Hi Chris, On Tue, Sep 03, 2019 at 08:31:51PM +0100, Chris Wilson wrote: > On my systems with lots of memdebug enabled, we would hit the oomkiller > 90% of the time during the initial mlock prior to allocating any objects > (and about 20% of the time lockup / panic). Tweak the target allocation > sizes, and include a few more breadcrumbs tracing the allocations so > that we can reliably start the tests. We still do hit our shrinker and > even the oom notifier, so still achieving its goal of exercising low > memory and swap pressure. > > To slightly compensate for the reduced mempressure (albeit we do not > remove the swapping, the raison d'etre of the test), we increase the > number of threads to force the system to reuse active fences, making it > more stressful on the fence code. Ah... OK :) > Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx> > Cc: Andi Shyti <andi.shyti@xxxxxxxxx> Reviewed-by: Andi Shyti <andi.shyti@xxxxxxxxx> Andi _______________________________________________ Intel-gfx mailing list Intel-gfx@xxxxxxxxxxxxxxxxxxxxx https://lists.freedesktop.org/mailman/listinfo/intel-gfx
