[PATCHSET 0/4] Last round of alloc side optimizations

[PATCHSET 0/4] Last round of alloc side optimizations

[Jens Axboe]

Hi,

We're doing pretty well on this front now, but there are still a few
tweaks we can make to improve it a bit. Here are 4 minor patches that
do just that.

-- 
Jens Axboe

[PATCH 1/4] block: add rq_flags to struct blk_mq_alloc_data

[Jens Axboe]

There's a hole here we can use, and it's faster to set this earlier
rather than need to check q->elevator multiple times.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c | 20 ++++++++++----------
 block/blk-mq.h |  8 ++++----
 2 files changed, 14 insertions(+), 14 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index d0fe86b46d1b..93b6912768ff 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -305,25 +305,22 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
 	struct blk_mq_ctx *ctx = data->ctx;
 	struct blk_mq_hw_ctx *hctx = data->hctx;
 	struct request_queue *q = data->q;
-	struct elevator_queue *e = q->elevator;
 	struct blk_mq_tags *tags = blk_mq_tags_from_data(data);
 	struct request *rq = tags->static_rqs[tag];
-	unsigned int rq_flags = 0;
 
-	if (e) {
-		rq_flags = RQF_ELV;
-		rq->tag = BLK_MQ_NO_TAG;
-		rq->internal_tag = tag;
-	} else {
+	if (!(data->rq_flags & RQF_ELV)) {
 		rq->tag = tag;
 		rq->internal_tag = BLK_MQ_NO_TAG;
+	} else {
+		rq->tag = BLK_MQ_NO_TAG;
+		rq->internal_tag = tag;
 	}
 
 	if (data->flags & BLK_MQ_REQ_PM)
-		rq_flags |= RQF_PM;
+		data->rq_flags |= RQF_PM;
 	if (blk_queue_io_stat(q))
-		rq_flags |= RQF_IO_STAT;
-	rq->rq_flags = rq_flags;
+		data->rq_flags |= RQF_IO_STAT;
+	rq->rq_flags = data->rq_flags;
 
 	if (blk_mq_need_time_stamp(rq))
 		rq->start_time_ns = ktime_get_ns();
@@ -474,6 +471,7 @@ struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
 		.q		= q,
 		.flags		= flags,
 		.cmd_flags	= op,
+		.rq_flags	= q->elevator ? RQF_ELV : 0,
 		.nr_tags	= 1,
 	};
 	struct request *rq;
@@ -503,6 +501,7 @@ struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
 		.q		= q,
 		.flags		= flags,
 		.cmd_flags	= op,
+		.rq_flags	= q->elevator ? RQF_ELV : 0,
 		.nr_tags	= 1,
 	};
 	u64 alloc_time_ns = 0;
@@ -2500,6 +2499,7 @@ void blk_mq_submit_bio(struct bio *bio)
 			.q		= q,
 			.nr_tags	= 1,
 			.cmd_flags	= bio->bi_opf,
+			.rq_flags	= q->elevator ? RQF_ELV : 0,
 		};
 
 		if (plug) {
diff --git a/block/blk-mq.h b/block/blk-mq.h
index d8ccb341e82e..10be317c1c13 100644
--- a/block/blk-mq.h
+++ b/block/blk-mq.h
@@ -148,6 +148,7 @@ struct blk_mq_alloc_data {
 	blk_mq_req_flags_t flags;
 	unsigned int shallow_depth;
 	unsigned int cmd_flags;
+	unsigned int rq_flags;
 
 	/* allocate multiple requests/tags in one go */
 	unsigned int nr_tags;
@@ -165,10 +166,9 @@ static inline bool blk_mq_is_shared_tags(unsigned int flags)
 
 static inline struct blk_mq_tags *blk_mq_tags_from_data(struct blk_mq_alloc_data *data)
 {
-	if (data->q->elevator)
-		return data->hctx->sched_tags;
-
-	return data->hctx->tags;
+	if (!(data->rq_flags & RQF_ELV))
+		return data->hctx->tags;
+	return data->hctx->sched_tags;
 }
 
 static inline bool blk_mq_hctx_stopped(struct blk_mq_hw_ctx *hctx)
-- 
2.33.1

[PATCH 2/4] block: pass in blk_mq_tags to blk_mq_rq_ctx_init()

[Jens Axboe]

Instead of getting this from data for every invocation of request
initialization, pass it in as an argument instead.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c | 25 ++++++++++++++-----------
 1 file changed, 14 insertions(+), 11 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 93b6912768ff..fbaecb6e6dd4 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -300,12 +300,11 @@ void blk_mq_wake_waiters(struct request_queue *q)
 }
 
 static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
-		unsigned int tag, u64 alloc_time_ns)
+		struct blk_mq_tags *tags, unsigned int tag, u64 alloc_time_ns)
 {
 	struct blk_mq_ctx *ctx = data->ctx;
 	struct blk_mq_hw_ctx *hctx = data->hctx;
 	struct request_queue *q = data->q;
-	struct blk_mq_tags *tags = blk_mq_tags_from_data(data);
 	struct request *rq = tags->static_rqs[tag];
 
 	if (!(data->rq_flags & RQF_ELV)) {
@@ -377,20 +376,22 @@ __blk_mq_alloc_requests_batch(struct blk_mq_alloc_data *data,
 		u64 alloc_time_ns)
 {
 	unsigned int tag, tag_offset;
+	struct blk_mq_tags *tags;
 	struct request *rq;
-	unsigned long tags;
+	unsigned long tag_mask;
 	int i, nr = 0;
 
-	tags = blk_mq_get_tags(data, data->nr_tags, &tag_offset);
-	if (unlikely(!tags))
+	tag_mask = blk_mq_get_tags(data, data->nr_tags, &tag_offset);
+	if (unlikely(!tag_mask))
 		return NULL;
 
-	for (i = 0; tags; i++) {
-		if (!(tags & (1UL << i)))
+	tags = blk_mq_tags_from_data(data);
+	for (i = 0; tag_mask; i++) {
+		if (!(tag_mask & (1UL << i)))
 			continue;
 		tag = tag_offset + i;
-		tags &= ~(1UL << i);
-		rq = blk_mq_rq_ctx_init(data, tag, alloc_time_ns);
+		tag_mask &= ~(1UL << i);
+		rq = blk_mq_rq_ctx_init(data, tags, tag, alloc_time_ns);
 		rq_list_add(data->cached_rq, rq);
 	}
 	data->nr_tags -= nr;
@@ -461,7 +462,8 @@ static struct request *__blk_mq_alloc_requests(struct blk_mq_alloc_data *data)
 		goto retry;
 	}
 
-	return blk_mq_rq_ctx_init(data, tag, alloc_time_ns);
+	return blk_mq_rq_ctx_init(data, blk_mq_tags_from_data(data), tag,
+					alloc_time_ns);
 }
 
 struct request *blk_mq_alloc_request(struct request_queue *q, unsigned int op,
@@ -547,7 +549,8 @@ struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
 	tag = blk_mq_get_tag(&data);
 	if (tag == BLK_MQ_NO_TAG)
 		goto out_queue_exit;
-	return blk_mq_rq_ctx_init(&data, tag, alloc_time_ns);
+	return blk_mq_rq_ctx_init(&data, blk_mq_tags_from_data(&data), tag,
+					alloc_time_ns);
 
 out_queue_exit:
 	blk_queue_exit(q);
-- 
2.33.1

[PATCH 3/4] block: prefetch request to be initialized

[Jens Axboe]

Now we have the tags available in __blk_mq_alloc_requests_batch(), we
can start fetching the first request cacheline before calling into the
request initialization.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c | 1 +
 1 file changed, 1 insertion(+)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index fbaecb6e6dd4..77c2c3220128 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -389,6 +389,7 @@ __blk_mq_alloc_requests_batch(struct blk_mq_alloc_data *data,
 	for (i = 0; tag_mask; i++) {
 		if (!(tag_mask & (1UL << i)))
 			continue;
+		prefetch(tags->static_rqs[tag]);
 		tag = tag_offset + i;
 		tag_mask &= ~(1UL << i);
 		rq = blk_mq_rq_ctx_init(data, tags, tag, alloc_time_ns);
-- 
2.33.1

[PATCH 4/4] block: re-flow blk_mq_rq_ctx_init()

[Jens Axboe]

Now that we have flags passed in, we can do a final re-arrange of the
flow of blk_mq_rq_ctx_init() so we're always writing request in the
order in which it is laid out.

Signed-off-by: Jens Axboe <axboe@kernel.dk>
---
 block/blk-mq.c | 24 ++++++++++++------------
 1 file changed, 12 insertions(+), 12 deletions(-)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 77c2c3220128..4a671d1d47ac 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -307,6 +307,17 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
 	struct request_queue *q = data->q;
 	struct request *rq = tags->static_rqs[tag];
 
+	rq->q = q;
+	rq->mq_ctx = ctx;
+	rq->mq_hctx = hctx;
+	rq->cmd_flags = data->cmd_flags;
+
+	if (data->flags & BLK_MQ_REQ_PM)
+		data->rq_flags |= RQF_PM;
+	if (blk_queue_io_stat(q))
+		data->rq_flags |= RQF_IO_STAT;
+	rq->rq_flags = data->rq_flags;
+
 	if (!(data->rq_flags & RQF_ELV)) {
 		rq->tag = tag;
 		rq->internal_tag = BLK_MQ_NO_TAG;
@@ -314,22 +325,12 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
 		rq->tag = BLK_MQ_NO_TAG;
 		rq->internal_tag = tag;
 	}
-
-	if (data->flags & BLK_MQ_REQ_PM)
-		data->rq_flags |= RQF_PM;
-	if (blk_queue_io_stat(q))
-		data->rq_flags |= RQF_IO_STAT;
-	rq->rq_flags = data->rq_flags;
+	rq->timeout = 0;
 
 	if (blk_mq_need_time_stamp(rq))
 		rq->start_time_ns = ktime_get_ns();
 	else
 		rq->start_time_ns = 0;
-	/* csd/requeue_work/fifo_time is initialized before use */
-	rq->q = q;
-	rq->mq_ctx = ctx;
-	rq->mq_hctx = hctx;
-	rq->cmd_flags = data->cmd_flags;
 	rq->rq_disk = NULL;
 	rq->part = NULL;
 #ifdef CONFIG_BLK_RQ_ALLOC_TIME
@@ -341,7 +342,6 @@ static struct request *blk_mq_rq_ctx_init(struct blk_mq_alloc_data *data,
 #if defined(CONFIG_BLK_DEV_INTEGRITY)
 	rq->nr_integrity_segments = 0;
 #endif
-	rq->timeout = 0;
 	rq->end_io = NULL;
 	rq->end_io_data = NULL;
 
-- 
2.33.1

Re: [PATCH 1/4] block: add rq_flags to struct blk_mq_alloc_data

[Keith Busch]

On Tue, Oct 19, 2021 at 09:32:57AM -0600, Jens Axboe wrote:
> There's a hole here we can use, and it's faster to set this earlier

<snip>

> @@ -148,6 +148,7 @@ struct blk_mq_alloc_data {
>  	blk_mq_req_flags_t flags;
>  	unsigned int shallow_depth;
>  	unsigned int cmd_flags;
> +	unsigned int rq_flags;
>  
>  	/* allocate multiple requests/tags in one go */
>  	unsigned int nr_tags;

The patch looks good, but the new field doesn't appear to occupy a hole.

Re: [PATCH 1/4] block: add rq_flags to struct blk_mq_alloc_data

[Jens Axboe]

On 10/19/21 12:42 PM, Keith Busch wrote:
> On Tue, Oct 19, 2021 at 09:32:57AM -0600, Jens Axboe wrote:
>> There's a hole here we can use, and it's faster to set this earlier
> 
> <snip>
> 
>> @@ -148,6 +148,7 @@ struct blk_mq_alloc_data {
>>  	blk_mq_req_flags_t flags;
>>  	unsigned int shallow_depth;
>>  	unsigned int cmd_flags;
>> +	unsigned int rq_flags;
>>  
>>  	/* allocate multiple requests/tags in one go */
>>  	unsigned int nr_tags;
> 
> The patch looks good, but the new field doesn't appear to occupy a hole.

Yes, not sure how I missed the first flags. I'll cut that from the commit
message.

-- 
Jens Axboe

Re: [PATCH 1/4] block: add rq_flags to struct blk_mq_alloc_data

[Guenter Roeck]

Hi,

On Tue, Oct 19, 2021 at 09:32:57AM -0600, Jens Axboe wrote:
> There's a hole here we can use, and it's faster to set this earlier
> rather than need to check q->elevator multiple times.
> 
> Signed-off-by: Jens Axboe <axboe@kernel.dk>

This patch results in a warning backtrace with one of my qemu
boot tests.

[   48.268501] ------------[ cut here ]------------
[   48.268620] WARNING: CPU: 0 PID: 1 at block/blk-mq-sched.c:432 blk_mq_sched_insert_request+0x3e/0x150
[   48.268771] Modules linked in:
[   48.268973] CPU: 0 PID: 1 Comm: swapper/0 Not tainted 5.15.0+ #1
[   48.269198] Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS rel-1.14.0-0-g155821a1990b-prebuilt.qemu.org 04/01/2014
[   48.269419] RIP: 0010:blk_mq_sched_insert_request+0x3e/0x150
[   48.269613] Code: ec 20 4c 8b 47 08 4c 8b 77 10 65 48 8b 04 25 28 00 00 00 48 89 44 24 18 48 8b 07 4c 8b 78 08 4d 85 ff 74 08 83 7f 20 ff 74 02 <0f> 0b f6 45 1c 10 0f 85 a6 00 00 00 0f b6 45 18 44 0f b6 e6 83 e8
[   48.269851] RSP: 0018:ffff991500013a00 EFLAGS: 00000217
[   48.269955] RAX: ffff88e4c36e8ce8 RBX: 0000000000000001 RCX: 0000000000000000
[   48.270131] RDX: 0000000000000001 RSI: 0000000000000001 RDI: ffff88e4c369d340
[   48.270232] RBP: ffff88e4c369d340 R08: ffffb914ffc02f80 R09: 0000000000000000
[   48.270328] R10: ffff88e4c2d419e0 R11: 0000000000000001 R12: 0000000000000001
[   48.270423] R13: 0000000000000000 R14: ffff88e4c36e4c00 R15: ffff88e4c3818400
[   48.270566] FS:  0000000000000000(0000) GS:ffff88e4fbc00000(0000) knlGS:0000000000000000
[   48.270679] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   48.270767] CR2: 0000000000000000 CR3: 0000000116422000 CR4: 00000000003506f0
[   48.270926] Call Trace:
[   48.271310]  ? lockdep_init_map_type+0x47/0x240
[   48.271469]  blk_execute_rq+0x5c/0xe0
[   48.271648]  __scsi_execute+0x158/0x2b0
[   48.271744]  __scsi_scan_target+0x2e1/0x4d0
[   48.271897]  scsi_scan_target+0xd4/0xf0
[   48.271990]  sas_rphy_add+0x123/0x180
[   48.272170]  mptsas_add_end_device.isra.0.cold+0x89/0xc8
[   48.272267]  ? mptsas_refreshing_device_handles+0x70/0xd0
[   48.272364]  mptsas_scan_sas_topology+0x147/0x290
[   48.272441]  ? _raw_spin_unlock_irqrestore+0x3d/0x70
[   48.272571]  mptsas_probe+0x3b2/0x470
[   48.272656]  local_pci_probe+0x3d/0x70
[   48.272739]  pci_device_probe+0xc6/0x150
[   48.272835]  really_probe.part.0+0xa0/0x290
[   48.272918]  __driver_probe_device+0x8b/0x120
[   48.272998]  driver_probe_device+0x19/0x90
[   48.273147]  __driver_attach+0x79/0x120
[   48.273212]  ? __device_attach_driver+0x90/0x90
[   48.273289]  bus_for_each_dev+0x76/0xc0
[   48.273376]  bus_add_driver+0x109/0x1b0
[   48.273455]  driver_register+0x8a/0xe0
[   48.273515]  ? fusion_init+0x103/0x103
[   48.273583]  mptsas_init+0x10f/0x128
[   48.273649]  do_one_initcall+0x56/0x2e0
[   48.273741]  ? rcu_read_lock_sched_held+0x3a/0x70
[   48.273828]  kernel_init_freeable+0x24c/0x299
[   48.273898]  ? rest_init+0x250/0x250
[   48.273969]  kernel_init+0x11/0x110
[   48.274099]  ret_from_fork+0x22/0x30
[   48.274304] irq event stamp: 10417319
[   48.274371] hardirqs last  enabled at (10417327): [<ffffffffa6ae0f22>] __up_console_sem+0x62/0x70
[   48.274523] hardirqs last disabled at (10417334): [<ffffffffa6ae0f07>] __up_console_sem+0x47/0x70
[   48.274637] softirqs last  enabled at (10416158): [<ffffffffa6a73db8>] irq_exit_rcu+0xd8/0x120
[   48.274756] softirqs last disabled at (10416145): [<ffffffffa6a73db8>] irq_exit_rcu+0xd8/0x120
[   48.274937] ---[ end trace 3376dfd19d6a9ecf ]---

Bisect log is attached.

Guenter

---
# bad: [dcd68326d29b62f3039e4f4d23d3e38f24d37360] Merge tag 'devicetree-for-5.16' of git://git.kernel.org/pub/scm/linux/kernel/git/robh/linux
# good: [8bb7eca972ad531c9b149c0a51ab43a417385813] Linux 5.15
git bisect start 'HEAD' 'v5.15'
# good: [84882cf72cd774cf16fd338bdbf00f69ac9f9194] Revert "net: avoid double accounting for pure zerocopy skbs"
git bisect good 84882cf72cd774cf16fd338bdbf00f69ac9f9194
# bad: [d2fac0afe89fe30c39eaa98dda71f7c4cea190c2] Merge tag 'audit-pr-20211101' of git://git.kernel.org/pub/scm/linux/kernel/git/pcmoore/audit
git bisect bad d2fac0afe89fe30c39eaa98dda71f7c4cea190c2
# bad: [8cb1ae19bfae92def42c985417cd6e894ddaa047] Merge tag 'x86-fpu-2021-11-01' of git://git.kernel.org/pub/scm/linux/kernel/git/tip/tip
git bisect bad 8cb1ae19bfae92def42c985417cd6e894ddaa047
# bad: [71ae42629e65edab618651c8ff9c88e1edd717aa] Merge tag 'for-5.16/passthrough-flag-2021-10-29' of git://git.kernel.dk/linux-block
git bisect bad 71ae42629e65edab618651c8ff9c88e1edd717aa
# bad: [33c8846c814c1c27c6e33af005042d15061f948b] Merge tag 'for-5.16/block-2021-10-29' of git://git.kernel.dk/linux-block
git bisect bad 33c8846c814c1c27c6e33af005042d15061f948b
# good: [057178cf518e699695a4b614a7a08c350b1fdcfd] fs: bdev: fix conflicting comment from lookup_bdev
git bisect good 057178cf518e699695a4b614a7a08c350b1fdcfd
# good: [203a31516616111b8eaaf00c16fa3fcaa25e6f81] mm/writeback: Add __folio_mark_dirty()
git bisect good 203a31516616111b8eaaf00c16fa3fcaa25e6f81
# bad: [02f7eab0095a47b45f48a4321d33de3569c59061] block: improve readability of blk_mq_end_request_batch()
git bisect bad 02f7eab0095a47b45f48a4321d33de3569c59061
# good: [54a88eb838d37af930c9f19e1930a4fba6789cb5] block: add single bio async direct IO helper
git bisect good 54a88eb838d37af930c9f19e1930a4fba6789cb5
# good: [842e39b013465a279fb60348427b9309427a29de] block: add async version of bio_set_polled
git bisect good 842e39b013465a279fb60348427b9309427a29de
# bad: [c7b84d4226adaa601e9f73574ef123d1500cf712] block: re-flow blk_mq_rq_ctx_init()
git bisect bad c7b84d4226adaa601e9f73574ef123d1500cf712
# bad: [fe6134f66906dfa16d4877cab60106275f48eef7] block: pass in blk_mq_tags to blk_mq_rq_ctx_init()
git bisect bad fe6134f66906dfa16d4877cab60106275f48eef7
# bad: [56f8da642bd827ef50a952e7bc3728c5830452be] block: add rq_flags to struct blk_mq_alloc_data
git bisect bad 56f8da642bd827ef50a952e7bc3728c5830452be
# first bad commit: [56f8da642bd827ef50a952e7bc3728c5830452be] block: add rq_flags to struct blk_mq_alloc_data

Re: [PATCH 1/4] block: add rq_flags to struct blk_mq_alloc_data

[Jens Axboe]

On 11/3/21 1:54 PM, Guenter Roeck wrote:
> Hi,
> 
> On Tue, Oct 19, 2021 at 09:32:57AM -0600, Jens Axboe wrote:
>> There's a hole here we can use, and it's faster to set this earlier
>> rather than need to check q->elevator multiple times.
>>
>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> 
> This patch results in a warning backtrace with one of my qemu
> boot tests.

Should be fixed in the current tree, will go out soonish. If you
have time, can you pull in:

git://git.kernel.dk/linux-block for-next

into -git and see if it fixes it for you?

-- 
Jens Axboe

Re: [PATCH 1/4] block: add rq_flags to struct blk_mq_alloc_data

[Guenter Roeck]

On Wed, Nov 03, 2021 at 01:56:29PM -0600, Jens Axboe wrote:
> On 11/3/21 1:54 PM, Guenter Roeck wrote:
> > Hi,
> > 
> > On Tue, Oct 19, 2021 at 09:32:57AM -0600, Jens Axboe wrote:
> >> There's a hole here we can use, and it's faster to set this earlier
> >> rather than need to check q->elevator multiple times.
> >>
> >> Signed-off-by: Jens Axboe <axboe@kernel.dk>
> > 
> > This patch results in a warning backtrace with one of my qemu
> > boot tests.
> 
> Should be fixed in the current tree, will go out soonish. If you
> have time, can you pull in:
> 
> git://git.kernel.dk/linux-block for-next
> 
> into -git and see if it fixes it for you?
>

Yes, it looks like the problem has been fixed in your for-next branch.

Thanks,
Guenter

Re: [PATCH 1/4] block: add rq_flags to struct blk_mq_alloc_data

[Jens Axboe]

On 11/3/21 3:41 PM, Guenter Roeck wrote:
> On Wed, Nov 03, 2021 at 01:56:29PM -0600, Jens Axboe wrote:
>> On 11/3/21 1:54 PM, Guenter Roeck wrote:
>>> Hi,
>>>
>>> On Tue, Oct 19, 2021 at 09:32:57AM -0600, Jens Axboe wrote:
>>>> There's a hole here we can use, and it's faster to set this earlier
>>>> rather than need to check q->elevator multiple times.
>>>>
>>>> Signed-off-by: Jens Axboe <axboe@kernel.dk>
>>>
>>> This patch results in a warning backtrace with one of my qemu
>>> boot tests.
>>
>> Should be fixed in the current tree, will go out soonish. If you
>> have time, can you pull in:
>>
>> git://git.kernel.dk/linux-block for-next
>>
>> into -git and see if it fixes it for you?
>>
> 
> Yes, it looks like the problem has been fixed in your for-next branch.

Great, thanks for testing.

-- 
Jens Axboe