On Fri, Sep 08, 2017 at 04:52:25PM -0700, Bart Van Assche wrote: > Instead of allowing request allocation to succeed for suspended > request queues and only to process power management requests, make > blk_get_request() wait until the request queue is resumed for > requests that are not power management requests. > > This patch avoids that resume does not occur if the maximum queue > depth is reached when a power management request is submitted. > > Note: this patch affects the behavior of scsi_device_quiesce() only > if that function is called from inside a power management callback. > This patch does not affect the behavior of scsi_device_quiesce() > when a call of that function is triggered by writing "quiesce" into > /sys/class/scsi_device/*/device/state. > > Signed-off-by: Bart Van Assche <[email protected]> > Cc: Christoph Hellwig <[email protected]> > Cc: Hannes Reinecke <[email protected]> > Cc: Johannes Thumshirn <[email protected]> > Cc: Rafael J. Wysocki <[email protected]> > Cc: Ming Lei <[email protected]> > --- > block/blk-core.c | 60 > +++++++++++++++++++++++++++----------------------- > block/blk.h | 12 ++++++++++ > include/linux/blkdev.h | 1 + > 3 files changed, 45 insertions(+), 28 deletions(-) > > diff --git a/block/blk-core.c b/block/blk-core.c > index bb53c6b58e8c..cd2700c763ed 100644 > --- a/block/blk-core.c > +++ b/block/blk-core.c > @@ -1325,6 +1325,24 @@ static struct request *__get_request(struct > request_list *rl, unsigned int op, > return ERR_PTR(-ENOMEM); > } > > +#ifdef CONFIG_PM > +static bool blk_wait_until_active(struct request_queue *q, bool wait) > + __releases(q->queue_lock) > + __acquires(q->queue_lock) > +{ > + if (wait) > + wait_event_lock_irq(q->rpm_active_wq, > + q->rpm_status == RPM_ACTIVE, > + *q->queue_lock); > + return q->rpm_status == RPM_ACTIVE;
If runtime PM is disabled via /sys/.../power/control, q->rpm_status can be always ACTIVE, even during system suspend, then you can't prevent any new request allocation at that time. -- Ming
