Re: [PATCH] soundwire: stream: fix bad unlock balance

[Sanyog Kale]

On Thu, May 23, 2019 at 09:43:14AM +0100, Srinivas Kandagatla wrote:
> 
> 
> On 22/05/2019 17:41, Pierre-Louis Bossart wrote:
> > 
> > 
> > On 5/22/19 11:25 AM, Srinivas Kandagatla wrote:
> > > This patch fixes below warning due to unlocking without locking.
> > > 
> > > ?? =====================================
> > > ?? WARNING: bad unlock balance detected!
> > > ?? 5.1.0-16506-gc1c383a6f0a2-dirty #1523 Tainted: G?????????????? W
> > > ?? -------------------------------------
> > > ?? aplay/2954 is trying to release lock (&bus->msg_lock) at:
> > > ?? do_bank_switch+0x21c/0x480
> > > ?? but there are no more locks to release!
> > > 
> > > Signed-off-by: Srinivas Kandagatla <srinivas.kandagatla@xxxxxxxxxx>
> > > ---
> > > ?? drivers/soundwire/stream.c | 3 ++-
> > > ?? 1 file changed, 2 insertions(+), 1 deletion(-)
> > > 
> > > diff --git a/drivers/soundwire/stream.c b/drivers/soundwire/stream.c
> > > index 544925ff0b40..d16268f30e4f 100644
> > > --- a/drivers/soundwire/stream.c
> > > +++ b/drivers/soundwire/stream.c
> > > @@ -814,7 +814,8 @@ static int do_bank_switch(struct
> > > sdw_stream_runtime *stream)
> > > ?????????????????????????? goto error;
> > > ?????????????????? }
> > > -?????????????? mutex_unlock(&bus->msg_lock);
> > > +?????????????? if (mutex_is_locked(&bus->msg_lock))
> > > +?????????????????????? utex_unlock(&bus->msg_lock);
> > 
> > Does this even compile? should be mutex_unlock, no?
> > 
> > We also may want to identify the issue in more details without pushing
> > it under the rug. The locking mechanism is far from simple and it's
> > likely there are a number of problems with it.
> > 
> msg_lock is taken conditionally during multi link bank switch cases, however
> the unlock is done unconditionally leading to this warning.
> 
> Having a closer look show that there could be a dead lock in this path while
> executing sdw_transfer(). And infact there is no need to take msg_lock in
> multi link switch cases as sdw_transfer should take care of this.
> 
> Vinod/Sanyog any reason why msg_lock is really required in this path?
>

In case of multi link we use sdw_transfer_defer instead of sdw_transfer
where lock is not acquired, hence lock is acquired in do_bank_switch for
multi link. we should add same check of multi link to release lock in
do_bank_switch.

> --srini
> 
> > > ?????????? }
> > > ?????????? return ret;
> > > 

--