From 0aed55af88345b5d673240f90e671d79662fb01e Mon Sep 17 00:00:00 2001 From: Dan Williams Date: Mon, 29 May 2017 12:22:50 -0700 Subject: x86, uaccess: introduce copy_from_iter_flushcache for pmem / cache-bypass operations The pmem driver has a need to transfer data with a persistent memory destination and be able to rely on the fact that the destination writes are not cached. It is sufficient for the writes to be flushed to a cpu-store-buffer (non-temporal / "movnt" in x86 terms), as we expect userspace to call fsync() to ensure data-writes have reached a power-fail-safe zone in the platform. The fsync() triggers a REQ_FUA or REQ_FLUSH to the pmem driver which will turn around and fence previous writes with an "sfence". Implement a __copy_from_user_inatomic_flushcache, memcpy_page_flushcache, and memcpy_flushcache, that guarantee that the destination buffer is not dirty in the cpu cache on completion. The new copy_from_iter_flushcache and sub-routines will be used to replace the "pmem api" (include/linux/pmem.h + arch/x86/include/asm/pmem.h). The availability of copy_from_iter_flushcache() and memcpy_flushcache() are gated by the CONFIG_ARCH_HAS_UACCESS_FLUSHCACHE config symbol, and fallback to copy_from_iter_nocache() and plain memcpy() otherwise. This is meant to satisfy the concern from Linus that if a driver wants to do something beyond the normal nocache semantics it should be something private to that driver [1], and Al's concern that anything uaccess related belongs with the rest of the uaccess code [2]. The first consumer of this interface is a new 'copy_from_iter' dax operation so that pmem can inject cache maintenance operations without imposing this overhead on other dax-capable drivers. [1]: https://lists.01.org/pipermail/linux-nvdimm/2017-January/008364.html [2]: https://lists.01.org/pipermail/linux-nvdimm/2017-April/009942.html Cc: Cc: Jan Kara Cc: Jeff Moyer Cc: Ingo Molnar Cc: Christoph Hellwig Cc: Toshi Kani Cc: "H. Peter Anvin" Cc: Al Viro Cc: Thomas Gleixner Cc: Matthew Wilcox Reviewed-by: Ross Zwisler Signed-off-by: Dan Williams --- include/linux/dax.h | 3 +++ 1 file changed, 3 insertions(+) (limited to 'include/linux/dax.h') diff --git a/include/linux/dax.h b/include/linux/dax.h index 5ec1f6c47716..bbe79ed90e2b 100644 --- a/include/linux/dax.h +++ b/include/linux/dax.h @@ -16,6 +16,9 @@ struct dax_operations { */ long (*direct_access)(struct dax_device *, pgoff_t, long, void **, pfn_t *); + /* copy_from_iter: dax-driver override for default copy_from_iter */ + size_t (*copy_from_iter)(struct dax_device *, pgoff_t, void *, size_t, + struct iov_iter *); }; #if IS_ENABLED(CONFIG_DAX) -- cgit v1.2.3 From 7e026c8c0a4200da86bc51edeaad79dcdccf78ca Mon Sep 17 00:00:00 2001 From: Dan Williams Date: Mon, 29 May 2017 12:57:56 -0700 Subject: dm: add ->copy_from_iter() dax operation support Allow device-mapper to route copy_from_iter operations to the per-target implementation. In order for the device stacking to work we need a dax_dev and a pgoff relative to that device. This gives each layer of the stack the information it needs to look up the operation pointer for the next level. This conceptually allows for an array of mixed device drivers with varying copy_from_iter implementations. Reviewed-by: Toshi Kani Reviewed-by: Mike Snitzer Signed-off-by: Dan Williams --- drivers/dax/super.c | 13 +++++++++++++ drivers/md/dm-linear.c | 15 +++++++++++++++ drivers/md/dm-stripe.c | 20 ++++++++++++++++++++ drivers/md/dm.c | 26 ++++++++++++++++++++++++++ include/linux/dax.h | 2 ++ include/linux/device-mapper.h | 3 +++ 6 files changed, 79 insertions(+) (limited to 'include/linux/dax.h') diff --git a/drivers/dax/super.c b/drivers/dax/super.c index 6ed32aac8bbe..dd299e55f65d 100644 --- a/drivers/dax/super.c +++ b/drivers/dax/super.c @@ -18,6 +18,7 @@ #include #include #include +#include #include #include @@ -172,6 +173,18 @@ long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, } EXPORT_SYMBOL_GPL(dax_direct_access); +size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, + size_t bytes, struct iov_iter *i) +{ + if (!dax_alive(dax_dev)) + return 0; + + if (!dax_dev->ops->copy_from_iter) + return copy_from_iter(addr, bytes, i); + return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i); +} +EXPORT_SYMBOL_GPL(dax_copy_from_iter); + bool dax_alive(struct dax_device *dax_dev) { lockdep_assert_held(&dax_srcu); diff --git a/drivers/md/dm-linear.c b/drivers/md/dm-linear.c index 7d42a9d9f406..0841ec1bfbad 100644 --- a/drivers/md/dm-linear.c +++ b/drivers/md/dm-linear.c @@ -159,6 +159,20 @@ static long linear_dax_direct_access(struct dm_target *ti, pgoff_t pgoff, return dax_direct_access(dax_dev, pgoff, nr_pages, kaddr, pfn); } +static size_t linear_dax_copy_from_iter(struct dm_target *ti, pgoff_t pgoff, + void *addr, size_t bytes, struct iov_iter *i) +{ + struct linear_c *lc = ti->private; + struct block_device *bdev = lc->dev->bdev; + struct dax_device *dax_dev = lc->dev->dax_dev; + sector_t dev_sector, sector = pgoff * PAGE_SECTORS; + + dev_sector = linear_map_sector(ti, sector); + if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(bytes, PAGE_SIZE), &pgoff)) + return 0; + return dax_copy_from_iter(dax_dev, pgoff, addr, bytes, i); +} + static struct target_type linear_target = { .name = "linear", .version = {1, 3, 0}, @@ -171,6 +185,7 @@ static struct target_type linear_target = { .prepare_ioctl = linear_prepare_ioctl, .iterate_devices = linear_iterate_devices, .direct_access = linear_dax_direct_access, + .dax_copy_from_iter = linear_dax_copy_from_iter, }; int __init dm_linear_init(void) diff --git a/drivers/md/dm-stripe.c b/drivers/md/dm-stripe.c index 75152482f3ad..1ef914f9ca72 100644 --- a/drivers/md/dm-stripe.c +++ b/drivers/md/dm-stripe.c @@ -332,6 +332,25 @@ static long stripe_dax_direct_access(struct dm_target *ti, pgoff_t pgoff, return dax_direct_access(dax_dev, pgoff, nr_pages, kaddr, pfn); } +static size_t stripe_dax_copy_from_iter(struct dm_target *ti, pgoff_t pgoff, + void *addr, size_t bytes, struct iov_iter *i) +{ + sector_t dev_sector, sector = pgoff * PAGE_SECTORS; + struct stripe_c *sc = ti->private; + struct dax_device *dax_dev; + struct block_device *bdev; + uint32_t stripe; + + stripe_map_sector(sc, sector, &stripe, &dev_sector); + dev_sector += sc->stripe[stripe].physical_start; + dax_dev = sc->stripe[stripe].dev->dax_dev; + bdev = sc->stripe[stripe].dev->bdev; + + if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(bytes, PAGE_SIZE), &pgoff)) + return 0; + return dax_copy_from_iter(dax_dev, pgoff, addr, bytes, i); +} + /* * Stripe status: * @@ -451,6 +470,7 @@ static struct target_type stripe_target = { .iterate_devices = stripe_iterate_devices, .io_hints = stripe_io_hints, .direct_access = stripe_dax_direct_access, + .dax_copy_from_iter = stripe_dax_copy_from_iter, }; int __init dm_stripe_init(void) diff --git a/drivers/md/dm.c b/drivers/md/dm.c index 37ccd73c79ec..7faaceb52819 100644 --- a/drivers/md/dm.c +++ b/drivers/md/dm.c @@ -19,6 +19,7 @@ #include #include #include +#include #include #include #include @@ -969,6 +970,30 @@ static long dm_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, return ret; } +static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, + void *addr, size_t bytes, struct iov_iter *i) +{ + struct mapped_device *md = dax_get_private(dax_dev); + sector_t sector = pgoff * PAGE_SECTORS; + struct dm_target *ti; + long ret = 0; + int srcu_idx; + + ti = dm_dax_get_live_target(md, sector, &srcu_idx); + + if (!ti) + goto out; + if (!ti->type->dax_copy_from_iter) { + ret = copy_from_iter(addr, bytes, i); + goto out; + } + ret = ti->type->dax_copy_from_iter(ti, pgoff, addr, bytes, i); + out: + dm_put_live_table(md, srcu_idx); + + return ret; +} + /* * A target may call dm_accept_partial_bio only from the map routine. It is * allowed for all bio types except REQ_PREFLUSH. @@ -2859,6 +2884,7 @@ static const struct block_device_operations dm_blk_dops = { static const struct dax_operations dm_dax_ops = { .direct_access = dm_dax_direct_access, + .copy_from_iter = dm_dax_copy_from_iter, }; /* diff --git a/include/linux/dax.h b/include/linux/dax.h index bbe79ed90e2b..28e398f8c59e 100644 --- a/include/linux/dax.h +++ b/include/linux/dax.h @@ -78,6 +78,8 @@ void kill_dax(struct dax_device *dax_dev); void *dax_get_private(struct dax_device *dax_dev); long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn); +size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, + size_t bytes, struct iov_iter *i); /* * We use lowest available bit in exceptional entry for locking, one bit for diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h index f4c639c0c362..11c8a0a92f9c 100644 --- a/include/linux/device-mapper.h +++ b/include/linux/device-mapper.h @@ -132,6 +132,8 @@ typedef int (*dm_busy_fn) (struct dm_target *ti); */ typedef long (*dm_dax_direct_access_fn) (struct dm_target *ti, pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn); +typedef size_t (*dm_dax_copy_from_iter_fn)(struct dm_target *ti, pgoff_t pgoff, + void *addr, size_t bytes, struct iov_iter *i); #define PAGE_SECTORS (PAGE_SIZE / 512) void dm_error(const char *message); @@ -181,6 +183,7 @@ struct target_type { dm_iterate_devices_fn iterate_devices; dm_io_hints_fn io_hints; dm_dax_direct_access_fn direct_access; + dm_dax_copy_from_iter_fn dax_copy_from_iter; /* For internal device-mapper use. */ struct list_head list; -- cgit v1.2.3 From 3c1cebff23cdca01c421411e953a9e239f2b9ef9 Mon Sep 17 00:00:00 2001 From: Dan Williams Date: Mon, 29 May 2017 12:58:19 -0700 Subject: dax, pmem: introduce an optional 'flush' dax_operation Filesystem-DAX flushes caches whenever it writes to the address returned through dax_direct_access() and when writing back dirty radix entries. That flushing is only required in the pmem case, so add a dax operation to allow pmem to take this extra action, but skip it for other dax capable devices that do not provide a flush routine. An example for this differentiation might be a volatile ram disk where there is no expectation of persistence. In fact the pmem driver itself might front such an address range specified by the NFIT. So, this "no flush" property might be something passed down by the bus / libnvdimm. Cc: Christoph Hellwig Cc: Matthew Wilcox Cc: Ross Zwisler Reviewed-by: Jan Kara Signed-off-by: Dan Williams --- drivers/nvdimm/pmem.c | 7 +++++++ include/linux/dax.h | 2 ++ 2 files changed, 9 insertions(+) (limited to 'include/linux/dax.h') diff --git a/drivers/nvdimm/pmem.c b/drivers/nvdimm/pmem.c index 2f3aefe565c6..823b07774244 100644 --- a/drivers/nvdimm/pmem.c +++ b/drivers/nvdimm/pmem.c @@ -242,9 +242,16 @@ static size_t pmem_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, return copy_from_iter_flushcache(addr, bytes, i); } +static void pmem_dax_flush(struct dax_device *dax_dev, pgoff_t pgoff, + void *addr, size_t size) +{ + wb_cache_pmem(addr, size); +} + static const struct dax_operations pmem_dax_ops = { .direct_access = pmem_dax_direct_access, .copy_from_iter = pmem_copy_from_iter, + .flush = pmem_dax_flush, }; static void pmem_release_queue(void *q) diff --git a/include/linux/dax.h b/include/linux/dax.h index 28e398f8c59e..407dd3ff6e54 100644 --- a/include/linux/dax.h +++ b/include/linux/dax.h @@ -19,6 +19,8 @@ struct dax_operations { /* copy_from_iter: dax-driver override for default copy_from_iter */ size_t (*copy_from_iter)(struct dax_device *, pgoff_t, void *, size_t, struct iov_iter *); + /* flush: optional driver-specific cache management after writes */ + void (*flush)(struct dax_device *, pgoff_t, void *, size_t); }; #if IS_ENABLED(CONFIG_DAX) -- cgit v1.2.3 From abebfbe2f7315dd3ec9a0c69596a76e32beb5749 Mon Sep 17 00:00:00 2001 From: Dan Williams Date: Mon, 29 May 2017 13:02:52 -0700 Subject: dm: add ->flush() dax operation support Allow device-mapper to route flush operations to the per-target implementation. In order for the device stacking to work we need a dax_dev and a pgoff relative to that device. This gives each layer of the stack the information it needs to look up the operation pointer for the next level. This conceptually allows for an array of mixed device drivers with varying flush implementations. Reviewed-by: Toshi Kani Reviewed-by: Mike Snitzer Signed-off-by: Dan Williams --- drivers/dax/super.c | 11 +++++++++++ drivers/md/dm-linear.c | 15 +++++++++++++++ drivers/md/dm-stripe.c | 20 ++++++++++++++++++++ drivers/md/dm.c | 19 +++++++++++++++++++ include/linux/dax.h | 2 ++ include/linux/device-mapper.h | 3 +++ 6 files changed, 70 insertions(+) (limited to 'include/linux/dax.h') diff --git a/drivers/dax/super.c b/drivers/dax/super.c index dd299e55f65d..b7729e4d351a 100644 --- a/drivers/dax/super.c +++ b/drivers/dax/super.c @@ -185,6 +185,17 @@ size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, } EXPORT_SYMBOL_GPL(dax_copy_from_iter); +void dax_flush(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, + size_t size) +{ + if (!dax_alive(dax_dev)) + return; + + if (dax_dev->ops->flush) + dax_dev->ops->flush(dax_dev, pgoff, addr, size); +} +EXPORT_SYMBOL_GPL(dax_flush); + bool dax_alive(struct dax_device *dax_dev) { lockdep_assert_held(&dax_srcu); diff --git a/drivers/md/dm-linear.c b/drivers/md/dm-linear.c index 0841ec1bfbad..25e661974319 100644 --- a/drivers/md/dm-linear.c +++ b/drivers/md/dm-linear.c @@ -173,6 +173,20 @@ static size_t linear_dax_copy_from_iter(struct dm_target *ti, pgoff_t pgoff, return dax_copy_from_iter(dax_dev, pgoff, addr, bytes, i); } +static void linear_dax_flush(struct dm_target *ti, pgoff_t pgoff, void *addr, + size_t size) +{ + struct linear_c *lc = ti->private; + struct block_device *bdev = lc->dev->bdev; + struct dax_device *dax_dev = lc->dev->dax_dev; + sector_t dev_sector, sector = pgoff * PAGE_SECTORS; + + dev_sector = linear_map_sector(ti, sector); + if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(size, PAGE_SIZE), &pgoff)) + return; + dax_flush(dax_dev, pgoff, addr, size); +} + static struct target_type linear_target = { .name = "linear", .version = {1, 3, 0}, @@ -186,6 +200,7 @@ static struct target_type linear_target = { .iterate_devices = linear_iterate_devices, .direct_access = linear_dax_direct_access, .dax_copy_from_iter = linear_dax_copy_from_iter, + .dax_flush = linear_dax_flush, }; int __init dm_linear_init(void) diff --git a/drivers/md/dm-stripe.c b/drivers/md/dm-stripe.c index 1ef914f9ca72..8e73517967b6 100644 --- a/drivers/md/dm-stripe.c +++ b/drivers/md/dm-stripe.c @@ -351,6 +351,25 @@ static size_t stripe_dax_copy_from_iter(struct dm_target *ti, pgoff_t pgoff, return dax_copy_from_iter(dax_dev, pgoff, addr, bytes, i); } +static void stripe_dax_flush(struct dm_target *ti, pgoff_t pgoff, void *addr, + size_t size) +{ + sector_t dev_sector, sector = pgoff * PAGE_SECTORS; + struct stripe_c *sc = ti->private; + struct dax_device *dax_dev; + struct block_device *bdev; + uint32_t stripe; + + stripe_map_sector(sc, sector, &stripe, &dev_sector); + dev_sector += sc->stripe[stripe].physical_start; + dax_dev = sc->stripe[stripe].dev->dax_dev; + bdev = sc->stripe[stripe].dev->bdev; + + if (bdev_dax_pgoff(bdev, dev_sector, ALIGN(size, PAGE_SIZE), &pgoff)) + return; + dax_flush(dax_dev, pgoff, addr, size); +} + /* * Stripe status: * @@ -471,6 +490,7 @@ static struct target_type stripe_target = { .io_hints = stripe_io_hints, .direct_access = stripe_dax_direct_access, .dax_copy_from_iter = stripe_dax_copy_from_iter, + .dax_flush = stripe_dax_flush, }; int __init dm_stripe_init(void) diff --git a/drivers/md/dm.c b/drivers/md/dm.c index 7faaceb52819..09b3efdc8abf 100644 --- a/drivers/md/dm.c +++ b/drivers/md/dm.c @@ -994,6 +994,24 @@ static size_t dm_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, return ret; } +static void dm_dax_flush(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, + size_t size) +{ + struct mapped_device *md = dax_get_private(dax_dev); + sector_t sector = pgoff * PAGE_SECTORS; + struct dm_target *ti; + int srcu_idx; + + ti = dm_dax_get_live_target(md, sector, &srcu_idx); + + if (!ti) + goto out; + if (ti->type->dax_flush) + ti->type->dax_flush(ti, pgoff, addr, size); + out: + dm_put_live_table(md, srcu_idx); +} + /* * A target may call dm_accept_partial_bio only from the map routine. It is * allowed for all bio types except REQ_PREFLUSH. @@ -2885,6 +2903,7 @@ static const struct block_device_operations dm_blk_dops = { static const struct dax_operations dm_dax_ops = { .direct_access = dm_dax_direct_access, .copy_from_iter = dm_dax_copy_from_iter, + .flush = dm_dax_flush, }; /* diff --git a/include/linux/dax.h b/include/linux/dax.h index 407dd3ff6e54..1f6b6072af64 100644 --- a/include/linux/dax.h +++ b/include/linux/dax.h @@ -82,6 +82,8 @@ long dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn); size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i); +void dax_flush(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, + size_t size); /* * We use lowest available bit in exceptional entry for locking, one bit for diff --git a/include/linux/device-mapper.h b/include/linux/device-mapper.h index 11c8a0a92f9c..67bfe8ddcb32 100644 --- a/include/linux/device-mapper.h +++ b/include/linux/device-mapper.h @@ -134,6 +134,8 @@ typedef long (*dm_dax_direct_access_fn) (struct dm_target *ti, pgoff_t pgoff, long nr_pages, void **kaddr, pfn_t *pfn); typedef size_t (*dm_dax_copy_from_iter_fn)(struct dm_target *ti, pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i); +typedef void (*dm_dax_flush_fn)(struct dm_target *ti, pgoff_t pgoff, void *addr, + size_t size); #define PAGE_SECTORS (PAGE_SIZE / 512) void dm_error(const char *message); @@ -184,6 +186,7 @@ struct target_type { dm_io_hints_fn io_hints; dm_dax_direct_access_fn direct_access; dm_dax_copy_from_iter_fn dax_copy_from_iter; + dm_dax_flush_fn dax_flush; /* For internal device-mapper use. */ struct list_head list; -- cgit v1.2.3 From 5d61e43b3975c0582003329d9de9d5e85abf5d33 Mon Sep 17 00:00:00 2001 From: Dan Williams Date: Tue, 27 Jun 2017 13:06:22 -0700 Subject: dax: remove default copy_from_iter fallback Require all dax-drivers to register a ->copy_from_iter() operation so that it is clear which dax_operations are optional and which must be implemented for filesystem-dax to operate. Cc: Gerald Schaefer Suggested-by: Christoph Hellwig Signed-off-by: Dan Williams --- arch/powerpc/sysdev/axonram.c | 8 ++++++++ drivers/block/brd.c | 8 ++++++++ drivers/dax/super.c | 2 -- drivers/s390/block/dcssblk.c | 8 ++++++++ include/linux/dax.h | 2 +- 5 files changed, 25 insertions(+), 3 deletions(-) (limited to 'include/linux/dax.h') diff --git a/arch/powerpc/sysdev/axonram.c b/arch/powerpc/sysdev/axonram.c index a7fe5fee744f..2799706106c6 100644 --- a/arch/powerpc/sysdev/axonram.c +++ b/arch/powerpc/sysdev/axonram.c @@ -45,6 +45,7 @@ #include #include #include +#include #include #include @@ -163,8 +164,15 @@ axon_ram_dax_direct_access(struct dax_device *dax_dev, pgoff_t pgoff, long nr_pa return __axon_ram_direct_access(bank, pgoff, nr_pages, kaddr, pfn); } +static size_t axon_ram_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, + void *addr, size_t bytes, struct iov_iter *i) +{ + return copy_from_iter(addr, bytes, i); +} + static const struct dax_operations axon_ram_dax_ops = { .direct_access = axon_ram_dax_direct_access, + .copy_from_iter = axon_ram_copy_from_iter, }; /** diff --git a/drivers/block/brd.c b/drivers/block/brd.c index 57b574f2f66a..f2a7ac350f6a 100644 --- a/drivers/block/brd.c +++ b/drivers/block/brd.c @@ -22,6 +22,7 @@ #ifdef CONFIG_BLK_DEV_RAM_DAX #include #include +#include #endif #include @@ -354,8 +355,15 @@ static long brd_dax_direct_access(struct dax_device *dax_dev, return __brd_direct_access(brd, pgoff, nr_pages, kaddr, pfn); } +static size_t brd_dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, + void *addr, size_t bytes, struct iov_iter *i) +{ + return copy_from_iter(addr, bytes, i); +} + static const struct dax_operations brd_dax_ops = { .direct_access = brd_dax_direct_access, + .copy_from_iter = brd_dax_copy_from_iter, }; #endif diff --git a/drivers/dax/super.c b/drivers/dax/super.c index b7729e4d351a..9e0160b950d7 100644 --- a/drivers/dax/super.c +++ b/drivers/dax/super.c @@ -179,8 +179,6 @@ size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, if (!dax_alive(dax_dev)) return 0; - if (!dax_dev->ops->copy_from_iter) - return copy_from_iter(addr, bytes, i); return dax_dev->ops->copy_from_iter(dax_dev, pgoff, addr, bytes, i); } EXPORT_SYMBOL_GPL(dax_copy_from_iter); diff --git a/drivers/s390/block/dcssblk.c b/drivers/s390/block/dcssblk.c index 36e5280af3e4..88fa7b3f7a9d 100644 --- a/drivers/s390/block/dcssblk.c +++ b/drivers/s390/block/dcssblk.c @@ -18,6 +18,7 @@ #include #include #include +#include #include #include #include @@ -43,8 +44,15 @@ static const struct block_device_operations dcssblk_devops = { .release = dcssblk_release, }; +static size_t dcssblk_dax_copy_from_iter(struct dax_device *dax_dev, + pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i) +{ + return copy_from_iter(addr, bytes, i); +} + static const struct dax_operations dcssblk_dax_ops = { .direct_access = dcssblk_dax_direct_access, + .copy_from_iter = dcssblk_dax_copy_from_iter, }; struct dcssblk_dev_info { diff --git a/include/linux/dax.h b/include/linux/dax.h index 1f6b6072af64..73fca1bebaf3 100644 --- a/include/linux/dax.h +++ b/include/linux/dax.h @@ -16,7 +16,7 @@ struct dax_operations { */ long (*direct_access)(struct dax_device *, pgoff_t, long, void **, pfn_t *); - /* copy_from_iter: dax-driver override for default copy_from_iter */ + /* copy_from_iter: required operation for fs-dax direct-i/o */ size_t (*copy_from_iter)(struct dax_device *, pgoff_t, void *, size_t, struct iov_iter *); /* flush: optional driver-specific cache management after writes */ -- cgit v1.2.3 From 6e0c90d691cd5d90569f5918ab03eb76c81f9c6e Mon Sep 17 00:00:00 2001 From: Dan Williams Date: Mon, 26 Jun 2017 21:28:41 -0700 Subject: libnvdimm, pmem, dax: export a cache control attribute The dax_flush() operation can be turned into a nop on platforms where firmware arranges for cpu caches to be flushed on a power-fail event. The ACPI 6.2 specification defines a mechanism for the platform to indicate this capability so the kernel can select the proper default. However, for other platforms, the administrator must toggle this setting manually. Given this flush setting is a dax-specific mechanism we advertise it through a 'dax' attribute group hanging off a host device. For example, a 'pmem0' block-device gets a 'dax' sysfs-subdirectory with a 'write_cache' attribute to control response to dax cache flush requests. This is similar to the 'queue/write_cache' attribute that appears under block devices. Cc: Jan Kara Cc: Jeff Moyer Cc: Matthew Wilcox Cc: Ross Zwisler Suggested-by: Christoph Hellwig Signed-off-by: Dan Williams --- drivers/dax/super.c | 79 +++++++++++++++++++++++++++++++++++++++++++++++++++ drivers/nvdimm/pmem.c | 10 +++++++ include/linux/dax.h | 3 ++ 3 files changed, 92 insertions(+) (limited to 'include/linux/dax.h') diff --git a/drivers/dax/super.c b/drivers/dax/super.c index 8bf71195921b..4827251782a1 100644 --- a/drivers/dax/super.c +++ b/drivers/dax/super.c @@ -119,6 +119,8 @@ EXPORT_SYMBOL_GPL(__bdev_dax_supported); enum dax_device_flags { /* !alive + rcu grace period == no new operations / mappings */ DAXDEV_ALIVE, + /* gate whether dax_flush() calls the low level flush routine */ + DAXDEV_WRITE_CACHE, }; /** @@ -139,6 +141,71 @@ struct dax_device { const struct dax_operations *ops; }; +static ssize_t write_cache_show(struct device *dev, + struct device_attribute *attr, char *buf) +{ + struct dax_device *dax_dev = dax_get_by_host(dev_name(dev)); + ssize_t rc; + + WARN_ON_ONCE(!dax_dev); + if (!dax_dev) + return -ENXIO; + + rc = sprintf(buf, "%d\n", !!test_bit(DAXDEV_WRITE_CACHE, + &dax_dev->flags)); + put_dax(dax_dev); + return rc; +} + +static ssize_t write_cache_store(struct device *dev, + struct device_attribute *attr, const char *buf, size_t len) +{ + bool write_cache; + int rc = strtobool(buf, &write_cache); + struct dax_device *dax_dev = dax_get_by_host(dev_name(dev)); + + WARN_ON_ONCE(!dax_dev); + if (!dax_dev) + return -ENXIO; + + if (rc) + len = rc; + else if (write_cache) + set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); + else + clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); + + put_dax(dax_dev); + return len; +} +static DEVICE_ATTR_RW(write_cache); + +static umode_t dax_visible(struct kobject *kobj, struct attribute *a, int n) +{ + struct device *dev = container_of(kobj, typeof(*dev), kobj); + struct dax_device *dax_dev = dax_get_by_host(dev_name(dev)); + + WARN_ON_ONCE(!dax_dev); + if (!dax_dev) + return 0; + + if (a == &dev_attr_write_cache.attr && !dax_dev->ops->flush) + return 0; + return a->mode; +} + +static struct attribute *dax_attributes[] = { + &dev_attr_write_cache.attr, + NULL, +}; + +struct attribute_group dax_attribute_group = { + .name = "dax", + .attrs = dax_attributes, + .is_visible = dax_visible, +}; +EXPORT_SYMBOL_GPL(dax_attribute_group); + /** * dax_direct_access() - translate a device pgoff to an absolute pfn * @dax_dev: a dax_device instance representing the logical memory range @@ -194,11 +261,23 @@ void dax_flush(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, if (!dax_alive(dax_dev)) return; + if (!test_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags)) + return; + if (dax_dev->ops->flush) dax_dev->ops->flush(dax_dev, pgoff, addr, size); } EXPORT_SYMBOL_GPL(dax_flush); +void dax_write_cache(struct dax_device *dax_dev, bool wc) +{ + if (wc) + set_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); + else + clear_bit(DAXDEV_WRITE_CACHE, &dax_dev->flags); +} +EXPORT_SYMBOL_GPL(dax_write_cache); + bool dax_alive(struct dax_device *dax_dev) { lockdep_assert_held(&dax_srcu); diff --git a/drivers/nvdimm/pmem.c b/drivers/nvdimm/pmem.c index 06f6c27ec1e9..7339d184070e 100644 --- a/drivers/nvdimm/pmem.c +++ b/drivers/nvdimm/pmem.c @@ -253,6 +253,11 @@ static const struct dax_operations pmem_dax_ops = { .flush = pmem_dax_flush, }; +static const struct attribute_group *pmem_attribute_groups[] = { + &dax_attribute_group, + NULL, +}; + static void pmem_release_queue(void *q) { blk_cleanup_queue(q); @@ -287,6 +292,7 @@ static int pmem_attach_disk(struct device *dev, struct pmem_device *pmem; struct resource pfn_res; struct request_queue *q; + struct device *gendev; struct gendisk *disk; void *addr; @@ -384,8 +390,12 @@ static int pmem_attach_disk(struct device *dev, put_disk(disk); return -ENOMEM; } + dax_write_cache(dax_dev, true); pmem->dax_dev = dax_dev; + gendev = disk_to_dev(disk); + gendev->groups = pmem_attribute_groups; + device_add_disk(dev, disk); if (devm_add_action_or_reset(dev, pmem_release_disk, pmem)) return -ENOMEM; diff --git a/include/linux/dax.h b/include/linux/dax.h index 73fca1bebaf3..8f39db7439c3 100644 --- a/include/linux/dax.h +++ b/include/linux/dax.h @@ -23,6 +23,8 @@ struct dax_operations { void (*flush)(struct dax_device *, pgoff_t, void *, size_t); }; +extern struct attribute_group dax_attribute_group; + #if IS_ENABLED(CONFIG_DAX) struct dax_device *dax_get_by_host(const char *host); void put_dax(struct dax_device *dax_dev); @@ -84,6 +86,7 @@ size_t dax_copy_from_iter(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, size_t bytes, struct iov_iter *i); void dax_flush(struct dax_device *dax_dev, pgoff_t pgoff, void *addr, size_t size); +void dax_write_cache(struct dax_device *dax_dev, bool wc); /* * We use lowest available bit in exceptional entry for locking, one bit for -- cgit v1.2.3 From baabda261424517110ea98c6651f632ebf2561e3 Mon Sep 17 00:00:00 2001 From: Dan Williams Date: Mon, 10 Jul 2017 15:48:25 -0700 Subject: mm: always enable thp for dax mappings The madvise policy for transparent huge pages is meant to avoid unwanted allocations of transparent huge pages. It allows a policy of disabling the extra memory pressure and effort to arrange for a huge page when it is not needed. DAX by definition never incurs this overhead since it is statically allocated. The policy choice makes even less sense for device-dax which tries to guarantee a given tlb-fault size. Specifically, the following setting: echo never > /sys/kernel/mm/transparent_hugepage/enabled ...violates that guarantee and silently disables all device-dax instances with a 2M or 1G alignment. So, let's avoid that non-obvious side effect by force enabling thp for dax mappings in all cases. It is worth noting that the reason this uses vma_is_dax(), and the resulting header include changes, is that previous attempts to add a VM_DAX flag were NAKd. Link: http://lkml.kernel.org/r/149739531127.20686.15813586620597484283.stgit@dwillia2-desk3.amr.corp.intel.com Signed-off-by: Dan Williams Reviewed-by: Ross Zwisler Cc: Jan Kara Cc: Christoph Hellwig Cc: "Kirill A. Shutemov" Cc: Michal Hocko Signed-off-by: Andrew Morton Signed-off-by: Linus Torvalds --- include/linux/dax.h | 5 ----- include/linux/fs.h | 6 ++++++ include/linux/huge_mm.h | 5 +++++ 3 files changed, 11 insertions(+), 5 deletions(-) (limited to 'include/linux/dax.h') diff --git a/include/linux/dax.h b/include/linux/dax.h index 8f39db7439c3..794811875732 100644 --- a/include/linux/dax.h +++ b/include/linux/dax.h @@ -154,11 +154,6 @@ static inline unsigned int dax_radix_order(void *entry) #endif int dax_pfn_mkwrite(struct vm_fault *vmf); -static inline bool vma_is_dax(struct vm_area_struct *vma) -{ - return vma->vm_file && IS_DAX(vma->vm_file->f_mapping->host); -} - static inline bool dax_mapping(struct address_space *mapping) { return mapping->host && IS_DAX(mapping->host); diff --git a/include/linux/fs.h b/include/linux/fs.h index 0cfa47125d52..78e1dbbe4cfd 100644 --- a/include/linux/fs.h +++ b/include/linux/fs.h @@ -18,6 +18,7 @@ #include #include #include +#include #include #include #include @@ -3127,6 +3128,11 @@ static inline bool io_is_direct(struct file *filp) return (filp->f_flags & O_DIRECT) || IS_DAX(filp->f_mapping->host); } +static inline bool vma_is_dax(struct vm_area_struct *vma) +{ + return vma->vm_file && IS_DAX(vma->vm_file->f_mapping->host); +} + static inline int iocb_flags(struct file *file) { int res = 0; diff --git a/include/linux/huge_mm.h b/include/linux/huge_mm.h index f4239d3c9c73..ee696347f928 100644 --- a/include/linux/huge_mm.h +++ b/include/linux/huge_mm.h @@ -3,6 +3,8 @@ #include +#include /* only for vma_is_dax() */ + extern int do_huge_pmd_anonymous_page(struct vm_fault *vmf); extern int copy_huge_pmd(struct mm_struct *dst_mm, struct mm_struct *src_mm, pmd_t *dst_pmd, pmd_t *src_pmd, unsigned long addr, @@ -103,6 +105,9 @@ static inline bool transparent_hugepage_enabled(struct vm_area_struct *vma) if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_FLAG)) return true; + if (vma_is_dax(vma)) + return true; + if (transparent_hugepage_flags & (1 << TRANSPARENT_HUGEPAGE_REQ_MADV_FLAG)) return !!(vma->vm_flags & VM_HUGEPAGE); -- cgit v1.2.3