On 11/17/19 6:47 PM, Kuninori Morimoto wrote:
Hi Pierre-Louis
Thank you for testing
But, these are just my guess. It works for me, but I can't re-produce the issue.
Below is the patch for "testing".
Morimoto-san, I haven't fully tested the logic and all the long description, but the patch suggested seems to work on all the platforms I tested on - and actually improves things on Baytrail/Cherrytrail.
I haven't seen any oopses in several hours now and no regression reported by Intel CI https://github.com/thesofproject/linux/pull/1504
Wow !! Nice to know !! I'm very happy about that !!
OK, I will post the patch officially with your tested by.
Let's see what others think or tested.
the only issue we found in additional testing is that we need to insert a ~20s delay after each of module insertion/removal sequences, or we get stuck in at some point. That is a likely sign of delayed work not properly canceled - it may however be a different topic altogether, I saw the same sighting with my initial patch which reverted the "ASoC: soc-core: add soc_unbind_dai_link()" patch.
Others at Intel redid my tests and also found that the module load/unload tests worked across the board on CI devices.
from the Intel side we're good.