Pass data temperature information to SCSI disk devices
Subsystems (7)
v3 cover Pass data temperature information to SCSI disk devices
Hi Jens,
UFS vendors need the data lifetime information to achieve good performance. Without this information there is significantly higher write amplification due to garbage collection. Hence this patch series that add support in F2FS and also in the block layer for data lifetime information. The SCSI disk (sd) driver is modified such that it passes write hint information to SCSI devices via the GROUP NUMBER field.
Please consider this patch series for the next merge window.
Thanks,
Bart.
Changes compared to v1:
- Instead of storing data lifetime information in bi_ioprio, introduce the new struct bio member bi_lifetime and also the struct request member 'lifetime'.
- Removed the bio_set_data_lifetime() and bio_get_data_lifetime() functions and replaced these with direct assignments.
- Dropped all changes related to I/O priority.
- Improved patch descriptions.
Changes compared to v1:
- Use six bits from the ioprio field for data lifetime information. The bio->bi_write_hint / req->write_hint / iocb->ki_hint members that were introduced in v1 have been removed again.
- The F_GET_FILE_RW_HINT and F_SET_FILE_RW_HINT fcntls have been removed.
- In the SCSI disk (sd) driver, query the stream status and check the PERM bit.
- The GET STREAM STATUS command has been implemented in the scsi_debug driver.
Bart Van Assche (14):
fs: Move enum rw_hint into a new header file
block: Restore data lifetime support in struct bio and struct request
fs: Restore write hint support
fs/f2fs: Restore data lifetime support
scsi: core: Query the Block Limits Extension VPD page
scsi_proto: Add structures and constants related to I/O groups and
streams
sd: Translate data lifetime information
scsi_debug: Reduce code duplication
scsi_debug: Support the block limits extension VPD page
scsi_debug: Rework page code error handling
scsi_debug: Rework subpage code error handling
scsi_debug: Implement the IO Advice Hints Grouping mode page
scsi_debug: Implement GET STREAM STATUS
scsi_debug: Maintain write statistics per group number
Documentation/filesystems/f2fs.rst | 70 ++++++++ block/bio.c | 2 + block/blk-crypto-fallback.c | 1 + block/blk-merge.c | 6 + block/blk-mq.c | 1 + block/bounce.c | 1 + block/fops.c | 3 + drivers/scsi/scsi.c | 2 + drivers/scsi/scsi_debug.c | 247 +++++++++++++++++++++-------- drivers/scsi/scsi_sysfs.c | 10 ++ drivers/scsi/sd.c | 111 ++++++++++++- drivers/scsi/sd.h | 3 + fs/f2fs/data.c | 2 + fs/f2fs/f2fs.h | 10 ++ fs/f2fs/segment.c | 95 +++++++++++ fs/f2fs/super.c | 32 +++- fs/fcntl.c | 1 + fs/inode.c | 1 + fs/iomap/buffered-io.c | 2 + fs/iomap/direct-io.c | 1 + fs/mpage.c | 1 + include/linux/blk-mq.h | 2 + include/linux/blk_types.h | 2 + include/linux/fs.h | 16 +- include/linux/rw_hint.h | 20 +++ include/scsi/scsi_device.h | 1 + include/scsi/scsi_proto.h | 75 +++++++++ 27 files changed, 635 insertions(+), 83 deletions(-) create mode 100644 include/linux/rw_hint.h
PATCH v3 1/14 fs: Move enum rw_hint into a new header file
Move enum rw_hint into a new header file to prepare for using this data type in the block layer. Add the attribute __packed to reduce the space occupied by instances of this data type from four bytes to one byte. Change the data type of i_write_hint from u8 into enum rw_hint. Change the RWH_* constants into literal constants to prevent that <uapi/linux/fcntl.h> would have to be included.
--- fs/f2fs/f2fs.h | 1 + fs/fcntl.c | 1 + fs/inode.c | 1 + include/linux/fs.h | 16 ++-------------- include/linux/rw_hint.h | 20 ++++++++++++++++++++ 5 files changed, 25 insertions(+), 14 deletions(-) create mode 100644 include/linux/rw_hint.h
fs/f2fs/f2fs.h +1 -0
@@ -24,6 +24,7 @@#include <linux/blkdev.h> #include <linux/quotaops.h> #include <linux/part_stat.h> +#include <linux/rw_hint.h> #include <crypto/hash.h> #include <linux/fscrypt.h>
fs/fcntl.c +1 -0
@@ -27,6 +27,7 @@#include <linux/memfd.h> #include <linux/compat.h> #include <linux/mount.h> +#include <linux/rw_hint.h> #include <linux/poll.h> #include <asm/siginfo.h>
fs/inode.c +1 -0
@@ -20,6 +20,7 @@#include <linux/ratelimit.h> #include <linux/list_lru.h> #include <linux/iversion.h> +#include <linux/rw_hint.h> #include <trace/events/writeback.h> #include "internal.h"
include/linux/fs.h +2 -14
@@ -43,6 +43,7 @@#include <linux/cred.h> #include <linux/mnt_idmapping.h> #include <linux/slab.h> +#include <linux/rw_hint.h> #include <asm/byteorder.h> #include <uapi/linux/fs.h>
@@ -309,19 +310,6 @@ struct address_space;struct writeback_control ;struct readahead_control ; -/* -* Write life time hint values. -* Stored in struct inode as u8. -*/ -enum rw_hint { -WRITE_LIFE_NOT_SET =0 , -WRITE_LIFE_NONE =RWH_WRITE_LIFE_NONE , -WRITE_LIFE_SHORT =RWH_WRITE_LIFE_SHORT , -WRITE_LIFE_MEDIUM =RWH_WRITE_LIFE_MEDIUM , -WRITE_LIFE_LONG =RWH_WRITE_LIFE_LONG , -WRITE_LIFE_EXTREME =RWH_WRITE_LIFE_EXTREME , -}; -/* Match RWF_* bits to IOCB bits */ #define IOCB_HIPRI (__force int) RWF_HIPRI#define IOCB_DSYNC (__force int) RWF_DSYNC
@@ -677,7 +665,7 @@ struct inode {spinlock_t i_lock;/* i_blocks, i_bytes, maybe i_size */ unsigned short i_bytes;u8 i_blkbits; -u8 i_write_hint; +enum rw_hint i_write_hint;blkcnt_t i_blocks;#ifdef __NEED_I_SIZE_ORDERED
include/linux/rw_hint.h new file +20 -0
@@ -0,0 +1,20 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _LINUX_RW_HINT_H +#define _LINUX_RW_HINT_H + +#include <linux/build_bug.h> +#include <linux/compiler_attributes.h> + +/* Block storage write lifetime hint values. */ +enum rw_hint { +WRITE_LIFE_NOT_SET =0 ,/* RWH_WRITE_LIFE_NOT_SET */ +WRITE_LIFE_NONE =1 ,/* RWH_WRITE_LIFE_NONE */ +WRITE_LIFE_SHORT =2 ,/* RWH_WRITE_LIFE_SHORT */ +WRITE_LIFE_MEDIUM =3 ,/* RWH_WRITE_LIFE_MEDIUM */ +WRITE_LIFE_LONG =4 ,/* RWH_WRITE_LIFE_LONG */ +WRITE_LIFE_EXTREME =5 ,/* RWH_WRITE_LIFE_EXTREME */ +} __packed; + +static_assert (sizeof (enum rw_hint ) ==1 ); + +#endif /* _LINUX_RW_HINT_H */
Re: [PATCH v3 01/14] fs: Move enum rw_hint into a new header file
Quoted 38 lines · Bart Van Assche
- * Write life time hint values.
- * Stored in struct inode as u8.
- */include/linux/fs.h +1
-10-enum rw_hint { -WRITE_LIFE_NOT_SET =0 , -WRITE_LIFE_NONE =RWH_WRITE_LIFE_NONE , -WRITE_LIFE_SHORT =RWH_WRITE_LIFE_SHORT , -WRITE_LIFE_MEDIUM =RWH_WRITE_LIFE_MEDIUM , -WRITE_LIFE_LONG =RWH_WRITE_LIFE_LONG , -WRITE_LIFE_EXTREME =RWH_WRITE_LIFE_EXTREME , -}; -/* Match RWF_* bits to IOCB bits */ #define IOCB_HIPRI (__force int) RWF_HIPRI#define IOCB_DSYNC (__force int) RWF_DSYNC@@ -677,7 +665,7 @@ struct inode {spinlock_t i_lock;/* i_blocks, i_bytes, maybe i_size */ unsigned short i_bytes;u8 i_blkbits; -u8 i_write_hint; +enum rw_hint i_write_hint;blkcnt_t i_blocks;#ifdef __NEED_I_SIZE_ORDEREDinclude/linux/rw_hint.h new file +8
-0@@ -0,0 +1,20 @@ +/* SPDX-License-Identifier: GPL-2.0 */ +#ifndef _LINUX_RW_HINT_H +#define _LINUX_RW_HINT_H + +#include <linux/build_bug.h> +#include <linux/compiler_attributes.h> + +/* Block storage write lifetime hint values. */
include/linux/rw_hint.h line 18
Quoted 10 lines
include/linux/rw_hint.h new file +10
-0+enum rw_hint { +WRITE_LIFE_NOT_SET =0 ,/* RWH_WRITE_LIFE_NOT_SET */ +WRITE_LIFE_NONE =1 ,/* RWH_WRITE_LIFE_NONE */ +WRITE_LIFE_SHORT =2 ,/* RWH_WRITE_LIFE_SHORT */ +WRITE_LIFE_MEDIUM =3 ,/* RWH_WRITE_LIFE_MEDIUM */ +WRITE_LIFE_LONG =4 ,/* RWH_WRITE_LIFE_LONG */ +WRITE_LIFE_EXTREME =5 ,/* RWH_WRITE_LIFE_EXTREME */ +} __packed; + +static_assert (sizeof (enum rw_hint ) ==1 );
Does it make sense to do away with these, and have temperature-neutral names instead e.g., WRITE_LIFE_1, WRITE_LIFE_2?
With the current choice:
- If the count goes up (beyond 5 hints), infra can scale fine but these names do not. Imagine ULTRA_EXTREME after EXTREME.
- Applications or in-kernel users can specify LONG hint with data that actually has a SHORT lifetime. Nothing really ensures that LONG is really LONG.
Temperature-neutral names seem more generic/scalable and do not present the unnecessary need to be accurate with relative temperatures.
Re: [PATCH v3 01/14] fs: Move enum rw_hint into a new header file
Quoted 23 lines · Kanchan Joshi
include/linux/rw_hint.h line 18
Quoted 11 lines · Bart Van Assche
include/linux/rw_hint.h new file +11
-0+/* Block storage write lifetime hint values. */ +enum rw_hint { +WRITE_LIFE_NOT_SET =0 ,/* RWH_WRITE_LIFE_NOT_SET */ +WRITE_LIFE_NONE =1 ,/* RWH_WRITE_LIFE_NONE */ +WRITE_LIFE_SHORT =2 ,/* RWH_WRITE_LIFE_SHORT */ +WRITE_LIFE_MEDIUM =3 ,/* RWH_WRITE_LIFE_MEDIUM */ +WRITE_LIFE_LONG =4 ,/* RWH_WRITE_LIFE_LONG */ +WRITE_LIFE_EXTREME =5 ,/* RWH_WRITE_LIFE_EXTREME */ +} __packed; + +static_assert (sizeof (enum rw_hint ) ==1 );Does it make sense to do away with these, and have temperature-neutral names instead e.g.,
WRITE_LIFE_1,WRITE_LIFE_2?With the current choice:
- If the count goes up (beyond 5 hints), infra can scale fine but these names do not. ImagineULTRA_EXTREMEafter EXTREME.
- Applications or in-kernel users can specify LONG hint with data that actually has a SHORT lifetime. Nothing really ensures that LONG is really LONG.
Quoted 2 lines
Temperature-neutral names seem more generic/scalable and do not present the unnecessary need to be accurate with relative temperatures.
Thanks for having taken a look at this patch series. Jens asked for data that shows that this patch series improves performance. Is this something Samsung can help with?
Thanks,
Bart.
RE:(2) [PATCH v3 01/14] fs: Move enum rw_hint into a new header file
Hi Bart,
Quoted 27 lines · Kanchan Joshi
Quoted 17 lines · Bart Van Assche
include/linux/rw_hint.h line 18
Quoted 11 lines
include/linux/rw_hint.h new file +11
-0+/* Block storage write lifetime hint values. */ +enum rw_hint { + WRITE_LIFE_NOT_SET = 0, /* RWH_WRITE_LIFE_NOT_SET */ + WRITE_LIFE_NONE = 1, /* RWH_WRITE_LIFE_NONE */ + WRITE_LIFE_SHORT = 2, /* RWH_WRITE_LIFE_SHORT */ + WRITE_LIFE_MEDIUM = 3, /* RWH_WRITE_LIFE_MEDIUM */ + WRITE_LIFE_LONG = 4, /* RWH_WRITE_LIFE_LONG */ + WRITE_LIFE_EXTREME = 5, /* RWH_WRITE_LIFE_EXTREME */ +} __packed; + +static_assert (sizeof (enum rw_hint ) ==1 );
Does it make sense to do away with these, and have temperature-neutral names instead e.g.,WRITE_LIFE_1,WRITE_LIFE_2?
With the current choice:Quoted 8 lines
- If the count goes up (beyond 5 hints), infra can scale fine but these names do not. Imagine
ULTRA_EXTREMEafter EXTREME.
- Applications or in-kernel users can specify LONG hint with data that actually has a SHORT lifetime. Nothing really ensures that LONG is really LONG.
Temperature-neutral names seem more generic/scalable and do not present the unnecessary need to be accurate with relative temperatures.
Quoted 3 lines
Thanks for having taken a look at this patch series. Jens asked for data that shows that this patch series improves performance. Is this something Samsung can help with?
We analyzed the NAND block erase counter with and without stream separation through a long-term workload in F2FS.
The analysis showed that the erase counter is reduced by approximately 40% with stream seperation.
Long-term workload is a scenario where erase and write are repeated by stream after performing precondition fill for each temperature of F2FS.
Thanks,
Daejun.
Quoted 7 lines
Thanks,
Bart.
Re: (2) [PATCH v3 01/14] fs: Move enum rw_hint into a new header file
Quoted 6 lines · Daejun Park
Quoted 5 lines · Kanchan Joshi
Quoted 1 line · Bart Van Assche
Thanks for having taken a look at this patch series. Jens asked for data that shows that this patch series improves performance. Is this something Samsung can help with?
Quoted 6 lines
We analyzed the NAND block erase counter with and without stream separation through a long-term workload in F2FS.
The analysis showed that the erase counter is reduced by approximately 40% with stream seperation.
Long-term workload is a scenario where erase and write are repeated by stream after performing precondition fill for each temperature of F2FS.
Hi Daejun,
Thank you for having shared this data. This is very helpful. Since I'm not familiar with the erase counter: does the above data perhaps mean that write amplification is reduced by 40% in the workload that has been examined?
Thanks,
Bart.
RE:(2) (2) [PATCH v3 01/14] fs: Move enum rw_hint into a new header file
Hi Bart,
Quoted 14 lines · Daejun Park
Quoted 5 lines
Quoted 5 lines · Kanchan Joshi
Quoted 1 line · Bart Van Assche
Thanks for having taken a look at this patch series. Jens asked for data that shows that this patch series improves performance. Is this something Samsung can help with?
Quoted 7 lines
We analyzed the NAND block erase counter with and without stream separation through a long-term workload in F2FS.
The analysis showed that the erase counter is reduced by approximately 40% with stream seperation.
Long-term workload is a scenario where erase and write are repeated by stream after performing precondition fill for each temperature of F2FS.
Quoted 6 lines
Hi Daejun,
Thank you for having shared this data. This is very helpful. Since I'm not familiar with the erase counter: does the above data perhaps mean that write amplification is reduced by 40% in the workload that has been examined?
WAF is not only caused by GC. It is also caused by other reasons. During device GC, the valid pages in the victim block are migrated, and a lower erase counter means that the effective GC is performed by selecting a victim block with a small number of invalid pages.
Thus, it can be said that the WAF can be decreased about 40% by selecting fewer victim blocks during device GC.
Thanks,
Daejun
Quoted 4 lines
Thanks,
Bart.
PATCH v3 2/14 block: Restore data lifetime support in struct bio and struct request
In reply to Bart Van AsscheProvide a mechanism for filesystems to pass data lifetime information to block drivers. Data lifetime information can be used by block devices with append/erase storage technology (NAND flash) to reduce garbage collection activity.
This patch restores a subset of the functionality that was removed by commit c75e707fe1aa ("block: remove the per-bio/request write hint").
--- block/bio.c | 2 ++ block/blk-crypto-fallback.c | 1 + block/blk-merge.c | 6 ++++++ block/blk-mq.c | 1 + block/bounce.c | 1 + include/linux/blk-mq.h | 2 ++ include/linux/blk_types.h | 2 ++ 7 files changed, 15 insertions(+)
block/bio.c +2 -0
@@ -251,6 +251,7 @@ void bio_init(struct bio *bio, struct block_device *bdev, struct bio_vec *table, bio->bi_opf = opf; bio->bi_flags =0 ; bio->bi_ioprio =0 ; + bio->bi_lifetime =0 ; bio->bi_status =0 ; bio->bi_iter.bi_sector =0 ; bio->bi_iter.bi_size =0 ;
@@ -813,6 +814,7 @@ 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_lifetime = bio_src->bi_lifetime; bio->bi_iter = bio_src->bi_iter;if (bio->bi_bdev) {
block/blk-crypto-fallback.c +1 -0
@@ -172,6 +172,7 @@ static struct bio *blk_crypto_fallback_clone_bio(struct bio *bio_src)if (bio_flagged (bio_src,BIO_REMAPPED ))bio_set_flag (bio,BIO_REMAPPED ); bio->bi_ioprio = bio_src->bi_ioprio; + bio->bi_lifetime = bio_src->bi_lifetime; bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
block/blk-merge.c +6 -0
@@ -814,6 +814,9 @@ static struct request *attempt_merge(struct request_queue *q,if (rq_data_dir (req) !=rq_data_dir (next))return NULL ; +if (req->lifetime != next->lifetime) +return NULL ; +if (req->ioprio != next->ioprio)return NULL ;
@@ -941,6 +944,9 @@ bool blk_rq_merge_ok(struct request *rq, struct bio *bio)if (!bio_crypt_rq_ctx_compatible (rq, bio))return false; +if (rq->lifetime != bio->bi_lifetime) +return NULL ; +if (rq->ioprio !=bio_prio (bio))return false;
block/blk-mq.c +1 -0
@@ -3148,6 +3148,7 @@ int blk_rq_prep_clone(struct request *rq, struct request *rq_src, } rq->nr_phys_segments = rq_src->nr_phys_segments; rq->ioprio = rq_src->ioprio; + rq->lifetime = rq_src->lifetime;if (rq->bio &&blk_crypto_rq_bio_prep (rq, rq->bio, gfp_mask) <0 ) goto free_and_out;
block/bounce.c +1 -0
@@ -169,6 +169,7 @@ static struct bio *bounce_clone_bio(struct bio *bio_src)if (bio_flagged (bio_src,BIO_REMAPPED ))bio_set_flag (bio,BIO_REMAPPED ); bio->bi_ioprio = bio_src->bi_ioprio; + bio->bi_lifetime = bio_src->bi_lifetime; bio->bi_iter.bi_sector = bio_src->bi_iter.bi_sector; bio->bi_iter.bi_size = bio_src->bi_iter.bi_size;
include/linux/blk-mq.h +2 -0
@@ -136,6 +136,7 @@ struct request {#endif unsigned short ioprio; +enum rw_hint lifetime;enum mq_rq_state state;atomic_t ref;
@@ -957,6 +958,7 @@ static inline void blk_rq_bio_prep(struct request *rq, struct bio *bio, rq->__data_len = bio->bi_iter.bi_size; rq->bio = rq->biotail = bio; rq->ioprio =bio_prio (bio); + rq->lifetime = bio->bi_lifetime; } void blk_mq_hctx_set_fq_lock_class(struct blk_mq_hw_ctx *hctx,
include/linux/blk_types.h +2 -0
@@ -10,6 +10,7 @@#include <linux/bvec.h> #include <linux/device.h> #include <linux/ktime.h> +#include <linux/rw_hint.h> struct bio_set ;struct bio ;
@@ -269,6 +270,7 @@ struct bio {*/ unsigned short bi_flags;/* BIO_* below */ unsigned short bi_ioprio; +enum rw_hint bi_lifetime;/* data lifetime */ blk_status_t bi_status;atomic_t __bi_remaining;
PATCH v3 3/14 fs: Restore write hint support
In reply to Bart Van AsscheInitialize the bio lifetime to the data lifetime information that is available in struct inode. This patch reverts a small subset of commit c75e707fe1aa ("block: remove the per-bio/request write hint").
--- block/fops.c | 3 +++ fs/iomap/buffered-io.c | 2 ++ fs/iomap/direct-io.c | 1 + fs/mpage.c | 1 + 4 files changed, 7 insertions(+)
block/fops.c +3 -0
@@ -74,6 +74,7 @@ static ssize_t __blkdev_direct_IO_simple(struct kiocb *iocb, } bio.bi_iter.bi_sector = pos >>SECTOR_SHIFT ; bio.bi_ioprio = iocb->ki_ioprio; + bio.bi_lifetime = iocb->ki_filp->f_inode->i_write_hint; ret =bio_iov_iter_get_pages (&bio, iter);if (unlikely (ret))
@@ -206,6 +207,7 @@ static ssize_t __blkdev_direct_IO(struct kiocb *iocb, struct iov_iter *iter, bio->bi_private = dio; bio->bi_end_io = blkdev_bio_end_io; bio->bi_ioprio = iocb->ki_ioprio; + bio->bi_lifetime = iocb->ki_filp->f_inode->i_write_hint; ret =bio_iov_iter_get_pages (bio, iter);if (unlikely (ret)) {
@@ -323,6 +325,7 @@ static ssize_t __blkdev_direct_IO_async(struct kiocb *iocb, bio->bi_iter.bi_sector = pos >>SECTOR_SHIFT ; bio->bi_end_io = blkdev_bio_end_io_async; bio->bi_ioprio = iocb->ki_ioprio; + bio->bi_lifetime = iocb->ki_filp->f_inode->i_write_hint;if (iov_iter_is_bvec (iter)) {/*
fs/iomap/buffered-io.c +2 -0
@@ -1660,6 +1660,7 @@ iomap_alloc_ioend(struct inode *inode, struct iomap_writepage_ctx *wpc,REQ_OP_WRITE |wbc_to_write_flags (wbc), GFP_NOFS, &iomap_ioend_bioset); bio->bi_iter.bi_sector = sector; + bio->bi_lifetime = inode->i_write_hint;wbc_init_bio (wbc, bio); ioend = container_of(bio, struct iomap_ioend, io_inline_bio);
@@ -1690,6 +1691,7 @@ iomap_chain_bio(struct bio *prev) new =bio_alloc (prev->bi_bdev,BIO_MAX_VECS , prev->bi_opf,GFP_NOFS );bio_clone_blkg_association (new, prev); new->bi_iter.bi_sector =bio_end_sector (prev); + new->bi_lifetime = prev->bi_lifetime;bio_chain (prev, new);bio_get (prev);/* for iomap_finish_ioend */
fs/iomap/direct-io.c +1 -0
@@ -381,6 +381,7 @@ static loff_t iomap_dio_bio_iter(const struct iomap_iter *iter, GFP_KERNEL); bio->bi_iter.bi_sector =iomap_sector (iomap, pos); bio->bi_ioprio = dio->iocb->ki_ioprio; + bio->bi_lifetime = dio->iocb->ki_filp->f_inode->i_write_hint; bio->bi_private = dio; bio->bi_end_io = iomap_dio_bio_end_io;
fs/mpage.c +1 -0
@@ -612,6 +612,7 @@ static int __mpage_writepage(struct folio *folio, struct writeback_control *wbc, GFP_NOFS); bio->bi_iter.bi_sector = blocks[0 ] << (blkbits -9 );wbc_init_bio (wbc, bio); + bio->bi_lifetime = inode->i_write_hint; }/*
PATCH v3 4/14 fs/f2fs: Restore data lifetime support
In reply to Bart Van AsscheRestore support for the whint_mode mount option by reverting commit 930e2607638d ("f2fs: remove obsolete whint_mode"). Additionally, restore the bio->bi_lifetime assignment in __bio_alloc() that was removed by commit c75e707fe1aa ("block: remove the per-bio/request write hint").
--- Documentation/filesystems/f2fs.rst | 70 ++++++++++++++++++++++ fs/f2fs/data.c | 2 + fs/f2fs/f2fs.h | 9 +++ fs/f2fs/segment.c | 95 ++++++++++++++++++++++++++++++ fs/f2fs/super.c | 32 +++++++++- 5 files changed, 207 insertions(+), 1 deletion(-)
Documentation/filesystems/f2fs.rst +70 -0
@@ -242,6 +242,12 @@ offgrpjquota Turn off group journalled quota. offprjjquota Turn off project journalled quota. quota Enable plain user disk quota accounting. noquota Disable all plain disk quota option. +whint_mode=%s Control which write hints are passed down to block + layer. This supports "off", "user-based", and + "fs-based". In "off" mode (default), f2fs does not pass + down hints. In "user-based" mode, f2fs tries to pass + down hints given by users. And in "fs-based" mode, f2fs + passes down hints with its policy. alloc_mode=%s Adjust block allocation policy, which supports "reuse" and "default". fsync_mode=%s Control the policy of fsync. Currently supports "posix",
@@ -776,6 +782,70 @@ In order to identify whether the data in the victim segment are valid or not, F2FS manages a bitmap. Each bit represents the validity of a block, and the bitmap is composed of a bit stream covering whole blocks in main area. +Write-hint Policy +----------------- + +1) whint_mode=off. F2FS only passes down WRITE_LIFE_NOT_SET. + +2) whint_mode=user-based. F2FS tries to pass down hints given by +users. + +===================== ======================== =================== +User F2FS Block +===================== ======================== =================== +N/A META WRITE_LIFE_NOT_SET +N/A HOT_NODE " +N/A WARM_NODE " +N/A COLD_NODE " +ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME +extension list " " + +-- buffered io +WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME +WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT +WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET +WRITE_LIFE_NONE " " +WRITE_LIFE_MEDIUM " " +WRITE_LIFE_LONG " " + +-- direct io +WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME +WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT +WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET +WRITE_LIFE_NONE " WRITE_LIFE_NONE +WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM +WRITE_LIFE_LONG " WRITE_LIFE_LONG +===================== ======================== =================== + +3) whint_mode=fs-based. F2FS passes down hints with its policy. + +===================== ======================== =================== +User F2FS Block +===================== ======================== =================== +N/A META WRITE_LIFE_MEDIUM; +N/A HOT_NODE WRITE_LIFE_NOT_SET +N/A WARM_NODE " +N/A COLD_NODE WRITE_LIFE_NONE +ioctl(COLD) COLD_DATA WRITE_LIFE_EXTREME +extension list " " + +-- buffered io +WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME +WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT +WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_LONG +WRITE_LIFE_NONE " " +WRITE_LIFE_MEDIUM " " +WRITE_LIFE_LONG " " + +-- direct io +WRITE_LIFE_EXTREME COLD_DATA WRITE_LIFE_EXTREME +WRITE_LIFE_SHORT HOT_DATA WRITE_LIFE_SHORT +WRITE_LIFE_NOT_SET WARM_DATA WRITE_LIFE_NOT_SET +WRITE_LIFE_NONE " WRITE_LIFE_NONE +WRITE_LIFE_MEDIUM " WRITE_LIFE_MEDIUM +WRITE_LIFE_LONG " WRITE_LIFE_LONG +===================== ======================== =================== + Fallocate(2) Policy -------------------
fs/f2fs/data.c +2 -0
@@ -478,6 +478,8 @@ static struct bio *__bio_alloc(struct f2fs_io_info *fio, int npages) }else { bio->bi_end_io = f2fs_write_end_io; bio->bi_private = sbi; + bio->bi_lifetime =f2fs_io_type_to_rw_hint (sbi, fio->type, + fio->temp); }iostat_alloc_and_bind_ctx (sbi, bio,NULL );
fs/f2fs/f2fs.h +9 -0
@@ -158,6 +158,7 @@ struct f2fs_mount_info {int s_jquota_fmt;/* Format of quota to use */ #endif /* For which write hints are passed down to block layer */ +int whint_mode;int alloc_mode;/* segment allocation policy */ int fsync_mode;/* fsync policy */ int fs_mode;/* fs mode: LFS or ADAPTIVE */
@@ -1344,6 +1345,12 @@ enum { FS_MODE_FRAGMENT_BLK, /* block fragmentation mode */ }; +enum { +WHINT_MODE_OFF ,/* not pass down write hints */ +WHINT_MODE_USER ,/* try to pass down hints given by users */ +WHINT_MODE_FS ,/* pass down hints with F2FS policy */ +}; +enum {ALLOC_MODE_DEFAULT ,/* stay default */ ALLOC_MODE_REUSE ,/* reuse segments as much as possible */
@@ -3728,6 +3735,8 @@ void f2fs_destroy_segment_manager(struct f2fs_sb_info *sbi);int __init f2fs_create_segment_manager_caches (void );void f2fs_destroy_segment_manager_caches (void );int f2fs_rw_hint_to_seg_type (enum rw_hint hint); +enum rw_hint f2fs_io_type_to_rw_hint (struct f2fs_sb_info *sbi, +enum page_type type,enum temp_type temp);unsigned int f2fs_usable_segs_in_sec (struct f2fs_sb_info *sbi,unsigned int segno); unsigned int f2fs_usable_blks_in_seg(struct f2fs_sb_info *sbi,
fs/f2fs/segment.c +95 -0
101 lines not on this page · raw message
fs/f2fs/super.c +31 -1
@@ -141,6 +141,7 @@ enum { Opt_jqfmt_vfsold, Opt_jqfmt_vfsv0, Opt_jqfmt_vfsv1, + Opt_whint, Opt_alloc, Opt_fsync, Opt_test_dummy_encryption,
@@ -220,6 +221,7 @@ static match_table_t f2fs_tokens = { {Opt_jqfmt_vfsold, "jqfmt=vfsold"}, {Opt_jqfmt_vfsv0, "jqfmt=vfsv0"}, {Opt_jqfmt_vfsv1, "jqfmt=vfsv1"}, + {Opt_whint, "whint_mode=%s"}, {Opt_alloc, "alloc_mode=%s"}, {Opt_fsync, "fsync_mode=%s"}, {Opt_test_dummy_encryption, "test_dummy_encryption=%s"},
@@ -988,6 +990,22 @@ static int parse_options(struct super_block *sb, char *options, bool is_remount)f2fs_info (sbi,"quota operations not supported" );break ;#endif +case Opt_whint: + name =match_strdup (&args[0 ]); +if (!name) +return -ENOMEM ; +if (!strcmp (name,"user-based" )) { +F2FS_OPTION (sbi).whint_mode =WHINT_MODE_USER ; + }else if (!strcmp (name,"off" )) { +F2FS_OPTION (sbi).whint_mode =WHINT_MODE_OFF ; + }else if (!strcmp (name,"fs-based" )) { +F2FS_OPTION (sbi).whint_mode =WHINT_MODE_FS ; + }else { +kfree (name); +return -EINVAL ; + } +kfree (name); +break ;case Opt_alloc: name =match_strdup (&args[0 ]);if (!name)
@@ -1389,6 +1407,12 @@ static int parse_options(struct super_block *sb, char *options, bool is_remount)return -EINVAL ; } +/* Not pass down write hints if the number of active logs is lesser +* than NR_CURSEG_PERSIST_TYPE. +*/ +if (F2FS_OPTION (sbi).active_logs !=NR_CURSEG_PERSIST_TYPE ) +F2FS_OPTION (sbi).whint_mode =WHINT_MODE_OFF ; +if (f2fs_sb_has_readonly (sbi) && !f2fs_readonly (sbi->sb)) {f2fs_err (sbi,"Allow to mount readonly mode only" );return -EROFS ;
@@ -2060,6 +2084,10 @@ static int f2fs_show_options(struct seq_file *seq, struct dentry *root)seq_puts (seq,",prjquota" );#endif f2fs_show_quota_options (seq, sbi->sb); +if (F2FS_OPTION (sbi).whint_mode ==WHINT_MODE_USER ) +seq_printf (seq,",whint_mode=%s" ,"user-based" ); +else if (F2FS_OPTION (sbi).whint_mode ==WHINT_MODE_FS ) +seq_printf (seq,",whint_mode=%s" ,"fs-based" );fscrypt_show_test_dummy_encryption (seq,',' , sbi->sb);
@@ -2129,6 +2157,7 @@ static void default_options(struct f2fs_sb_info *sbi, bool remount)F2FS_OPTION (sbi).active_logs =NR_CURSEG_PERSIST_TYPE ;F2FS_OPTION (sbi).inline_xattr_size =DEFAULT_INLINE_XATTR_ADDRS ; +F2FS_OPTION (sbi).whint_mode =WHINT_MODE_OFF ;if (le32_to_cpu (F2FS_RAW_SUPER (sbi)->segment_count_main) <=SMALL_VOLUME_SEGMENTS )F2FS_OPTION (sbi).alloc_mode =ALLOC_MODE_REUSE ;
@@ -2443,7 +2472,8 @@ static int f2fs_remount(struct super_block *sb, int *flags, char *data) need_stop_gc = true; } -if (*flags &SB_RDONLY ) { +if (*flags &SB_RDONLY || +F2FS_OPTION (sbi).whint_mode != org_mount_opt.whint_mode) {sync_inodes_sb (sb);set_sbi_flag (sbi,SBI_IS_DIRTY );
PATCH v3 5/14 scsi: core: Query the Block Limits Extension VPD page
In reply to Bart Van AsscheParse the Reduced Stream Control Supported (RSCS) bit from the block limits extension VPD page. The RSCS bit is defined in SBC-5 r05
(https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf).
--- drivers/scsi/scsi.c | 2 ++ drivers/scsi/scsi_sysfs.c | 10 ++++++++++ drivers/scsi/sd.c | 13 +++++++++++++ drivers/scsi/sd.h | 1 + include/scsi/scsi_device.h | 1 + 5 files changed, 27 insertions(+)
drivers/scsi/scsi.c +2 -0
@@ -499,6 +499,8 @@ void scsi_attach_vpd(struct scsi_device *sdev)scsi_update_vpd_page (sdev,0xb1 , &sdev->vpd_pgb1);if (vpd_buf->data[i] ==0xb2 )scsi_update_vpd_page (sdev,0xb2 , &sdev->vpd_pgb2); +if (vpd_buf->data[i] ==0xb7 ) +scsi_update_vpd_page (sdev,0xb7 , &sdev->vpd_pgb7); }kfree (vpd_buf); }
drivers/scsi/scsi_sysfs.c +10 -0
@@ -449,6 +449,7 @@ static void scsi_device_dev_release(struct device *dev)struct scsi_vpd *vpd_pg80 =NULL , *vpd_pg83 =NULL ;struct scsi_vpd *vpd_pg0 =NULL , *vpd_pg89 =NULL ;struct scsi_vpd *vpd_pgb0 =NULL , *vpd_pgb1 =NULL , *vpd_pgb2 =NULL ; +struct scsi_vpd *vpd_pgb7 =NULL ;unsigned long flags;might_sleep ();
@@ -494,6 +495,8 @@ static void scsi_device_dev_release(struct device *dev) lockdep_is_held(&sdev->inquiry_mutex)); vpd_pgb2 =rcu_replace_pointer (sdev->vpd_pgb2, vpd_pgb2,lockdep_is_held (&sdev->inquiry_mutex)); + vpd_pgb7 =rcu_replace_pointer (sdev->vpd_pgb7, vpd_pgb7, +lockdep_is_held (&sdev->inquiry_mutex));mutex_unlock (&sdev->inquiry_mutex);if (vpd_pg0)
@@ -510,6 +513,8 @@ static void scsi_device_dev_release(struct device *dev)kfree_rcu (vpd_pgb1, rcu);if (vpd_pgb2)kfree_rcu (vpd_pgb2, rcu); +if (vpd_pgb7) +kfree_rcu (vpd_pgb7, rcu);kfree (sdev->inquiry);kfree (sdev);
@@ -921,6 +926,7 @@ sdev_vpd_pg_attr(pg89);sdev_vpd_pg_attr (pgb0);sdev_vpd_pg_attr (pgb1);sdev_vpd_pg_attr (pgb2); +sdev_vpd_pg_attr (pgb7);sdev_vpd_pg_attr (pg0); static ssize_t show_inquiry(struct file *filep, struct kobject *kobj,
@@ -1295,6 +1301,9 @@ static umode_t scsi_sdev_bin_attr_is_visible(struct kobject *kobj,if (attr == &dev_attr_vpd_pgb2 && !sdev->vpd_pgb2)return 0 ; +if (attr == &dev_attr_vpd_pgb7 && !sdev->vpd_pgb7) +return 0 ; +return S_IRUGO ; }
@@ -1347,6 +1356,7 @@ static struct bin_attribute *scsi_sdev_bin_attrs[] = { &dev_attr_vpd_pgb0, &dev_attr_vpd_pgb1, &dev_attr_vpd_pgb2, + &dev_attr_vpd_pgb7, &dev_attr_inquiry, NULL };
drivers/scsi/sd.c +13 -0
@@ -3019,6 +3019,18 @@ static void sd_read_block_limits(struct scsi_disk *sdkp)rcu_read_unlock (); } +/* Parse the Block Limits Extension VPD page (0xb7) */ +static void sd_read_block_limits_ext (struct scsi_disk *sdkp) +{ +struct scsi_vpd *vpd; + +rcu_read_lock (); + vpd =rcu_dereference (sdkp->device->vpd_pgb7); +if (vpd && vpd->len >=2 ) + sdkp->rscs = vpd->data[5 ] &1 ; +rcu_read_unlock (); +} +/** * sd_read_block_characteristics - Query block dev. characteristics * @sdkp: disk to query
@@ -3373,6 +3385,7 @@ static int sd_revalidate_disk(struct gendisk *disk)if (scsi_device_supports_vpd (sdp)) {sd_read_block_provisioning (sdkp);sd_read_block_limits (sdkp); +sd_read_block_limits_ext (sdkp);sd_read_block_characteristics (sdkp);sd_zbc_read_zones (sdkp, buffer);sd_read_cpr (sdkp);
drivers/scsi/sd.h +1 -0
@@ -150,6 +150,7 @@ struct scsi_disk { unsigned urswrz : 1; unsigned security : 1; unsigned ignore_medium_access_errors : 1; + bool rscs : 1; /* reduced stream control support */ };#define to_scsi_disk (obj) container_of(obj, struct scsi_disk, disk_dev)
include/scsi/scsi_device.h +1 -0
@@ -153,6 +153,7 @@ struct scsi_device {struct scsi_vpd __rcu *vpd_pgb0;struct scsi_vpd __rcu *vpd_pgb1;struct scsi_vpd __rcu *vpd_pgb2; +struct scsi_vpd __rcu *vpd_pgb7;struct scsi_target *sdev_target;
PATCH v3 6/14 scsi_proto: Add structures and constants related to I/O groups and streams
In reply to Bart Van AsschePrepare for adding code that will query the I/O advice hints group descriptors and for adding code that will retrieve the stream status.
--- include/scsi/scsi_proto.h | 75 +++++++++++++++++++++++++++++++++++++++ 1 file changed, 75 insertions(+)
include/scsi/scsi_proto.h +75 -0
@@ -10,6 +10,7 @@#ifndef _SCSI_PROTO_H_#define _SCSI_PROTO_H_ +#include <linux/build_bug.h> #include <linux/types.h> /*
@@ -126,6 +127,7 @@#define SAI_READ_CAPACITY_16 0x10#define SAI_GET_LBA_STATUS 0x12#define SAI_REPORT_REFERRALS 0x13 +#define SAI_GET_STREAM_STATUS 0x16/* values for maintenance in */ #define MI_REPORT_IDENTIFYING_INFORMATION 0x05#define MI_REPORT_TARGET_PGS 0x0a
@@ -275,6 +277,79 @@ struct scsi_lun {__u8 scsi_lun[8 ]; }; +/* SBC-5 IO advice hints group descriptor */ +struct scsi_io_group_descriptor { +#if defined(__BIG_ENDIAN) +u8 io_advice_hints_mode:2 ; +u8 reserved1:3 ; +u8 st_enble:1 ; +u8 cs_enble:1 ; +u8 ic_enable:1 ; +#elif defined(__LITTLE_ENDIAN) +u8 ic_enable:1 ; +u8 cs_enble:1 ; +u8 st_enble:1 ; +u8 reserved1:3 ; +u8 io_advice_hints_mode:2 ; +#else +#error +#endif +u8 reserved2[3 ]; +/* Logical block markup descriptor */ +#if defined(__BIG_ENDIAN) +u8 acdlu:1 ; +u8 reserved3:1 ; +u8 rlbsr:2 ; +u8 lbm_descriptor_type:4 ; +#elif defined(__LITTLE_ENDIAN) +u8 lbm_descriptor_type:4 ; +u8 rlbsr:2 ; +u8 reserved3:1 ; +u8 acdlu:1 ; +#else +#error +#endif +u8 params[2 ]; +u8 reserved4; +u8 reserved5[8 ]; +}; + +static_assert (sizeof (struct scsi_io_group_descriptor ) ==16 ); + +struct scsi_stream_status { +#if defined(__BIG_ENDIAN) +u16 perm:1 ; +u16 reserved1:15 ; +#elif defined(__LITTLE_ENDIAN) +u16 reserved1:15 ; +u16 perm:1 ; +#else +#error +#endif +__be16 stream_identifier; +#if defined(__BIG_ENDIAN) +u8 reserved2:2 ; +u8 rel_lifetime:6 ; +#elif defined(__LITTLE_ENDIAN) +u8 rel_lifetime:6 ; +u8 reserved2:2 ; +#else +#error +#endif +u8 reserved3[3 ]; +}; + +static_assert (sizeof (struct scsi_stream_status ) ==8 ); + +struct scsi_stream_status_header { +__be32 len;/* length in bytes of stream_status[] array. */ +u16 reserved; +u16 number_of_open_streams; + DECLARE_FLEX_ARRAY(struct scsi_stream_status, stream_status); +}; + +static_assert (sizeof (struct scsi_stream_status_header ) ==8 ); +/* SPC asymmetric access states */ #define SCSI_ACCESS_STATE_OPTIMAL 0x00#define SCSI_ACCESS_STATE_ACTIVE 0x01
PATCH v3 7/14 sd: Translate data lifetime information
In reply to Bart Van AsscheRecently T10 standardized SBC constrained streams. This mechanism allows to pass data lifetime information to SCSI devices in the group number field. Add support for translating write hint information into a permanent stream number in the sd driver. Use WRITE(10) instead of WRITE(6) if data lifetime information is present because the WRITE(6) command does not have a GROUP NUMBER field.
--- drivers/scsi/sd.c | 98 +++++++++++++++++++++++++++++++++++++++++++++-- drivers/scsi/sd.h | 2 + 2 files changed, 97 insertions(+), 3 deletions(-)
drivers/scsi/sd.c +95 -3
@@ -47,6 +47,7 @@#include <linux/blkpg.h> #include <linux/blk-pm.h> #include <linux/delay.h> +#include <linux/rw_hint.h> #include <linux/major.h> #include <linux/mutex.h> #include <linux/string_helpers.h>
@@ -1001,12 +1002,38 @@ static blk_status_t sd_setup_flush_cmnd(struct scsi_cmnd *cmd)return BLK_STS_OK ; } +/** +* sd_group_number() - Compute the GROUP NUMBER field +* @cmd: SCSI command for which to compute the value of the six-bit GROUP NUMBER +* field. +* +* From SBC-5 r05 (https://www.t10.org/cgi-bin/ac.pl?t=f&f=sbc5r05.pdf): +* 0: no relative lifetime. +* 1: shortest relative lifetime. +* 2: second shortest relative lifetime. +* 3 - 0x3d: intermediate relative lifetimes. +* 0x3e: second longest relative lifetime. +* 0x3f: longest relative lifetime. +*/ +static u8 sd_group_number (struct scsi_cmnd *cmd) +{ +const struct request *rq =scsi_cmd_to_rq (cmd); +struct scsi_disk *sdkp =scsi_disk (rq->q->disk); + +if (!sdkp->rscs) +return 0 ; + +return min3 ((u32 )rq->lifetime, (u32 )sdkp->permanent_stream_count, +0x3fu ); +} +static blk_status_t sd_setup_rw32_cmnd (struct scsi_cmnd *cmd,bool write,sector_t lba,unsigned int nr_blocks,unsigned char flags,unsigned int dld) { cmd->cmd_len =SD_EXT_CDB_SIZE ; cmd->cmnd[0 ] =VARIABLE_LENGTH_CMD ; + cmd->cmnd[6 ] =sd_group_number (cmd); cmd->cmnd[7 ] =0x18 ;/* Additional CDB len */ cmd->cmnd[9 ] = write ?WRITE_32 :READ_32 ; cmd->cmnd[10 ] = flags;
@@ -1025,7 +1052,7 @@ static blk_status_t sd_setup_rw16_cmnd(struct scsi_cmnd *cmd, bool write, cmd->cmd_len =16 ; cmd->cmnd[0 ] = write ?WRITE_16 :READ_16 ; cmd->cmnd[1 ] = flags | ((dld >>2 ) &0x01 ); - cmd->cmnd[14 ] = (dld &0x03 ) <<6 ; + cmd->cmnd[14 ] = ((dld &0x03 ) <<6 ) |sd_group_number (cmd); cmd->cmnd[15 ] =0 ;put_unaligned_be64 (lba, &cmd->cmnd[2 ]);put_unaligned_be32 (nr_blocks, &cmd->cmnd[10 ]);
@@ -1040,7 +1067,7 @@ static blk_status_t sd_setup_rw10_cmnd(struct scsi_cmnd *cmd, bool write, cmd->cmd_len =10 ; cmd->cmnd[0 ] = write ?WRITE_10 :READ_10 ; cmd->cmnd[1 ] = flags; - cmd->cmnd[6 ] =0 ; + cmd->cmnd[6 ] =sd_group_number (cmd); cmd->cmnd[9 ] =0 ;put_unaligned_be32 (lba, &cmd->cmnd[2 ]);put_unaligned_be16 (nr_blocks, &cmd->cmnd[7 ]);
@@ -1177,7 +1204,7 @@ static blk_status_t sd_setup_read_write_cmnd(struct scsi_cmnd *cmd) ret =sd_setup_rw16_cmnd (cmd, write, lba, nr_blocks, protect | fua, dld); }else if ((nr_blocks >0xff ) || (lba >0x1fffff ) || - sdp->use_10_for_rw || protect) { + sdp->use_10_for_rw || protect || rq->lifetime) { ret =sd_setup_rw10_cmnd (cmd, write, lba, nr_blocks, protect | fua); }else {
@@ -2912,6 +2939,70 @@ sd_read_cache_type(struct scsi_disk *sdkp, unsigned char *buffer) sdkp->DPOFUA =0 ; } +static bool sd_is_perm_stream (struct scsi_disk *sdkp,unsigned stream_id) +{ +u8 cdb[16 ] = {SERVICE_ACTION_IN_16 ,SAI_GET_STREAM_STATUS }; +struct { +struct scsi_stream_status_header h; +struct scsi_stream_status s; + } buf; +struct scsi_device *sdev = sdkp->device; +struct scsi_sense_hdr sshdr; +const struct scsi_exec_args exec_args = { + .sshdr = &sshdr, + }; +int res; + +put_unaligned_be16 (stream_id, &cdb[4 ]); +put_unaligned_be32 (sizeof (buf), &cdb[10 ]); + + res =scsi_execute_cmd (sdev, cdb,REQ_OP_DRV_IN , &buf,sizeof (buf), +SD_TIMEOUT , sdkp->max_retries, &exec_args); +if (res <0 ) +return false; +if (scsi_status_is_check_condition (res) &&scsi_sense_valid (&sshdr)) +sd_print_sense_hdr (sdkp, &sshdr); +if (res) +return false; +if (get_unaligned_be32 (&buf.h.len) <sizeof (struct scsi_stream_status )) +return false; +return buf.h.stream_status[0 ].perm; +} + +static void sd_read_io_hints (struct scsi_disk *sdkp,unsigned char *buffer) +{ +struct scsi_device *sdp = sdkp->device; +const struct scsi_io_group_descriptor *desc, *start, *end; +struct scsi_sense_hdr sshdr; +struct scsi_mode_data data; +int res; + + res =scsi_mode_sense (sdp,/*dbd=*/ 0x8 ,/*modepage=*/ 0x0a , +/*subpage=*/ 0x05 , buffer,SD_BUF_SIZE , +SD_TIMEOUT , sdkp->max_retries, &data, &sshdr); +if (res <0 ) +return ; + start = (void *)buffer + data.header_length +16 ; + end = (void *)buffer +ALIGN_DOWN (data.header_length + data.length, +sizeof (*end)); +/* +* From "SBC-5 Constrained Streams with Data Lifetimes": Device severs +* should assign the lowest numbered stream identifiers to permanent +* streams. +*/ +for (desc = start; desc < end; desc++) +if (!desc->st_enble || !sd_is_perm_stream (sdkp, desc - start)) +break ; + sdkp->permanent_stream_count = desc - start; +if (sdkp->rscs && sdkp->permanent_stream_count <2 ) +sd_printk (KERN_INFO , sdkp, +"Unexpected: RSCS has been set and the permanent stream count is %u\n" , + sdkp->permanent_stream_count); +else if (sdkp->permanent_stream_count) +sd_printk (KERN_INFO , sdkp,"permanent stream count = %d\n" , + sdkp->permanent_stream_count); +} +/* * The ATO bit indicates whether the DIF application tag is available * for use by the operating system.
@@ -3395,6 +3486,7 @@ static int sd_revalidate_disk(struct gendisk *disk)sd_read_write_protect_flag (sdkp, buffer);sd_read_cache_type (sdkp, buffer); +sd_read_io_hints (sdkp, buffer);sd_read_app_tag_own (sdkp, buffer);sd_read_write_same (sdkp, buffer);sd_read_security (sdkp, buffer);
drivers/scsi/sd.h +2 -0
@@ -125,6 +125,8 @@ struct scsi_disk {unsigned int physical_block_size;unsigned int max_medium_access_timeouts;unsigned int medium_access_timed_out; +/* number of permanent streams */ +u16 permanent_stream_count;u8 media_present;u8 write_prot;u8 protection_type;/* Data Integrity Field */
kernel test robot
build botRe: [PATCH v3 07/14] sd: Translate data lifetime information
kernel test robot noticed the following build warnings:
Hi Bart,
kernel test robot noticed the following build warnings:
[auto build test WARNING on jejb-scsi/for-next]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Bart-Van-Assche/fs-Move-enum-rw_hint-into-a-new-header-file/20231018-045042 base: https://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git for-next patch link: https://lore.kernel.org/r/20231017204739.3409052-8-bvanassche%40acm.org patch subject: [PATCH v3 07/14] sd: Translate data lifetime information reproduce: (https://download.01.org/0day-ci/archive/20231018/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
# many are suggestions rather than must-fix
WARNING:UNSPECIFIED_INT: Prefer 'unsigned int' to bare use of 'unsigned' #104: FILE: drivers/scsi/sd.c:2944:
+static bool sd_is_perm_stream(struct scsi_disk *sdkp, unsigned stream_id)
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
PATCH v3 8/14 scsi_debug: Reduce code duplication
In reply to Bart Van AsscheAll VPD pages have the page code in byte one. Reduce code duplication by storing the VPD page code once.
--- drivers/scsi/scsi_debug.c | 16 ++-------------- 1 file changed, 2 insertions(+), 14 deletions(-)
drivers/scsi/scsi_debug.c +2 -14
@@ -1598,7 +1598,8 @@ static int resp_inquiry(struct scsi_cmnd *scp, struct sdebug_dev_info *devip)u32 len;char lu_id_str[6 ];int host_no = devip->sdbg_host->shost->host_no; - + + arr[1 ] = cmd[2 ]; port_group_id = (((host_no +1 ) &0x7f ) <<8 ) + (devip->channel &0x7f );if (sdebug_vpd_use_hostno ==0 )
@@ -1609,7 +1610,6 @@ static int resp_inquiry(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) (devip->target *1000 ) -3 ; len =scnprintf (lu_id_str,6 ,"%d" , lu_id_num);if (0 == cmd[2 ]) {/* supported vital product data pages */ - arr[1 ] = cmd[2 ];/*sanity */ n =4 ; arr[n++] =0x0 ;/* this page */ arr[n++] =0x80 ;/* unit serial number */
@@ -1630,23 +1630,18 @@ static int resp_inquiry(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) } arr[3 ] = n -4 ;/* number of supported VPD pages */ }else if (0x80 == cmd[2 ]) {/* unit serial number */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] = len;memcpy (&arr[4 ], lu_id_str, len); }else if (0x83 == cmd[2 ]) {/* device identification */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_83 (&arr[4 ], port_group_id, target_dev_id, lu_id_num, lu_id_str, len, &devip->lu_name); }else if (0x84 == cmd[2 ]) {/* Software interface ident. */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_84 (&arr[4 ]); }else if (0x85 == cmd[2 ]) {/* Management network addresses */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_85 (&arr[4 ]); }else if (0x86 == cmd[2 ]) {/* extended inquiry */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =0x3c ;/* number of following entries */ if (sdebug_dif ==T10_PI_TYPE3_PROTECTION ) arr[4 ] =0x4 ;/* SPT: GRD_CHK:1 */
@@ -1656,30 +1651,23 @@ static int resp_inquiry(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) arr[4 ] =0x0 ;/* no protection stuff */ arr[5 ] =0x7 ;/* head of q, ordered + simple q's */ }else if (0x87 == cmd[2 ]) {/* mode page policy */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =0x8 ;/* number of following entries */ arr[4 ] =0x2 ;/* disconnect-reconnect mp */ arr[6 ] =0x80 ;/* mlus, shared */ arr[8 ] =0x18 ;/* protocol specific lu */ arr[10 ] =0x82 ;/* mlus, per initiator port */ }else if (0x88 == cmd[2 ]) {/* SCSI Ports */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_88 (&arr[4 ], target_dev_id); }else if (is_disk_zbc &&0x89 == cmd[2 ]) {/* ATA info */ - arr[1 ] = cmd[2 ];/*sanity */ n =inquiry_vpd_89 (&arr[4 ]);put_unaligned_be16 (n, arr +2 ); }else if (is_disk_zbc &&0xb0 == cmd[2 ]) {/* Block limits */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_b0 (&arr[4 ]); }else if (is_disk_zbc &&0xb1 == cmd[2 ]) {/* Block char. */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_b1 (devip, &arr[4 ]); }else if (is_disk &&0xb2 == cmd[2 ]) {/* LB Prov. */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_b2 (&arr[4 ]); }else if (is_zbc && cmd[2 ] ==0xb6 ) {/* ZB dev. charact. */ - arr[1 ] = cmd[2 ];/*sanity */ arr[3 ] =inquiry_vpd_b6 (devip, &arr[4 ]); }else {mk_sense_invalid_fld (scp,SDEB_IN_CDB ,2 ,-1 );
PATCH v3 9/14 scsi_debug: Support the block limits extension VPD page
In reply to Bart Van AsscheFrom SBC-5 r05:
"Reduced stream control:
a) reduces the maximum number of streams that the device server supports; and
b) increases the number of write commands that are able to specify a stream to be written in any write command that contains the GROUP NUMBER field in its CDB.
If the RSCS bit (see 6.6.5) is set to one, then the device server shall:
a) support per group stream identifier usage as described in 4.32.2;
b) support the IO Advice Hints Grouping mode page (see 6.5.7); and
c) set the MAXIMUM NUMBER OF STREAMS field (see 6.6.5) to a value that is less than 64.
Device servers that set the RSCS bit to one may support other features (e.g., permanent streams (see 4.32.4)).
4.32.4 Permanent streams
A permanent stream is a stream for which the device server does not allow closing or otherwise modifying the configuration of that stream. The PERM bit (see 5.9.2.3) indicates whether a stream is a permanent stream. If a STREAM CONTROL command (see 5.32) specifies the closing of a permanent stream, the device server terminates that command with CHECK CONDITION status instead of closing the specified stream. A permanent stream is always an open stream. Device severs should assign the lowest numbered stream identifiers to permanent streams."
Report that reduced stream control is supported.
--- drivers/scsi/scsi_debug.c | 4 ++++ 1 file changed, 4 insertions(+)
drivers/scsi/scsi_debug.c +4 -0
@@ -1627,6 +1627,7 @@ static int resp_inquiry(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) arr[n++] =0xb2 ;/* LB Provisioning */ if (is_zbc) arr[n++] =0xb6 ;/* ZB dev. char. */ + arr[n++] =0xb7 ;/* Block limits extension */ } arr[3 ] = n -4 ;/* number of supported VPD pages */ }else if (0x80 == cmd[2 ]) {/* unit serial number */
@@ -1669,6 +1670,9 @@ static int resp_inquiry(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) arr[3 ] =inquiry_vpd_b2 (&arr[4 ]); }else if (is_zbc && cmd[2 ] ==0xb6 ) {/* ZB dev. charact. */ arr[3 ] =inquiry_vpd_b6 (devip, &arr[4 ]); + }else if (cmd[2 ] ==0xb7 ) {/* block limits extension page */ + arr[3 ] =2 ;/* page length */ + arr[5 ] =1 ;/* Reduced stream control support (RSCS) */ }else {mk_sense_invalid_fld (scp,SDEB_IN_CDB ,2 ,-1 );kfree (arr);
PATCH v3 10/14 scsi_debug: Rework page code error handling
In reply to Bart Van AsscheInstead of tracking whether or not the page code is valid in a boolean variable, jump to error handling code if an unsupported page code is encountered.
--- drivers/scsi/scsi_debug.c | 24 ++++++++++++------------ 1 file changed, 12 insertions(+), 12 deletions(-)
drivers/scsi/scsi_debug.c +12 -12
@@ -2327,7 +2327,7 @@ static int resp_mode_sense(struct scsi_cmnd *scp,unsigned char *ap;unsigned char arr[SDEBUG_MAX_MSENSE_SZ ];unsigned char *cmd = scp->cmnd; -bool dbd, llbaa, msense_6, is_disk, is_zbc, bad_pcode; +bool dbd, llbaa, msense_6, is_disk, is_zbc; dbd = !!(cmd[1 ] &0x8 );/* disable block descriptors */ pcontrol = (cmd[2 ] &0xc0 ) >>6 ;
@@ -2391,7 +2391,6 @@ static int resp_mode_sense(struct scsi_cmnd *scp,mk_sense_invalid_fld (scp,SDEB_IN_CDB ,3 ,-1 );return check_condition_result; } - bad_pcode = false;switch (pcode) {case 0x1 :/* Read-Write error recovery page, direct access */
@@ -2406,15 +2405,17 @@ static int resp_mode_sense(struct scsi_cmnd *scp,if (is_disk) { len =resp_format_pg (ap, pcontrol, target); offset += len; - }else - bad_pcode = true; + }else { + goto bad_pcode; + }break ;case 0x8 :/* Caching page, direct access */ if (is_disk || is_zbc) { len =resp_caching_pg (ap, pcontrol, target); offset += len; - }else - bad_pcode = true; + }else { + goto bad_pcode; + }break ;case 0xa :/* Control Mode page, all devices */ len =resp_ctrl_m_pg (ap, pcontrol, target);
@@ -2467,18 +2468,17 @@ static int resp_mode_sense(struct scsi_cmnd *scp, }break ;default : - bad_pcode = true; -break ; - } -if (bad_pcode) { -mk_sense_invalid_fld (scp,SDEB_IN_CDB ,2 ,5 ); -return check_condition_result; + goto bad_pcode; }if (msense_6) arr[0 ] = offset -1 ;else put_unaligned_be16 ((offset -2 ), arr +0 );return fill_from_dev_buffer (scp, arr,min_t (u32, alloc_len, offset)); + +bad_pcode: +mk_sense_invalid_fld (scp,SDEB_IN_CDB ,2 ,5 ); +return check_condition_result; }#define SDEBUG_MAX_MSELECT_SZ 512
PATCH v3 11/14 scsi_debug: Rework subpage code error handling
In reply to Bart Van AsscheMove the subpage code checks into the switch statement to make it easier to add support for new page code / subpage code combinations.
--- drivers/scsi/scsi_debug.c | 70 ++++++++++++++++++++------------------- 1 file changed, 36 insertions(+), 34 deletions(-)
drivers/scsi/scsi_debug.c +36 -34
@@ -2386,22 +2386,22 @@ static int resp_mode_sense(struct scsi_cmnd *scp, ap = arr + offset; } -if ((subpcode >0x0 ) && (subpcode <0xff ) && (0x19 != pcode)) { -/* TODO: Control Extension page */ -mk_sense_invalid_fld (scp,SDEB_IN_CDB ,3 ,-1 ); -return check_condition_result; - } -switch (pcode) {case 0x1 :/* Read-Write error recovery page, direct access */ +if (subpcode >0x0 && subpcode <0xff ) + goto bad_subpcode; len =resp_err_recov_pg (ap, pcontrol, target); offset += len;break ;case 0x2 :/* Disconnect-Reconnect page, all devices */ +if (subpcode >0x0 && subpcode <0xff ) + goto bad_subpcode; len =resp_disconnect_pg (ap, pcontrol, target); offset += len;break ;case 0x3 :/* Format device page, direct access */ +if (subpcode >0x0 && subpcode <0xff ) + goto bad_subpcode;if (is_disk) { len =resp_format_pg (ap, pcontrol, target); offset += len;
@@ -2410,6 +2410,8 @@ static int resp_mode_sense(struct scsi_cmnd *scp, }break ;case 0x8 :/* Caching page, direct access */ +if (subpcode >0x0 && subpcode <0xff ) + goto bad_subpcode;if (is_disk || is_zbc) { len =resp_caching_pg (ap, pcontrol, target); offset += len;
@@ -2418,14 +2420,14 @@ static int resp_mode_sense(struct scsi_cmnd *scp, }break ;case 0xa :/* Control Mode page, all devices */ +if (subpcode >0x0 && subpcode <0xff ) + goto bad_subpcode; len =resp_ctrl_m_pg (ap, pcontrol, target); offset += len;break ;case 0x19 :/* if spc==1 then sas phy, control+discover */ -if ((subpcode >0x2 ) && (subpcode <0xff )) { -mk_sense_invalid_fld (scp,SDEB_IN_CDB ,3 ,-1 ); -return check_condition_result; - } +if (subpcode >0x2 && subpcode <0xff ) + goto bad_subpcode; len =0 ;if ((0x0 == subpcode) || (0xff == subpcode)) len +=resp_sas_sf_m_pg (ap + len, pcontrol, target);
@@ -2437,35 +2439,31 @@ static int resp_mode_sense(struct scsi_cmnd *scp, offset += len;break ;case 0x1c :/* Informational Exceptions Mode page, all devices */ +if (subpcode >0x0 && subpcode <0xff ) + goto bad_subpcode; len =resp_iec_m_pg (ap, pcontrol, target); offset += len;break ;case 0x3f :/* Read all Mode pages */ -if ((0 == subpcode) || (0xff == subpcode)) { - len =resp_err_recov_pg (ap, pcontrol, target); - len +=resp_disconnect_pg (ap + len, pcontrol, target); -if (is_disk) { - len +=resp_format_pg (ap + len, pcontrol, - target); - len +=resp_caching_pg (ap + len, pcontrol, - target); - }else if (is_zbc) { - len +=resp_caching_pg (ap + len, pcontrol, - target); - } - len +=resp_ctrl_m_pg (ap + len, pcontrol, target); - len +=resp_sas_sf_m_pg (ap + len, pcontrol, target); -if (0xff == subpcode) { - len +=resp_sas_pcd_m_spg (ap + len, pcontrol, - target, target_dev_id); - len +=resp_sas_sha_m_spg (ap + len, pcontrol); - } - len +=resp_iec_m_pg (ap + len, pcontrol, target); - offset += len; - }else { -mk_sense_invalid_fld (scp,SDEB_IN_CDB ,3 ,-1 ); -return check_condition_result; +if (subpcode >0x0 && subpcode <0xff ) + goto bad_subpcode; + len =resp_err_recov_pg (ap, pcontrol, target); + len +=resp_disconnect_pg (ap + len, pcontrol, target); +if (is_disk) { + len +=resp_format_pg (ap + len, pcontrol, target); + len +=resp_caching_pg (ap + len, pcontrol, target); + }else if (is_zbc) { + len +=resp_caching_pg (ap + len, pcontrol, target); + } + len +=resp_ctrl_m_pg (ap + len, pcontrol, target); + len +=resp_sas_sf_m_pg (ap + len, pcontrol, target); +if (0xff == subpcode) { + len +=resp_sas_pcd_m_spg (ap + len, pcontrol, target, + target_dev_id); + len +=resp_sas_sha_m_spg (ap + len, pcontrol); } + len +=resp_iec_m_pg (ap + len, pcontrol, target); + offset += len;break ;default : goto bad_pcode;
@@ -2479,6 +2477,10 @@ static int resp_mode_sense(struct scsi_cmnd *scp, bad_pcode:mk_sense_invalid_fld (scp,SDEB_IN_CDB ,2 ,5 );return check_condition_result; + +bad_subpcode: +mk_sense_invalid_fld (scp,SDEB_IN_CDB ,3 ,-1 ); +return check_condition_result; }#define SDEBUG_MAX_MSELECT_SZ 512
kernel test robot
build botRe: [PATCH v3 11/14] scsi_debug: Rework subpage code error handling
kernel test robot noticed the following build warnings:
Hi Bart,
kernel test robot noticed the following build warnings:
[auto build test WARNING on jejb-scsi/for-next]
[If your patch is applied to the wrong git tree, kindly drop us a note. And when submitting patch, we suggest to use '--base' as documented in
https://git-scm.com/docs/git-format-patch#_base_tree_information]
url: https://github.com/intel-lab-lkp/linux/commits/Bart-Van-Assche/fs-Move-enum-rw_hint-into-a-new-header-file/20231018-045042 base: https://git.kernel.org/pub/scm/linux/kernel/git/jejb/scsi.git for-next patch link: https://lore.kernel.org/r/20231017204739.3409052-12-bvanassche%40acm.org patch subject: [PATCH v3 11/14] scsi_debug: Rework subpage code error handling reproduce: (https://download.01.org/0day-ci/archive/20231018/[email protected]/reproduce)
If you fix the issue in a separate patch/commit (i.e. not just a new version of the same patch/commit), kindly add following tags
| Reported-by: kernel test robot <[email protected]>
| Closes: https://lore.kernel.org/oe-kbuild-all/[email protected]/
# many are suggestions rather than must-fix
WARNING:CONSTANT_COMPARISON: Comparisons should place the constant on the right side of the test #123: FILE: drivers/scsi/scsi_debug.c:2460:
+ if (0xff == subpcode) {
Signature · 2 lines
0-DAY CI Kernel Test Service https://github.com/intel/lkp-tests/wiki
PATCH v3 12/14 scsi_debug: Implement the IO Advice Hints Grouping mode page
In reply to Bart Van AsscheImplement an IO Advice Hints Grouping mode page with three permanent streams. A permanent stream is a stream for which the device server does not allow closing or otherwise modifying the configuration of that stream. The stream identifier enable (ST_ENBLE) bit specifies whether the stream identifier may be used in the GROUP NUMBER field of SCSI WRITE commands.
--- drivers/scsi/scsi_debug.c | 42 +++++++++++++++++++++++++++++++++++++-- 1 file changed, 40 insertions(+), 2 deletions(-)
drivers/scsi/scsi_debug.c +40 -2
@@ -2241,6 +2241,36 @@ static int resp_ctrl_m_pg(unsigned char *p, int pcontrol, int target)return sizeof (ctrl_m_pg); } +enum {MAXIMUM_NUMBER_OF_STREAMS =4 }; + +/* IO Advice Hints Grouping mode page */ +static int resp_grouping_m_pg (unsigned char *p,int pcontrol,int target) +{ +/* IO Advice Hints Grouping mode page */ +struct grouping_m_pg { +u8 page_code; +u8 subpage_code; +__be16 page_length; +u8 reserved[12 ]; +struct scsi_io_group_descriptor + descr[MAXIMUM_NUMBER_OF_STREAMS ]; + }; +static const struct grouping_m_pg gr_m_pg = { + .page_code =0xa , + .subpage_code =5 , + .page_length =cpu_to_be16 (sizeof (gr_m_pg) -4 ), + .descr = { + { .st_enble =1 }, + { .st_enble =1 }, + { .st_enble =1 }, + { .st_enble =0 }, + } + }; + +BUILD_BUG_ON (sizeof (struct grouping_m_pg ) !=16 +4 *16 ); +memcpy (p, &gr_m_pg,sizeof (gr_m_pg)); +return sizeof (gr_m_pg); +}static int resp_iec_m_pg (unsigned char *p,int pcontrol,int target) {/* Informational Exceptions control mode page for mode_sense */
@@ -2420,9 +2450,17 @@ static int resp_mode_sense(struct scsi_cmnd *scp, }break ;case 0xa :/* Control Mode page, all devices */ -if (subpcode >0x0 && subpcode <0xff ) +switch (subpcode) { +case 0 : +case 0xff : + len =resp_ctrl_m_pg (ap, pcontrol, target); +break ; +case 0x05 : + len =resp_grouping_m_pg (ap, pcontrol, target); +break ; +default : goto bad_subpcode; - len =resp_ctrl_m_pg (ap, pcontrol, target); + } offset += len;break ;case 0x19 :/* if spc==1 then sas phy, control+discover */
PATCH v3 13/14 scsi_debug: Implement GET STREAM STATUS
In reply to Bart Van AsscheImplement the GET STREAM STATUS SCSI command. Report that the first three stream indexes correspond to permanent streams.
--- drivers/scsi/scsi_debug.c | 44 ++++++++++++++++++++++++++++++++++++++- 1 file changed, 43 insertions(+), 1 deletion(-)
drivers/scsi/scsi_debug.c +43 -1
@@ -481,6 +481,8 @@ static int resp_write_scat(struct scsi_cmnd *, struct sdebug_dev_info *);static int resp_start_stop (struct scsi_cmnd *,struct sdebug_dev_info *);static int resp_readcap16 (struct scsi_cmnd *,struct sdebug_dev_info *);static int resp_get_lba_status (struct scsi_cmnd *,struct sdebug_dev_info *); +static int resp_get_stream_status (struct scsi_cmnd *scp, +struct sdebug_dev_info *devip);static int resp_report_tgtpgs (struct scsi_cmnd *,struct sdebug_dev_info *);static int resp_unmap (struct scsi_cmnd *,struct sdebug_dev_info *);static int resp_rsup_opcodes (struct scsi_cmnd *,struct sdebug_dev_info *);
@@ -555,6 +557,9 @@ static const struct opcode_info_t sa_in_16_iarr[] = { {0 ,0x9e ,0x12 ,F_SA_LOW |F_D_IN , resp_get_lba_status,NULL , {16, 0x12, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0xff, 0, 0xc7} }, /* GET LBA STATUS(16) */ + {0, 0x9e, 0x16, F_SA_LOW | F_D_IN, resp_get_stream_status, NULL, + {16, 0x16, 0, 0, 0xff, 0xff, 0, 0, 0, 0, 0xff, 0xff, 0xff, 0xff, + 0, 0} }, /* GET STREAM STATUS */ }; static const struct opcode_info_t vl_iarr[] = { /* VARIABLE LENGTH */
@@ -2241,7 +2246,7 @@ static int resp_ctrl_m_pg(unsigned char *p, int pcontrol, int target)return sizeof (ctrl_m_pg); } -enum {MAXIMUM_NUMBER_OF_STREAMS =4 }; +enum {MAXIMUM_NUMBER_OF_STREAMS =4 ,PERMANENT_STREAM_COUNT =3 };/* IO Advice Hints Grouping mode page */ static int resp_grouping_m_pg (unsigned char *p,int pcontrol,int target)
@@ -4236,6 +4241,43 @@ static int resp_get_lba_status(struct scsi_cmnd *scp,return fill_from_dev_buffer (scp, arr,SDEBUG_GET_LBA_STATUS_LEN ); } +static int resp_get_stream_status (struct scsi_cmnd *scp, +struct sdebug_dev_info *devip) +{ +u16 starting_stream_id, stream_id; +const u8 *cmd = scp->cmnd; +u32 alloc_len, offset; +u8 arr[256 ]; + + starting_stream_id =get_unaligned_be16 (cmd +4 ); + alloc_len =get_unaligned_be32 (cmd +10 ); + +if (alloc_len <8 ) { +mk_sense_invalid_fld (scp,SDEB_IN_CDB ,10 ,-1 ); +return check_condition_result; + } + +if (starting_stream_id >=MAXIMUM_NUMBER_OF_STREAMS ) { +mk_sense_invalid_fld (scp,SDEB_IN_CDB ,4 ,-1 ); +return check_condition_result; + } + +for (offset =8 , stream_id = starting_stream_id; + offset +8 <=min_t (u32, alloc_len,sizeof (arr)) && + stream_id <MAXIMUM_NUMBER_OF_STREAMS ; + offset +=8 , stream_id++) { +struct scsi_stream_status *stream_status = (void *)arr + offset; + + stream_status->perm = stream_id <PERMANENT_STREAM_COUNT ; +put_unaligned_be16 (stream_id, + &stream_status->stream_identifier); + stream_status->rel_lifetime = stream_id +1 ; + } +put_unaligned_be32 (offset -8 , arr +0 );/* PARAMETER DATA LENGTH */ + +return fill_from_dev_buffer (scp, arr,min (offset, alloc_len)); +} +static int resp_sync_cache (struct scsi_cmnd *scp,struct sdebug_dev_info *devip) {
PATCH v3 14/14 scsi_debug: Maintain write statistics per group number
In reply to Bart Van AsscheTrack per GROUP NUMBER how many write commands have been processed. Make this information available in sysfs. Reset these statistics if any data is written into the sysfs attribute.
Note: SCSI devices should only interpret the information in the GROUP NUMBER field as a stream identifier if the ST_ENBLE bit has been set to one. This patch follows a simpler approach: count the number of writes per GROUP NUMBER whether or not the group number represents a stream identifier.
--- drivers/scsi/scsi_debug.c | 51 ++++++++++++++++++++++++++++++++++++--- 1 file changed, 47 insertions(+), 4 deletions(-)
drivers/scsi/scsi_debug.c +47 -4
@@ -846,6 +846,8 @@ static int sdeb_zbc_nr_conv = DEF_ZBC_NR_CONV_ZONES;static int submit_queues =DEF_SUBMIT_QUEUES ;/* > 1 for multi-queue (mq) */ static int poll_queues;/* iouring iopoll interface.*/ +static atomic_long_t writes_by_group_number[64 ]; +static char sdebug_proc_name[] =MY_NAME ;static const char *my_name =MY_NAME ;
@@ -3040,7 +3042,8 @@ static inline struct sdeb_store_info *devip2sip(struct sdebug_dev_info *devip,/* Returns number of bytes copied or -1 if error. */ static int do_device_access (struct sdeb_store_info *sip,struct scsi_cmnd *scp, -u32 sg_skip,u64 lba,u32 num,bool do_write) +u32 sg_skip,u64 lba,u32 num,bool do_write, +u8 group_number) {int ret;u64 block, rest =0 ;
@@ -3059,6 +3062,10 @@ static int do_device_access(struct sdeb_store_info *sip, struct scsi_cmnd *scp,return 0 ;if (scp->sc_data_direction != dir)return -1 ; + +if (do_write && group_number <ARRAY_SIZE (writes_by_group_number)) +atomic_long_inc (&writes_by_group_number[group_number]); + fsp = sip->storep; block =do_div (lba, sdebug_store_sectors);
@@ -3432,7 +3439,7 @@ static int resp_read_dt0(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) } } - ret =do_device_access (sip, scp,0 , lba, num, false); + ret =do_device_access (sip, scp,0 , lba, num, false,0 );sdeb_read_unlock (sip);if (unlikely (ret ==-1 ))return DID_ERROR <<16 ;
@@ -3617,6 +3624,7 @@ static int resp_write_dt0(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) {bool check_prot;u32 num; +u8 group =0 ;u32 ei_lba;int ret;u64 lba;
@@ -3628,11 +3636,13 @@ static int resp_write_dt0(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) ei_lba =0 ; lba =get_unaligned_be64 (cmd +2 ); num =get_unaligned_be32 (cmd +10 ); + group = cmd[14 ] &0x3f ; check_prot = true;break ;case WRITE_10 : ei_lba =0 ; lba =get_unaligned_be32 (cmd +2 ); + group = cmd[6 ] &0x3f ; num =get_unaligned_be16 (cmd +7 ); check_prot = true;break ;
@@ -3647,15 +3657,18 @@ static int resp_write_dt0(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) ei_lba =0 ; lba =get_unaligned_be32 (cmd +2 ); num =get_unaligned_be32 (cmd +6 ); + group = cmd[6 ] &0x3f ; check_prot = true;break ;case 0x53 :/* XDWRITEREAD(10) */ ei_lba =0 ; lba =get_unaligned_be32 (cmd +2 ); + group = cmd[6 ] &0x1f ; num =get_unaligned_be16 (cmd +7 ); check_prot = false;break ;default :/* assume WRITE(32) */ + group = cmd[6 ] &0x3f ; lba =get_unaligned_be64 (cmd +12 ); ei_lba =get_unaligned_be32 (cmd +20 ); num =get_unaligned_be32 (cmd +28 );
@@ -3710,7 +3723,7 @@ static int resp_write_dt0(struct scsi_cmnd *scp, struct sdebug_dev_info *devip) } } - ret =do_device_access (sip, scp,0 , lba, num, true); + ret =do_device_access (sip, scp,0 , lba, num, true, group);if (unlikely (scsi_debug_lbp ()))map_region (sip, lba, num);/* If ZBC zone then bump its write pointer */
@@ -3762,12 +3775,14 @@ static int resp_write_scat(struct scsi_cmnd *scp,u32 lb_size = sdebug_sector_size;u32 ei_lba;u64 lba; +u8 group;int ret, res;bool is_16;static const u32 lrd_size =32 ;/* + parameter list header size */ if (cmd[0 ] ==VARIABLE_LENGTH_CMD ) { is_16 = false; + group = cmd[6 ] &0x3f ; wrprotect = (cmd[10 ] >>5 ) &0x7 ; lbdof =get_unaligned_be16 (cmd +12 ); num_lrd =get_unaligned_be16 (cmd +16 );
@@ -3778,6 +3793,7 @@ static int resp_write_scat(struct scsi_cmnd *scp, lbdof =get_unaligned_be16 (cmd +4 ); num_lrd =get_unaligned_be16 (cmd +8 ); bt_len =get_unaligned_be32 (cmd +10 ); + group = cmd[14 ] &0x3f ;if (unlikely (have_dif_prot)) {if (sdebug_dif ==T10_PI_TYPE2_PROTECTION && wrprotect) {
@@ -3866,7 +3882,8 @@ static int resp_write_scat(struct scsi_cmnd *scp, } } - ret =do_device_access (sip, scp, sg_off, lba, num, true); + ret =do_device_access (sip, scp, sg_off, lba, num, true, + group);/* If ZBC zone then bump its write pointer */ if (sdebug_dev_is_zoned (devip))zbc_inc_wp (devip, lba, num);
@@ -6828,6 +6845,31 @@ static ssize_t tur_ms_to_ready_show(struct device_driver *ddp, char *buf) }static DRIVER_ATTR_RO (tur_ms_to_ready); +static ssize_t group_number_stats_show (struct device_driver *ddp,char *buf) +{ +char *p = buf, *end = buf +PAGE_SIZE ; +int i; + +for (i =0 ; i <ARRAY_SIZE (writes_by_group_number); i++) + p +=scnprintf (p, end - p,"%d %ld\n" , i, +atomic_long_read (&writes_by_group_number[i])); + +return p - buf; +} + +static ssize_t group_number_stats_store (struct device_driver *ddp, +const char *buf, +size_t count) +{ +int i; + +for (i =0 ; i <ARRAY_SIZE (writes_by_group_number); i++) +atomic_long_set (&writes_by_group_number[i],0 ); + +return 0 ; +} +static DRIVER_ATTR_RW (group_number_stats); +/* Note: The following array creates attribute files in the /sys/bus/pseudo/drivers/scsi_debug directory. The advantage of these files (over those found in the /sys/module/scsi_debug/parameters
@@ -6874,6 +6916,7 @@ static struct attribute *sdebug_drv_attrs[] = { &driver_attr_cdb_len.attr, &driver_attr_tur_ms_to_ready.attr, &driver_attr_zbc.attr, + &driver_attr_group_number_stats.attr,NULL , };ATTRIBUTE_GROUPS (sdebug_drv);
Jens Axboe
maintainerQuoted 10 lines · Bart Van Assche
Hi Jens,
UFS vendors need the data lifetime information to achieve good performance. Without this information there is significantly higher write amplification due to garbage collection. Hence this patch series that add support in F2FS and also in the block layer for data lifetime information. The SCSI disk (sd) driver is modified such that it passes write hint information to SCSI devices via the GROUP NUMBER field.
Please consider this patch series for the next merge window.
My main hesitation with this is that there's a big gap between what makes theoretical sense and practical sense. When we previously tried this, turns out devices retained the data temperature on media, as expected, but tossed it out when data was GC'ed. That made it more of a benchmarking case than anything else. How do we know that things are better now? In previous postings I've seen you point at some papers, but I'm mostly concerned with practical use cases and devices. Are there any results, at all, from that? Or is this a case of vendors asking for something to check some marketing boxes or have value add?
I can take a closer look once this is fully understood. Not adding something like this without proper justification.
I'm also really against growing struct bio just for this. Why is patch 2 not just using the ioprio field at least?
Signature · 1 line
Jens Axboe
Quoted 9 lines · Jens Axboe
My main hesitation with this is that there's a big gap between what makes theoretical sense and practical sense. When we previously tried this, turns out devices retained the data temperature on media, as expected, but tossed it out when data was GC'ed. That made it more of a benchmarking case than anything else. How do we know that things are better now? In previous postings I've seen you point at some papers, but I'm mostly concerned with practical use cases and devices. Are there any results, at all, from that? Or is this a case of vendors asking for something to check some marketing boxes or have value add?
Hi Jens,
Multiple UFS vendors made it clear to me that this feature is essential for their UFS devices to perform well. I will reach out to some of these vendors off-list and will ask them to share performance numbers.
A note: persistent stream support is a feature that was only added
recently in the latest SCSI SBC-5 draft. This SCSI specification change allows SCSI device vendors to interpret the GROUP NUMBER field as a data lifetime. UFS device vendors interpret the GROUP NUMBER field as a data lifetime since a long time - long before this was allowed by the SCSI standards. See also the "ContextID" feature in the UFS specification.
That feature is mentioned in every version of the UFS specification I
have access to. The oldest version of the UFS specification I have
access to is version 2.2, published in 2016.
(https://www.jedec.org/system/files/docs/JESD220C-2_2.pdf). This
document is available free of charge after an account has been created on the JEDEC website.
Quoted 2 lines
I'm also really against growing struct bio just for this. Why is patch 2 not just using the ioprio field at least?
Hmm ... shouldn't the bits in the ioprio field in struct bio have the same meaning as in the ioprio fields used in interfaces between user space and the kernel? Damien Le Moal asked me not to use any of the ioprio bits passing data lifetime information from user space to the kernel.
Is it clear that the size of struct bio has not been changed because the new bi_lifetime member fills a hole in struct bio?
Thanks,
Bart.
Quoted 31 lines · Bart Van Assche
Quoted 9 lines · Jens Axboe
My main hesitation with this is that there's a big gap between what makes theoretical sense and practical sense. When we previously tried this, turns out devices retained the data temperature on media, as expected, but tossed it out when data was GC'ed. That made it more of a benchmarking case than anything else. How do we know that things are better now? In previous postings I've seen you point at some papers, but I'm mostly concerned with practical use cases and devices. Are there any results, at all, from that? Or is this a case of vendors asking for something to check some marketing boxes or have value add?
Hi Jens,
Multiple UFS vendors made it clear to me that this feature is essential for their UFS devices to perform well. I will reach out to some of these vendors off-list and will ask them to share performance numbers.
A note: persistent stream support is a feature that was only added recently in the latest SCSI SBC-5 draft. This SCSI specification change allows SCSI device vendors to interpret the GROUP NUMBER field as a data lifetime. UFS device vendors interpret the GROUP NUMBER field as a data lifetime since a long time - long before this was allowed by the SCSI standards. See also the "ContextID" feature in the UFS specification. That feature is mentioned in every version of the UFS specification I have access to. The oldest version of the UFS specification I have access to is version 2.2, published in 2016.
(https://www.jedec.org/system/files/docs/JESD220C-2_2.pdf). This document is available free of charge after an account has been created on the JEDEC website.
Quoted 7 lines
Quoted 2 lines
I'm also really against growing struct bio just for this. Why is patch 2 not just using the ioprio field at least?
Hmm ... shouldn't the bits in the ioprio field in struct bio have the same meaning as in the ioprio fields used in interfaces between user space and the kernel? Damien Le Moal asked me not to use any of the ioprio bits passing data lifetime information from user space to the kernel.
I said so in the context that if lifetime is a per-inode property, then ioprio is the wrong interface since the ioprio API is per process or per IO. There is a mismatch.
One version of your patch series used fnctl() to set the lifetime per inoe, which is fine, and then used the BIO ioprio to pass the lifetime down to the device driver. That is in theory a nice trick, but that creates conflicts with the userspace ioprio API if the user uses that at the same time.
So may be we should change bio ioprio from int to u16 and use the freedup u16 for lifetime. With that, things are cleanly separated without growing struct bio.
Quoted 3 lines
Is it clear that the size of struct bio has not been changed because the new
bi_lifetimemember fills a hole in struct bio?
When the struct is randomized, holes move or disappear. Don't count on that...
Quoted 6 lines
Thanks,
Bart.
Signature · 2 lines
Damien Le Moal Western Digital Research
Quoted 14 lines · Damien Le Moal
Quoted 8 lines · Bart Van Assche
Quoted 2 lines · Jens Axboe
I'm also really against growing struct bio just for this. Why is patch 2 not just using the ioprio field at least?
Hmm ... shouldn't the bits in the ioprio field in struct bio have the same meaning as in the ioprio fields used in interfaces between user space and the kernel? Damien Le Moal asked me not to use any of the ioprio bits passing data lifetime information from user space to the kernel.
I said so in the context that if lifetime is a per-inode property, then ioprio is the wrong interface since the ioprio API is per process or per IO. There is a mismatch.
Quoted 7 lines
One version of your patch series used
fnctl()to set the lifetime per inode, which is fine, and then used the BIO ioprio to pass the lifetime down to the device driver. That is in theory a nice trick, but that creates conflicts with the userspace ioprio API if the user uses that at the same time.So may be we should change bio ioprio from int to u16 and use the freedup u16 for lifetime. With that, things are cleanly separated without growing struct bio.
Hmm ... I think that bi_ioprio has been 16 bits wide since the introduction of that data structure member in 2016?
Quoted 4 lines
Quoted 2 lines
Is it clear that the size of struct bio has not been changed because the new
bi_lifetimemember fills a hole in struct bio?When the struct is randomized, holes move or disappear. Don't count on that...
We should aim to maximize performance for users who do not use data structure layout randomization.
Additionally, I doubt that anyone is using full structure layout randomization for SCSI devices. No SCSI driver has any __no_randomize_layout / __randomize_layout annotations although I'm sure there are plenty of data structures in SCSI drivers for which the layout matters.
Thanks,
Bart.
Quoted 20 lines · Bart Van Assche
Quoted 19 lines · Damien Le Moal
Quoted 8 lines · Bart Van Assche
Quoted 2 lines · Jens Axboe
I'm also really against growing struct bio just for this. Why is patch 2 not just using the ioprio field at least?
Hmm ... shouldn't the bits in the ioprio field in struct bio have the same meaning as in the ioprio fields used in interfaces between user space and the kernel? Damien Le Moal asked me not to use any of the ioprio bits passing data lifetime information from user space to the kernel.
I said so in the context that if lifetime is a per-inode property, then ioprio is the wrong interface since the ioprio API is per process or per IO. There is a mismatch.
One version of your patch series used
fnctl()to set the lifetime per inode, which is fine, and then used the BIO ioprio to pass the lifetime down to the device driver. That is in theory a nice trick, but that creates conflicts with the userspace ioprio API if the user uses that at the same time.
Quoted 5 lines
Quoted 2 lines
So may be we should change bio ioprio from int to u16 and use the freedup u16 for lifetime. With that, things are cleanly separated without growing struct bio.
Hmm ... I think that
bi_iopriohas been 16 bits wide since the introduction of that data structure member in 2016?
My bad. struct bio->bi_ioprio is an unsigned short. I got confused with the user API and kernel functions using an int in many places. We really should change the kernel functions to use unsigned short for ioprio everywhere.
Quoted 5 lines
Quoted 4 lines
Quoted 2 lines
Is it clear that the size of struct bio has not been changed because the new
bi_lifetimemember fills a hole in struct bio?When the struct is randomized, holes move or disappear. Don't count on that...
Quoted 8 lines
We should aim to maximize performance for users who do not use data structure layout randomization.
Additionally, I doubt that anyone is using full structure layout randomization for SCSI devices. No SCSI driver has any
__no_randomize_layout/__randomize_layoutannotations although I'm sure there are plenty of data structures in SCSI drivers for which the layout matters.
Well, if Jens is OK with adding another "unsigned short bi_lifetime" in a hole in struct bio, that's fine with me. Otherwise, we are back to discussing how to pack bi_ioprio in a sensible manner so that we do not create a mess between the use cases and APIs:
1) inode based lifetime with FS setting up the bi_ioprio field
2) Direct IOs to files of an FS with lifetime set by user per IO (e.g. aio/io_uring/ioprio_set()) and/or fcntl()
3) Direct IOs to raw block devices with lifetime set by user per IO (e.g. aio/io_uring/ioprio_set())
Any of the above case should also allow using ioprio class/level and CDL hint.
I think the most problematic part is (2) when lifetime are set with both fcntl() and per IO: which lifetime is the valid one ? The one set with fcntl() or the one specified for the IO ? I think the former is the one we want here.
If we can clarify that, then I guess using 3 or 4 bits from the 10 bits ioprio hint should be OK. That would give you 7 or 15 lifetime values. Enough no ?
Signature · 2 lines
Damien Le Moal Western Digital Research
Quoted 55 lines · Damien Le Moal
Quoted 20 lines · Bart Van Assche
Quoted 19 lines · Damien Le Moal
Quoted 8 lines · Bart Van Assche
Quoted 2 lines · Jens Axboe
I'm also really against growing struct bio just for this. Why is patch 2 not just using the ioprio field at least?
Hmm ... shouldn't the bits in the ioprio field in struct bio have the same meaning as in the ioprio fields used in interfaces between user space and the kernel? Damien Le Moal asked me not to use any of the ioprio bits passing data lifetime information from user space to the kernel.
I said so in the context that if lifetime is a per-inode property, then ioprio is the wrong interface since the ioprio API is per process or per IO. There is a mismatch.
One version of your patch series used
fnctl()to set the lifetime per inode, which is fine, and then used the BIO ioprio to pass the lifetime down to the device driver. That is in theory a nice trick, but that creates conflicts with the userspace ioprio API if the user uses that at the same time.Quoted 5 lines
Quoted 2 lines
So may be we should change bio ioprio from int to u16 and use the freedup u16 for lifetime. With that, things are cleanly separated without growing struct bio.
Hmm ... I think that
bi_iopriohas been 16 bits wide since the introduction of that data structure member in 2016?My bad. struct
bio->bi_iopriois an unsigned short. I got confused with the user API and kernel functions using an int in many places. We really should change the kernel functions to use unsigned short for ioprio everywhere.Quoted 5 lines
Quoted 4 lines
Quoted 2 lines
Is it clear that the size of struct bio has not been changed because the new
bi_lifetimemember fills a hole in struct bio?When the struct is randomized, holes move or disappear. Don't count on that...
Quoted 8 lines
We should aim to maximize performance for users who do not use data structure layout randomization.
Additionally, I doubt that anyone is using full structure layout randomization for SCSI devices. No SCSI driver has any
__no_randomize_layout/__randomize_layoutannotations although I'm sure there are plenty of data structures in SCSI drivers for which the layout matters.Well, if Jens is OK with adding another "unsigned short
bi_lifetime" in a hole in struct bio, that's fine with me. Otherwise, we are back to discussing how to packbi_ioprioin a sensible manner so that we do not create a mess between the use cases and APIs:
1) inode based lifetime with FS setting up thebi_iopriofield
2) Direct IOs to files of an FS with lifetime set by user per IO (e.g. aio/io_uring/ioprio_set()) and/orfcntl()
3) Direct IOs to raw block devices with lifetime set by user per IO (e.g. aio/io_uring/ioprio_set())
Quoted 8 lines
Any of the above case should also allow using ioprio class/level and CDL hint.
I think the most problematic part is (2) when lifetime are set with both
fcntl()and per IO: which lifetime is the valid one ? The one set withfcntl()or the one specified for the IO ? I think the former is the one we want here.If we can clarify that, then I guess using 3 or 4 bits from the 10 bits ioprio hint should be OK. That would give you 7 or 15 lifetime values. Enough no ?
To be clear, we have to deal with these cases:
1) File IOs
- User uses fcntl() only for lifetime
- User uses per direct IO ioprio with lifetime (and maybe class/level/cdl)
- User uses all of the above
2) Raw block device direct IOs
- Per IO ioprio with lifetime (and maybe class/level/cdl)
(2) is easy. No real change needed beside the UFS driver bits. But the cases for (1) need clarification about how things should work.
Signature · 2 lines
Damien Le Moal Western Digital Research
Quoted 2 lines · Jens Axboe
I'm also really against growing struct bio just for this. Why is patch 2 not just using the ioprio field at least?
Hi Jens,
Can you please clarify whether your concern is about the size of struct bio only or also about the runtime impact of the comparisons that have been added in attempt_merge() and blk_rq_merge_ok()? It may be possible to eliminate the overhead of the new comparisons as follows:
* Introduce a union of struct { I/O priority; data lifetime; } and u32.
* Use that union in struct bio instead of bi_ioprio and bi_lifetime.
* Use that union in struct request instead of the ioprio and lifetime
members.
* In attempt_merge() and blk_rq_merge_ok(), compare the u32 union member
instead of comparing the I/O priority and data lifetime separately.
Thanks,
Bart.