On Wed, 2025-08-06 at 14:13 -0400, Harry Wentland wrote: > > > On 2025-08-06 13:45, Timur Kristóf wrote: > > > > Harry Wentland > > <harry.wentl...@amd.com <mailto:harry.wentl...@amd.com>> ezt írta > > (időpont: 2025. aug. 6., Sze 16:56): > > > > this patch regresses the kms_bw IGT test with Navi 31 and 48 > > with a single 4k60 DP display connected. These subtests fail > > when they should pass: > > > > linear-tiling-2-displays-1920x1080p > > linear-tiling-2-displays-2160x1440p > > linear-tiling-2-displays-2560x1440p > > linear-tiling-2-displays-3840x2160p > > linear-tiling-3-displays-1920x1080p > > linear-tiling-3-displays-2160x1440p > > linear-tiling-3-displays-2560x1440p > > linear-tiling-3-displays-3840x2160p > > linear-tiling-4-displays-1920x1080p > > linear-tiling-4-displays-2160x1440p > > linear-tiling-4-displays-2560x1440p > > linear-tiling-4-displays-3840x2160p > > > > We confirmed with a revert of this patch. > > > > Harry > > > At the moment I don't see what is there that would make a > > difference to Navi 31 or 48. > > > > Not sure either. But that's why we have these tests because > they often catch things that aren't obvious. > > Harry
Hi Harry, I took a look at this issue. I found the code for this igt test here: https://gitlab.freedesktop.org/drm/igt-gpu-tools/-/blob/master/tests/kms_bw.c?ref_type=heads It seems to me that the test sets up a very simple display scenario and tries to check that the CRC isn't zero. However, the variable "igt_crc_t zero" is never initialized (and is also of the wrong type). So what the test is actually doing is comparing the "captured" CRC with a piece of uninitialized memory. That being said, I decided to rewrite my patch and only check the DC connector type for DVI signals to distinguish between DVI-D and DVI-I specifically, and otherwise leave the rest of the code as-is in order to avoid any potential regressions. Does that sound OK to you? Thanks & best regards, Timur
