Skip to contentSkip to conversation

blk-mq: misc changes

PATCH In review Ming Lei 5 participants 13 messages Apr 17, 2014 – Apr 19, 2014 Block layer

lore
Share & tools
lore

Ming Lei

author
#0 lore raw

cover blk-mq: misc changes

Hi,

Three are fixes, and another one is cleanup.

Thanks,
--
Ming Lei

author
#1 lore raw

PATCH 1/4 blk-mq: free hctx->ctx_map when init failed

Avoid memory leak in the failure path.

---
 block/blk-mq.c |    1 +
 1 file changed, 1 insertion(+)
block/blk-mq.c +1 -0
index ee225cc..5fbbb22 100644
@@ -1289,6 +1289,7 @@ static int blk_mq_init_hw_queues(struct request_queue *q,
 
 		blk_mq_unregister_cpu_notifier(&hctx->cpu_notifier);
 		kfree(hctx->ctxs);
+		kfree(hctx->ctx_map);
 	}
 
 	return 1;
Signature · 1 line
1.7.9.5
author
#3 lore raw

PATCH 2/4 blk-mq: fix allocation of set->tags

In reply to Ming Lei

type of set->tags is struct blk_mq_tags **.

---
 block/blk-mq.c |    2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
block/blk-mq.c +1 -1
index 5fbbb22..baff2d0 100644
@@ -1537,7 +1537,7 @@ int blk_mq_alloc_tag_set(struct blk_mq_tag_set *set)
 		return -EINVAL;
 
 
-	set->tags = kmalloc_node(set->nr_hw_queues * sizeof(struct blk_mq_tags),
+	set->tags = kmalloc_node(set->nr_hw_queues * sizeof(struct blk_mq_tags *),
 				 GFP_KERNEL, set->numa_node);
 	if (!set->tags)
 		goto out;
Signature · 1 line
1.7.9.5
author
#5 lore raw

PATCH 3/4 blk-mq: user (1 << order) to implement order_to_size()

In reply to Ming Lei
---
 block/blk-mq.c |    7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
block/blk-mq.c +1 -6
index baff2d0..1eff3d9 100644
@@ -1130,12 +1130,7 @@ static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
 
 static size_t order_to_size(unsigned int order)
 {
-	size_t ret = PAGE_SIZE;
-
-	while (order--)
-		ret *= 2;
-
-	return ret;
+	return (1 << order) * PAGE_SIZE;
 }
 
 static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
Signature · 1 line
1.7.9.5
#6 lore raw

Re: [PATCH next 3/4] blk-mq: user (1 << order) to implement order_to_size()

Ming Lei wrote, on 04/17/2014 17:49:

Quoted 10 lines
---
 block/blk-mq.c |    7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
block/blk-mq.c +0 -0
index baff2d0..1eff3d9 100644

block/blk-mq.c static void blk_mq_free_rq_map(struct blk_mq_tag_set *set, · line 1133

Quoted 10 lines
block/blk-mq.c +1 -6
@@ -1130,12 +1130,7 @@ static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
 
 static size_t order_to_size(unsigned int order)
 {
-	size_t ret = PAGE_SIZE;
-
-	while (order--)
-		ret *= 2;
-
-	return ret;
+	return (1 << order) * PAGE_SIZE;

Shouldn't this be

        return ((size_t)1 << order) * PAGE_SIZE;

?

block/blk-mq.c static void blk_mq_free_rq_map(struct blk_mq_tag_set *set, · line 1136

Quoted 4 lines
block/blk-mq.c +0 -0
 }
 
 static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,

Regards,
jvp.

#7 lore raw

Re: [PATCH next 3/4] blk-mq: user (1 << order) to implement order_to_size()

Quoted 24 lines · Jörg-Volker Peetz

Ming Lei wrote, on 04/17/2014 17:49:

Quoted 10 lines
---
 block/blk-mq.c |    7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
block/blk-mq.c +0 -0
index baff2d0..1eff3d9 100644

block/blk-mq.c static void blk_mq_free_rq_map(struct blk_mq_tag_set *set, · line 1133

Quoted 10 lines
block/blk-mq.c +1 -6
@@ -1130,12 +1130,7 @@ static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
 
 static size_t order_to_size(unsigned int order)
 {
-     size_t ret = PAGE_SIZE;
-
-     while (order--)
-             ret *= 2;
-
-     return ret;
+     return (1 << order) * PAGE_SIZE;

Shouldn't this be

Quoted 3 lines
        return ((size_t)1 << order) * PAGE_SIZE;

?

Maybe even

return (size_t)PAGE_SIZE << order;

?

Signature · 2 lines
Thanks.
-- Max
author
#8 lore raw

Re: [PATCH next 3/4] blk-mq: user (1 << order) to implement order_to_size()

In reply to Jörg-Volker Peetz
Quoted 24 lines · Jörg-Volker Peetz

Ming Lei wrote, on 04/17/2014 17:49:

Quoted 10 lines
---
 block/blk-mq.c |    7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
block/blk-mq.c +0 -0
index baff2d0..1eff3d9 100644

block/blk-mq.c static void blk_mq_free_rq_map(struct blk_mq_tag_set *set, · line 1133

Quoted 10 lines
block/blk-mq.c +1 -6
@@ -1130,12 +1130,7 @@ static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
 
 static size_t order_to_size(unsigned int order)
 {
-     size_t ret = PAGE_SIZE;
-
-     while (order--)
-             ret *= 2;
-
-     return ret;
+     return (1 << order) * PAGE_SIZE;

Shouldn't this be

Quoted 3 lines
        return ((size_t)1 << order) * PAGE_SIZE;

?

IMO, it isn't necessary since compiler will cast type of 1 and the expression to size_t automatically.

Thanks,

Signature · 1 line
Ming Lei
#9 lore raw

Re: [PATCH next 3/4] blk-mq: user (1 << order) to implement order_to_size()

Ming Lei wrote, on 04/18/2014 17:47:

Quoted 27 lines · Jörg-Volker Peetz
Quoted 26 lines

Ming Lei wrote, on 04/17/2014 17:49:

Quoted 10 lines
---
 block/blk-mq.c |    7 +------
 1 file changed, 1 insertion(+), 6 deletions(-)
block/blk-mq.c +0 -0
index baff2d0..1eff3d9 100644

block/blk-mq.c static void blk_mq_free_rq_map(struct blk_mq_tag_set *set, · line 1133

Quoted 10 lines
block/blk-mq.c +1 -6
@@ -1130,12 +1130,7 @@ static void blk_mq_free_rq_map(struct blk_mq_tag_set *set,
 
 static size_t order_to_size(unsigned int order)
 {
-     size_t ret = PAGE_SIZE;
-
-     while (order--)
-             ret *= 2;
-
-     return ret;
+     return (1 << order) * PAGE_SIZE;

Shouldn't this be

        return ((size_t)1 << order) * PAGE_SIZE;
Quoted 7 lines
Quoted 1 line

?

IMO, it isn't necessary since compiler will cast type of 1 and the expression to size_t automatically.

Thanks,

Testing this with gcc 4.8.2 with order > 30 confirms that the automatic casting is done operator-wise and the explicit casting is necessary.

Regards,
jvp.

author
#10 lore raw

Re: [PATCH next 3/4] blk-mq: user (1 << order) to implement order_to_size()

Quoted 3 lines · Jörg-Volker Peetz

Testing this with gcc 4.8.2 with order > 30 confirms that the automatic casting is done operator-wise and the explicit casting is necessary.

My gcc 4.8.2 doesn't complain the change, see below:

ming@ming:~/ming/linux$ gcc --version
gcc (Ubuntu/Linaro 4.8.2-19ubuntu1) 4.8.2

ming@ming:~/ming/linux$

ming@ming:~/ming/linux$ vim block/blk-mq.c
ming@ming:~/ming/linux$ ./build
  CHK     include/config/kernel.release
  CHK     include/generated/uapi/linux/version.h
  CHK     include/generated/utsrelease.h
  CALL    scripts/checksyscalls.sh
  CHK     include/generated/compile.h
  CC      block/blk-mq.o
  CHK     kernel/config_data.h
  LD      block/built-in.o
  LINK    vmlinux
  LD      vmlinux.o
  MODPOST vmlinux.o

Thanks,

Signature · 1 line
Ming Lei
maintainer
#11 lore raw

Re: [PATCH next 3/4] blk-mq: user (1 << order) to implement order_to_size()

Quoted 7 lines · Ming Lei
Quoted 5 lines · Jörg-Volker Peetz

Testing this with gcc 4.8.2 with order > 30 confirms that the automatic casting is done operator-wise and the explicit casting is necessary.

My gcc 4.8.2 doesn't complain the change, see below:

It won't complain, but that doesn't mean that it won't get it wrong. The cast is necessary.

author
#12 lore raw

PATCH 4/4 blk-mq: initialize req->q in allocation

In reply to Ming Lei

The patch basically reverts the patch of(blk-mq:
initialize request on allocation) in Jens's tree(already in -next), and only initialize req->q in allocation for two reasons:

- presumed cache hotness on completion
- blk_rq_tagged(rq) depends on reset of req->mq_ctx

---
 block/blk-mq.c |    4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)
block/blk-mq.c +3 -1
index 1eff3d9..b7a51ed 100644
@@ -82,7 +82,6 @@ static struct request *__blk_mq_alloc_request(struct blk_mq_hw_ctx *hctx,
 	tag = blk_mq_get_tag(hctx->tags, gfp, reserved);
 	if (tag != BLK_MQ_TAG_FAIL) {
 		rq = hctx->tags->rqs[tag];
-		blk_rq_init(hctx->queue, rq);
 		rq->tag = tag;
 
 		return rq;
@@ -187,6 +186,7 @@ static void blk_mq_rq_ctx_init(struct request_queue *q, struct blk_mq_ctx *ctx,
 	if (blk_queue_io_stat(q))
 		rw_flags |= REQ_IO_STAT;
 
+	rq->q = q;
 	rq->mq_ctx = ctx;
 	rq->cmd_flags = rw_flags;
 	rq->start_time = jiffies;
@@ -258,6 +258,7 @@ static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 	const int tag = rq->tag;
 	struct request_queue *q = rq->q;
 
+	blk_rq_init(hctx->queue, rq);
 	blk_mq_put_tag(hctx->tags, tag);
 	blk_mq_queue_exit(q);
 }
@@ -1194,6 +1195,7 @@ static struct blk_mq_tags *blk_mq_init_rq_map(struct blk_mq_tag_set *set,
 		left -= to_do * rq_size;
 		for (j = 0; j < to_do; j++) {
 			tags->rqs[i] = p;
+			blk_rq_init(NULL, tags->rqs[i]);
 			if (set->ops->init_request) {
 				if (set->ops->init_request(set->driver_data,
 						tags->rqs[i], hctx_idx, i,
Signature · 1 line
1.7.9.5