Linux-NVME Archive on lore.kernel.org
 help / color / Atom feed
* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
       [not found] <CALCePG0y1REA8xXX1ymTwAZhrbSyUh41zfpOmFQViuyPGf5ePg@mail.gmail.com>
@ 2021-04-17  2:39 ` Keith Busch
  0 siblings, 0 replies; 11+ messages in thread
From: Keith Busch @ 2021-04-17  2:39 UTC (permalink / raw)
  To: Casey Chen; +Cc: axboe, hch, linux-block, linux-nvme, sagi, yzhong

On Fri, Apr 16, 2021 at 07:04:34PM -0700, Casey Chen wrote:
> > On Fri, Apr 16, 2021 at 10:12:11AM -0700, Yuanyuan Zhong wrote:
> > > >         if (poll)
> > > >                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
> > > You may need to audit other completion handlers for blk_execute_rq_nowait().
> >
> > Why? Those callers already provide their own callback that directly get
> > the error.
> 
> We should make sure all callbacks provided to blk_execute_rq_nowait()
> carry error back. i.e. by reusing rq->end_io_data.
>
> > > How to get error ret from polled rq?
> >
> > Please see nvme_end_sync_rq() for that driver's polled handler callback.
> > It already has the error.
> 
> nvme_end_sync_rq() currently doesn't store error in rq->end_io_data as
> you proposed in patch 1.

The question was how the error gets back to the caller, and they already
have it.

Patch 1 is specific to the sync execution. All the async users' handling
of the provided error are implementation specific. If they're not using
it correctly, then they can be fixed too.

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-23 21:57 [PATCH 1/2] block: return errors " Keith Busch
@ 2021-04-23 21:58 ` Keith Busch
  0 siblings, 0 replies; 11+ messages in thread
From: Keith Busch @ 2021-04-23 21:58 UTC (permalink / raw)
  To: linux-nvme, sagi, hch, axboe, linux-block
  Cc: Yuanyuan Zhong, Casey Chen, Keith Busch

We don't have an nvme status to report if the driver's .queue_rq()
returns an error without dispatching the requested nvme command. Use the
return value from blk_execute_rq() so the caller may know their command
was not successful.

Reported-by: Yuanyuan Zhong <yzhong@purestorage.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
---
 drivers/nvme/host/core.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 85538c38aae9..b57157106cac 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1000,12 +1000,12 @@ int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 	if (poll)
 		nvme_execute_rq_polled(req->q, NULL, req, at_head);
 	else
-		blk_execute_rq(NULL, req, at_head);
+		ret = blk_execute_rq(NULL, req, at_head);
 	if (result)
 		*result = nvme_req(req)->result;
 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 		ret = -EINTR;
-	else
+	else if (nvme_req(req)->status)
 		ret = nvme_req(req)->status;
  out:
 	blk_mq_free_request(req);
-- 
2.25.4


_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-19 17:48             ` Keith Busch
@ 2021-04-19 18:20               ` Yuanyuan Zhong
  0 siblings, 0 replies; 11+ messages in thread
From: Yuanyuan Zhong @ 2021-04-19 18:20 UTC (permalink / raw)
  To: Keith Busch; +Cc: Christoph Hellwig, linux-nvme, sagi, axboe, linux-block

On Mon, Apr 19, 2021 at 10:48 AM Keith Busch <kbusch@kernel.org> wrote:
>

> diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
> index b57157106cac..a0fb9ad132af 100644
> --- a/drivers/nvme/host/core.c
> +++ b/drivers/nvme/host/core.c
> @@ -949,6 +949,9 @@ static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
>         struct completion *waiting = rq->end_io_data;
>
>         rq->end_io_data = NULL;
> +       if (error && !nvme_req(rq)->status)
Is setting nvme_req(rq)->status for each error return in ->queue_rq()
going to gradually roll out, and eventually skipping the fallback here?

> +               nvme_req(rq)->status = blk_status_to_errno(error);
Casting int negative errno to u16 will get 0xfff., meaning NVME_SC_DNR is set.
Is that always right?

--
Regards,
Yuanyuan Zhong

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-19 17:27           ` Yuanyuan Zhong
@ 2021-04-19 17:48             ` Keith Busch
  2021-04-19 18:20               ` Yuanyuan Zhong
  0 siblings, 1 reply; 11+ messages in thread
From: Keith Busch @ 2021-04-19 17:48 UTC (permalink / raw)
  To: Yuanyuan Zhong; +Cc: Christoph Hellwig, linux-nvme, sagi, axboe, linux-block

On Mon, Apr 19, 2021 at 10:27:42AM -0700, Yuanyuan Zhong wrote:
> On Mon, Apr 19, 2021 at 8:14 AM Keith Busch <kbusch@kernel.org> wrote:
> >
> > On Mon, Apr 19, 2021 at 09:16:05AM +0200, Christoph Hellwig wrote:
> > > On Sat, Apr 17, 2021 at 02:17:35AM +0900, Keith Busch wrote:
> > > > On Fri, Apr 16, 2021 at 10:12:11AM -0700, Yuanyuan Zhong wrote:
> > > > > >         if (poll)
> > > > > >                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
> > > > > You may need to audit other completion handlers for blk_execute_rq_nowait().
> > > >
> > > > Why? Those callers already provide their own callback that directly get
> > > > the error.
> 
> See below clarification. I was wondering whether the way you were going to
> propose for nvme_end_sync_rq() would establish new convention for other
> blk_execute_rq_nowait() completion handlers implementation.

I'm not sure it can be turned into a common pattern. The callbacks are
implementation specific.
 
> > > >
> > > > > How to get error ret from polled rq?
> > > >
> > > > Please see nvme_end_sync_rq() for that driver's polled handler callback.
> > > > It already has the error.
> > >
> > > But it never looks at it..
> >
> > The question was how to get ret. I didn't mean to imply the example was
> > actually using it. :)
> 
> The question was how to let nvme_end_sync_rq() propagate the blk_status_t
> error to the ret for __nvme_submit_sync_cmd(). That's part of the problem
> here: __nvme_submit_sync_cmd() may return success for a command that
> failed submission.

---
diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index b57157106cac..a0fb9ad132af 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -949,6 +949,9 @@ static void nvme_end_sync_rq(struct request *rq, blk_status_t error)
 	struct completion *waiting = rq->end_io_data;
 
 	rq->end_io_data = NULL;
+	if (error && !nvme_req(rq)->status)
+		nvme_req(rq)->status = blk_status_to_errno(error);
 	complete(waiting);
 }
 
--

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-19 15:14         ` Keith Busch
@ 2021-04-19 17:27           ` Yuanyuan Zhong
  2021-04-19 17:48             ` Keith Busch
  0 siblings, 1 reply; 11+ messages in thread
From: Yuanyuan Zhong @ 2021-04-19 17:27 UTC (permalink / raw)
  To: Keith Busch; +Cc: Christoph Hellwig, linux-nvme, sagi, axboe, linux-block

On Mon, Apr 19, 2021 at 8:14 AM Keith Busch <kbusch@kernel.org> wrote:
>
> On Mon, Apr 19, 2021 at 09:16:05AM +0200, Christoph Hellwig wrote:
> > On Sat, Apr 17, 2021 at 02:17:35AM +0900, Keith Busch wrote:
> > > On Fri, Apr 16, 2021 at 10:12:11AM -0700, Yuanyuan Zhong wrote:
> > > > >         if (poll)
> > > > >                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
> > > > You may need to audit other completion handlers for blk_execute_rq_nowait().
> > >
> > > Why? Those callers already provide their own callback that directly get
> > > the error.

See below clarification. I was wondering whether the way you were going to
propose for nvme_end_sync_rq() would establish new convention for other
blk_execute_rq_nowait() completion handlers implementation.

> > >
> > > > How to get error ret from polled rq?
> > >
> > > Please see nvme_end_sync_rq() for that driver's polled handler callback.
> > > It already has the error.
> >
> > But it never looks at it..
>
> The question was how to get ret. I didn't mean to imply the example was
> actually using it. :)

The question was how to let nvme_end_sync_rq() propagate the blk_status_t
error to the ret for __nvme_submit_sync_cmd(). That's part of the problem
here: __nvme_submit_sync_cmd() may return success for a command that
failed submission.

-- 
Regards,
Yuanyuan Zhong

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-19  7:16       ` Christoph Hellwig
@ 2021-04-19 15:14         ` Keith Busch
  2021-04-19 17:27           ` Yuanyuan Zhong
  0 siblings, 1 reply; 11+ messages in thread
From: Keith Busch @ 2021-04-19 15:14 UTC (permalink / raw)
  To: Christoph Hellwig; +Cc: Yuanyuan Zhong, linux-nvme, sagi, axboe, linux-block

On Mon, Apr 19, 2021 at 09:16:05AM +0200, Christoph Hellwig wrote:
> On Sat, Apr 17, 2021 at 02:17:35AM +0900, Keith Busch wrote:
> > On Fri, Apr 16, 2021 at 10:12:11AM -0700, Yuanyuan Zhong wrote:
> > > >         if (poll)
> > > >                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
> > > You may need to audit other completion handlers for blk_execute_rq_nowait().
> > 
> > Why? Those callers already provide their own callback that directly get
> > the error.
> > 
> > > How to get error ret from polled rq?
> > 
> > Please see nvme_end_sync_rq() for that driver's polled handler callback.
> > It already has the error.
> 
> But it never looks at it..

The question was how to get ret. I didn't mean to imply the example was
actually using it. :)

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-16 17:17     ` Keith Busch
@ 2021-04-19  7:16       ` Christoph Hellwig
  2021-04-19 15:14         ` Keith Busch
  0 siblings, 1 reply; 11+ messages in thread
From: Christoph Hellwig @ 2021-04-19  7:16 UTC (permalink / raw)
  To: Keith Busch; +Cc: Yuanyuan Zhong, linux-nvme, sagi, hch, axboe, linux-block

On Sat, Apr 17, 2021 at 02:17:35AM +0900, Keith Busch wrote:
> On Fri, Apr 16, 2021 at 10:12:11AM -0700, Yuanyuan Zhong wrote:
> > >         if (poll)
> > >                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
> > You may need to audit other completion handlers for blk_execute_rq_nowait().
> 
> Why? Those callers already provide their own callback that directly get
> the error.
> 
> > How to get error ret from polled rq?
> 
> Please see nvme_end_sync_rq() for that driver's polled handler callback.
> It already has the error.

But it never looks at it..

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-16 16:53 ` [PATCH 2/2] nvme: use return value " Keith Busch
  2021-04-16 17:12   ` Yuanyuan Zhong
@ 2021-04-18  4:17   ` Chaitanya Kulkarni
  1 sibling, 0 replies; 11+ messages in thread
From: Chaitanya Kulkarni @ 2021-04-18  4:17 UTC (permalink / raw)
  To: Keith Busch, linux-nvme, sagi, hch, axboe, linux-block; +Cc: Yuanyuan Zhong

On 4/16/21 09:54, Keith Busch wrote:
> We don't have an nvme status to report if the driver's .queue_rq()
> returns an error without dispatching the requested nvme command. Use the
> return value from blk_execute_rq() so the caller may know their command
> was not successful.
>
> Reported-by: Yuanyuan Zhong <yzhong@purestorage.com>
> Signed-off-by: Keith Busch <kbusch@kernel.org>


Looks good.

Reviewed-by: Chaitanya Kulkarni <chaitanya.kulkarni@wdc.com>




_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-16 17:12   ` Yuanyuan Zhong
@ 2021-04-16 17:17     ` Keith Busch
  2021-04-19  7:16       ` Christoph Hellwig
  0 siblings, 1 reply; 11+ messages in thread
From: Keith Busch @ 2021-04-16 17:17 UTC (permalink / raw)
  To: Yuanyuan Zhong; +Cc: linux-nvme, sagi, hch, axboe, linux-block

On Fri, Apr 16, 2021 at 10:12:11AM -0700, Yuanyuan Zhong wrote:
> >         if (poll)
> >                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
> You may need to audit other completion handlers for blk_execute_rq_nowait().

Why? Those callers already provide their own callback that directly get
the error.

> How to get error ret from polled rq?

Please see nvme_end_sync_rq() for that driver's polled handler callback.
It already has the error.

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* Re: [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-16 16:53 ` [PATCH 2/2] nvme: use return value " Keith Busch
@ 2021-04-16 17:12   ` Yuanyuan Zhong
  2021-04-16 17:17     ` Keith Busch
  2021-04-18  4:17   ` Chaitanya Kulkarni
  1 sibling, 1 reply; 11+ messages in thread
From: Yuanyuan Zhong @ 2021-04-16 17:12 UTC (permalink / raw)
  To: Keith Busch; +Cc: linux-nvme, sagi, hch, axboe, linux-block

>         if (poll)
>                 nvme_execute_rq_polled(req->q, NULL, req, at_head);
You may need to audit other completion handlers for blk_execute_rq_nowait().
How to get error ret from polled rq?
>         else
> -               blk_execute_rq(NULL, req, at_head);
> +               ret = blk_execute_rq(NULL, req, at_head);
>         if (result)
>                 *result = nvme_req(req)->result;
>         if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
>                 ret = -EINTR;
> -       else
> +       else if (nvme_req(req)->status)
>                 ret = nvme_req(req)->status;


-- 
Regards,
Yuanyuan Zhong

_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

* [PATCH 2/2] nvme: use return value from blk_execute_rq()
  2021-04-16 16:53 [PATCH 1/2] block: return errors " Keith Busch
@ 2021-04-16 16:53 ` Keith Busch
  2021-04-16 17:12   ` Yuanyuan Zhong
  2021-04-18  4:17   ` Chaitanya Kulkarni
  0 siblings, 2 replies; 11+ messages in thread
From: Keith Busch @ 2021-04-16 16:53 UTC (permalink / raw)
  To: linux-nvme, sagi, hch, axboe, linux-block; +Cc: Keith Busch, Yuanyuan Zhong

We don't have an nvme status to report if the driver's .queue_rq()
returns an error without dispatching the requested nvme command. Use the
return value from blk_execute_rq() so the caller may know their command
was not successful.

Reported-by: Yuanyuan Zhong <yzhong@purestorage.com>
Signed-off-by: Keith Busch <kbusch@kernel.org>
---
 drivers/nvme/host/core.c | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/drivers/nvme/host/core.c b/drivers/nvme/host/core.c
index 85538c38aae9..b57157106cac 100644
--- a/drivers/nvme/host/core.c
+++ b/drivers/nvme/host/core.c
@@ -1000,12 +1000,12 @@ int __nvme_submit_sync_cmd(struct request_queue *q, struct nvme_command *cmd,
 	if (poll)
 		nvme_execute_rq_polled(req->q, NULL, req, at_head);
 	else
-		blk_execute_rq(NULL, req, at_head);
+		ret = blk_execute_rq(NULL, req, at_head);
 	if (result)
 		*result = nvme_req(req)->result;
 	if (nvme_req(req)->flags & NVME_REQ_CANCELLED)
 		ret = -EINTR;
-	else
+	else if (nvme_req(req)->status)
 		ret = nvme_req(req)->status;
  out:
 	blk_mq_free_request(req);
-- 
2.25.4


_______________________________________________
Linux-nvme mailing list
Linux-nvme@lists.infradead.org
http://lists.infradead.org/mailman/listinfo/linux-nvme

^ permalink raw reply	[flat|nested] 11+ messages in thread

end of thread, back to index

Thread overview: 11+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
     [not found] <CALCePG0y1REA8xXX1ymTwAZhrbSyUh41zfpOmFQViuyPGf5ePg@mail.gmail.com>
2021-04-17  2:39 ` [PATCH 2/2] nvme: use return value from blk_execute_rq() Keith Busch
2021-04-23 21:57 [PATCH 1/2] block: return errors " Keith Busch
2021-04-23 21:58 ` [PATCH 2/2] nvme: use return value " Keith Busch
  -- strict thread matches above, loose matches on Subject: below --
2021-04-16 16:53 [PATCH 1/2] block: return errors " Keith Busch
2021-04-16 16:53 ` [PATCH 2/2] nvme: use return value " Keith Busch
2021-04-16 17:12   ` Yuanyuan Zhong
2021-04-16 17:17     ` Keith Busch
2021-04-19  7:16       ` Christoph Hellwig
2021-04-19 15:14         ` Keith Busch
2021-04-19 17:27           ` Yuanyuan Zhong
2021-04-19 17:48             ` Keith Busch
2021-04-19 18:20               ` Yuanyuan Zhong
2021-04-18  4:17   ` Chaitanya Kulkarni

Linux-NVME Archive on lore.kernel.org

Archives are clonable:
	git clone --mirror https://lore.kernel.org/linux-nvme/0 linux-nvme/git/0.git

	# If you have public-inbox 1.1+ installed, you may
	# initialize and index your mirror using the following commands:
	public-inbox-init -V2 linux-nvme linux-nvme/ https://lore.kernel.org/linux-nvme \
		linux-nvme@lists.infradead.org
	public-inbox-index linux-nvme

Example config snippet for mirrors

Newsgroup available over NNTP:
	nntp://nntp.lore.kernel.org/org.infradead.lists.linux-nvme


AGPL code for this site: git clone https://public-inbox.org/public-inbox.git