On Wed, Jun 21, 2023 at 03:39:12PM +0100, Mark Brown wrote:
On Wed, Jun 21, 2023 at 04:08:47PM +0200, Jaroslav Kysela wrote:
I think that the problem is somewhere else here. The overall test timeout should be calculated dynamically. All tests may be queried for the maximal expected interval based on the hardware/software capabilities. It's a bit pitfall to have a fixed time limit where the realtime tests depend on the number of devices.
I tend to agree here, unfortunately Shuah hasn't responded to queries from Nícolas about this which I imagine is what inspired this patch. We also have problems with mixer-test on one of the Dialog CODECs with a couple of 64k value controls and no cache only mode.
Yes, exactly. I've tried increasing the timeout for this test to a larger fixed value previously, and later asked for more information on how to deal with the kselftest timeout. [1]
Since I didn't hear back, I thought this patch would be a way to at least mitigate the issue for now, without limiting the test coverage, which was a concern with having limited scopes for the test.
I've just noticed that in the mean time a way to override the timeout when running kselftest has been introduced [2], so I suppose we could use that to work around the timeout limitation in CI systems and be able to run through completion on the different hardware at the lab. But I still believe, like you do, that calculating the timeout at runtime based on the hardware would make much more sense, though if there's such a desire to keep kselftests under the 45s mark, I'm not sure if it would be acceptable.
[1] https://lore.kernel.org/all/5302e70d-cb58-4e70-b44f-ff81b138a2e1@notapiano/ [2] https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/commit/?i...
Thanks, Nícolas