Groups | Search | Server Info | Keyboard shortcuts | Login | Register [http] [https] [nntp] [nntps]

Groups > linux.kernel > #1415464 > unrolled thread

[PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

Back to article view | Back to linux.kernel

This discussion starts older than the indexed window; earlier articles aren't shown. The article labeled Started by below is the oldest one visible, not the original post.

Contents

  [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Christoph Hellwig <hch@lst.de> - 2016-06-06 23:30 +0200
    Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Keith Busch <keith.busch@intel.com> - 2016-06-07 17:00 +0200
      Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Ming Lin <mlin@kernel.org> - 2016-06-07 17:30 +0200
        Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Ming Lin <mlin@kernel.org> - 2016-06-08 01:30 +0200
    Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Jens Axboe <axboe@kernel.dk> - 2016-06-08 06:50 +0200
      Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Ming Lin <mlin@kernel.org> - 2016-06-08 07:30 +0200
        Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Christoph Hellwig <hch@lst.de> - 2016-06-08 14:00 +0200
      Re: [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx Christoph Hellwig <hch@lst.de> - 2016-06-08 14:00 +0200

#1415464 — [PATCH 1/8] blk-mq: add blk_mq_alloc_request_hctx

From: Ming Lin <ming.l@ssi.samsung.com>

For some protocols like NVMe over Fabrics we need to be able to send
initialization commands to a specific queue.

Based on an earlier patch from Christoph Hellwig <hch@lst.de>.

Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/blk-mq.c         | 33 +++++++++++++++++++++++++++++++++
 include/linux/blk-mq.h |  2 ++
 2 files changed, 35 insertions(+)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 29cbc1b..7bb45ed 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -266,6 +266,39 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 }
 EXPORT_SYMBOL(blk_mq_alloc_request);
 
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
+		unsigned int flags, unsigned int hctx_idx)
+{
+	struct blk_mq_hw_ctx *hctx;
+	struct blk_mq_ctx *ctx;
+	struct request *rq;
+	struct blk_mq_alloc_data alloc_data;
+	int ret;
+
+	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
+	if (ret)
+		return ERR_PTR(ret);
+
+	hctx = q->queue_hw_ctx[hctx_idx];
+	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
+
+	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
+
+	rq = __blk_mq_alloc_request(&alloc_data, rw);
+	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
+		__blk_mq_run_hw_queue(hctx);
+
+		rq =  __blk_mq_alloc_request(&alloc_data, rw);
+	}
+	if (!rq) {
+		blk_queue_exit(q);
+		return ERR_PTR(-EWOULDBLOCK);
+	}
+
+	return rq;
+}
+EXPORT_SYMBOL(blk_mq_alloc_request_hctx);
+
 static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 				  struct blk_mq_ctx *ctx, struct request *rq)
 {
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index 2498fdf..6bf8735 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -196,6 +196,8 @@ enum {
 
 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 		unsigned int flags);
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
+		unsigned int flags, unsigned int hctx_idx);
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
 
-- 
2.1.4

[toc] | [next] | [standalone]

#1416281

On Mon, Jun 06, 2016 at 11:21:52PM +0200, Christoph Hellwig wrote:
> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
> +		unsigned int flags, unsigned int hctx_idx)
> +{
> +	struct blk_mq_hw_ctx *hctx;
> +	struct blk_mq_ctx *ctx;
> +	struct request *rq;
> +	struct blk_mq_alloc_data alloc_data;
> +	int ret;
> +
> +	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
> +	if (ret)
> +		return ERR_PTR(ret);
> +
> +	hctx = q->queue_hw_ctx[hctx_idx];

We probably want to check 'if (hctx_idx < q->nr_hw_queues)' before
getting the hctx. Even if hctx_idx was origially valid, it's possible
(though unlikely) blk_queue_enter waits on reallocating h/w contexts,
which can make hctx_idx invalid.

[toc] | [prev] | [next] | [standalone]

#1416318

On Tue, Jun 7, 2016 at 7:57 AM, Keith Busch <keith.busch@intel.com> wrote:
> On Mon, Jun 06, 2016 at 11:21:52PM +0200, Christoph Hellwig wrote:
>> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
>> +             unsigned int flags, unsigned int hctx_idx)
>> +{
>> +     struct blk_mq_hw_ctx *hctx;
>> +     struct blk_mq_ctx *ctx;
>> +     struct request *rq;
>> +     struct blk_mq_alloc_data alloc_data;
>> +     int ret;
>> +
>> +     ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
>> +     if (ret)
>> +             return ERR_PTR(ret);
>> +
>> +     hctx = q->queue_hw_ctx[hctx_idx];
>
> We probably want to check 'if (hctx_idx < q->nr_hw_queues)' before
> getting the hctx. Even if hctx_idx was origially valid, it's possible
> (though unlikely) blk_queue_enter waits on reallocating h/w contexts,
> which can make hctx_idx invalid.

Yes, I'll update it.

[toc] | [prev] | [next] | [standalone]

#1416727

On Tue, Jun 7, 2016 at 8:27 AM, Ming Lin <mlin@kernel.org> wrote:
> On Tue, Jun 7, 2016 at 7:57 AM, Keith Busch <keith.busch@intel.com> wrote:
>> On Mon, Jun 06, 2016 at 11:21:52PM +0200, Christoph Hellwig wrote:
>>> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
>>> +             unsigned int flags, unsigned int hctx_idx)
>>> +{
>>> +     struct blk_mq_hw_ctx *hctx;
>>> +     struct blk_mq_ctx *ctx;
>>> +     struct request *rq;
>>> +     struct blk_mq_alloc_data alloc_data;
>>> +     int ret;
>>> +
>>> +     ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
>>> +     if (ret)
>>> +             return ERR_PTR(ret);
>>> +
>>> +     hctx = q->queue_hw_ctx[hctx_idx];
>>
>> We probably want to check 'if (hctx_idx < q->nr_hw_queues)' before
>> getting the hctx. Even if hctx_idx was origially valid, it's possible
>> (though unlikely) blk_queue_enter waits on reallocating h/w contexts,
>> which can make hctx_idx invalid.
>
> Yes, I'll update it.

How about this?

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 7bb45ed..b59d2ef 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -279,6 +279,11 @@ struct request *blk_mq_alloc_request_hctx(struct
request_queue *q, int rw,
        if (ret)
                return ERR_PTR(ret);

+       if (hctx_idx >= q->nr_hw_queues) {
+               blk_queue_exit(q);
+               return ERR_PTR(-EINVAL);
+       }
+
        hctx = q->queue_hw_ctx[hctx_idx];
        ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));

[toc] | [prev] | [next] | [standalone]

#1416862

On 06/06/2016 03:21 PM, Christoph Hellwig wrote:
> From: Ming Lin <ming.l@ssi.samsung.com>
>
> For some protocols like NVMe over Fabrics we need to be able to send
> initialization commands to a specific queue.
>
> Based on an earlier patch from Christoph Hellwig <hch@lst.de>.
>
> Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
> Signed-off-by: Christoph Hellwig <hch@lst.de>
> ---
>   block/blk-mq.c         | 33 +++++++++++++++++++++++++++++++++
>   include/linux/blk-mq.h |  2 ++
>   2 files changed, 35 insertions(+)
>
> diff --git a/block/blk-mq.c b/block/blk-mq.c
> index 29cbc1b..7bb45ed 100644
> --- a/block/blk-mq.c
> +++ b/block/blk-mq.c
> @@ -266,6 +266,39 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
>   }
>   EXPORT_SYMBOL(blk_mq_alloc_request);
>
> +struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
> +		unsigned int flags, unsigned int hctx_idx)
> +{
> +	struct blk_mq_hw_ctx *hctx;
> +	struct blk_mq_ctx *ctx;
> +	struct request *rq;
> +	struct blk_mq_alloc_data alloc_data;
> +	int ret;
> +
> +	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
> +	if (ret)
> +		return ERR_PTR(ret);
> +
> +	hctx = q->queue_hw_ctx[hctx_idx];
> +	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
> +
> +	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
> +
> +	rq = __blk_mq_alloc_request(&alloc_data, rw);
> +	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
> +		__blk_mq_run_hw_queue(hctx);
> +
> +		rq =  __blk_mq_alloc_request(&alloc_data, rw);
> +	}

Why are we duplicating this code here? If NOWAIT isn't set, then we'll
always return a request. bt_get() will run the queue for us, if it needs
to. blk_mq_alloc_request() does this too, and I'm guessing that code was
just copied. I'll fix that up. Looks like this should just be:

	rq = __blk_mq_alloc_request(&alloc_data, rw);
	if (rq)
		return rq;

	blk_queue_exit(q);
	return ERR_PTR(-EWOULDBLOCK);

for this case.

-- 
Jens Axboe

[toc] | [prev] | [next] | [standalone]

#1416876

On Tue, 2016-06-07 at 22:49 -0600, Jens Axboe wrote:
> On 06/06/2016 03:21 PM, Christoph Hellwig wrote:
> > From: Ming Lin <ming.l@ssi.samsung.com>
> > 
> > For some protocols like NVMe over Fabrics we need to be able to
> > send
> > initialization commands to a specific queue.
> > 
> > Based on an earlier patch from Christoph Hellwig <hch@lst.de>.
> > 
> > Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
> > Signed-off-by: Christoph Hellwig <hch@lst.de>
> > ---
> >   block/blk-mq.c         | 33 +++++++++++++++++++++++++++++++++
> >   include/linux/blk-mq.h |  2 ++
> >   2 files changed, 35 insertions(+)
> > 
> > diff --git a/block/blk-mq.c b/block/blk-mq.c
> > index 29cbc1b..7bb45ed 100644
> > --- a/block/blk-mq.c
> > +++ b/block/blk-mq.c
> > @@ -266,6 +266,39 @@ struct request *blk_mq_alloc_request(struct
> > request_queue *q, int rw,
> >   }
> >   EXPORT_SYMBOL(blk_mq_alloc_request);
> > 
> > +struct request *blk_mq_alloc_request_hctx(struct request_queue *q,
> > int rw,
> > +		unsigned int flags, unsigned int hctx_idx)
> > +{
> > +	struct blk_mq_hw_ctx *hctx;
> > +	struct blk_mq_ctx *ctx;
> > +	struct request *rq;
> > +	struct blk_mq_alloc_data alloc_data;
> > +	int ret;
> > +
> > +	ret = blk_queue_enter(q, flags & BLK_MQ_REQ_NOWAIT);
> > +	if (ret)
> > +		return ERR_PTR(ret);
> > +
> > +	hctx = q->queue_hw_ctx[hctx_idx];
> > +	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
> > +
> > +	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
> > +
> > +	rq = __blk_mq_alloc_request(&alloc_data, rw);
> > +	if (!rq && !(flags & BLK_MQ_REQ_NOWAIT)) {
> > +		__blk_mq_run_hw_queue(hctx);
> > +
> > +		rq =  __blk_mq_alloc_request(&alloc_data, rw);
> > +	}
> 
> Why are we duplicating this code here? If NOWAIT isn't set, then
> we'll
> always return a request. bt_get() will run the queue for us, if it
> needs
> to. blk_mq_alloc_request() does this too, and I'm guessing that code
> was
> just copied. I'll fix that up. Looks like this should just be:
> 
> 	rq = __blk_mq_alloc_request(&alloc_data, rw);
> 	if (rq)
> 		return rq;
> 
> 	blk_queue_exit(q);
> 	return ERR_PTR(-EWOULDBLOCK);
> 
> for this case.

Yes,

But the bt_get() reminds me that this patch actually has a problem.

blk_mq_alloc_request_hctx() ->
  __blk_mq_alloc_request() ->
    blk_mq_get_tag() -> 
      __blk_mq_get_tag() ->
        bt_get() ->
          blk_mq_put_ctx(data->ctx);

Here are blk_mq_get_ctx() and blk_mq_put_ctx().

static inline struct blk_mq_ctx *blk_mq_get_ctx(struct request_queue *q)
{       
        return __blk_mq_get_ctx(q, get_cpu());
} 

static inline void blk_mq_put_ctx(struct blk_mq_ctx *ctx)
{
        put_cpu();
}

blk_mq_alloc_request_hctx() calls __blk_mq_get_ctx() instead
of blk_mq_get_ctx(). Then reason is the "hctx" could belong to other
cpu. So blk_mq_get_ctx() doesn't work.

But then above put_cpu() in blk_mq_put_ctx() will trigger a WARNING
because we didn't do get_cpu() in blk_mq_alloc_request_hctx()

[toc] | [prev] | [next] | [standalone]

#1417328

What we really nee to do is to always set the NOWAIT flag (we have a
reserved tag for connect anyway), and thus never trigger the code
deep down in bt_get that might switch to a different hctx.

Below is the version that I've tested together with the NVMe change
to use the NOWAIT flag:

---
From 2346a4fdcc57c70a671e1329f7550d91d4e9d8a8 Mon Sep 17 00:00:00 2001
From: Ming Lin <ming.l@ssi.samsung.com>
Date: Mon, 30 Nov 2015 19:45:48 +0100
Subject: blk-mq: add blk_mq_alloc_request_hctx

For some protocols like NVMe over Fabrics we need to be able to send
initialization commands to a specific queue.

Based on an earlier patch from Christoph Hellwig <hch@lst.de>.

Signed-off-by: Ming Lin <ming.l@ssi.samsung.com>
[hch: disallow sleeping allocation, req_op fixes]
Signed-off-by: Christoph Hellwig <hch@lst.de>
---
 block/blk-mq.c         | 39 +++++++++++++++++++++++++++++++++++++++
 include/linux/blk-mq.h |  2 ++
 2 files changed, 41 insertions(+)

diff --git a/block/blk-mq.c b/block/blk-mq.c
index 13f4603..7aa60c4 100644
--- a/block/blk-mq.c
+++ b/block/blk-mq.c
@@ -267,6 +267,45 @@ struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 }
 EXPORT_SYMBOL(blk_mq_alloc_request);
 
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int rw,
+		unsigned int flags, unsigned int hctx_idx)
+{
+	struct blk_mq_hw_ctx *hctx;
+	struct blk_mq_ctx *ctx;
+	struct request *rq;
+	struct blk_mq_alloc_data alloc_data;
+	int ret;
+
+	/*
+	 * If the tag allocator sleeps we could get an allocation for a
+	 * different hardware context.  No need to complicate the low level
+	 * allocator for this for the rare use case of a command tied to
+	 * a specific queue.
+	 */
+	if (WARN_ON_ONCE(!(flags & BLK_MQ_REQ_NOWAIT)))
+		return ERR_PTR(-EINVAL);
+
+	if (hctx_idx >= q->nr_hw_queues)
+		return ERR_PTR(-EIO);
+
+	ret = blk_queue_enter(q, true);
+	if (ret)
+		return ERR_PTR(ret);
+
+	hctx = q->queue_hw_ctx[hctx_idx];
+	ctx = __blk_mq_get_ctx(q, cpumask_first(hctx->cpumask));
+
+	blk_mq_set_alloc_data(&alloc_data, q, flags, ctx, hctx);
+	rq = __blk_mq_alloc_request(&alloc_data, rw, 0);
+	if (!rq) {
+		blk_queue_exit(q);
+		return ERR_PTR(-EWOULDBLOCK);
+	}
+
+	return rq;
+}
+EXPORT_SYMBOL_GPL(blk_mq_alloc_request_hctx);
+
 static void __blk_mq_free_request(struct blk_mq_hw_ctx *hctx,
 				  struct blk_mq_ctx *ctx, struct request *rq)
 {
diff --git a/include/linux/blk-mq.h b/include/linux/blk-mq.h
index faa7d5c2..e43bbff 100644
--- a/include/linux/blk-mq.h
+++ b/include/linux/blk-mq.h
@@ -198,6 +198,8 @@ enum {
 
 struct request *blk_mq_alloc_request(struct request_queue *q, int rw,
 		unsigned int flags);
+struct request *blk_mq_alloc_request_hctx(struct request_queue *q, int op,
+		unsigned int flags, unsigned int hctx_idx);
 struct request *blk_mq_tag_to_rq(struct blk_mq_tags *tags, unsigned int tag);
 struct cpumask *blk_mq_tags_cpumask(struct blk_mq_tags *tags);
 
-- 
2.1.4

[toc] | [prev] | [next] | [standalone]

#1417323

On Tue, Jun 07, 2016 at 10:49:22PM -0600, Jens Axboe wrote:
> Why are we duplicating this code here? If NOWAIT isn't set, then we'll
> always return a request. bt_get() will run the queue for us, if it needs
> to. blk_mq_alloc_request() does this too, and I'm guessing that code was
> just copied. I'll fix that up. Looks like this should just be:

The generic code is a bit of a mess as it does the nowait optimization
in two different cases, I'll send you a separate cleanup patch for that.

[toc] | [prev] | [standalone]

Back to top | Article view | linux.kernel

csiph-web