmb_pmem();
                        count_vm_event(PGMAJFAULT);
                        mem_cgroup_count_vm_event(vma->vm_mm, PGMAJFAULT);
                        result |= VM_FAULT_MAJOR;

Ross Zwisler (7):
  brd: make rd_size static
  pmem, x86: move x86 PMEM API to new pmem.h header
  pmem: remove layer when calling arch_has_wmb_pmem()
  pmem, x86: clean up conditional pmem includes
  pmem: add copy_from_iter_pmem() and clear_pmem()
  dax: update I/O path to do proper PMEM flushing
  pmem, dax: have direct_access use __pmem annotation

 Documentation/filesystems/Locking |   3 +-
 MAINTAINERS                       |   1 +
 arch/powerpc/sysdev/axonram.c     |   7 +-
 arch/x86/include/asm/cacheflush.h |  71 -----------------
 arch/x86/include/asm/pmem.h       | 158 ++++++++++++++++++++++++++++++++++++++
 drivers/block/brd.c               |   6 +-
 drivers/nvdimm/pmem.c             |   4 +-
 drivers/s390/block/dcssblk.c      |  10 ++-
 fs/block_dev.c                    |   2 +-
 fs/dax.c                          |  63 +++++++++------
 include/linux/blkdev.h            |   8 +-
 include/linux/pmem.h              |  77 ++++++++++++++++---
 12 files changed, 285 insertions(+), 125 deletions(-)
 create mode 100644 arch/x86/include/asm/pmem.h

-- 
2.1.0

[PATCH v5 1/7] brd: make rd_size static

[Ross Zwisler]

Make rd_size static because it is local to drivers/block/brd.c

This was reported by sparse:

drivers/block/brd.c:445:5: warning: symbol 'rd_size' was not declared.
Should it be static?

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 drivers/block/brd.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/block/brd.c b/drivers/block/brd.c
index 64ab495..5750b39 100644
--- a/drivers/block/brd.c
+++ b/drivers/block/brd.c
@@ -442,7 +442,7 @@ static int rd_nr = CONFIG_BLK_DEV_RAM_COUNT;
 module_param(rd_nr, int, S_IRUGO);
 MODULE_PARM_DESC(rd_nr, "Maximum number of brd devices");
 
-int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
+static int rd_size = CONFIG_BLK_DEV_RAM_SIZE;
 module_param(rd_size, int, S_IRUGO);
 MODULE_PARM_DESC(rd_size, "Size of each RAM disk in kbytes.");
 
-- 
2.1.0

[PATCH v5 2/7] pmem, x86: move x86 PMEM API to new pmem.h header

[Ross Zwisler]

Move the x86 PMEM API implementation out of asm/cacheflush.h and into
its own header asm/pmem.h.  This will allow members of the PMEM API to
be more easily identified on this and other architectures.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Suggested-by: Christoph Hellwig <hch@lst.de>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 MAINTAINERS                       |  1 +
 arch/x86/include/asm/cacheflush.h | 71 ------------------------------
 arch/x86/include/asm/pmem.h       | 92 +++++++++++++++++++++++++++++++++++++++
 include/linux/pmem.h              |  2 +-
 4 files changed, 94 insertions(+), 72 deletions(-)
 create mode 100644 arch/x86/include/asm/pmem.h

diff --git a/MAINTAINERS b/MAINTAINERS
index 9289ecb..8fcde371 100644
--- a/MAINTAINERS
+++ b/MAINTAINERS
@@ -6161,6 +6161,7 @@ Q:	https://patchwork.kernel.org/project/linux-nvdimm/list/
 S:	Supported
 F:	drivers/nvdimm/pmem.c
 F:	include/linux/pmem.h
+F:	arch/*/include/asm/pmem.h
 
 LINUX FOR IBM pSERIES (RS/6000)
 M:	Paul Mackerras <paulus@au.ibm.com>
diff --git a/arch/x86/include/asm/cacheflush.h b/arch/x86/include/asm/cacheflush.h
index 9bf3ea1..471418a 100644
--- a/arch/x86/include/asm/cacheflush.h
+++ b/arch/x86/include/asm/cacheflush.h
@@ -109,75 +109,4 @@ static inline int rodata_test(void)
 }
 #endif
 
-#ifdef ARCH_HAS_NOCACHE_UACCESS
-
-/**
- * arch_memcpy_to_pmem - copy data to persistent memory
- * @dst: destination buffer for the copy
- * @src: source buffer for the copy
- * @n: length of the copy in bytes
- *
- * Copy data to persistent memory media via non-temporal stores so that
- * a subsequent arch_wmb_pmem() can flush cpu and memory controller
- * write buffers to guarantee durability.
- */
-static inline void arch_memcpy_to_pmem(void __pmem *dst, const void *src,
-		size_t n)
-{
-	int unwritten;
-
-	/*
-	 * We are copying between two kernel buffers, if
-	 * __copy_from_user_inatomic_nocache() returns an error (page
-	 * fault) we would have already reported a general protection fault
-	 * before the WARN+BUG.
-	 */
-	unwritten = __copy_from_user_inatomic_nocache((void __force *) dst,
-			(void __user *) src, n);
-	if (WARN(unwritten, "%s: fault copying %p <- %p unwritten: %d\n",
-				__func__, dst, src, unwritten))
-		BUG();
-}
-
-/**
- * arch_wmb_pmem - synchronize writes to persistent memory
- *
- * After a series of arch_memcpy_to_pmem() operations this drains data
- * from cpu write buffers and any platform (memory controller) buffers
- * to ensure that written data is durable on persistent memory media.
- */
-static inline void arch_wmb_pmem(void)
-{
-	/*
-	 * wmb() to 'sfence' all previous writes such that they are
-	 * architecturally visible to 'pcommit'.  Note, that we've
-	 * already arranged for pmem writes to avoid the cache via
-	 * arch_memcpy_to_pmem().
-	 */
-	wmb();
-	pcommit_sfence();
-}
-
-static inline bool __arch_has_wmb_pmem(void)
-{
-#ifdef CONFIG_X86_64
-	/*
-	 * We require that wmb() be an 'sfence', that is only guaranteed on
-	 * 64-bit builds
-	 */
-	return static_cpu_has(X86_FEATURE_PCOMMIT);
-#else
-	return false;
-#endif
-}
-#else /* ARCH_HAS_NOCACHE_UACCESS i.e. ARCH=um */
-extern void arch_memcpy_to_pmem(void __pmem *dst, const void *src, size_t n);
-extern void arch_wmb_pmem(void);
-
-static inline bool __arch_has_wmb_pmem(void)
-{
-	return false;
-}
-#endif
-
 #endif /* _ASM_X86_CACHEFLUSH_H */
diff --git a/arch/x86/include/asm/pmem.h b/arch/x86/include/asm/pmem.h
new file mode 100644
index 0000000..f43462c
--- /dev/null
+++ b/arch/x86/include/asm/pmem.h
@@ -0,0 +1,92 @@
+/*
+ * Copyright(c) 2015 Intel Corporation. All rights reserved.
+ *
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of version 2 of the GNU General Public License as
+ * published by the Free Software Foundation.
+ *
+ * This program is distributed in the hope that it will be useful, but
+ * WITHOUT ANY WARRANTY; without even the implied warranty of
+ * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
+ * General Public License for more details.
+ */
+#ifndef __ASM_X86_PMEM_H__
+#define __ASM_X86_PMEM_H__
+
+#include <linux/uaccess.h>
+#include <asm/cacheflush.h>
+#include <asm/cpufeature.h>
+#include <asm/special_insns.h>
+
+#ifdef ARCH_HAS_NOCACHE_UACCESS
+
+/**
+ * arch_memcpy_to_pmem - copy data to persistent memory
+ * @dst: destination buffer for the copy
+ * @src: source buffer for the copy
+ * @n: length of the copy in bytes
+ *
+ * Copy data to persistent memory media via non-temporal stores so that
+ * a subsequent arch_wmb_pmem() can flush cpu and memory controller
+ * write buffers to guarantee durability.
+ */
+static inline void arch_memcpy_to_pmem(void __pmem *dst, const void *src,
+		size_t n)
+{
+	int unwritten;
+
+	/*
+	 * We are copying between two kernel buffers, if
+	 * __copy_from_user_inatomic_nocache() returns an error (page
+	 * fault) we would have already reported a general protection fault
+	 * before the WARN+BUG.
+	 */
+	unwritten = __copy_from_user_inatomic_nocache((void __force *) dst,
+			(void __user *) src, n);
+	if (WARN(unwritten, "%s: fault copying %p <- %p unwritten: %d\n",
+				__func__, dst, src, unwritten))
+		BUG();
+}
+
+/**
+ * arch_wmb_pmem - synchronize writes to persistent memory
+ *
+ * After a series of arch_memcpy_to_pmem() operations this drains data
+ * from cpu write buffers and any platform (memory controller) buffers
+ * to ensure that written data is durable on persistent memory media.
+ */
+static inline void arch_wmb_pmem(void)
+{
+	/*
+	 * wmb() to 'sfence' all previous writes such that they are
+	 * architecturally visible to 'pcommit'.  Note, that we've
+	 * already arranged for pmem writes to avoid the cache via
+	 * arch_memcpy_to_pmem().
+	 */
+	wmb();
+	pcommit_sfence();
+}
+
+static inline bool __arch_has_wmb_pmem(void)
+{
+#ifdef CONFIG_X86_64
+	/*
+	 * We require that wmb() be an 'sfence', that is only guaranteed on
+	 * 64-bit builds
+	 */
+	return static_cpu_has(X86_FEATURE_PCOMMIT);
+#else
+	return false;
+#endif
+}
+#else /* ARCH_HAS_NOCACHE_UACCESS i.e. ARCH=um */
+extern void arch_memcpy_to_pmem(void __pmem *dst, const void *src, size_t n);
+extern void arch_wmb_pmem(void);
+
+static inline bool __arch_has_wmb_pmem(void)
+{
+	return false;
+}
+#endif
+
+#endif /* __ASM_X86_PMEM_H__ */
diff --git a/include/linux/pmem.h b/include/linux/pmem.h
index 20c367c..c2af613 100644
--- a/include/linux/pmem.h
+++ b/include/linux/pmem.h
@@ -16,7 +16,7 @@
 #include <linux/io.h>
 
 #ifdef CONFIG_ARCH_HAS_PMEM_API
-#include <asm/cacheflush.h>
+#include <asm/pmem.h>
 #else
 static inline void arch_wmb_pmem(void)
 {
-- 
2.1.0

[PATCH v5 3/7] pmem: remove layer when calling arch_has_wmb_pmem()

[Ross Zwisler]

Prior to this change arch_has_wmb_pmem() was only called by
arch_has_pmem_api().  Both arch_has_wmb_pmem() and arch_has_pmem_api()
checked to make sure that CONFIG_ARCH_HAS_PMEM_API was enabled.

Instead, remove the old arch_has_wmb_pmem() wrapper to be rid of one
extra layer of indirection and the redundant CONFIG_ARCH_HAS_PMEM_API
check. Rename __arch_has_wmb_pmem() to arch_has_wmb_pmem() since we no
longer have a wrapper, and just have arch_has_pmem_api() call the
architecture specific arch_has_wmb_pmem() directly.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 arch/x86/include/asm/pmem.h |  2 +-
 include/linux/pmem.h        | 13 +++----------
 2 files changed, 4 insertions(+), 11 deletions(-)

diff --git a/arch/x86/include/asm/pmem.h b/arch/x86/include/asm/pmem.h
index f43462c..1e8dbb7 100644
--- a/arch/x86/include/asm/pmem.h
+++ b/arch/x86/include/asm/pmem.h
@@ -67,7 +67,7 @@ static inline void arch_wmb_pmem(void)
 	pcommit_sfence();
 }
 
-static inline bool __arch_has_wmb_pmem(void)
+static inline bool arch_has_wmb_pmem(void)
 {
 #ifdef CONFIG_X86_64
 	/*
diff --git a/include/linux/pmem.h b/include/linux/pmem.h
index c2af613..a0706ea 100644
--- a/include/linux/pmem.h
+++ b/include/linux/pmem.h
@@ -23,7 +23,7 @@ static inline void arch_wmb_pmem(void)
 	BUG();
 }
 
-static inline bool __arch_has_wmb_pmem(void)
+static inline bool arch_has_wmb_pmem(void)
 {
 	return false;
 }
@@ -38,7 +38,7 @@ static inline void arch_memcpy_to_pmem(void __pmem *dst, const void *src,
 /*
  * Architectures that define ARCH_HAS_PMEM_API must provide
  * implementations for arch_memcpy_to_pmem(), arch_wmb_pmem(), and
- * __arch_has_wmb_pmem().
+ * arch_has_wmb_pmem().
  */
 
 static inline void memcpy_from_pmem(void *dst, void __pmem const *src, size_t size)
@@ -52,7 +52,7 @@ static inline void memunmap_pmem(struct device *dev, void __pmem *addr)
 }
 
 /**
- * arch_has_wmb_pmem - true if wmb_pmem() ensures durability
+ * arch_has_pmem_api - true if wmb_pmem() ensures durability
  *
  * For a given cpu implementation within an architecture it is possible
  * that wmb_pmem() resolves to a nop.  In the case this returns
@@ -60,13 +60,6 @@ static inline void memunmap_pmem(struct device *dev, void __pmem *addr)
  * fall back to a different data consistency model, or otherwise notify
  * the user.
  */
-static inline bool arch_has_wmb_pmem(void)
-{
-	if (IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API))
-		return __arch_has_wmb_pmem();
-	return false;
-}
-
 static inline bool arch_has_pmem_api(void)
 {
 	return IS_ENABLED(CONFIG_ARCH_HAS_PMEM_API) && arch_has_wmb_pmem();
-- 
2.1.0

[PATCH v5 4/7] pmem, x86: clean up conditional pmem includes

[Ross Zwisler]

Prior to this change x86_64 used the pmem defines in
arch/x86/include/asm/pmem.h, and UM used the default ones at the
top of include/linux/pmem.h.  The inclusion or exclusion in linux/pmem.h
was controlled by CONFIG_ARCH_HAS_PMEM_API, but the ones in asm/pmem.h
were controlled by ARCH_HAS_NOCACHE_UACCESS.

Instead, control them both with CONFIG_ARCH_HAS_PMEM_API so that it's
clear that they are related and we don't run into the possibility where
they are both included or excluded.  Also remove a bunch of stale
function prototypes meant for UM in asm/pmem.h - these just conflicted
with the inline defaults in linux/pmem.h and gave compile errors.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 arch/x86/include/asm/pmem.h | 13 ++-----------
 1 file changed, 2 insertions(+), 11 deletions(-)

diff --git a/arch/x86/include/asm/pmem.h b/arch/x86/include/asm/pmem.h
index 1e8dbb7..7f3413f 100644
--- a/arch/x86/include/asm/pmem.h
+++ b/arch/x86/include/asm/pmem.h
@@ -18,8 +18,7 @@
 #include <asm/cpufeature.h>
 #include <asm/special_insns.h>
 
-#ifdef ARCH_HAS_NOCACHE_UACCESS
-
+#ifdef CONFIG_ARCH_HAS_PMEM_API
 /**
  * arch_memcpy_to_pmem - copy data to persistent memory
  * @dst: destination buffer for the copy
@@ -79,14 +78,6 @@ static inline bool arch_has_wmb_pmem(void)
 	return false;
 #endif
 }
-#else /* ARCH_HAS_NOCACHE_UACCESS i.e. ARCH=um */
-extern void arch_memcpy_to_pmem(void __pmem *dst, const void *src, size_t n);
-extern void arch_wmb_pmem(void);
-
-static inline bool __arch_has_wmb_pmem(void)
-{
-	return false;
-}
-#endif
+#endif /* CONFIG_ARCH_HAS_PMEM_API */
 
 #endif /* __ASM_X86_PMEM_H__ */
-- 
2.1.0

[PATCH v5 5/7] pmem: add copy_from_iter_pmem() and clear_pmem()

[Ross Zwisler]

Add support for two new PMEM APIs, copy_from_iter_pmem() and
clear_pmem().  copy_from_iter_pmem() is used to copy data from an
iterator into a PMEM buffer.  clear_pmem() zeros a PMEM memory range.

Both of these new APIs must be explicitly ordered using a wmb_pmem()
function call and are implemented in such a way that the wmb_pmem()
will make the stores to PMEM durable.  Because both APIs are unordered
they can be called as needed without introducing any unwanted memory
barriers.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
---
 arch/x86/include/asm/pmem.h | 75 +++++++++++++++++++++++++++++++++++++++++++++
 include/linux/pmem.h        | 64 ++++++++++++++++++++++++++++++++++++--
 2 files changed, 137 insertions(+), 2 deletions(-)

diff --git a/arch/x86/include/asm/pmem.h b/arch/x86/include/asm/pmem.h
index 7f3413f..a3a0df6 100644
--- a/arch/x86/include/asm/pmem.h
+++ b/arch/x86/include/asm/pmem.h
@@ -66,6 +66,81 @@ static inline void arch_wmb_pmem(void)
 	pcommit_sfence();
 }
 
+/**
+ * __arch_wb_cache_pmem - write back a cache range with CLWB
+ * @vaddr:	virtual start address
+ * @size:	number of bytes to write back
+ *
+ * Write back a cache range using the CLWB (cache line write back)
+ * instruction.  This function requires explicit ordering with an
+ * arch_wmb_pmem() call.  This API is internal to the x86 PMEM implementation.
+ */
+static inline void __arch_wb_cache_pmem(void *vaddr, size_t size)
+{
+	u16 x86_clflush_size = boot_cpu_data.x86_clflush_size;
+	unsigned long clflush_mask = x86_clflush_size - 1;
+	void *vend = vaddr + size;
+	void *p;
+
+	for (p = (void *)((unsigned long)vaddr & ~clflush_mask);
+	     p < vend; p += x86_clflush_size)
+		clwb(p);
+}
+
+/*
+ * copy_from_iter_nocache() on x86 only uses non-temporal stores for iovec
+ * iterators, so for other types (bvec & kvec) we must do a cache write-back.
+ */
+static inline bool __iter_needs_pmem_wb(struct iov_iter *i)
+{
+	return iter_is_iovec(i) == false;
+}
+
+/**
+ * arch_copy_from_iter_pmem - copy data from an iterator to PMEM
+ * @addr:	PMEM destination address
+ * @bytes:	number of bytes to copy
+ * @i:		iterator with source data
+ *
+ * Copy data from the iterator 'i' to the PMEM buffer starting at 'addr'.
+ * This function requires explicit ordering with an arch_wmb_pmem() call.
+ */
+static inline size_t arch_copy_from_iter_pmem(void __pmem *addr, size_t bytes,
+		struct iov_iter *i)
+{
+	void *vaddr = (void __force *)addr;
+	size_t len;
+
+	/* TODO: skip the write-back by always using non-temporal stores */
+	len = copy_from_iter_nocache(vaddr, bytes, i);
+
+	if (__iter_needs_pmem_wb(i))
+		__arch_wb_cache_pmem(vaddr, bytes);
+
+	return len;
+}
+
+/**
+ * arch_clear_pmem - zero a PMEM memory range
+ * @addr:	virtual start address
+ * @size:	number of bytes to zero
+ *
+ * Write zeros into the memory range starting at 'addr' for 'size' bytes.
+ * This function requires explicit ordering with an arch_wmb_pmem() call.
+ */
+static inline void arch_clear_pmem(void __pmem *addr, size_t size)
+{
+	void *vaddr = (void __force *)addr;
+
+	/* TODO: implement the zeroing via non-temporal writes */
+	if (size == PAGE_SIZE && ((unsigned long)vaddr & ~PAGE_MASK) == 0)
+		clear_page(vaddr);
+	else
+		memset(vaddr, 0, size);
+
+	__arch_wb_cache_pmem(vaddr, size);
+}
+
 static inline bool arch_has_wmb_pmem(void)
 {
 #ifdef CONFIG_X86_64
diff --git a/include/linux/pmem.h b/include/linux/pmem.h
index a0706ea..a9d84bf 100644
--- a/include/linux/pmem.h
+++ b/include/linux/pmem.h
@@ -14,6 +14,7 @@
 #define __PMEM_H__
 
 #include <linux/io.h>
+#include <linux/uio.h>
 
 #ifdef CONFIG_ARCH_HAS_PMEM_API
 #include <asm/pmem.h>
@@ -33,12 +34,24 @@ static inline void arch_memcpy_to_pmem(void __pmem *dst, const void *src,
 {
 	BUG();
 }
+
+static inline size_t arch_copy_from_iter_pmem(void __pmem *addr, size_t bytes,
+		struct iov_iter *i)
+{
+	BUG();
+	return 0;
+}
+
+static inline void arch_clear_pmem(void __pmem *addr, size_t size)
+{
+	BUG();
+}
 #endif
 
 /*
  * Architectures that define ARCH_HAS_PMEM_API must provide
- * implementations for arch_memcpy_to_pmem(), arch_wmb_pmem(), and
- * arch_has_wmb_pmem().
+ * implementations for arch_memcpy_to_pmem(), arch_wmb_pmem(),
+ * arch_copy_from_iter_pmem(), arch_clear_pmem() and arch_has_wmb_pmem().
  */
 
 static inline void memcpy_from_pmem(void *dst, void __pmem const *src, size_t size)
@@ -78,6 +91,20 @@ static inline void default_memcpy_to_pmem(void __pmem *dst, const void *src,
 	memcpy((void __force *) dst, src, size);
 }
 
+static inline size_t default_copy_from_iter_pmem(void __pmem *addr,
+		size_t bytes, struct iov_iter *i)
+{
+	return copy_from_iter_nocache((void __force *)addr, bytes, i);
+}
+
+static inline void default_clear_pmem(void __pmem *addr, size_t size)
+{
+	if (size == PAGE_SIZE && ((unsigned long)addr & ~PAGE_MASK) == 0)
+		clear_page((void __force *)addr);
+	else
+		memset((void __force *)addr, 0, size);
+}
+
 /**
  * memremap_pmem - map physical persistent memory for pmem api
  * @offset: physical address of persistent memory
@@ -134,4 +161,37 @@ static inline void wmb_pmem(void)
 	if (arch_has_pmem_api())
 		arch_wmb_pmem();
 }
+
+/**
+ * copy_from_iter_pmem - copy data from an iterator to PMEM
+ * @addr:	PMEM destination address
+ * @bytes:	number of bytes to copy
+ * @i:		iterator with source data
+ *
+ * Copy data from the iterator 'i' to the PMEM buffer starting at 'addr'.
+ * This function requires explicit ordering with a wmb_pmem() call.
+ */
+static inline size_t copy_from_iter_pmem(void __pmem *addr, size_t bytes,
+		struct iov_iter *i)
+{
+	if (arch_has_pmem_api())
+		return arch_copy_from_iter_pmem(addr, bytes, i);
+	return default_copy_from_iter_pmem(addr, bytes, i);
+}
+
+/**
+ * clear_pmem - zero a PMEM memory range
+ * @addr:	virtual start address
+ * @size:	number of bytes to zero
+ *
+ * Write zeros into the memory range starting at 'addr' for 'size' bytes.
+ * This function requires explicit ordering with a wmb_pmem() call.
+ */
+static inline void clear_pmem(void __pmem *addr, size_t size)
+{
+	if (arch_has_pmem_api())
+		arch_clear_pmem(addr, size);
+	else
+		default_clear_pmem(addr, size);
+}
 #endif /* __PMEM_H__ */
-- 
2.1.0

[PATCH v5 6/7] dax: update I/O path to do proper PMEM flushing

[Ross Zwisler]

Update the DAX I/O path so that all operations that store data (I/O
writes, zeroing blocks, punching holes, etc.) properly synchronize the
stores to media using the PMEM API.  This ensures that the data DAX is
writing is durable on media before the operation completes.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 fs/dax.c | 40 ++++++++++++++++++++++++++--------------
 1 file changed, 26 insertions(+), 14 deletions(-)

diff --git a/fs/dax.c b/fs/dax.c
index c3e21cc..cb02a61 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -17,12 +17,14 @@
 #include <linux/atomic.h>
 #include <linux/blkdev.h>
 #include <linux/buffer_head.h>
+#include <linux/dax.h>
 #include <linux/fs.h>
 #include <linux/genhd.h>
 #include <linux/highmem.h>
 #include <linux/memcontrol.h>
 #include <linux/mm.h>
 #include <linux/mutex.h>
+#include <linux/pmem.h>
 #include <linux/sched.h>
 #include <linux/uio.h>
 #include <linux/vmstat.h>
@@ -46,10 +48,7 @@ int dax_clear_blocks(struct inode *inode, sector_t block, long size)
 			unsigned pgsz = PAGE_SIZE - offset_in_page(addr);
 			if (pgsz > count)
 				pgsz = count;
-			if (pgsz < PAGE_SIZE)
-				memset(addr, 0, pgsz);
-			else
-				clear_page(addr);
+			clear_pmem((void __pmem *)addr, pgsz);
 			addr += pgsz;
 			size -= pgsz;
 			count -= pgsz;
@@ -59,6 +58,7 @@ int dax_clear_blocks(struct inode *inode, sector_t block, long size)
 		}
 	} while (size);
 
+	wmb_pmem();
 	return 0;
 }
 EXPORT_SYMBOL_GPL(dax_clear_blocks);
@@ -70,15 +70,16 @@ static long dax_get_addr(struct buffer_head *bh, void **addr, unsigned blkbits)
 	return bdev_direct_access(bh->b_bdev, sector, addr, &pfn, bh->b_size);
 }
 
+/* the clear_pmem() calls are ordered by a wmb_pmem() in the caller */
 static void dax_new_buf(void *addr, unsigned size, unsigned first, loff_t pos,
 			loff_t end)
 {
 	loff_t final = end - pos + first; /* The final byte of the buffer */
 
 	if (first > 0)
-		memset(addr, 0, first);
+		clear_pmem((void __pmem *)addr, first);
 	if (final < size)
-		memset(addr + final, 0, size - final);
+		clear_pmem((void __pmem *)addr + final, size - final);
 }
 
 static bool buffer_written(struct buffer_head *bh)
@@ -108,12 +109,13 @@ static ssize_t dax_io(struct inode *inode, struct iov_iter *iter,
 	loff_t bh_max = start;
 	void *addr;
 	bool hole = false;
+	bool need_wmb = false;
 
 	if (iov_iter_rw(iter) != WRITE)
 		end = min(end, i_size_read(inode));
 
 	while (pos < end) {
-		unsigned len;
+		size_t len;
 		if (pos == max) {
 			unsigned blkbits = inode->i_blkbits;
 			sector_t block = pos >> blkbits;
@@ -145,18 +147,22 @@ static ssize_t dax_io(struct inode *inode, struct iov_iter *iter,
 				retval = dax_get_addr(bh, &addr, blkbits);
 				if (retval < 0)
 					break;
-				if (buffer_unwritten(bh) || buffer_new(bh))
+				if (buffer_unwritten(bh) || buffer_new(bh)) {
 					dax_new_buf(addr, retval, first, pos,
 									end);
+					need_wmb = true;
+				}
 				addr += first;
 				size = retval - first;
 			}
 			max = min(pos + size, end);
 		}
 
-		if (iov_iter_rw(iter) == WRITE)
-			len = copy_from_iter_nocache(addr, max - pos, iter);
-		else if (!hole)
+		if (iov_iter_rw(iter) == WRITE) {
+			len = copy_from_iter_pmem((void __pmem *)addr,
+					max - pos, iter);
+			need_wmb = true;
+		} else if (!hole)
 			len = copy_to_iter(addr, max - pos, iter);
 		else
 			len = iov_iter_zero(max - pos, iter);
@@ -168,6 +174,9 @@ static ssize_t dax_io(struct inode *inode, struct iov_iter *iter,
 		addr += len;
 	}
 
+	if (need_wmb)
+		wmb_pmem();
+
 	return (pos == start) ? retval : pos - start;
 }
 
@@ -303,8 +312,10 @@ static int dax_insert_mapping(struct inode *inode, struct buffer_head *bh,
 		goto out;
 	}
 
-	if (buffer_unwritten(bh) || buffer_new(bh))
-		clear_page(addr);
+	if (buffer_unwritten(bh) || buffer_new(bh)) {
+		clear_pmem((void __pmem *)addr, PAGE_SIZE);
+		wmb_pmem();
+	}
 
 	error = vm_insert_mixed(vma, vaddr, pfn);
 
@@ -542,7 +553,8 @@ int dax_zero_page_range(struct inode *inode, loff_t from, unsigned length,
 		err = dax_get_addr(&bh, &addr, inode->i_blkbits);
 		if (err < 0)
 			return err;
-		memset(addr + offset, 0, length);
+		clear_pmem((void __pmem *)addr + offset, length);
+		wmb_pmem();
 	}
 
 	return 0;
-- 
2.1.0

[PATCH v5 7/7] pmem, dax: have direct_access use __pmem annotation

[Ross Zwisler]

Update the annotation for the kaddr pointer returned by direct_access()
so that it is a __pmem pointer.  This is consistent with the PMEM driver
and with how this direct_access() pointer is used in the DAX code.

Signed-off-by: Ross Zwisler <ross.zwisler@linux.intel.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
---
 Documentation/filesystems/Locking |  3 ++-
 arch/powerpc/sysdev/axonram.c     |  7 ++++---
 drivers/block/brd.c               |  4 ++--
 drivers/nvdimm/pmem.c             |  4 ++--
 drivers/s390/block/dcssblk.c      | 10 ++++++----
 fs/block_dev.c                    |  2 +-
 fs/dax.c                          | 37 ++++++++++++++++++++-----------------
 include/linux/blkdev.h            |  8 ++++----
 8 files changed, 41 insertions(+), 34 deletions(-)

diff --git a/Documentation/filesystems/Locking b/Documentation/filesystems/Locking
index 6a34a0f..06d4434 100644
--- a/Documentation/filesystems/Locking
+++ b/Documentation/filesystems/Locking
@@ -397,7 +397,8 @@ prototypes:
 	int (*release) (struct gendisk *, fmode_t);
 	int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
 	int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
-	int (*direct_access) (struct block_device *, sector_t, void **, unsigned long *);
+	int (*direct_access) (struct block_device *, sector_t, void __pmem **,
+				unsigned long *);
 	int (*media_changed) (struct gendisk *);
 	void (*unlock_native_capacity) (struct gendisk *);
 	int (*revalidate_disk) (struct gendisk *);
diff --git a/arch/powerpc/sysdev/axonram.c b/arch/powerpc/sysdev/axonram.c
index ee90db1..a2be2a6 100644
--- a/arch/powerpc/sysdev/axonram.c
+++ b/arch/powerpc/sysdev/axonram.c
@@ -141,13 +141,14 @@ axon_ram_make_request(struct request_queue *queue, struct bio *bio)
  */
 static long
 axon_ram_direct_access(struct block_device *device, sector_t sector,
-		       void **kaddr, unsigned long *pfn, long size)
+		       void __pmem **kaddr, unsigned long *pfn, long size)
 {
 	struct axon_ram_bank *bank = device->bd_disk->private_data;
 	loff_t offset = (loff_t)sector << AXON_RAM_SECTOR_SHIFT;
+	void *addr = (void *)(bank->ph_addr + offset);
 
-	*kaddr = (void *)(bank->ph_addr + offset);
-	*pfn = virt_to_phys(*kaddr) >> PAGE_SHIFT;
+	*kaddr = (void __pmem *)addr;
+	*pfn = virt_to_phys(addr) >> PAGE_SHIFT;
 
 	return bank->size - offset;
 }
diff --git a/drivers/block/brd.c b/drivers/block/brd.c
index 5750b39..2691bb6 100644
--- a/drivers/block/brd.c
+++ b/drivers/block/brd.c
@@ -371,7 +371,7 @@ static int brd_rw_page(struct block_device *bdev, sector_t sector,
 
 #ifdef CONFIG_BLK_DEV_RAM_DAX
 static long brd_direct_access(struct block_device *bdev, sector_t sector,
-			void **kaddr, unsigned long *pfn, long size)
+			void __pmem **kaddr, unsigned long *pfn, long size)
 {
 	struct brd_device *brd = bdev->bd_disk->private_data;
 	struct page *page;
@@ -381,7 +381,7 @@ static long brd_direct_access(struct block_device *bdev, sector_t sector,
 	page = brd_insert_page(brd, sector);
 	if (!page)
 		return -ENOSPC;
-	*kaddr = page_address(page);
+	*kaddr = (void __pmem *)page_address(page);
 	*pfn = page_to_pfn(page);
 
 	/*
diff --git a/drivers/nvdimm/pmem.c b/drivers/nvdimm/pmem.c
index eb7552d..f3b6297 100644
--- a/drivers/nvdimm/pmem.c
+++ b/drivers/nvdimm/pmem.c
@@ -92,7 +92,7 @@ static int pmem_rw_page(struct block_device *bdev, sector_t sector,
 }
 
 static long pmem_direct_access(struct block_device *bdev, sector_t sector,
-			      void **kaddr, unsigned long *pfn, long size)
+		      void __pmem **kaddr, unsigned long *pfn, long size)
 {
 	struct pmem_device *pmem = bdev->bd_disk->private_data;
 	size_t offset = sector << 9;
@@ -101,7 +101,7 @@ static long pmem_direct_access(struct block_device *bdev, sector_t sector,
 		return -ENODEV;
 
 	/* FIXME convert DAX to comprehend that this mapping has a lifetime */
-	*kaddr = (void __force *) pmem->virt_addr + offset;
+	*kaddr = pmem->virt_addr + offset;
 	*pfn = (pmem->phys_addr + offset) >> PAGE_SHIFT;
 
 	return pmem->size - offset;
diff --git a/drivers/s390/block/dcssblk.c b/drivers/s390/block/dcssblk.c
index da21281..2c5a397 100644
--- a/drivers/s390/block/dcssblk.c
+++ b/drivers/s390/block/dcssblk.c
@@ -29,7 +29,7 @@ static int dcssblk_open(struct block_device *bdev, fmode_t mode);
 static void dcssblk_release(struct gendisk *disk, fmode_t mode);
 static void dcssblk_make_request(struct request_queue *q, struct bio *bio);
 static long dcssblk_direct_access(struct block_device *bdev, sector_t secnum,
-				 void **kaddr, unsigned long *pfn, long size);
+			 void __pmem **kaddr, unsigned long *pfn, long size);
 
 static char dcssblk_segments[DCSSBLK_PARM_LEN] = "\0";
 
@@ -879,18 +879,20 @@ fail:
 
 static long
 dcssblk_direct_access (struct block_device *bdev, sector_t secnum,
-			void **kaddr, unsigned long *pfn, long size)
+			void __pmem **kaddr, unsigned long *pfn, long size)
 {
 	struct dcssblk_dev_info *dev_info;
 	unsigned long offset, dev_sz;
+	void *addr;
 
 	dev_info = bdev->bd_disk->private_data;
 	if (!dev_info)
 		return -ENODEV;
 	dev_sz = dev_info->end - dev_info->start;
 	offset = secnum * 512;
-	*kaddr = (void *) (dev_info->start + offset);
-	*pfn = virt_to_phys(*kaddr) >> PAGE_SHIFT;
+	addr = (void *) (dev_info->start + offset);
+	*pfn = virt_to_phys(addr) >> PAGE_SHIFT;
+	*kaddr = (void __pmem *) addr;
 
 	return dev_sz - offset;
 }
diff --git a/fs/block_dev.c b/fs/block_dev.c
index 1982437..2345a98 100644
--- a/fs/block_dev.c
+++ b/fs/block_dev.c
@@ -441,7 +441,7 @@ EXPORT_SYMBOL_GPL(bdev_write_page);
  * accessible at this address.
  */
 long bdev_direct_access(struct block_device *bdev, sector_t sector,
-			void **addr, unsigned long *pfn, long size)
+			void __pmem **addr, unsigned long *pfn, long size)
 {
 	long avail;
 	const struct block_device_operations *ops = bdev->bd_disk->fops;
diff --git a/fs/dax.c b/fs/dax.c
index cb02a61..57e8867 100644
--- a/fs/dax.c
+++ b/fs/dax.c
@@ -36,7 +36,7 @@ int dax_clear_blocks(struct inode *inode, sector_t block, long size)
 
 	might_sleep();
 	do {
-		void *addr;
+		void __pmem *addr;
 		unsigned long pfn;
 		long count;
 
@@ -48,7 +48,7 @@ int dax_clear_blocks(struct inode *inode, sector_t block, long size)
 			unsigned pgsz = PAGE_SIZE - offset_in_page(addr);
 			if (pgsz > count)
 				pgsz = count;
-			clear_pmem((void __pmem *)addr, pgsz);
+			clear_pmem(addr, pgsz);
 			addr += pgsz;
 			size -= pgsz;
 			count -= pgsz;
@@ -63,7 +63,8 @@ int dax_clear_blocks(struct inode *inode, sector_t block, long size)
 }
 EXPORT_SYMBOL_GPL(dax_clear_blocks);
 
-static long dax_get_addr(struct buffer_head *bh, void **addr, unsigned blkbits)
+static long dax_get_addr(struct buffer_head *bh, void __pmem **addr,
+		unsigned blkbits)
 {
 	unsigned long pfn;
 	sector_t sector = bh->b_blocknr << (blkbits - 9);
@@ -71,15 +72,15 @@ static long dax_get_addr(struct buffer_head *bh, void **addr, unsigned blkbits)
 }
 
 /* the clear_pmem() calls are ordered by a wmb_pmem() in the caller */
-static void dax_new_buf(void *addr, unsigned size, unsigned first, loff_t pos,
-			loff_t end)
+static void dax_new_buf(void __pmem *addr, unsigned size, unsigned first,
+		loff_t pos, loff_t end)
 {
 	loff_t final = end - pos + first; /* The final byte of the buffer */
 
 	if (first > 0)
-		clear_pmem((void __pmem *)addr, first);
+		clear_pmem(addr, first);
 	if (final < size)
-		clear_pmem((void __pmem *)addr + final, size - final);
+		clear_pmem(addr + final, size - final);
 }
 
 static bool buffer_written(struct buffer_head *bh)
@@ -107,7 +108,7 @@ static ssize_t dax_io(struct inode *inode, struct iov_iter *iter,
 	loff_t pos = start;
 	loff_t max = start;
 	loff_t bh_max = start;
-	void *addr;
+	void __pmem *addr;
 	bool hole = false;
 	bool need_wmb = false;
 
@@ -159,11 +160,11 @@ static ssize_t dax_io(struct inode *inode, struct iov_iter *iter,
 		}
 
 		if (iov_iter_rw(iter) == WRITE) {
-			len = copy_from_iter_pmem((void __pmem *)addr,
-					max - pos, iter);
+			len = copy_from_iter_pmem(addr, max - pos, iter);
 			need_wmb = true;
 		} else if (!hole)
-			len = copy_to_iter(addr, max - pos, iter);
+			len = copy_to_iter((void __force *)addr, max - pos,
+					iter);
 		else
 			len = iov_iter_zero(max - pos, iter);
 
@@ -269,11 +270,13 @@ static int dax_load_hole(struct address_space *mapping, struct page *page,
 static int copy_user_bh(struct page *to, struct buffer_head *bh,
 			unsigned blkbits, unsigned long vaddr)
 {
-	void *vfrom, *vto;
+	void __pmem *vfrom;
+	void *vto;
+
 	if (dax_get_addr(bh, &vfrom, blkbits) < 0)
 		return -EIO;
 	vto = kmap_atomic(to);
-	copy_user_page(vto, vfrom, vaddr, to);
+	copy_user_page(vto, (void __force *)vfrom, vaddr, to);
 	kunmap_atomic(vto);
 	return 0;
 }
@@ -284,7 +287,7 @@ static int dax_insert_mapping(struct inode *inode, struct buffer_head *bh,
 	struct address_space *mapping = inode->i_mapping;
 	sector_t sector = bh->b_blocknr << (inode->i_blkbits - 9);
 	unsigned long vaddr = (unsigned long)vmf->virtual_address;
-	void *addr;
+	void __pmem *addr;
 	unsigned long pfn;
 	pgoff_t size;
 	int error;
@@ -313,7 +316,7 @@ static int dax_insert_mapping(struct inode *inode, struct buffer_head *bh,
 	}
 
 	if (buffer_unwritten(bh) || buffer_new(bh)) {
-		clear_pmem((void __pmem *)addr, PAGE_SIZE);
+		clear_pmem(addr, PAGE_SIZE);
 		wmb_pmem();
 	}
 
@@ -549,11 +552,11 @@ int dax_zero_page_range(struct inode *inode, loff_t from, unsigned length,
 	if (err < 0)
 		return err;
 	if (buffer_written(&bh)) {
-		void *addr;
+		void __pmem *addr;
 		err = dax_get_addr(&bh, &addr, inode->i_blkbits);
 		if (err < 0)
 			return err;
-		clear_pmem((void __pmem *)addr + offset, length);
+		clear_pmem(addr + offset, length);
 		wmb_pmem();
 	}
 
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index d4068c1..c401ecd 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -1555,8 +1555,8 @@ struct block_device_operations {
 	int (*rw_page)(struct block_device *, sector_t, struct page *, int rw);
 	int (*ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
 	int (*compat_ioctl) (struct block_device *, fmode_t, unsigned, unsigned long);
-	long (*direct_access)(struct block_device *, sector_t,
-					void **, unsigned long *pfn, long size);
+	long (*direct_access)(struct block_device *, sector_t, void __pmem **,
+			unsigned long *pfn, long size);
 	unsigned int (*check_events) (struct gendisk *disk,
 				      unsigned int clearing);
 	/* ->media_changed() is DEPRECATED, use ->check_events() instead */
@@ -1574,8 +1574,8 @@ extern int __blkdev_driver_ioctl(struct block_device *, fmode_t, unsigned int,
 extern int bdev_read_page(struct block_device *, sector_t, struct page *);
 extern int bdev_write_page(struct block_device *, sector_t, struct page *,
 						struct writeback_control *);
-extern long bdev_direct_access(struct block_device *, sector_t, void **addr,
-						unsigned long *pfn, long size);
+extern long bdev_direct_access(struct block_device *, sector_t,
+		void __pmem **addr, unsigned long *pfn, long size);
 #else /* CONFIG_BLOCK */
 
 struct block_device;
-- 
2.1.0

Re: [PATCH v5 5/7] pmem: add copy_from_iter_pmem() and clear_pmem()

[Christoph Hellwig]

Looks good,

Reviewed-by: Christoph Hellwig <hch@lst.de>