Hi Krzysztof, > The i915_active selftests operate on an active attached to a mock Instead of "selftests", you could mention explicitly the two selftests that depend on this call. It took me a second to figure out that not all of the three live_active* tests suffer from this race.
Apart from that, the change looks good: Reviewed-by: Krzysztof Karas <[email protected]> > request. This request is created during the test, and a fence is > attached to it for the test to control when the request is processed. > However, because signalling the fence happens inside the same thread as > the test, and before a call to __i915_active_wait(), a race condition is > created where if the request is processed before the aforementioned > call, the active callback will not be called. > > Defer signalling the request's fence, so that the request completes only > after the test's call to __i915_active_wait(). > > Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/issues/14808 > Signed-off-by: Krzysztof Niemiec <[email protected]> > --- -- Best Regards, Krzysztof
