[alsa-devel] [PATCH 1/4] ASoC: compress: Pass error out of soc_compr_pointer

[Takashi Iwai]

On Fri, 11 Mar 2016 11:04:25 +0100,
Charles Keepax wrote:
> 
> On Fri, Mar 11, 2016 at 01:18:43PM +0530, Vinod Koul wrote:
> > On Thu, Mar 10, 2016 at 10:44:51AM +0000, Charles Keepax wrote:
> > > The soc_compr_pointer does not correctly pass any errors returned by the
> > > driver callback back up the stack. This patch corrects this issue.
> > 
> > Should we do that :) I am not too sure. Pointer query is supposed to read
> > the current value and return. You are trying to indicate that stream has
> > gone bad which is not the same as read faced an error...
> > 
> > Also please use cover letter for these things to describe problem you are
> > trying to solve.
> 
> Apologies for not doing so, I had been viewing this as more of a
> simple oversight in the framework rather than a design choice.
> 
> The problem I am looking at is the DSP suffers an unrecoverable
> error. We can find out about this error in our driver because the
> DSP returns some error status to us.  This is fine if user-space
> is doing a read as reads return error status back to user-space
> so the user can find out that things have gone bad. However, if
> user-space is doing an avail request there is no path for the
> error to come back up to user-space. The pointer request returns
> zero available data, so a read never happens and we basically
> just end up sitting waiting for data on a stream that we know
> full well has died.
> 
> I will have a look at other options for propagating this error,
> I guess it should also be possible to push it through the poll so
> I can look at why that isn't happening.
>
> I guess my return question would be I can imagine many reasons
> why a pointer query might fail, especially for off chip DSPs.
> What is the reasoning for wanting to hide those errors from the
> rest of the system? It seems to me it would be best to handle an
> error as soon as it is noticed, and if a particular system has a
> pointer request that never fails then it can just not return an
> error.

IMO, propagating the error immediately is a good thing.  I guess it
wasn't checked in the pointer callback just because the pointer
callback was supposed to be a simple state copy without involving the
state change.

OTOH, another question is whether it's enough just to tell the error
there as is.  When such an error is detected, it essentially means
that the whole DSP got wrong.  What we can do at best is to prepare
for recovery.  And this requires the state change.

Actually, PCM pointer callback may return a special value indicating
an XRUN error.  The PCM core reacts for it, stops the stream and
changes the stream state, so that further accesses get the error
consistently.  Similar mechanism would be needed for compress API, I
suppose.


Takashi