Igt-dev Archive on lore.kernel.org
 help / color / mirror / Atom feed
* [PATCH i-g-t] tests/kms_async_flip: skip subtest if invalid driver-specific condition
@ 2025-04-15 19:47 Melissa Wen
  2025-04-16  1:22 ` ✓ i915.CI.BAT: success for " Patchwork
                   ` (4 more replies)
  0 siblings, 5 replies; 9+ messages in thread
From: Melissa Wen @ 2025-04-15 19:47 UTC (permalink / raw)
  To: Development mailing list for IGT GPU Tools, Petri Latvala,
	Arkadiusz Hiler, Kamil Konieczny, Juha-Pekka Heikkila,
	Bhanuprakash Modem, Ashutosh Dixit, Karthik B S
  Cc: Thadeu Lima de Souza Cascardo, Rodrigo Siqueira, Andre Almeida,
	Alex Hung, Leo Li, Simon Ser, Dmitry Baryshkov, Ville Syrjala,
	Santhosh Reddy Guddati, Chaitanya Kumar Borah, Xaver Hugl,
	kernel-dev

Async page flips can fail for any driver-specific reasons, i.e. when the
driver can't commit to async page flip request for hw limitations, the
commit is rejected with -EINVAL since there is one or more conditions
that make the request invalid. These limitations varies between vendors
and hw versions. Skip a subtest if any invalid driver-specific condition
happens. This behavior is in line with userspace expectations and fits
better the plurality of drivers in the DRM subsystem.

Signed-off-by: Melissa Wen <mwen@igalia.com>

---

Hi,

Recent changes to async page flip conditions and limitations in the DRM
subsystem shed light on unexpected restrictions of async page flip
tests. The current approach doesn't make the test generic enough to
correctly deal with the diversity of reasons for a async page flip being
rejected by drivers of any HW vendor.

As an example, AMD display driver can't perform async flip when the
memory type changes [1][2][3]. After further investigation performed by
my colleague Thadeu Cascardo, he elucidated that this situation can
happen whenever there is not enough space for FBs in VRAM, so they are
allocated in GTT. Moreover, the space available in VRAM varies between
hw families and other system needs. Therefore, the driver rejects the
async flip due to memory type change, instead of silently downgrade the
page flip from async to sync - which seems correct.

The userspace expects the rejection (-EINVAL), but not the downgrade.
So, I understand the test_async_flip should just validate if async page
flips aren't downgrade in acceptable conditions and skip for any invalid
driver-specific condition.

[1] https://lore.kernel.org/amd-gfx/20230621202459.979661-2-andrealmeid@igalia.com/
[2] https://lore.kernel.org/amd-gfx/20230804182054.142988-1-hamza.mahfooz@amd.com/
[3] https://lore.kernel.org/amd-gfx/20250107152855.2953302-17-chiahsuan.chung@amd.com/

Let me know your thoughts!

Melissa

---
 tests/kms_async_flips.c | 8 +++++---
 1 file changed, 5 insertions(+), 3 deletions(-)

diff --git a/tests/kms_async_flips.c b/tests/kms_async_flips.c
index 031417ed8..0de299c2b 100644
--- a/tests/kms_async_flips.c
+++ b/tests/kms_async_flips.c
@@ -397,8 +397,10 @@ static void test_async_flip(data_t *data)
 
 		ret = perform_flip(data, frame, flags);
 
-		if (frame == 1 && data->allow_fail)
-			igt_skip_on(ret == -EINVAL);
+		if (data->allow_fail)
+			igt_skip_on_f(ret == -EINVAL,
+				      "Skipping, async flip not supported at frame %d " \
+				      "due to invalid driver-specific conditions.\n", frame);
 		else
 			igt_assert_eq(ret, 0);
 
@@ -768,7 +770,7 @@ static void run_test(data_t *data, void (*test)(data_t *))
 			continue;
 
 		test_init(data);
-		data->allow_fail = false;
+		data->allow_fail = true;
 		data->modifier = default_modifier(data);
 		igt_dynamic_f("pipe-%s-%s", kmstest_pipe_name(data->pipe), data->output->name) {
 			/*
-- 
2.47.2


^ permalink raw reply related	[flat|nested] 9+ messages in thread
end of thread, other threads:[~2025-05-01 17:59 UTC | newest]

Thread overview: 9+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2025-04-15 19:47 [PATCH i-g-t] tests/kms_async_flip: skip subtest if invalid driver-specific condition Melissa Wen
2025-04-16  1:22 ` ✓ i915.CI.BAT: success for " Patchwork
2025-04-16  1:43 ` ✓ Xe.CI.BAT: " Patchwork
2025-04-16  6:05 ` ✗ i915.CI.Full: failure " Patchwork
2025-04-16 11:59 ` ✗ Xe.CI.Full: " Patchwork
2025-04-17 11:01 ` [PATCH i-g-t] " Karthik B S
2025-04-21 15:04   ` Melissa Wen
2025-04-21 16:59     ` Borah, Chaitanya Kumar
2025-05-01 17:13       ` Melissa Wen

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox