[PATCH v16 00/11] Block write streams with nvme fdp

[PATCH v16 00/11] Block write streams with nvme fdp

[Kanchan Joshi]

The series enables FDP support for block IO.
The patches
- Add ki_write_stream in kiocb (patch 1), and bi_write_stream in bio (patch 2).
- Introduce two new queue limits - max_write_streams and
  write_stream_granularity (patch 3, 4)
- Pass write stream (either from kiocb, or from inode write hints)
  for block device (patch 5)
- Per I/O write stream interface in io_uring (patch 6)
- Register nvme fdp via write stream queue limits (patch 10, 11)

Changes since v15:
- Merged to latest for-next (Jens)

Previous discussions:
v15: https://lore.kernel.org/linux-nvme/20250203184129.1829324-1-kbusch@meta.com/T/#u
v14: https://lore.kernel.org/linux-nvme/20241211183514.64070-1-kbusch@meta.com/T/#u
v13: https://lore.kernel.org/linux-nvme/20241210194722.1905732-1-kbusch@meta.com/T/#u
v12: https://lore.kernel.org/linux-nvme/20241206221801.790690-1-kbusch@meta.com/T/#u
v11: https://lore.kernel.org/linux-nvme/20241206015308.3342386-1-kbusch@meta.com/T/#u
v10: https://lore.kernel.org/linux-nvme/20241029151922.459139-1-kbusch@meta.com/T/#u
v9: https://lore.kernel.org/linux-nvme/20241025213645.3464331-1-kbusch@meta.com/T/#u
v8: https://lore.kernel.org/linux-nvme/20241017160937.2283225-1-kbusch@meta.com/T/#u
v7: https://lore.kernel.org/linux-nvme/20240930181305.17286-1-joshi.k@samsung.com/T/#u
v6: https://lore.kernel.org/linux-nvme/20240924092457.7846-1-joshi.k@samsung.com/T/#u
v5: https://lore.kernel.org/linux-nvme/20240910150200.6589-1-joshi.k@samsung.com/T/#u
v4: https://lore.kernel.org/linux-nvme/20240826170606.255718-1-joshi.k@samsung.com/T/#u
v3: https://lore.kernel.org/linux-nvme/20240702102619.164170-1-joshi.k@samsung.com/T/#u
v2: https://lore.kernel.org/linux-nvme/20240528150233.55562-1-joshi.k@samsung.com/T/#u
v1: https://lore.kernel.org/linux-nvme/20240510134015.29717-1-joshi.k@samsung.com/T/#u


Christoph Hellwig (7):
  fs: add a write stream field to the kiocb
  block: add a bi_write_stream field
  block: introduce a write_stream_granularity queue limit
  block: expose write streams for block device nodes
  nvme: add a nvme_get_log_lsi helper
  nvme: pass a void pointer to nvme_get/set_features for the result
  nvme: add FDP definitions

Keith Busch (4):
  block: introduce max_write_streams queue limit
  io_uring: enable per-io write streams
  nvme: register fdp parameters with the block layer
  nvme: use fdp streams if write stream is provided

 Documentation/ABI/stable/sysfs-block |  15 +++
 block/bio.c                          |   2 +
 block/blk-crypto-fallback.c          |   1 +
 block/blk-merge.c                    |   4 +
 block/blk-sysfs.c                    |   6 +
 block/fops.c                         |  23 ++++
 drivers/nvme/host/core.c             | 191 ++++++++++++++++++++++++++-
 drivers/nvme/host/nvme.h             |   7 +-
 include/linux/blk_types.h            |   1 +
 include/linux/blkdev.h               |  10 ++
 include/linux/fs.h                   |   1 +
 include/linux/nvme.h                 |  77 +++++++++++
 include/uapi/linux/io_uring.h        |   4 +
 io_uring/io_uring.c                  |   2 +
 io_uring/rw.c                        |   1 +
 15 files changed, 339 insertions(+), 6 deletions(-)


base-commit: e6d9dcfdc0c53b87cfe86163bfbd14f6457ef2b7
-- 
2.25.1

[PATCH v16 01/11] fs: add a write stream field to the kiocb

[Kanchan Joshi]

From: Christoph Hellwig <hch@lst.de>

Prepare for io_uring passthrough of write streams. The write stream
field in the kiocb structure fits into an existing 2-byte hole, so its
size is not changed.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 include/linux/fs.h | 1 +
 1 file changed, 1 insertion(+)

diff --git a/include/linux/fs.h b/include/linux/fs.h
index 016b0fe1536e..d5988867fe31 100644
--- a/include/linux/fs.h
+++ b/include/linux/fs.h
@@ -408,6 +408,7 @@ struct kiocb {
 	void			*private;
 	int			ki_flags;
 	u16			ki_ioprio; /* See linux/ioprio.h */
+	u8			ki_write_stream;
 	union {
 		/*
 		 * Only used for async buffered reads, where it denotes the
-- 
2.25.1

[PATCH v16 02/11] block: add a bi_write_stream field

[Kanchan Joshi]

From: Christoph Hellwig <hch@lst.de>

Add the ability to pass a write stream for placement control in the bio.
The new field fits in an existing hole, so does not change the size of
the struct.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 block/bio.c                 | 2 ++
 block/blk-crypto-fallback.c | 1 +
 block/blk-merge.c           | 4 ++++
 include/linux/blk_types.h   | 1 +
 4 files changed, 8 insertions(+)

diff --git a/block/bio.c b/block/bio.c
index 4e6c85a33d74..1e42aefc7377 100644
--- a/block/bio.c
+++ b/block/bio.c
@@ -251,6 +251,7 @@ void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table,
 	bio->bi_flags = 0;
 	bio->bi_ioprio = 0;
 	bio->bi_write_hint = 0;
+	bio->bi_write_stream = 0;
 	bio->bi_status = 0;
 	bio->bi_iter.bi_sector = 0;
 	bio->bi_iter.bi_size = 0;
@@ -827,6 +828,7 @@ static int __bio_clone(struct bio *bio, struct bio *bio_src, gfp_t gfp)
 	bio_set_flag(bio, BIO_CLONED);
 	bio->bi_ioprio = bio_src->bi_ioprio;
 	bio->bi_write_hint = bio_src->bi_write_hint;
+	bio->bi_write_stream = bio_src->bi_write_stream;
 	bio->bi_iter = bio_src->bi_iter;
 
 	if (bio->bi_bdev) {
diff --git a/block/blk-crypto-fallback.c b/block/blk-crypto-fallback.c
index f154be0b575a..005c9157ffb3 100644
--- a/block/blk-crypto-fallback.c
+++ b/block/blk-crypto-fallback.c
@@ -173,6 +173,7 @@ static struct bio *blk_crypto_fallback_clone_bio(struct bio *bio_src)
 		bio_set_flag(bio, BIO_REMAPPED);
 	bio->bi_ioprio		= bio_src->bi_ioprio;
 	bio->bi_write_hint	= bio_src->bi_write_hint;
+	bio->bi_write_stream	= bio_src->bi_write_stream;
 	bio->bi_iter.bi_sector	= bio_src->bi_iter.bi_sector;
 	bio->bi_iter.bi_size	= bio_src->bi_iter.bi_size;
 
diff --git a/block/blk-merge.c b/block/blk-merge.c
index fdd4efb54c6c..782308b73b53 100644
--- a/block/blk-merge.c
+++ b/block/blk-merge.c
@@ -832,6 +832,8 @@ static struct request *attempt_merge(struct request_queue *q,
 
 	if (req->bio->bi_write_hint != next->bio->bi_write_hint)
 		return NULL;
+	if (req->bio->bi_write_stream != next->bio->bi_write_stream)
+		return NULL;
 	if (req->bio->bi_ioprio != next->bio->bi_ioprio)
 		return NULL;
 	if (!blk_atomic_write_mergeable_rqs(req, next))
@@ -953,6 +955,8 @@ bool blk_rq_merge_ok(struct request *rq, struct bio *bio)
 		return false;
 	if (rq->bio->bi_write_hint != bio->bi_write_hint)
 		return false;
+	if (rq->bio->bi_write_stream != bio->bi_write_stream)
+		return false;
 	if (rq->bio->bi_ioprio != bio->bi_ioprio)
 		return false;
 	if (blk_atomic_write_mergeable_rq_bio(rq, bio) == false)
diff --git a/include/linux/blk_types.h b/include/linux/blk_types.h
index 5a46067e85b1..f38425338c3f 100644
--- a/include/linux/blk_types.h
+++ b/include/linux/blk_types.h
@@ -220,6 +220,7 @@ struct bio {
 	unsigned short		bi_flags;	/* BIO_* below */
 	unsigned short		bi_ioprio;
 	enum rw_hint		bi_write_hint;
+	u8			bi_write_stream;
 	blk_status_t		bi_status;
 	atomic_t		__bi_remaining;
 
-- 
2.25.1

[PATCH v16 03/11] block: introduce max_write_streams queue limit

[Kanchan Joshi]

From: Keith Busch <kbusch@kernel.org>

Drivers with hardware that support write streams need a way to export how
many are available so applications can generically query this.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
[hch: renamed hints to streams, removed stacking]
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 Documentation/ABI/stable/sysfs-block | 7 +++++++
 block/blk-sysfs.c                    | 3 +++
 include/linux/blkdev.h               | 9 +++++++++
 3 files changed, 19 insertions(+)

diff --git a/Documentation/ABI/stable/sysfs-block b/Documentation/ABI/stable/sysfs-block
index 11545c9e2e93..8bbe1eca28df 100644
--- a/Documentation/ABI/stable/sysfs-block
+++ b/Documentation/ABI/stable/sysfs-block
@@ -547,6 +547,13 @@ Description:
 		[RO] Maximum size in bytes of a single element in a DMA
 		scatter/gather list.
 
+What:		/sys/block/<disk>/queue/max_write_streams
+Date:		November 2024
+Contact:	linux-block@vger.kernel.org
+Description:
+		[RO] Maximum number of write streams supported, 0 if not
+		supported. If supported, valid values are 1 through
+		max_write_streams, inclusive.
 
 What:		/sys/block/<disk>/queue/max_segments
 Date:		March 2010
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 1f9b45b0b9ee..986cdba4f550 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -134,6 +134,7 @@ QUEUE_SYSFS_LIMIT_SHOW(max_segments)
 QUEUE_SYSFS_LIMIT_SHOW(max_discard_segments)
 QUEUE_SYSFS_LIMIT_SHOW(max_integrity_segments)
 QUEUE_SYSFS_LIMIT_SHOW(max_segment_size)
+QUEUE_SYSFS_LIMIT_SHOW(max_write_streams)
 QUEUE_SYSFS_LIMIT_SHOW(logical_block_size)
 QUEUE_SYSFS_LIMIT_SHOW(physical_block_size)
 QUEUE_SYSFS_LIMIT_SHOW(chunk_sectors)
@@ -488,6 +489,7 @@ QUEUE_LIM_RO_ENTRY(queue_max_hw_sectors, "max_hw_sectors_kb");
 QUEUE_LIM_RO_ENTRY(queue_max_segments, "max_segments");
 QUEUE_LIM_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
 QUEUE_LIM_RO_ENTRY(queue_max_segment_size, "max_segment_size");
+QUEUE_LIM_RO_ENTRY(queue_max_write_streams, "max_write_streams");
 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
 
 QUEUE_LIM_RO_ENTRY(queue_logical_block_size, "logical_block_size");
@@ -642,6 +644,7 @@ static struct attribute *queue_attrs[] = {
 	&queue_max_discard_segments_entry.attr,
 	&queue_max_integrity_segments_entry.attr,
 	&queue_max_segment_size_entry.attr,
+	&queue_max_write_streams_entry.attr,
 	&queue_hw_sector_size_entry.attr,
 	&queue_logical_block_size_entry.attr,
 	&queue_physical_block_size_entry.attr,
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index a9bd945e87b9..3747fbbd65fa 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -402,6 +402,8 @@ struct queue_limits {
 	unsigned short		max_integrity_segments;
 	unsigned short		max_discard_segments;
 
+	unsigned short		max_write_streams;
+
 	unsigned int		max_open_zones;
 	unsigned int		max_active_zones;
 
@@ -1285,6 +1287,13 @@ static inline unsigned int bdev_max_segments(struct block_device *bdev)
 	return queue_max_segments(bdev_get_queue(bdev));
 }
 
+static inline unsigned short bdev_max_write_streams(struct block_device *bdev)
+{
+	if (bdev_is_partition(bdev))
+		return 0;
+	return bdev_limits(bdev)->max_write_streams;
+}
+
 static inline unsigned queue_logical_block_size(const struct request_queue *q)
 {
 	return q->limits.logical_block_size;
-- 
2.25.1

[PATCH v16 04/11] block: introduce a write_stream_granularity queue limit

[Kanchan Joshi]

From: Christoph Hellwig <hch@lst.de>

Export the granularity that write streams should be discarded with,
as it is essential for making good use of them.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 Documentation/ABI/stable/sysfs-block | 8 ++++++++
 block/blk-sysfs.c                    | 3 +++
 include/linux/blkdev.h               | 1 +
 3 files changed, 12 insertions(+)

diff --git a/Documentation/ABI/stable/sysfs-block b/Documentation/ABI/stable/sysfs-block
index 8bbe1eca28df..4ba771b56b3b 100644
--- a/Documentation/ABI/stable/sysfs-block
+++ b/Documentation/ABI/stable/sysfs-block
@@ -555,6 +555,14 @@ Description:
 		supported. If supported, valid values are 1 through
 		max_write_streams, inclusive.
 
+What:		/sys/block/<disk>/queue/write_stream_granularity
+Date:		November 2024
+Contact:	linux-block@vger.kernel.org
+Description:
+		[RO] Granularity of a write stream in bytes.  The granularity
+		of a write stream is the size that should be discarded or
+		overwritten together to avoid write amplification in the device.
+
 What:		/sys/block/<disk>/queue/max_segments
 Date:		March 2010
 Contact:	linux-block@vger.kernel.org
diff --git a/block/blk-sysfs.c b/block/blk-sysfs.c
index 986cdba4f550..ed00dedfb9ce 100644
--- a/block/blk-sysfs.c
+++ b/block/blk-sysfs.c
@@ -135,6 +135,7 @@ QUEUE_SYSFS_LIMIT_SHOW(max_discard_segments)
 QUEUE_SYSFS_LIMIT_SHOW(max_integrity_segments)
 QUEUE_SYSFS_LIMIT_SHOW(max_segment_size)
 QUEUE_SYSFS_LIMIT_SHOW(max_write_streams)
+QUEUE_SYSFS_LIMIT_SHOW(write_stream_granularity)
 QUEUE_SYSFS_LIMIT_SHOW(logical_block_size)
 QUEUE_SYSFS_LIMIT_SHOW(physical_block_size)
 QUEUE_SYSFS_LIMIT_SHOW(chunk_sectors)
@@ -490,6 +491,7 @@ QUEUE_LIM_RO_ENTRY(queue_max_segments, "max_segments");
 QUEUE_LIM_RO_ENTRY(queue_max_integrity_segments, "max_integrity_segments");
 QUEUE_LIM_RO_ENTRY(queue_max_segment_size, "max_segment_size");
 QUEUE_LIM_RO_ENTRY(queue_max_write_streams, "max_write_streams");
+QUEUE_LIM_RO_ENTRY(queue_write_stream_granularity, "write_stream_granularity");
 QUEUE_RW_ENTRY(elv_iosched, "scheduler");
 
 QUEUE_LIM_RO_ENTRY(queue_logical_block_size, "logical_block_size");
@@ -645,6 +647,7 @@ static struct attribute *queue_attrs[] = {
 	&queue_max_integrity_segments_entry.attr,
 	&queue_max_segment_size_entry.attr,
 	&queue_max_write_streams_entry.attr,
+	&queue_write_stream_granularity_entry.attr,
 	&queue_hw_sector_size_entry.attr,
 	&queue_logical_block_size_entry.attr,
 	&queue_physical_block_size_entry.attr,
diff --git a/include/linux/blkdev.h b/include/linux/blkdev.h
index 3747fbbd65fa..886009b6c3e5 100644
--- a/include/linux/blkdev.h
+++ b/include/linux/blkdev.h
@@ -403,6 +403,7 @@ struct queue_limits {
 	unsigned short		max_discard_segments;
 
 	unsigned short		max_write_streams;
+	unsigned int		write_stream_granularity;
 
 	unsigned int		max_open_zones;
 	unsigned int		max_active_zones;
-- 
2.25.1

[PATCH v16 05/11] block: expose write streams for block device nodes

[Kanchan Joshi]

From: Christoph Hellwig <hch@lst.de>

Use the per-kiocb write stream if provided, or map temperature hints to
write streams (which is a bit questionable, but this shows how it is
done).

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
[kbusch: removed statx reporting]
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 block/fops.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/block/fops.c b/block/fops.c
index b6d7cdd96b54..1309861d4c2c 100644
--- a/block/fops.c
+++ b/block/fops.c
@@ -73,6 +73,7 @@ static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb,
 	}
 	bio.bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 	bio.bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
+	bio.bi_write_stream = iocb->ki_write_stream;
 	bio.bi_ioprio = iocb->ki_ioprio;
 	if (iocb->ki_flags & IOCB_ATOMIC)
 		bio.bi_opf |= REQ_ATOMIC;
@@ -206,6 +207,7 @@ static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter,
 	for (;;) {
 		bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 		bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
+		bio->bi_write_stream = iocb->ki_write_stream;
 		bio->bi_private = dio;
 		bio->bi_end_io = blkdev_bio_end_io;
 		bio->bi_ioprio = iocb->ki_ioprio;
@@ -333,6 +335,7 @@ static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb,
 	dio->iocb = iocb;
 	bio->bi_iter.bi_sector = pos >> SECTOR_SHIFT;
 	bio->bi_write_hint = file_inode(iocb->ki_filp)->i_write_hint;
+	bio->bi_write_stream = iocb->ki_write_stream;
 	bio->bi_end_io = blkdev_bio_end_io_async;
 	bio->bi_ioprio = iocb->ki_ioprio;
 
@@ -398,6 +401,26 @@ static ssize_t blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter)
 	if (blkdev_dio_invalid(bdev, iocb, iter))
 		return -EINVAL;
 
+	if (iov_iter_rw(iter) == WRITE) {
+		u16 max_write_streams = bdev_max_write_streams(bdev);
+
+		if (iocb->ki_write_stream) {
+			if (iocb->ki_write_stream > max_write_streams)
+				return -EINVAL;
+		} else if (max_write_streams) {
+			enum rw_hint write_hint =
+				file_inode(iocb->ki_filp)->i_write_hint;
+
+			/*
+			 * Just use the write hint as write stream for block
+			 * device writes.  This assumes no file system is
+			 * mounted that would use the streams differently.
+			 */
+			if (write_hint <= max_write_streams)
+				iocb->ki_write_stream = write_hint;
+		}
+	}
+
 	nr_pages = bio_iov_vecs_to_alloc(iter, BIO_MAX_VECS + 1);
 	if (likely(nr_pages <= BIO_MAX_VECS)) {
 		if (is_sync_kiocb(iocb))
-- 
2.25.1

[PATCH v16 06/11] io_uring: enable per-io write streams

[Kanchan Joshi]

From: Keith Busch <kbusch@kernel.org>

Allow userspace to pass a per-I/O write stream in the SQE:

      __u8 write_stream;

The __u8 type matches the size the filesystems and block layer support.

Application can query the supported values from the block devices
max_write_streams sysfs attribute. Unsupported values are ignored by
file operations that do not support write streams or rejected with an
error by those that support them.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 include/uapi/linux/io_uring.h | 4 ++++
 io_uring/io_uring.c           | 2 ++
 io_uring/rw.c                 | 1 +
 3 files changed, 7 insertions(+)

diff --git a/include/uapi/linux/io_uring.h b/include/uapi/linux/io_uring.h
index 5ce096090b0c..cfd17e382082 100644
--- a/include/uapi/linux/io_uring.h
+++ b/include/uapi/linux/io_uring.h
@@ -94,6 +94,10 @@ struct io_uring_sqe {
 			__u16	addr_len;
 			__u16	__pad3[1];
 		};
+		struct {
+			__u8	write_stream;
+			__u8	__pad4[3];
+		};
 	};
 	union {
 		struct {
diff --git a/io_uring/io_uring.c b/io_uring/io_uring.c
index 703251f6f4d8..36c689a50126 100644
--- a/io_uring/io_uring.c
+++ b/io_uring/io_uring.c
@@ -3916,6 +3916,8 @@ static int __init io_uring_init(void)
 	BUILD_BUG_SQE_ELEM(44, __s32,  splice_fd_in);
 	BUILD_BUG_SQE_ELEM(44, __u32,  file_index);
 	BUILD_BUG_SQE_ELEM(44, __u16,  addr_len);
+	BUILD_BUG_SQE_ELEM(44, __u8,   write_stream);
+	BUILD_BUG_SQE_ELEM(45, __u8,   __pad4[0]);
 	BUILD_BUG_SQE_ELEM(46, __u16,  __pad3[0]);
 	BUILD_BUG_SQE_ELEM(48, __u64,  addr3);
 	BUILD_BUG_SQE_ELEM_SIZE(48, 0, cmd);
diff --git a/io_uring/rw.c b/io_uring/rw.c
index 17a12a1cf3a6..303fdded3758 100644
--- a/io_uring/rw.c
+++ b/io_uring/rw.c
@@ -279,6 +279,7 @@ static int __io_prep_rw(struct io_kiocb *req, const struct io_uring_sqe *sqe,
 	}
 	rw->kiocb.dio_complete = NULL;
 	rw->kiocb.ki_flags = 0;
+	rw->kiocb.ki_write_stream = READ_ONCE(sqe->write_stream);
 
 	if (req->ctx->flags & IORING_SETUP_IOPOLL)
 		rw->kiocb.ki_complete = io_complete_rw_iopoll;
-- 
2.25.1

[PATCH v16 07/11] nvme: add a nvme_get_log_lsi helper

[Kanchan Joshi]

From: Christoph Hellwig <hch@lst.de>

For log pages that need to pass in a LSI value, while at the same time
not touching all the existing nvme_get_log callers.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 drivers/nvme/host/core.c | 14 ++++++++++++--
 1 file changed, 12 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index eb6ea8acb3cc..0d834ca606d9 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -150,6 +150,8 @@ static void nvme_remove_invalid_namespaces(struct nvme_ctrl *ctrl,
 					   unsigned nsid);
 static void nvme_update_keep_alive(struct nvme_ctrl *ctrl,
 				   struct nvme_command *cmd);
+static int nvme_get_log_lsi(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page,
+		u8 lsp, u8 csi, void *log, size_t size, u64 offset, u16 lsi);
 
 void nvme_queue_scan(struct nvme_ctrl *ctrl)
 {
@@ -3084,8 +3086,8 @@ static int nvme_init_subsystem(struct nvme_ctrl *ctrl, struct nvme_id_ctrl *id)
 	return ret;
 }
 
-int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
-		void *log, size_t size, u64 offset)
+static int nvme_get_log_lsi(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page,
+		u8 lsp, u8 csi, void *log, size_t size, u64 offset, u16 lsi)
 {
 	struct nvme_command c = { };
 	u32 dwlen = nvme_bytes_to_numd(size);
@@ -3099,10 +3101,18 @@ int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
 	c.get_log_page.lpol = cpu_to_le32(lower_32_bits(offset));
 	c.get_log_page.lpou = cpu_to_le32(upper_32_bits(offset));
 	c.get_log_page.csi = csi;
+	c.get_log_page.lsi = cpu_to_le16(lsi);
 
 	return nvme_submit_sync_cmd(ctrl->admin_q, &c, log, size);
 }
 
+int nvme_get_log(struct nvme_ctrl *ctrl, u32 nsid, u8 log_page, u8 lsp, u8 csi,
+		void *log, size_t size, u64 offset)
+{
+	return nvme_get_log_lsi(ctrl, nsid, log_page, lsp, csi, log, size,
+			offset, 0);
+}
+
 static int nvme_get_effects_log(struct nvme_ctrl *ctrl, u8 csi,
 				struct nvme_effects_log **log)
 {
-- 
2.25.1

[PATCH v16 08/11] nvme: pass a void pointer to nvme_get/set_features for the result

[Kanchan Joshi]

From: Christoph Hellwig <hch@lst.de>

That allows passing in structures instead of the u32 result, and thus
reduce the amount of bit shifting and masking required to parse the
result.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 drivers/nvme/host/core.c | 4 ++--
 drivers/nvme/host/nvme.h | 4 ++--
 2 files changed, 4 insertions(+), 4 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 0d834ca606d9..dd71b4c2b7b7 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1676,7 +1676,7 @@ static int nvme_features(struct nvme_ctrl *dev, u8 op, unsigned int fid,
 
 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
 		      unsigned int dword11, void *buffer, size_t buflen,
-		      u32 *result)
+		      void *result)
 {
 	return nvme_features(dev, nvme_admin_set_features, fid, dword11, buffer,
 			     buflen, result);
@@ -1685,7 +1685,7 @@ EXPORT_SYMBOL_GPL(nvme_set_features);
 
 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
 		      unsigned int dword11, void *buffer, size_t buflen,
-		      u32 *result)
+		      void *result)
 {
 	return nvme_features(dev, nvme_admin_get_features, fid, dword11, buffer,
 			     buflen, result);
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index 51e078642127..aedb734283b8 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -896,10 +896,10 @@ int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 		int qid, nvme_submit_flags_t flags);
 int nvme_set_features(struct nvme_ctrl *dev, unsigned int fid,
 		      unsigned int dword11, void *buffer, size_t buflen,
-		      u32 *result);
+		      void *result);
 int nvme_get_features(struct nvme_ctrl *dev, unsigned int fid,
 		      unsigned int dword11, void *buffer, size_t buflen,
-		      u32 *result);
+		      void *result);
 int nvme_set_queue_count(struct nvme_ctrl *ctrl, int *count);
 void nvme_stop_keep_alive(struct nvme_ctrl *ctrl);
 int nvme_reset_ctrl(struct nvme_ctrl *ctrl);
-- 
2.25.1

[PATCH v16 09/11] nvme: add FDP definitions

[Kanchan Joshi]

From: Christoph Hellwig <hch@lst.de>

Add the config feature result, config log page, and management receive
commands needed for FDP.

Partially based on a patch from Kanchan Joshi <joshi.k@samsung.com>.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 include/linux/nvme.h | 77 ++++++++++++++++++++++++++++++++++++++++++++
 1 file changed, 77 insertions(+)

diff --git a/include/linux/nvme.h b/include/linux/nvme.h
index 2479ed10f53e..51308f65b72f 100644
--- a/include/linux/nvme.h
+++ b/include/linux/nvme.h
@@ -303,6 +303,7 @@ enum nvme_ctrl_attr {
 	NVME_CTRL_ATTR_TBKAS		= (1 << 6),
 	NVME_CTRL_ATTR_ELBAS		= (1 << 15),
 	NVME_CTRL_ATTR_RHII		= (1 << 18),
+	NVME_CTRL_ATTR_FDPS		= (1 << 19),
 };
 
 struct nvme_id_ctrl {
@@ -689,6 +690,44 @@ struct nvme_rotational_media_log {
 	__u8	rsvd24[488];
 };
 
+struct nvme_fdp_config {
+	__u8			flags;
+#define FDPCFG_FDPE	(1U << 0)
+	__u8			fdpcidx;
+	__le16			reserved;
+};
+
+struct nvme_fdp_ruh_desc {
+	__u8			ruht;
+	__u8			reserved[3];
+};
+
+struct nvme_fdp_config_desc {
+	__le16			dsze;
+	__u8			fdpa;
+	__u8			vss;
+	__le32			nrg;
+	__le16			nruh;
+	__le16			maxpids;
+	__le32			nns;
+	__le64			runs;
+	__le32			erutl;
+	__u8			rsvd28[36];
+	struct nvme_fdp_ruh_desc ruhs[];
+};
+
+struct nvme_fdp_config_log {
+	__le16			numfdpc;
+	__u8			ver;
+	__u8			rsvd3;
+	__le32			sze;
+	__u8			rsvd8[8];
+	/*
+	 * This is followed by variable number of nvme_fdp_config_desc
+	 * structures, but sparse doesn't like nested variable sized arrays.
+	 */
+};
+
 struct nvme_smart_log {
 	__u8			critical_warning;
 	__u8			temperature[2];
@@ -915,6 +954,7 @@ enum nvme_opcode {
 	nvme_cmd_resv_register	= 0x0d,
 	nvme_cmd_resv_report	= 0x0e,
 	nvme_cmd_resv_acquire	= 0x11,
+	nvme_cmd_io_mgmt_recv	= 0x12,
 	nvme_cmd_resv_release	= 0x15,
 	nvme_cmd_zone_mgmt_send	= 0x79,
 	nvme_cmd_zone_mgmt_recv	= 0x7a,
@@ -936,6 +976,7 @@ enum nvme_opcode {
 		nvme_opcode_name(nvme_cmd_resv_register),	\
 		nvme_opcode_name(nvme_cmd_resv_report),		\
 		nvme_opcode_name(nvme_cmd_resv_acquire),	\
+		nvme_opcode_name(nvme_cmd_io_mgmt_recv),	\
 		nvme_opcode_name(nvme_cmd_resv_release),	\
 		nvme_opcode_name(nvme_cmd_zone_mgmt_send),	\
 		nvme_opcode_name(nvme_cmd_zone_mgmt_recv),	\
@@ -1087,6 +1128,7 @@ enum {
 	NVME_RW_PRINFO_PRCHK_GUARD	= 1 << 12,
 	NVME_RW_PRINFO_PRACT		= 1 << 13,
 	NVME_RW_DTYPE_STREAMS		= 1 << 4,
+	NVME_RW_DTYPE_DPLCMT		= 2 << 4,
 	NVME_WZ_DEAC			= 1 << 9,
 };
 
@@ -1174,6 +1216,38 @@ struct nvme_zone_mgmt_recv_cmd {
 	__le32			cdw14[2];
 };
 
+struct nvme_io_mgmt_recv_cmd {
+	__u8			opcode;
+	__u8			flags;
+	__u16			command_id;
+	__le32			nsid;
+	__le64			rsvd2[2];
+	union nvme_data_ptr	dptr;
+	__u8			mo;
+	__u8			rsvd11;
+	__u16			mos;
+	__le32			numd;
+	__le32			cdw12[4];
+};
+
+enum {
+	NVME_IO_MGMT_RECV_MO_RUHS	= 1,
+};
+
+struct nvme_fdp_ruh_status_desc {
+	__le16			pid;
+	__le16			ruhid;
+	__le32			earutr;
+	__le64			ruamw;
+	__u8			reserved[16];
+};
+
+struct nvme_fdp_ruh_status {
+	__u8			rsvd0[14];
+	__le16			nruhsd;
+	struct nvme_fdp_ruh_status_desc ruhsd[];
+};
+
 enum {
 	NVME_ZRA_ZONE_REPORT		= 0,
 	NVME_ZRASF_ZONE_REPORT_ALL	= 0,
@@ -1309,6 +1383,7 @@ enum {
 	NVME_FEAT_PLM_WINDOW	= 0x14,
 	NVME_FEAT_HOST_BEHAVIOR	= 0x16,
 	NVME_FEAT_SANITIZE	= 0x17,
+	NVME_FEAT_FDP		= 0x1d,
 	NVME_FEAT_SW_PROGRESS	= 0x80,
 	NVME_FEAT_HOST_ID	= 0x81,
 	NVME_FEAT_RESV_MASK	= 0x82,
@@ -1329,6 +1404,7 @@ enum {
 	NVME_LOG_ANA		= 0x0c,
 	NVME_LOG_FEATURES	= 0x12,
 	NVME_LOG_RMI		= 0x16,
+	NVME_LOG_FDP_CONFIGS	= 0x20,
 	NVME_LOG_DISC		= 0x70,
 	NVME_LOG_RESERVATION	= 0x80,
 	NVME_FWACT_REPL		= (0 << 3),
@@ -1923,6 +1999,7 @@ struct nvme_command {
 		struct nvmf_auth_receive_command auth_receive;
 		struct nvme_dbbuf dbbuf;
 		struct nvme_directive_cmd directive;
+		struct nvme_io_mgmt_recv_cmd imr;
 	};
 };
 
-- 
2.25.1

[PATCH v16 10/11] nvme: register fdp parameters with the block layer

[Kanchan Joshi]

From: Keith Busch <kbusch@kernel.org>

Register the device data placement limits if supported. This is just
registering the limits with the block layer. Nothing beyond reporting
these attributes is happening in this patch.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 drivers/nvme/host/core.c | 144 +++++++++++++++++++++++++++++++++++++++
 drivers/nvme/host/nvme.h |   2 +
 2 files changed, 146 insertions(+)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index dd71b4c2b7b7..f25e03ff03df 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -38,6 +38,8 @@ struct nvme_ns_info {
 	u32 nsid;
 	__le32 anagrpid;
 	u8 pi_offset;
+	u16 endgid;
+	u64 runs;
 	bool is_shared;
 	bool is_readonly;
 	bool is_ready;
@@ -1611,6 +1613,7 @@ static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
 	info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
 	info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
 	info->is_ready = true;
+	info->endgid = le16_to_cpu(id->endgid);
 	if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
 		dev_info(ctrl->device,
 			 "Ignoring bogus Namespace Identifiers\n");
@@ -1651,6 +1654,7 @@ static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
 		info->is_ready = id->nstat & NVME_NSTAT_NRDY;
 		info->is_rotational = id->nsfeat & NVME_NS_ROTATIONAL;
 		info->no_vwc = id->nsfeat & NVME_NS_VWC_NOT_PRESENT;
+		info->endgid = le16_to_cpu(id->endgid);
 	}
 	kfree(id);
 	return ret;
@@ -2155,6 +2159,132 @@ static int nvme_update_ns_info_generic(struct nvme_ns *ns,
 	return ret;
 }
 
+static int nvme_query_fdp_granularity(struct nvme_ctrl *ctrl,
+				      struct nvme_ns_info *info, u8 fdp_idx)
+{
+	struct nvme_fdp_config_log hdr, *h;
+	struct nvme_fdp_config_desc *desc;
+	size_t size = sizeof(hdr);
+	void *log, *end;
+	int i, n, ret;
+
+	ret = nvme_get_log_lsi(ctrl, 0, NVME_LOG_FDP_CONFIGS, 0,
+			       NVME_CSI_NVM, &hdr, size, 0, info->endgid);
+	if (ret) {
+		dev_warn(ctrl->device,
+			 "FDP configs log header status:0x%x endgid:%d\n", ret,
+			 info->endgid);
+		return ret;
+	}
+
+	size = le32_to_cpu(hdr.sze);
+	if (size > PAGE_SIZE * MAX_ORDER_NR_PAGES) {
+		dev_warn(ctrl->device, "FDP config size too large:%zu\n",
+			 size);
+		return 0;
+	}
+
+	h = kvmalloc(size, GFP_KERNEL);
+	if (!h)
+		return -ENOMEM;
+
+	ret = nvme_get_log_lsi(ctrl, 0, NVME_LOG_FDP_CONFIGS, 0,
+			       NVME_CSI_NVM, h, size, 0, info->endgid);
+	if (ret) {
+		dev_warn(ctrl->device,
+			 "FDP configs log status:0x%x endgid:%d\n", ret,
+			 info->endgid);
+		goto out;
+	}
+
+	n = le16_to_cpu(h->numfdpc) + 1;
+	if (fdp_idx > n) {
+		dev_warn(ctrl->device, "FDP index:%d out of range:%d\n",
+			 fdp_idx, n);
+		/* Proceed without registering FDP streams */
+		ret = 0;
+		goto out;
+	}
+
+	log = h + 1;
+	desc = log;
+	end = log + size - sizeof(*h);
+	for (i = 0; i < fdp_idx; i++) {
+		log += le16_to_cpu(desc->dsze);
+		desc = log;
+		if (log >= end) {
+			dev_warn(ctrl->device,
+				 "FDP invalid config descriptor list\n");
+			ret = 0;
+			goto out;
+		}
+	}
+
+	if (le32_to_cpu(desc->nrg) > 1) {
+		dev_warn(ctrl->device, "FDP NRG > 1 not supported\n");
+		ret = 0;
+		goto out;
+	}
+
+	info->runs = le64_to_cpu(desc->runs);
+out:
+	kvfree(h);
+	return ret;
+}
+
+static int nvme_query_fdp_info(struct nvme_ns *ns, struct nvme_ns_info *info)
+{
+	struct nvme_ns_head *head = ns->head;
+	struct nvme_ctrl *ctrl = ns->ctrl;
+	struct nvme_fdp_ruh_status *ruhs;
+	struct nvme_fdp_config fdp;
+	struct nvme_command c = {};
+	size_t size;
+	int ret;
+
+	/*
+	 * The FDP configuration is static for the lifetime of the namespace,
+	 * so return immediately if we've already registered this namespace's
+	 * streams.
+	 */
+	if (head->nr_plids)
+		return 0;
+
+	ret = nvme_get_features(ctrl, NVME_FEAT_FDP, info->endgid, NULL, 0,
+				&fdp);
+	if (ret) {
+		dev_warn(ctrl->device, "FDP get feature status:0x%x\n", ret);
+		return ret;
+	}
+
+	if (!(fdp.flags & FDPCFG_FDPE))
+		return 0;
+
+	ret = nvme_query_fdp_granularity(ctrl, info, fdp.fdpcidx);
+	if (!info->runs)
+		return ret;
+
+	size = struct_size(ruhs, ruhsd, S8_MAX - 1);
+	ruhs = kzalloc(size, GFP_KERNEL);
+	if (!ruhs)
+		return -ENOMEM;
+
+	c.imr.opcode = nvme_cmd_io_mgmt_recv;
+	c.imr.nsid = cpu_to_le32(head->ns_id);
+	c.imr.mo = NVME_IO_MGMT_RECV_MO_RUHS;
+	c.imr.numd = cpu_to_le32(nvme_bytes_to_numd(size));
+	ret = nvme_submit_sync_cmd(ns->queue, &c, ruhs, size);
+	if (ret) {
+		dev_warn(ctrl->device, "FDP io-mgmt status:0x%x\n", ret);
+		goto free;
+	}
+
+	head->nr_plids = le16_to_cpu(ruhs->nruhsd);
+free:
+	kfree(ruhs);
+	return ret;
+}
+
 static int nvme_update_ns_info_block(struct nvme_ns *ns,
 		struct nvme_ns_info *info)
 {
@@ -2192,6 +2322,12 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns,
 			goto out;
 	}
 
+	if (ns->ctrl->ctratt & NVME_CTRL_ATTR_FDPS) {
+		ret = nvme_query_fdp_info(ns, info);
+		if (ret < 0)
+			goto out;
+	}
+
 	lim = queue_limits_start_update(ns->disk->queue);
 
 	memflags = blk_mq_freeze_queue(ns->disk->queue);
@@ -2225,6 +2361,12 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns,
 	if (!nvme_init_integrity(ns->head, &lim, info))
 		capacity = 0;
 
+	lim.max_write_streams = ns->head->nr_plids;
+	if (lim.max_write_streams)
+		lim.write_stream_granularity = max(info->runs, U32_MAX);
+	else
+		lim.write_stream_granularity = 0;
+
 	ret = queue_limits_commit_update(ns->disk->queue, &lim);
 	if (ret) {
 		blk_mq_unfreeze_queue(ns->disk->queue, memflags);
@@ -2328,6 +2470,8 @@ static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
 			ns->head->disk->flags |= GENHD_FL_HIDDEN;
 		else
 			nvme_init_integrity(ns->head, &lim, info);
+		lim.max_write_streams = ns_lim->max_write_streams;
+		lim.write_stream_granularity = ns_lim->write_stream_granularity;
 		ret = queue_limits_commit_update(ns->head->disk->queue, &lim);
 
 		set_capacity_and_notify(ns->head->disk, get_capacity(ns->disk));
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index aedb734283b8..3e14daa4ed3e 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -496,6 +496,8 @@ struct nvme_ns_head {
 	struct device		cdev_device;
 
 	struct gendisk		*disk;
+
+	u16			nr_plids;
 #ifdef CONFIG_NVME_MULTIPATH
 	struct bio_list		requeue_list;
 	spinlock_t		requeue_lock;
-- 
2.25.1

Re: [PATCH v16 10/11] nvme: register fdp parameters with the block layer

[Caleb Sander Mateos]

On Tue, May 6, 2025 at 5:31 AM Kanchan Joshi <joshi.k@samsung.com> wrote:
>
> From: Keith Busch <kbusch@kernel.org>
>
> Register the device data placement limits if supported. This is just
> registering the limits with the block layer. Nothing beyond reporting
> these attributes is happening in this patch.
>
> Reviewed-by: Hannes Reinecke <hare@suse.de>
> Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
> Reviewed-by: Christoph Hellwig <hch@lst.de>
> Signed-off-by: Keith Busch <kbusch@kernel.org>
> Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
> ---
>  drivers/nvme/host/core.c | 144 +++++++++++++++++++++++++++++++++++++++
>  drivers/nvme/host/nvme.h |   2 +
>  2 files changed, 146 insertions(+)
>
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index dd71b4c2b7b7..f25e03ff03df 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -38,6 +38,8 @@ struct nvme_ns_info {
>         u32 nsid;
>         __le32 anagrpid;
>         u8 pi_offset;
> +       u16 endgid;
> +       u64 runs;
>         bool is_shared;
>         bool is_readonly;
>         bool is_ready;
> @@ -1611,6 +1613,7 @@ static int nvme_ns_info_from_identify(struct nvme_ctrl *ctrl,
>         info->is_shared = id->nmic & NVME_NS_NMIC_SHARED;
>         info->is_readonly = id->nsattr & NVME_NS_ATTR_RO;
>         info->is_ready = true;
> +       info->endgid = le16_to_cpu(id->endgid);
>         if (ctrl->quirks & NVME_QUIRK_BOGUS_NID) {
>                 dev_info(ctrl->device,
>                          "Ignoring bogus Namespace Identifiers\n");
> @@ -1651,6 +1654,7 @@ static int nvme_ns_info_from_id_cs_indep(struct nvme_ctrl *ctrl,
>                 info->is_ready = id->nstat & NVME_NSTAT_NRDY;
>                 info->is_rotational = id->nsfeat & NVME_NS_ROTATIONAL;
>                 info->no_vwc = id->nsfeat & NVME_NS_VWC_NOT_PRESENT;
> +               info->endgid = le16_to_cpu(id->endgid);
>         }
>         kfree(id);
>         return ret;
> @@ -2155,6 +2159,132 @@ static int nvme_update_ns_info_generic(struct nvme_ns *ns,
>         return ret;
>  }
>
> +static int nvme_query_fdp_granularity(struct nvme_ctrl *ctrl,
> +                                     struct nvme_ns_info *info, u8 fdp_idx)
> +{
> +       struct nvme_fdp_config_log hdr, *h;
> +       struct nvme_fdp_config_desc *desc;
> +       size_t size = sizeof(hdr);
> +       void *log, *end;
> +       int i, n, ret;
> +
> +       ret = nvme_get_log_lsi(ctrl, 0, NVME_LOG_FDP_CONFIGS, 0,
> +                              NVME_CSI_NVM, &hdr, size, 0, info->endgid);
> +       if (ret) {
> +               dev_warn(ctrl->device,
> +                        "FDP configs log header status:0x%x endgid:%d\n", ret,
> +                        info->endgid);
> +               return ret;
> +       }
> +
> +       size = le32_to_cpu(hdr.sze);
> +       if (size > PAGE_SIZE * MAX_ORDER_NR_PAGES) {
> +               dev_warn(ctrl->device, "FDP config size too large:%zu\n",
> +                        size);
> +               return 0;
> +       }
> +
> +       h = kvmalloc(size, GFP_KERNEL);
> +       if (!h)
> +               return -ENOMEM;
> +
> +       ret = nvme_get_log_lsi(ctrl, 0, NVME_LOG_FDP_CONFIGS, 0,
> +                              NVME_CSI_NVM, h, size, 0, info->endgid);
> +       if (ret) {
> +               dev_warn(ctrl->device,
> +                        "FDP configs log status:0x%x endgid:%d\n", ret,
> +                        info->endgid);
> +               goto out;
> +       }
> +
> +       n = le16_to_cpu(h->numfdpc) + 1;
> +       if (fdp_idx > n) {
> +               dev_warn(ctrl->device, "FDP index:%d out of range:%d\n",
> +                        fdp_idx, n);
> +               /* Proceed without registering FDP streams */
> +               ret = 0;
> +               goto out;
> +       }
> +
> +       log = h + 1;
> +       desc = log;
> +       end = log + size - sizeof(*h);
> +       for (i = 0; i < fdp_idx; i++) {
> +               log += le16_to_cpu(desc->dsze);
> +               desc = log;
> +               if (log >= end) {
> +                       dev_warn(ctrl->device,
> +                                "FDP invalid config descriptor list\n");
> +                       ret = 0;
> +                       goto out;
> +               }
> +       }
> +
> +       if (le32_to_cpu(desc->nrg) > 1) {
> +               dev_warn(ctrl->device, "FDP NRG > 1 not supported\n");
> +               ret = 0;
> +               goto out;
> +       }
> +
> +       info->runs = le64_to_cpu(desc->runs);
> +out:
> +       kvfree(h);
> +       return ret;
> +}
> +
> +static int nvme_query_fdp_info(struct nvme_ns *ns, struct nvme_ns_info *info)
> +{
> +       struct nvme_ns_head *head = ns->head;
> +       struct nvme_ctrl *ctrl = ns->ctrl;
> +       struct nvme_fdp_ruh_status *ruhs;
> +       struct nvme_fdp_config fdp;
> +       struct nvme_command c = {};
> +       size_t size;
> +       int ret;
> +
> +       /*
> +        * The FDP configuration is static for the lifetime of the namespace,
> +        * so return immediately if we've already registered this namespace's
> +        * streams.
> +        */
> +       if (head->nr_plids)
> +               return 0;
> +
> +       ret = nvme_get_features(ctrl, NVME_FEAT_FDP, info->endgid, NULL, 0,
> +                               &fdp);
> +       if (ret) {
> +               dev_warn(ctrl->device, "FDP get feature status:0x%x\n", ret);
> +               return ret;
> +       }
> +
> +       if (!(fdp.flags & FDPCFG_FDPE))
> +               return 0;
> +
> +       ret = nvme_query_fdp_granularity(ctrl, info, fdp.fdpcidx);
> +       if (!info->runs)
> +               return ret;
> +
> +       size = struct_size(ruhs, ruhsd, S8_MAX - 1);
> +       ruhs = kzalloc(size, GFP_KERNEL);
> +       if (!ruhs)
> +               return -ENOMEM;
> +
> +       c.imr.opcode = nvme_cmd_io_mgmt_recv;
> +       c.imr.nsid = cpu_to_le32(head->ns_id);
> +       c.imr.mo = NVME_IO_MGMT_RECV_MO_RUHS;
> +       c.imr.numd = cpu_to_le32(nvme_bytes_to_numd(size));
> +       ret = nvme_submit_sync_cmd(ns->queue, &c, ruhs, size);
> +       if (ret) {
> +               dev_warn(ctrl->device, "FDP io-mgmt status:0x%x\n", ret);
> +               goto free;
> +       }
> +
> +       head->nr_plids = le16_to_cpu(ruhs->nruhsd);
> +free:
> +       kfree(ruhs);
> +       return ret;
> +}
> +
>  static int nvme_update_ns_info_block(struct nvme_ns *ns,
>                 struct nvme_ns_info *info)
>  {
> @@ -2192,6 +2322,12 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns,
>                         goto out;
>         }
>
> +       if (ns->ctrl->ctratt & NVME_CTRL_ATTR_FDPS) {
> +               ret = nvme_query_fdp_info(ns, info);
> +               if (ret < 0)
> +                       goto out;
> +       }
> +
>         lim = queue_limits_start_update(ns->disk->queue);
>
>         memflags = blk_mq_freeze_queue(ns->disk->queue);
> @@ -2225,6 +2361,12 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns,
>         if (!nvme_init_integrity(ns->head, &lim, info))
>                 capacity = 0;
>
> +       lim.max_write_streams = ns->head->nr_plids;
> +       if (lim.max_write_streams)
> +               lim.write_stream_granularity = max(info->runs, U32_MAX);

What is the purpose of this max(..., U32_MAX)? Should it be min() instead?

Best,
Caleb

> +       else
> +               lim.write_stream_granularity = 0;
> +
>         ret = queue_limits_commit_update(ns->disk->queue, &lim);
>         if (ret) {
>                 blk_mq_unfreeze_queue(ns->disk->queue, memflags);
> @@ -2328,6 +2470,8 @@ static int nvme_update_ns_info(struct nvme_ns *ns, struct nvme_ns_info *info)
>                         ns->head->disk->flags |= GENHD_FL_HIDDEN;
>                 else
>                         nvme_init_integrity(ns->head, &lim, info);
> +               lim.max_write_streams = ns_lim->max_write_streams;
> +               lim.write_stream_granularity = ns_lim->write_stream_granularity;
>                 ret = queue_limits_commit_update(ns->head->disk->queue, &lim);
>
>                 set_capacity_and_notify(ns->head->disk, get_capacity(ns->disk));
> diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
> index aedb734283b8..3e14daa4ed3e 100644
> --- a/drivers/nvme/host/nvme.h
> +++ b/drivers/nvme/host/nvme.h
> @@ -496,6 +496,8 @@ struct nvme_ns_head {
>         struct device           cdev_device;
>
>         struct gendisk          *disk;
> +
> +       u16                     nr_plids;
>  #ifdef CONFIG_NVME_MULTIPATH
>         struct bio_list         requeue_list;
>         spinlock_t              requeue_lock;
> --
> 2.25.1
>
>

Re: [PATCH v16 10/11] nvme: register fdp parameters with the block layer

[Keith Busch]

On Tue, May 06, 2025 at 09:13:33AM -0700, Caleb Sander Mateos wrote:
> On Tue, May 6, 2025 at 5:31 AM Kanchan Joshi <joshi.k@samsung.com> wrote:
> > @@ -2225,6 +2361,12 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns,
> >         if (!nvme_init_integrity(ns->head, &lim, info))
> >                 capacity = 0;
> >
> > +       lim.max_write_streams = ns->head->nr_plids;
> > +       if (lim.max_write_streams)
> > +               lim.write_stream_granularity = max(info->runs, U32_MAX);
> 
> What is the purpose of this max(..., U32_MAX)? Should it be min() instead?

You're right, should have been min. Because "runs" is a u64 and the
queue_limit is a u32, so U32_MAX is the upper limit, but it's not
supposed to exceed "runs". 

Re: [PATCH v16 10/11] nvme: register fdp parameters with the block layer

[Kanchan Joshi]

On Tue, May 6, 2025 at 9:56 PM Keith Busch <kbusch@kernel.org> wrote:
>
> On Tue, May 06, 2025 at 09:13:33AM -0700, Caleb Sander Mateos wrote:
> > On Tue, May 6, 2025 at 5:31 AM Kanchan Joshi <joshi.k@samsung.com> wrote:
> > > @@ -2225,6 +2361,12 @@ static int nvme_update_ns_info_block(struct nvme_ns *ns,
> > >         if (!nvme_init_integrity(ns->head, &lim, info))
> > >                 capacity = 0;
> > >
> > > +       lim.max_write_streams = ns->head->nr_plids;
> > > +       if (lim.max_write_streams)
> > > +               lim.write_stream_granularity = max(info->runs, U32_MAX);
> >
> > What is the purpose of this max(..., U32_MAX)? Should it be min() instead?
>
> You're right, should have been min. Because "runs" is a u64 and the
> queue_limit is a u32, so U32_MAX is the upper limit, but it's not
> supposed to exceed "runs".

Would it be better to change write_stream_granularity to "long
unsigned int" so that it matches with what is possible in nvme?

Re: [PATCH v16 10/11] nvme: register fdp parameters with the block layer

[Keith Busch]

On Tue, May 06, 2025 at 11:44:27PM +0530, Kanchan Joshi wrote:
> On Tue, May 6, 2025 at 9:56 PM Keith Busch <kbusch@kernel.org> wrote:
> >
> > You're right, should have been min. Because "runs" is a u64 and the
> > queue_limit is a u32, so U32_MAX is the upper limit, but it's not
> > supposed to exceed "runs".
> 
> Would it be better to change write_stream_granularity to "long
> unsigned int" so that it matches with what is possible in nvme?

That type is still 4 bytes on many 32-bit archs, but I know what you
mean (unsigned long long). I didn't think we'd see reclaim units
approach 4GB, but if you think it's possible, may as well have the
queue_limit type be large enough to report it.

[PATCH v16 11/11] nvme: use fdp streams if write stream is provided

[Kanchan Joshi]

From: Keith Busch <kbusch@kernel.org>

Maps a user requested write stream to an FDP placement ID if possible.

Reviewed-by: Hannes Reinecke <hare@suse.de>
Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
Reviewed-by: Christoph Hellwig <hch@lst.de>
Signed-off-by: Keith Busch <kbusch@kernel.org>
Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
---
 drivers/nvme/host/core.c | 31 ++++++++++++++++++++++++++++++-
 drivers/nvme/host/nvme.h |  1 +
 2 files changed, 31 insertions(+), 1 deletion(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index f25e03ff03df..52331a14bce1 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -672,6 +672,7 @@ static void nvme_free_ns_head(struct kref *ref)
 	ida_free(&head->subsys->ns_ida, head->instance);
 	cleanup_srcu_struct(&head->srcu);
 	nvme_put_subsystem(head->subsys);
+	kfree(head->plids);
 	kfree(head);
 }
 
@@ -995,6 +996,18 @@ static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
 	if (req->cmd_flags & REQ_RAHEAD)
 		dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
 
+	if (op == nvme_cmd_write && ns->head->nr_plids) {
+		u16 write_stream = req->bio->bi_write_stream;
+
+		if (WARN_ON_ONCE(write_stream > ns->head->nr_plids))
+			return BLK_STS_INVAL;
+
+		if (write_stream) {
+			dsmgmt |= ns->head->plids[write_stream - 1] << 16;
+			control |= NVME_RW_DTYPE_DPLCMT;
+		}
+	}
+
 	if (req->cmd_flags & REQ_ATOMIC && !nvme_valid_atomic_write(req))
 		return BLK_STS_INVAL;
 
@@ -2240,7 +2253,7 @@ static int nvme_query_fdp_info(struct nvme_ns *ns, struct nvme_ns_info *info)
 	struct nvme_fdp_config fdp;
 	struct nvme_command c = {};
 	size_t size;
-	int ret;
+	int i, ret;
 
 	/*
 	 * The FDP configuration is static for the lifetime of the namespace,
@@ -2280,6 +2293,22 @@ static int nvme_query_fdp_info(struct nvme_ns *ns, struct nvme_ns_info *info)
 	}
 
 	head->nr_plids = le16_to_cpu(ruhs->nruhsd);
+	if (!head->nr_plids)
+		goto free;
+
+	head->plids = kcalloc(head->nr_plids, sizeof(head->plids),
+			      GFP_KERNEL);
+	if (!head->plids) {
+		dev_warn(ctrl->device,
+			 "failed to allocate %u FDP placement IDs\n",
+			 head->nr_plids);
+		head->nr_plids = 0;
+		ret = -ENOMEM;
+		goto free;
+	}
+
+	for (i = 0; i < head->nr_plids; i++)
+		head->plids[i] = le16_to_cpu(ruhs->ruhsd[i].pid);
 free:
 	kfree(ruhs);
 	return ret;
diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
index 3e14daa4ed3e..7aad581271c7 100644
--- a/drivers/nvme/host/nvme.h
+++ b/drivers/nvme/host/nvme.h
@@ -498,6 +498,7 @@ struct nvme_ns_head {
 	struct gendisk		*disk;
 
 	u16			nr_plids;
+	u16			*plids;
 #ifdef CONFIG_NVME_MULTIPATH
 	struct bio_list		requeue_list;
 	spinlock_t		requeue_lock;
-- 
2.25.1

Re: [PATCH v16 11/11] nvme: use fdp streams if write stream is provided

[Caleb Sander Mateos]

On Tue, May 6, 2025 at 5:31 AM Kanchan Joshi <joshi.k@samsung.com> wrote:
>
> From: Keith Busch <kbusch@kernel.org>
>
> Maps a user requested write stream to an FDP placement ID if possible.
>
> Reviewed-by: Hannes Reinecke <hare@suse.de>
> Reviewed-by: Nitesh Shetty <nj.shetty@samsung.com>
> Reviewed-by: Christoph Hellwig <hch@lst.de>
> Signed-off-by: Keith Busch <kbusch@kernel.org>
> Signed-off-by: Kanchan Joshi <joshi.k@samsung.com>
> ---
>  drivers/nvme/host/core.c | 31 ++++++++++++++++++++++++++++++-
>  drivers/nvme/host/nvme.h |  1 +
>  2 files changed, 31 insertions(+), 1 deletion(-)
>
> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index f25e03ff03df..52331a14bce1 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -672,6 +672,7 @@ static void nvme_free_ns_head(struct kref *ref)
>         ida_free(&head->subsys->ns_ida, head->instance);
>         cleanup_srcu_struct(&head->srcu);
>         nvme_put_subsystem(head->subsys);
> +       kfree(head->plids);
>         kfree(head);
>  }
>
> @@ -995,6 +996,18 @@ static inline blk_status_t nvme_setup_rw(struct nvme_ns *ns,
>         if (req->cmd_flags & REQ_RAHEAD)
>                 dsmgmt |= NVME_RW_DSM_FREQ_PREFETCH;
>
> +       if (op == nvme_cmd_write && ns->head->nr_plids) {
> +               u16 write_stream = req->bio->bi_write_stream;
> +
> +               if (WARN_ON_ONCE(write_stream > ns->head->nr_plids))
> +                       return BLK_STS_INVAL;
> +
> +               if (write_stream) {
> +                       dsmgmt |= ns->head->plids[write_stream - 1] << 16;
> +                       control |= NVME_RW_DTYPE_DPLCMT;
> +               }
> +       }
> +
>         if (req->cmd_flags & REQ_ATOMIC && !nvme_valid_atomic_write(req))
>                 return BLK_STS_INVAL;
>
> @@ -2240,7 +2253,7 @@ static int nvme_query_fdp_info(struct nvme_ns *ns, struct nvme_ns_info *info)
>         struct nvme_fdp_config fdp;
>         struct nvme_command c = {};
>         size_t size;
> -       int ret;
> +       int i, ret;
>
>         /*
>          * The FDP configuration is static for the lifetime of the namespace,
> @@ -2280,6 +2293,22 @@ static int nvme_query_fdp_info(struct nvme_ns *ns, struct nvme_ns_info *info)
>         }
>
>         head->nr_plids = le16_to_cpu(ruhs->nruhsd);
> +       if (!head->nr_plids)
> +               goto free;
> +
> +       head->plids = kcalloc(head->nr_plids, sizeof(head->plids),
> +                             GFP_KERNEL);

Should this be sizeof(*head->plids)?

Best,
Caleb

> +       if (!head->plids) {
> +               dev_warn(ctrl->device,
> +                        "failed to allocate %u FDP placement IDs\n",
> +                        head->nr_plids);
> +               head->nr_plids = 0;
> +               ret = -ENOMEM;
> +               goto free;
> +       }
> +
> +       for (i = 0; i < head->nr_plids; i++)
> +               head->plids[i] = le16_to_cpu(ruhs->ruhsd[i].pid);
>  free:
>         kfree(ruhs);
>         return ret;
> diff --git a/drivers/nvme/host/nvme.h b/drivers/nvme/host/nvme.h
> index 3e14daa4ed3e..7aad581271c7 100644
> --- a/drivers/nvme/host/nvme.h
> +++ b/drivers/nvme/host/nvme.h
> @@ -498,6 +498,7 @@ struct nvme_ns_head {
>         struct gendisk          *disk;
>
>         u16                     nr_plids;
> +       u16                     *plids;
>  #ifdef CONFIG_NVME_MULTIPATH
>         struct bio_list         requeue_list;
>         spinlock_t              requeue_lock;
> --
> 2.25.1
>
>

Re: [PATCH v16 11/11] nvme: use fdp streams if write stream is provided

[Keith Busch]

On Tue, May 06, 2025 at 09:14:19AM -0700, Caleb Sander Mateos wrote:
> > +       head->plids = kcalloc(head->nr_plids, sizeof(head->plids),
> > +                             GFP_KERNEL);
> 
> Should this be sizeof(*head->plids)?

Indeed it should. This as-is overallocates the array size, so wouldn't
have easily found it at runtime.

Re: [PATCH v16 00/11] Block write streams with nvme fdp

[Jens Axboe]

On Tue, 06 May 2025 17:47:21 +0530, Kanchan Joshi wrote:
> The series enables FDP support for block IO.
> The patches
> - Add ki_write_stream in kiocb (patch 1), and bi_write_stream in bio (patch 2).
> - Introduce two new queue limits - max_write_streams and
>   write_stream_granularity (patch 3, 4)
> - Pass write stream (either from kiocb, or from inode write hints)
>   for block device (patch 5)
> - Per I/O write stream interface in io_uring (patch 6)
> - Register nvme fdp via write stream queue limits (patch 10, 11)
> 
> [...]

Applied, thanks!

[01/11] fs: add a write stream field to the kiocb
        commit: 732f25a2895a8c1c54fb56544f0b1e23770ef4d7
[02/11] block: add a bi_write_stream field
        commit: 5006f85ea23ea0bda9a8e31fdda126f4fca48f20
[03/11] block: introduce max_write_streams queue limit
        commit: d2f526ba27d29c442542f7c5df0a86ef0b576716
[04/11] block: introduce a write_stream_granularity queue limit
        commit: c23acfac10786ac5062a0615e23e68b913ac8da0
[05/11] block: expose write streams for block device nodes
        commit: c27683da6406031d47a65b344d04a40736490d95
[06/11] io_uring: enable per-io write streams
        commit: 02040353f4fedb823f011f27962325f328d0689f
[07/11] nvme: add a nvme_get_log_lsi helper
        commit: d4f8359eaecf0f8b0a9f631e6652b60ae61f3016
[08/11] nvme: pass a void pointer to nvme_get/set_features for the result
        commit: 7a044d34b1e21fc4e04d4e48dae1dc3795621570
[09/11] nvme: add FDP definitions
        commit: ee203d3d86113559b77b1723e0d10909ebbd66ad
[10/11] nvme: register fdp parameters with the block layer
        commit: 30b5f20bb2ddab013035399e5c7e6577da49320a
[11/11] nvme: use fdp streams if write stream is provided
        commit: 38e8397dde6338c76593ddb17ccf3118fc3f5203

Best regards,
-- 
Jens Axboe