Quoting Mika Kuoppala (2018-02-08 12:34:08) > Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> writes: > > > Quoting Mika Kuoppala (2018-02-08 11:43:50) > >> Any thoughts of starting to log the reset attempts > >> with timeout, even if the subsequent reset succeeds? > > > > If it succeeds, do we care? Capturing why it fails, sure. > > > > The question being what do we want to gain from it? Faster reset by > > removing timeout loops -- but if it does take X attempts, we can't > > really make it faster, just swap out one delay for another? > > > > It's a challenge, trying to provide the right information to solve a > > user's problem without their intervention and without any burden. > > Easier when you are chasing a problem down to know what you need. And > > likely need again in future? > > I was thinking of gauging the rest robustness. To check if > the level we are at, through CI. I would keep the timeouts and would > keep retries. > > Mainly the intent would be to find a pattern. Like if on some platform, > after test x, the first reset always timeouts would be a sign > to further robustify. But unless you automate such pattern finding, it will be lost and just annoy the next person trying to extract signal from the noise :) If you can think of it, write a test for it and let it run for a few months in CI. -Chris _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
