Skip to contentSkip to conversation

Cleanup patches

PATCH Applied Damien Le Moal 6 participants 15 messages Jan 29 – Jan 30

Subsystems (3)
lore
Share & tools
lore
#0 lore raw

cover Cleanup patches

Jens,

Here is a couple of cleanup patches for the block layer. No functional changes are introduced by these patches.

Damien Le Moal (2):
block: cleanup queue limit features definition
block: introduce blk_queue_rot()

 block/bfq-iosched.c    | 20 ++++++++++----------
 block/blk-iocost.c     |  2 +-
 block/blk-iolatency.c  |  5 +----
 block/blk-wbt.c        |  5 ++---
 include/linux/blkdev.h | 11 +++++------
 5 files changed, 19 insertions(+), 24 deletions(-)
Signature · 1 line
2.52.0
#1 lore raw

PATCH 1/2 block: cleanup queue limit features definition

Unwrap the definition of BLK_FEAT_ATOMIC_WRITES and
renumber this feature to be sequential with BLK_FEAT_SKIP_TAGSET_QUIESCE.

---
 include/linux/blkdev.h | 7 +++----
 1 file changed, 3 insertions(+), 4 deletions(-)
include/linux/blkdev.h +3 -4
index 251e0f538c4c..4536211ff33c 100644
@@ -340,14 +340,13 @@ typedef unsigned int __bitwise blk_features_t;
 /* skip this queue in blk_mq_(un)quiesce_tagset */
 #define BLK_FEAT_SKIP_TAGSET_QUIESCE	((__force blk_features_t)(1u << 13))
 
+/* atomic writes enabled */
+#define BLK_FEAT_ATOMIC_WRITES		((__force blk_features_t)(1u << 14))
+
 /* undocumented magic for bcache */
 #define BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE \
 	((__force blk_features_t)(1u << 15))
 
-/* atomic writes enabled */
-#define BLK_FEAT_ATOMIC_WRITES \
-	((__force blk_features_t)(1u << 16))
-
 /*
  * Flags automatically inherited when stacking limits.
  */
Signature · 1 line
2.52.0
Reviewed-by on v1 1/2
#3 lore raw

Re: [PATCH 1/2] block: cleanup queue limit features definition

In reply to Damien Le Moal
Quoted 4 lines · Damien Le Moal

Unwrap the definition of BLK_FEAT_ATOMIC_WRITES and
renumber this feature to be sequential with BLK_FEAT_SKIP_TAGSET_QUIESCE.

Quoted 9 lines
---
  include/linux/blkdev.h | 7 +++----
  1 file changed, 3 insertions(+), 4 deletions(-)
include/linux/blkdev.h +0 -0
index 251e0f538c4c..4536211ff33c 100644

include/linux/blkdev.h typedef unsigned int __bitwise blk_features_t; · line 344

Quoted 5 lines
include/linux/blkdev.h +2 -0
@@ -340,14 +340,13 @@ typedef unsigned int __bitwise blk_features_t;
  /* skip this queue in blk_mq_(un)quiesce_tagset */
  #define BLK_FEAT_SKIP_TAGSET_QUIESCE	((__force blk_features_t)(1u << 13))
 
+/* atomic writes enabled */
+#define BLK_FEAT_ATOMIC_WRITES		((__force blk_features_t)(1u << 14))

I think that BLK_FEAT_BOUNCE_HIGH used to be numbered here

Quoted 12 lines
include/linux/blkdev.h +1 -4
+
  /* undocumented magic for bcache */
  #define BLK_FEAT_RAID_PARTIAL_STRIPES_EXPENSIVE \
  	((__force blk_features_t)(1u << 15))
 
-/* atomic writes enabled */
-#define BLK_FEAT_ATOMIC_WRITES \
-	((__force blk_features_t)(1u << 16))
-
  /*
   * Flags automatically inherited when stacking limits.
   */
#5 lore raw

PATCH 2/2 block: introduce blk_queue_rot()

In reply to Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

No functional changes.

---
 block/bfq-iosched.c    | 20 ++++++++++----------
 block/blk-iocost.c     |  2 +-
 block/blk-iolatency.c  |  5 +----
 block/blk-wbt.c        |  5 ++---
 include/linux/blkdev.h |  4 ++--
 5 files changed, 16 insertions(+), 20 deletions(-)
block/bfq-iosched.c +10 -10
index 6e54b1d3d8bc..3ebdec40e758 100644
@@ -231,7 +231,7 @@ static struct kmem_cache *bfq_pool;
 #define BFQ_RQ_SEEKY(bfqd, last_pos, rq) \
 	(get_sdist(last_pos, rq) >			\
 	 BFQQ_SEEK_THR &&				\
-	 (!blk_queue_nonrot(bfqd->queue) ||		\
+	 (blk_queue_rot(bfqd->queue) ||			\
 	  blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT))
 #define BFQQ_CLOSE_THR		(sector_t)(8 * 1024)
 #define BFQQ_SEEKY(bfqq)	(hweight32(bfqq->seek_history) > 19)
@@ -4165,7 +4165,7 @@ static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
 
 	/* don't use too short time intervals */
 	if (delta_usecs < 1000) {
-		if (blk_queue_nonrot(bfqd->queue))
+		if (!blk_queue_rot(bfqd->queue))
 			 /*
 			  * give same worst-case guarantees as idling
 			  * for seeky
@@ -4487,7 +4487,7 @@ static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
 					     struct bfq_queue *bfqq)
 {
 	bool rot_without_queueing =
-		!blk_queue_nonrot(bfqd->queue) && !bfqd->hw_tag,
+		blk_queue_rot(bfqd->queue) && !bfqd->hw_tag,
 		bfqq_sequential_and_IO_bound,
 		idling_boosts_thr;
 
@@ -4521,7 +4521,7 @@ static bool idling_boosts_thr_without_issues(struct bfq_data *bfqd,
 	 * flash-based device.
 	 */
 	idling_boosts_thr = rot_without_queueing ||
-		((!blk_queue_nonrot(bfqd->queue) || !bfqd->hw_tag) &&
+		((blk_queue_rot(bfqd->queue) || !bfqd->hw_tag) &&
 		 bfqq_sequential_and_IO_bound);
 
 	/*
@@ -4722,7 +4722,7 @@ bfq_choose_bfqq_for_injection(struct bfq_data *bfqd)
 			 * there is only one in-flight large request
 			 * at a time.
 			 */
-			if (blk_queue_nonrot(bfqd->queue) &&
+			if (!blk_queue_rot(bfqd->queue) &&
 			    blk_rq_sectors(bfqq->next_rq) >=
 			    BFQQ_SECT_THR_NONROT &&
 			    bfqd->tot_rq_in_driver >= 1)
@@ -6340,7 +6340,7 @@ static void bfq_update_hw_tag(struct bfq_data *bfqd)
 	bfqd->hw_tag_samples = 0;
 
 	bfqd->nonrot_with_queueing =
-		blk_queue_nonrot(bfqd->queue) && bfqd->hw_tag;
+		!blk_queue_rot(bfqd->queue) && bfqd->hw_tag;
 }
 
 static void bfq_completed_request(struct bfq_queue *bfqq, struct bfq_data *bfqd)
@@ -7293,7 +7293,7 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_queue *eq)
 	INIT_HLIST_HEAD(&bfqd->burst_list);
 
 	bfqd->hw_tag = -1;
-	bfqd->nonrot_with_queueing = blk_queue_nonrot(bfqd->queue);
+	bfqd->nonrot_with_queueing = !blk_queue_rot(bfqd->queue);
 
 	bfqd->bfq_max_budget = bfq_default_max_budget;
 
@@ -7328,9 +7328,9 @@ static int bfq_init_queue(struct request_queue *q, struct elevator_queue *eq)
 	 * Begin by assuming, optimistically, that the device peak
 	 * rate is equal to 2/3 of the highest reference rate.
 	 */
-	bfqd->rate_dur_prod = ref_rate[blk_queue_nonrot(bfqd->queue)] *
-		ref_wr_duration[blk_queue_nonrot(bfqd->queue)];
-	bfqd->peak_rate = ref_rate[blk_queue_nonrot(bfqd->queue)] * 2 / 3;
+	bfqd->rate_dur_prod = ref_rate[!blk_queue_rot(bfqd->queue)] *
+		ref_wr_duration[!blk_queue_rot(bfqd->queue)];
+	bfqd->peak_rate = ref_rate[!blk_queue_rot(bfqd->queue)] * 2 / 3;
 
 	/* see comments on the definition of next field inside bfq_data */
 	bfqd->actuator_load_threshold = 4;
block/blk-iocost.c +1 -1
index a0416927d33d..ef543d163d46 100644
@@ -812,7 +812,7 @@ static int ioc_autop_idx(struct ioc *ioc, struct gendisk *disk)
 	u64 now_ns;
 
 	/* rotational? */
-	if (!blk_queue_nonrot(disk->queue))
+	if (blk_queue_rot(disk->queue))
 		return AUTOP_HDD;
 
 	/* handle SATA SSDs w/ broken NCQ */
block/blk-iolatency.c +1 -4
index 45bd18f68541..f7434278cd29 100644
@@ -988,10 +988,7 @@ static void iolatency_pd_init(struct blkg_policy_data *pd)
 	u64 now = blk_time_get_ns();
 	int cpu;
 
-	if (blk_queue_nonrot(blkg->q))
-		iolat->ssd = true;
-	else
-		iolat->ssd = false;
+	iolat->ssd = !blk_queue_rot(blkg->q);
 
 	for_each_possible_cpu(cpu) {
 		struct latency_stat *stat;
block/blk-wbt.c +2 -3
index 0974875f77bd..8e025834f2fb 100644
@@ -747,10 +747,9 @@ u64 wbt_default_latency_nsec(struct request_queue *q)
 	 * We default to 2msec for non-rotational storage, and 75msec
 	 * for rotational storage.
 	 */
-	if (blk_queue_nonrot(q))
-		return 2000000ULL;
-	else
+	if (blk_queue_rot(q))
 		return 75000000ULL;
+	return 2000000ULL;
 }
 
 static int wbt_data_dir(const struct request *rq)
include/linux/blkdev.h +2 -2
index 4536211ff33c..1e5b5547929f 100644
@@ -680,7 +680,7 @@ void blk_queue_flag_clear(unsigned int flag, struct request_queue *q);
 #define blk_queue_nomerges(q)	test_bit(QUEUE_FLAG_NOMERGES, &(q)->queue_flags)
 #define blk_queue_noxmerges(q)	\
 	test_bit(QUEUE_FLAG_NOXMERGES, &(q)->queue_flags)
-#define blk_queue_nonrot(q)	(!((q)->limits.features & BLK_FEAT_ROTATIONAL))
+#define blk_queue_rot(q)	((q)->limits.features & BLK_FEAT_ROTATIONAL)
 #define blk_queue_io_stat(q)	((q)->limits.features & BLK_FEAT_IO_STAT)
 #define blk_queue_passthrough_stat(q)	\
 	((q)->limits.flags & BLK_FLAG_IOSTATS_PASSTHROUGH)
@@ -1463,7 +1463,7 @@ bdev_write_zeroes_unmap_sectors(struct block_device *bdev)
 
 static inline bool bdev_nonrot(struct block_device *bdev)
 {
-	return blk_queue_nonrot(bdev_get_queue(bdev));
+	return !blk_queue_rot(bdev_get_queue(bdev));
 }
 
 static inline bool bdev_synchronous(struct block_device *bdev)
Signature · 1 line
2.52.0
#6 lore raw

Re: [PATCH 2/2] block: introduce blk_queue_rot()

Quoted 32 lines · Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

No functional changes.

---

block/bfq-iosched.c    | 20 ++++++++++----------
block/blk-iocost.c     |  2 +-
block/blk-iolatency.c  |  5 +----
block/blk-wbt.c        |  5 ++---
include/linux/blkdev.h |  4 ++--
5 files changed, 16 insertions(+), 20 deletions(-)
block/bfq-iosched.c +1 -1
index 6e54b1d3d8bc..3ebdec40e758 100644
@@ -231,7 +231,7 @@ static struct kmem_cache *bfq_pool;
 #define BFQ_RQ_SEEKY(bfqd, last_pos, rq) \
 	(get_sdist(last_pos, rq) >			\
 	 BFQQ_SEEK_THR &&				\
-	 (!blk_queue_nonrot(bfqd->queue) ||		\
+	 (blk_queue_rot(bfqd->queue) ||			\
 	  blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT))
 #define BFQQ_CLOSE_THR		(sector_t)(8 * 1024)
 #define BFQQ_SEEKY(bfqq)	(hweight32(bfqq->seek_history) > 19)

block/bfq-iosched.c static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq, · line 4168

Quoted 5 lines
block/bfq-iosched.c +1 -1
@@ -4165,7 +4165,7 @@ static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
 
 	/* don't use too short time intervals */
 	if (delta_usecs < 1000) {
-		if (blk_queue_nonrot(bfqd->queue))
+		if (!blk_queue_rot(bfqd->queue))

blk_queue_rot looks fine to me in other places.
Keeping blk_queue_nonrot would avoid the double negation here and few places. I see you made the changes for bdev_nonrot down, but missed to add here ?

Thanks,
Nitesh

#7 lore raw

Re: [PATCH 2/2] block: introduce blk_queue_rot()

Quoted 34 lines · Nitesh Shetty
Quoted 33 lines · Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

No functional changes.

---

block/bfq-iosched.c    | 20 ++++++++++----------
block/blk-iocost.c     |  2 +-
block/blk-iolatency.c  |  5 +----
block/blk-wbt.c        |  5 ++---
include/linux/blkdev.h |  4 ++--
5 files changed, 16 insertions(+), 20 deletions(-)
block/bfq-iosched.c +1 -1
index 6e54b1d3d8bc..3ebdec40e758 100644
@@ -231,7 +231,7 @@ static struct kmem_cache *bfq_pool;
 #define BFQ_RQ_SEEKY(bfqd, last_pos, rq) \
 	(get_sdist(last_pos, rq) >			\
 	 BFQQ_SEEK_THR &&				\
-	 (!blk_queue_nonrot(bfqd->queue) ||		\
+	 (blk_queue_rot(bfqd->queue) ||			\
 	  blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT))
 #define BFQQ_CLOSE_THR		(sector_t)(8 * 1024)
 #define BFQQ_SEEKY(bfqq)	(hweight32(bfqq->seek_history) > 19)
@@ -4165,7 +4165,7 @@ static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
 
Quoted 6 lines

block/bfq-iosched.c static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq, · line 4168

Quoted 4 lines
block/bfq-iosched.c +1 -1
 	/* don't use too short time intervals */
 	if (delta_usecs < 1000) {
-		if (blk_queue_nonrot(bfqd->queue))
+		if (!blk_queue_rot(bfqd->queue))

blk_queue_rot looks fine to me in other places.
Keeping blk_queue_nonrot would avoid the double negation here and few places.

There is no more double negation. blk_queue_rot() tests for a feature flag being set. The !blk_queue_rot(q) tests the opposite.

Quoted 2 lines

I see you made the changes for bdev_nonrot down, but missed to add here ?

Here where ? What did I miss ?

Signature · 2 lines
Damien Le Moal
Western Digital Research
Reviewed-by
#8 lore raw

Re: [PATCH 2/2] block: introduce blk_queue_rot()

Quoted 42 lines · Damien Le Moal
Quoted 34 lines · Nitesh Shetty
Quoted 33 lines · Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

No functional changes.

---

block/bfq-iosched.c    | 20 ++++++++++----------
block/blk-iocost.c     |  2 +-
block/blk-iolatency.c  |  5 +----
block/blk-wbt.c        |  5 ++---
include/linux/blkdev.h |  4 ++--
5 files changed, 16 insertions(+), 20 deletions(-)
block/bfq-iosched.c +1 -1
index 6e54b1d3d8bc..3ebdec40e758 100644
@@ -231,7 +231,7 @@ static struct kmem_cache *bfq_pool;
 #define BFQ_RQ_SEEKY(bfqd, last_pos, rq) \
 	(get_sdist(last_pos, rq) >			\
 	 BFQQ_SEEK_THR &&				\
-	 (!blk_queue_nonrot(bfqd->queue) ||		\
+	 (blk_queue_rot(bfqd->queue) ||			\
 	  blk_rq_sectors(rq) < BFQQ_SECT_THR_NONROT))
 #define BFQQ_CLOSE_THR		(sector_t)(8 * 1024)
 #define BFQQ_SEEKY(bfqq)	(hweight32(bfqq->seek_history) > 19)
@@ -4165,7 +4165,7 @@ static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq,
 
Quoted 6 lines

block/bfq-iosched.c static bool bfq_bfqq_is_slow(struct bfq_data *bfqd, struct bfq_queue *bfqq, · line 4168

Quoted 4 lines
block/bfq-iosched.c +1 -1
 	/* don't use too short time intervals */
 	if (delta_usecs < 1000) {
-		if (blk_queue_nonrot(bfqd->queue))
+		if (!blk_queue_rot(bfqd->queue))

blk_queue_rot looks fine to me in other places.
Keeping blk_queue_nonrot would avoid the double negation here and few places.

Quoted 8 lines

There is no more double negation. blk_queue_rot() tests for a feature flag being set. The !blk_queue_rot(q) tests the opposite.

Quoted 2 lines

I see you made the changes for bdev_nonrot down, but missed to add here ?

Here where ? What did I miss ?

My bad, I thought you added a new helper for bdev_rot at the end, which check for rotation device.

Reviewed-by on v1 2/2
#9 lore raw

Re: [PATCH 2/2] block: introduce blk_queue_rot()

In reply to Damien Le Moal
Quoted 6 lines · Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

Looks good, and I should have done that soon after inverting the underlying flag:

We should also do the same with bdev_nonrot, maybe staged because of a lot of callers outside the block code.

#10 lore raw

Re: [PATCH 2/2] block: introduce blk_queue_rot()

Quoted 8 lines · Christoph Hellwig
Quoted 6 lines · Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

Quoted 7 lines

Looks good, and I should have done that soon after inverting the underlying flag:

We should also do the same with bdev_nonrot, maybe staged because of a lot of callers outside the block code.

Yes, that is why I did not send a patch for that one as the changes are mostly in file systems. Will send something later.

Signature · 2 lines
Damien Le Moal
Western Digital Research
#11 lore raw

Re: [PATCH 2/2] block: introduce blk_queue_rot()

Quoted 12 lines · Damien Le Moal
Quoted 11 lines · Christoph Hellwig
Quoted 6 lines · Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

Looks good, and I should have done that soon after inverting the underlying flag:

Quoted 7 lines
Quoted 4 lines

We should also do the same with bdev_nonrot, maybe staged because of a lot of callers outside the block code.

Yes, that is why I did not send a patch for that one as the changes are mostly in file systems. Will send something later.

It might be a good idea to just add it and use in "block tree" code, and then we can convert other callers in the next merge window and eventually kill off bdev_nonrot.

#12 lore raw

Re: [PATCH 2/2] block: introduce blk_queue_rot()

Quoted 18 lines · Christoph Hellwig
Quoted 17 lines · Damien Le Moal
Quoted 15 lines · Christoph Hellwig
Quoted 6 lines · Damien Le Moal

To check if a request queue is for a rotational device, a double negation is needed with the pattern "!blk_queue_nonrot(q)". Simplify this with the introduction of the helper blk_queue_rot() which tests if a requests queue limit has the BLK_FEAT_ROTATIONAL feature set. All call sites of blk_queue_nonrot() are modified to use blk_queue_rot() and blk_queue_nonrot() definition removed.

Looks good, and I should have done that soon after inverting the underlying flag:

We should also do the same with bdev_nonrot, maybe staged because of a lot of callers outside the block code.

Quoted 6 lines
Quoted 2 lines

Yes, that is why I did not send a patch for that one as the changes are mostly in file systems. Will send something later.

It might be a good idea to just add it and use in "block tree" code, and then we can convert other callers in the next merge window and eventually kill off bdev_nonrot.

Indeed, that will avoid potential problems with new use in linux-next.

Signature · 2 lines
Damien Le Moal
Western Digital Research
maintainer Applied
#14 lore raw
In reply to Damien Le Moal
Quoted 10 lines · Damien Le Moal

Jens,

Here is a couple of cleanup patches for the block layer. No functional changes are introduced by these patches.

Damien Le Moal (2):
block: cleanup queue limit features definition
block: introduce blk_queue_rot()

[...]

Applied, thanks!

[1/2] block: cleanup queue limit features definition

      commit: 068f5b5ef5bf97e25568950f06ba32325bdc660b
[2/2] block: introduce blk_queue_rot()
      commit: 2719bd1ee1a1cd0535bc62e89b52822f2bbd14eb

Best regards,

Signature · 1 line
Jens Axboe