[PATCH REPOST] dm rq: remove BUG_ON(!irqs_disabled) check

[PATCH REPOST] dm rq: remove BUG_ON(!irqs_disabled) check

[Sebastian Andrzej Siewior]

In commit 052189a2ec95 ("dm: remove superfluous irq disablement in
dm_request_fn") the spin_lock_irq() was replaced with spin_lock() + a
check for disabled interrupts. Later the locking part was removed in
commit 2eb6e1e3aa87 ("dm: submit stacked requests in irq enabled
context") but the BUG_ON() check remained.

Since the original purpose for the "are-irqs-off" check is gone (the
->queue_lock has been removed) remove it.

Cc: Keith Busch <keith.busch@intel.com>
Cc: Mike Snitzer <snitzer@redhat.com>
Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 drivers/md/dm-rq.c | 1 -
 1 file changed, 1 deletion(-)

diff --git a/drivers/md/dm-rq.c b/drivers/md/dm-rq.c
index bf0b840645cc..1a524f992f72 100644
--- a/drivers/md/dm-rq.c
+++ b/drivers/md/dm-rq.c
@@ -688,7 +688,6 @@ static void dm_old_request_fn(struct request_queue *q)
 		/* Establish tio->ti before queuing work (map_tio_request) */
 		tio->ti = ti;
 		kthread_queue_work(&md->kworker, &tio->work);
-		BUG_ON(!irqs_disabled());
 	}
 }
 
-- 
2.17.0

Re: [PATCH REPOST] dm rq: remove BUG_ON(!irqs_disabled) check

[Mike Snitzer]

On Fri, May 04 2018 at 10:36am -0400,
Sebastian Andrzej Siewior <bigeasy@linutronix.de> wrote:

> In commit 052189a2ec95 ("dm: remove superfluous irq disablement in
> dm_request_fn") the spin_lock_irq() was replaced with spin_lock() + a
> check for disabled interrupts. Later the locking part was removed in
> commit 2eb6e1e3aa87 ("dm: submit stacked requests in irq enabled
> context") but the BUG_ON() check remained.
> 
> Since the original purpose for the "are-irqs-off" check is gone (the
> ->queue_lock has been removed) remove it.
> 
> Cc: Keith Busch <keith.busch@intel.com>
> Cc: Mike Snitzer <snitzer@redhat.com>
> Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
> ---
>  drivers/md/dm-rq.c | 1 -
>  1 file changed, 1 deletion(-)
> 
> diff --git a/drivers/md/dm-rq.c b/drivers/md/dm-rq.c
> index bf0b840645cc..1a524f992f72 100644
> --- a/drivers/md/dm-rq.c
> +++ b/drivers/md/dm-rq.c
> @@ -688,7 +688,6 @@ static void dm_old_request_fn(struct request_queue *q)
>  		/* Establish tio->ti before queuing work (map_tio_request) */
>  		tio->ti = ti;
>  		kthread_queue_work(&md->kworker, &tio->work);
> -		BUG_ON(!irqs_disabled());
>  	}
>  }
>  
> -- 
> 2.17.0
> 

the queue_lock hasn't been removed for the old .request_fn path in block
core -- which dm_old_request_fn is providing the hook for.

This BUG_ON() documents that reality.

Commit 2eb6e1e3aa87 made it such that work is queued into an irq enabled
context.   But it never changed the fact that old block core's
.request_fn (and as such dm_old_request_fn) is called with irqs
disabled.

So all said:

Nacked-by: Mike Snitzer <snitzer@redhat.com>

Re: [PATCH REPOST] dm rq: remove BUG_ON(!irqs_disabled) check

[Sebastian Andrzej Siewior]

On 2018-05-04 11:53:12 [-0400], Mike Snitzer wrote:
> the queue_lock hasn't been removed for the old .request_fn path in block
> core -- which dm_old_request_fn is providing the hook for.
> 
> This BUG_ON() documents that reality.
> 
> Commit 2eb6e1e3aa87 made it such that work is queued into an irq enabled
> context.   But it never changed the fact that old block core's
> .request_fn (and as such dm_old_request_fn) is called with irqs
> disabled.

As per history it looked different, sorry. Is there a lock that is
supposed to be held?

Sebastian

[PATCH] dm rq: replace BUG_ON(!irqs_disabled) with lockdep's lock annotation

[Sebastian Andrzej Siewior]

In commit 052189a2ec95 ("dm: remove superfluous irq disablement in
dm_request_fn") the spin_lock_irq() was replaced with spin_lock() + a
check for disabled interrupts. Later the locking part was removed in
commit 2eb6e1e3aa87 ("dm: submit stacked requests in irq enabled
context") but the BUG_ON() check remained.
Mike Snitzer explained that:
|the queue_lock hasn't been removed for the old .request_fn path in block
|core -- which dm_old_request_fn is providing the hook for.
|
|This BUG_ON() documents that reality.

The BUG_ON() statement checks for disabled interrupts. The better check
would to let lockdep check if the required lock (queue_lock) is held.
Lockdep will then validate that the lock is held an complain if this is
not the case. Since lockdep will see this lock held in IRQ-context it
will also complain should the interrupts be enabled while lock still
held or acquired with enabled interrupts. So this !irq_disabled() is not
required.
This change helps to avoid the BUG_ON() statement on preempt-rt where
the lock is a sleeping lock and interrupts are never really disabled.

Signed-off-by: Sebastian Andrzej Siewior <bigeasy@linutronix.de>
---
 drivers/md/dm-rq.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/drivers/md/dm-rq.c b/drivers/md/dm-rq.c
index bf0b840645cc..78a44bb7537f 100644
--- a/drivers/md/dm-rq.c
+++ b/drivers/md/dm-rq.c
@@ -688,7 +688,7 @@ static void dm_old_request_fn(struct request_queue *q)
 		/* Establish tio->ti before queuing work (map_tio_request) */
 		tio->ti = ti;
 		kthread_queue_work(&md->kworker, &tio->work);
-		BUG_ON(!irqs_disabled());
+		lockdep_assert_held(q->queue_lock);
 	}
 }
 
-- 
2.17.0