On 12/06/2023 15:28, Oswald Buddenhagen wrote:
On Mon, Jun 12, 2023 at 02:16:15PM +0100, Jon Hunter wrote:
On 12/06/2023 13:18, Jon Hunter wrote:
On 05/04/2023 21:12, Oswald Buddenhagen wrote:
... in wait_for_avail() and snd_pcm_drain().
Sorry for not catching this sooner, but I have just noticed that one
of our audio tests for Tegra is failing on v6.4-rc and bisect is
pointing to this commit. Reverting this on top of the current
mainline fixes it.
If I enable the debug prints, I do see the following messages ...
tegra-audio-graph-card sound: capture read timeout (DMA or IRQ trouble?)
yes, this is the kind of fallout one would expect from this change, as
it significantly shortened the effective timeout under most circumstances.
first check that there isn't a genuine underlying bug, that is, that the
unusually slow timings match expectations.
if everything looks right, then properly codify the timeout in the
driver by setting substream->wait_time as required.
the lazy approach of more or less restoring the previous status quo
would be setting it to 10000 in the `open` callback.
fwiw, soc/sof sets it to 500, which may actually be a bad idea (it's
short enough that a very long period time would exceed it, if such is
permitted). and it's not obvious why it does that.
Yes I can see now that reducing the timeout is causing the problem here.
I can try a few settings to see what works.
Thanks
Jon
--
nvpublic
