The timer delta in the driver is calculated using hardware written value to PPHWSP/LRC image and previously stored value. As seen in [1] this difference may become negative for graphics version 12 as a cause of quirk in updating mechanism explained in [2]. This, in turn, causes -EOVERFLOW to be thrown inside live_pphwsp_runtime test.
Similar problem with gen 12 has been observed [3] and addressed in [4], so follow suit and disable the test here as well. [1] https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/16521 [2] HSD #16010904313 [3] https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/13697 [4] "73782fc64793 (i915/gt/selftests: Disable lrc_timestamp test)" Closes: https://gitlab.freedesktop.org/drm/i915/kernel/-/work_items/16521 Signed-off-by: Krzysztof Karas <[email protected]> --- drivers/gpu/drm/i915/gt/selftest_lrc.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/drivers/gpu/drm/i915/gt/selftest_lrc.c b/drivers/gpu/drm/i915/gt/selftest_lrc.c index 23f04f6f8fba..3be59324fdc1 100644 --- a/drivers/gpu/drm/i915/gt/selftest_lrc.c +++ b/drivers/gpu/drm/i915/gt/selftest_lrc.c @@ -1955,6 +1955,15 @@ static int live_pphwsp_runtime(void *arg) enum intel_engine_id id; int err = 0; + /* + * On Gen12 (TGL/DG2) a hardware timestamp updating mechanism + * could write a smaller than previously value to the register, + * so CTX_TIMESTAMP would appear to go backwards, causing + * sporadic underflow detections. + */ + if (GRAPHICS_VER(gt->i915) == 12) + return 0; + /* * Check that cumulative context runtime as stored in the pphwsp[16] * is monotonic. -- 2.34.1
