Yair K. wrote:
> On Tuesday 19 August 2008 14:28:24 Hannu Savolainen wrote:
>
>> Yair K. wrote:
>>
>>> Hi,
>>>
>>> I noticed pulseaudio has problems starting when OSSv4 is loaded, and
>>> traced it down to using select() and GET{I,O}SPACE with an mmap()'d device
>>>
> To be more accurate, PulseAudio uses poll() not select().
>
>
>>> (PulseAudio tries using mmap unless explicitly told otherwise). OSS
>>> doesn't support that, and returns -EIO, which is an error code PulseAudio
>>> doesn't check for and eventually leads to it failing to start. One might
>>> consider blaming Pulseaudio, except:
>>>
>>> A) These limitations aren't documented anywhere I can see. AFAICT
>>> FreeBSD's implementation does support using select(),etc. with mmap().
>>> B) OSS's return value isn't entirely consistent: It returns OSS_EIO when
>>> select() is used (oss_audio_core.c / oss_audio_chpoll() ), but OSS_EPERM
>>> when GETOSPACE is used with mmap. OSS_EIO seems wrong to me - the device is
>>> okay, it's just select() isn't supported in the current mode.
>>>
>>>
>> The attached diff comments out the error checking. However the result is
>> probably that select/poll will never report data available. However
>> could you try if it makes any difference with pulseaudio.
>>
>
> What happens is exactly as described - poll() never reports data (it does
> work in non-mmaped mode).
>
Ok.
I split oss_audio_chpoll() to separate input and output handlers. Now
poll/select will report available data if dmap->interrupt_counter > 0.
This counter gets incremented at each fragment boundary. Calling
GETIPTR/GETOPTR clears the counter to zero. I think approach should work
because every mmap application will make these calls.
Best regards,
Hannu
_______________________________________________
oss-devel mailing list
oss-devel@mailman.opensound.com
http://mailman.opensound.com/mailman/listinfo/oss-devel
