lkml.org 
[lkml]   [2013]   [Jul]   [1]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
SubjectRe: [PATCH] block: Fix possible sleep in invalid context
From
Date
On Mon, 2013-07-01 at 15:17 -0700, Andrew Morton wrote:
> On Mon, 1 Jul 2013 20:58:35 +0530 Sujit Reddy Thumma <sthumma@codeaurora.org> wrote:
>
> > When block runtime PM is enabled following warning is seen
> > while resuming the device.
> >
> > BUG: sleeping function called from invalid context at
> > .../drivers/base/power/runtime.c:923
> > in_atomic(): 1, irqs_disabled(): 128, pid: 12, name: kworker/0:1
> > [<c0014448>] (unwind_backtrace+0x0/0x120) from
> > [<c03120e4>] (__pm_runtime_suspend+0x34/0xa0) from
> > [<c021c33c>] (blk_post_runtime_resume+0x4c/0x5c) from
> > [<c03297cc>] (scsi_runtime_resume+0x90/0xb4) from
> > [<c0310940>] (__rpm_callback+0x30/0x58) from
> > [<c0310980>] (rpm_callback+0x18/0x28) from
> > [<c0311ab0>] (rpm_resume+0x3dc/0x540) from
> > [<c03120a4>] (pm_runtime_work+0x8c/0x98) from
> > [<c007767c>] (process_one_work+0x238/0x3e4) from
> > [<c0077b90>] (worker_thread+0x1ac/0x2ac) from
> > [<c007cfdc>] (kthread+0x88/0x94) from
> > [<c000ece0>] (kernel_thread_exit+0x0/0x8)
> >
> > Fix this by releasing spin_lock_irq() before calling
> > pm_runtime_autosuspend() in blk_post_runtime_resume().
> >
> > --- a/block/blk-core.c
> > +++ b/block/blk-core.c
> > @@ -3159,16 +3159,18 @@ EXPORT_SYMBOL(blk_pre_runtime_resume);
> > */
> > void blk_post_runtime_resume(struct request_queue *q, int err)
> > {
> > - spin_lock_irq(q->queue_lock);
> > if (!err) {
> > + spin_lock_irq(q->queue_lock);
> > q->rpm_status = RPM_ACTIVE;
> > __blk_run_queue(q);
> > pm_runtime_mark_last_busy(q->dev);
> > + spin_unlock_irq(q->queue_lock);
> > pm_runtime_autosuspend(q->dev);
> > } else {
> > + spin_lock_irq(q->queue_lock);
> > q->rpm_status = RPM_SUSPENDED;
> > + spin_unlock_irq(q->queue_lock);
> > }
> > - spin_unlock_irq(q->queue_lock);
> > }
> > EXPORT_SYMBOL(blk_post_runtime_resume);
> > #endif
>
> I suppose we can do this cleanly enough:
>
> --- a/block/blk-core.c~block-fix-possible-sleep-in-invalid-context-fix
> +++ a/block/blk-core.c
> @@ -3159,15 +3159,14 @@ EXPORT_SYMBOL(blk_pre_runtime_resume);
> */
> void blk_post_runtime_resume(struct request_queue *q, int err)
> {
> + spin_lock_irq(q->queue_lock);
> if (!err) {
> - spin_lock_irq(q->queue_lock);
> q->rpm_status = RPM_ACTIVE;
> __blk_run_queue(q);
> pm_runtime_mark_last_busy(q->dev);
> spin_unlock_irq(q->queue_lock);
> pm_request_autosuspend(q->dev);
> } else {
> - spin_lock_irq(q->queue_lock);
> q->rpm_status = RPM_SUSPENDED;
> spin_unlock_irq(q->queue_lock);
> }
> _
>
>
> I wonder if we actually need locking around that second write to
> q->rpm_status.

Shouldn't: it's an int, which makes it a 32 bit quantity we believe to
have atomic write properties on every platform.

James




\
 
 \ /
  Last update: 2013-07-02 01:41    [W:0.070 / U:0.100 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site