Re: HDA record fails with FIFO error

[David Henningsson <david.henningsson@canonical.com>]

On 2011-04-07 16:30, David Henningsson wrote:
> On 2011-04-07 12:10, Takashi Iwai wrote:
>> At Thu, 7 Apr 2011 16:54:50 +0800,
>> Xu, Andiry wrote:
>>>
>>> Hi Takashi,
>>>
>>> Currently I encountered an HD audio issue reported by customer on a
>>> Lenovo x100e system. Sometimes, but not always, when starting a
>>> recording stream through the HDA controller, the controller generates a
>>> large amount of interrupts (~40 000 interrupts per second). After this
>>> has happened, jack sense (i e unsolicited events from the codec) stops
>>> working until the next system reboot.
>
> This seems more reproducible now than it was a while ago, now it seems
> to happen more often than not.
>
>>> SD0STS returns a FIFO error (0x28) in interrupt handler. The interrupt
>>> service routine acknowledges this error but does not do anything to
>>> counteract the root cause to the problem, so it appears again and again.
>>> Restarting the stream does not seem to help. Enable MSI or not does not
>>> help either.
>>>
>>> The issue occurs on 2.6.35 and 2.6.38-rc8+, have not tried latest kernel
>>> yet but I think it's also there. FIFO error indicates FIFO overrun
>>> occurring while the RUN bit is set, but the driver simply acknowledge
>>> and clear the error. I wonder what the root cause and the right
>>> treatment are in this case. Any suggestions?
>
> Brainstorming:
> 1) We could try adding udelays at random places.
> 2) There are a few workarounds for different chips already in
> hda_intel.c, maybe try applying them here as well.
>
> Does that make sense to either of you?
>
>> FIFO_ERR is actually never handled, so basically we can ignore.
>> Simply disabling it like below works around your problem?
>>
>> Of course, a proper handling of FIFO error would be better, but
>> this may need more code rewrites.
>
> This does not fix the error in question, it removes the interrupts all
> right, but neither recording nor jack sense starts to work.

Hi Andiry,

I'm still trying to get a grip of what the error could be. Having 
re-enabled the FIFO interrupt again and some debug printk's the first 
time the FIFO error happens, I notice that it happens after a while, 
where "a while" ranges from 100 ms to 4 s or so. CBL is 65536 and LPIB 
seems to be a reasonable value (a value below 65536, approximately 
corresponding to the time the stream has been running). BDL entries look 
correct.

Can it be something else than a chipset bug in this case? I'm trying to 
rule out every possible driver problem I can think of.

Btw - how can a recording FIFO overrun in the first place? I mean, if we 
have two BDLE buffers with 32768 bytes in each (assume bl_pos_adj=0 for 
this example), the FIFO should just write to the first, then the second, 
then the first again, and so on, without having overrun errors.

-- 
David Henningsson, Canonical Ltd.
http://launchpad.net/~diwic