| Age | Commit message (Collapse) | Author |
|---|
|
The test is doing multiple iterations (50). Each iteration tries to
squeeze target number of cursor updates in half a second worth of flips.
If we don't hit the target in any given iteration we bail out early.
Because of that we don't have the data on the number of iterations that
have failed and/or succeeded any given run, which makes hunting down
this elusive issue hard.
Let's change that so we always go through all the iterations and fail at
the end printing out the number of iterations we haven't met the target.
Each failed iteration also logs how many cursor updates it has managed
to do.
Since our target is generated run-time and is also load sensitive, let's
bump the log level of message containing our target to "info" so we can
compare those values across both passes and failures.
Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Cc: Martin Peres <martin.peres@linux.intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Reviewed-by: Martin Peres <martin.peres@linux.intel.com>
|
|
There is no guarantee that spinners are and will be implemented
using batches. As we have igt_spin_t, manipulate it through
igt_spin_* functions consistently and hide the batch nature.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
For the !modeset case, after a pageflip is performed on both crtc's,
vblank counter is advanced by 1 when compared to its original value.
For the modeset case, when pageflips and modesets are interleaved,
vblank counter does not always advance by 1 as nonblocking modeset
could fail, in which case we do not need vblank_matches.
v2: Remove the assert statement instead of changing the condition
Cc: Maarten Lankhorst <maarten.lankhorst@intel.com>
Signed-off-by: Nischala Yelchuri <nischala.yelchuri@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=105454
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Added munmap and replaced hard-coded values with PAGE_SIZE macro.
Cc: Easwar Hariharan <easwar.hariharan@intel.com>
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Signed-off-by: Nischala Yelchuri <nischala.yelchuri@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
Simple rule of thumb, if a kms_* test calls igt_display_init() in its
global fixture, skip the entire test if the driver has disabled KMS.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Antonio Argenziano <antonio.argenziano@intel.com>
|
|
igt_pipe_crc_get_current()
This is a more race free of accomplishing the same.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
|
|
In order to make adding more options easier, expose the full set of
options to the caller.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Petri Latvala <petri.latvala@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
|
|
Collecting CRC may result in a modeset and extra vblank waits. On some
tests this will increase the runtime a lot, so it makes sense to
keep it enabled, and only collect the most recent CRC when needed.
Changes since v1:
- Fix read_crc semantics. (Ville)
Changes since v2:
- Remove EAGAIN assert, can be reached from drain_crc.
Changes since v3:
- Do not infinitely loop in igt_pipe_crc_drain().
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com> #v2
Tested-by: Vidya Srinivas <vidya.srinivas@intel.com>
|
|
Using the fancy new DRM_CAP_CRTC_IN_VBLANK_EVENT cap I can finally
make this test the work I originally intended to.
For the !modeset case that means performing a pageflip on both crtc's,
then requeueing as soon as the event is delivered and then check the
vblank counter against the original value, it should be advanced by 1.
The modeset case is slightly more complicated, ideally it's handled
the same, but if we can't perform a modeset and pageflip at the same
time, fall back to queueing both in a single commit, in which case
we can say nothing about the vblank counter.
There is a small race between flip_done and hw_done, so make
flip_nonblocking retry for a second when encountering -EBUSY.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101634
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
Instead of assuming each subtest cleans up after itself, assume it
fails and doesn't. Now that igt_kms can clean up stale events, we
can just force each subtest to only clean up its framebuffers,
which isn't harmful if it failed.
The nonblocking modeset test is removed, this was a workaround only
needed for intel because it supported nonblocking commit only for the
!modeset case, this has been fixed.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
General update to reflect current state of things.
Signed-off-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Acked-by: Radoslaw Szwichtenberg <radoslaw.szwichtenberg@intel.com>
Acked-by: Petri Latvala <petri.latvala@intel.com>
|
|
igt_output_set_pipe with PIPE_ANY used to mean that we bind the output
to any pipe, but this is now a deprecated alias for PIPE_NONE, and
means the output will be unbound.
Because of this it's better to change output->pending_crtc_idx_mask to
an enum pipe, because only a single choice may be given for a pipe.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
Most of these tests have no reason to look at those members,
so try other ways of getting the information.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
Remove the local recursive spinner in favour of igt_spin_t.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
|
|
Collecting CRC may force a modeset, which is a bad idea after we just
forced a hang. The hang is intended to make sure the page flip doesn't
complete before the cursor, making sure that works.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102259
Cc: Marta Lofstedt <marta.lofstedt@intel.com>
Tested-by: Marta Lofstedt <marta.lofstedt@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[mlankhorst: Prevent overflow when collecting crcs, based on Villes feedback]
|
|
We are, the build system takes care of that.
Reviewed-by: Eric Anholt <eric@anholt.net>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Acked-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Acked-by: Petri Latvala <petri.latvala@intel.com>
Acked-by: Daniel Stone <daniels@collabora.com>
Acked-by: Radoslaw Szwichtenberg <radoslaw.szwichtenberg@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
|
|
for_each_pipe cannot be used for enumerating testcases, so provide
something that can.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Petri Latvala <petri.latvala@intel.com>
|
|
Signed-off-by: Gustavo Padovan <gustavo.padovan@collabora.com>
Reviewed-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
|
|
Name should not be driver-specific.
Signed-off-by: Leo (Sunpeng) Li <sunpeng.li@amd.com>
Reviewed-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
|
|
This reverts commit 7c8703fb02b248c2bcf9756bba8812bcfe7ed5d3.
If we expect it to fail until we find a solution, let the hw fail and
continue to track the known failure in CI/bugs.
Cc: Martin Peres <martin.peres@linux.intel.com>
Acked-by: Martin Peres <martin.peres@linux.intel.com>
|
|
This appears to be required on SNB, or else we miss a lot of interrupts.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Also allow the initial vblank evasion to fail. This was the most
likely cause of the test failure, so be sure to fix it too.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
flip-before/after-cursor.
Run the test 25 times and fail if we miss too many vblanks.
CI receives too many false positives so we have to fix it.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
If we can't use the GPU to make a framebuffer busy, skip the test.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
This reverts commit 25fbae15262cf570e207e62f50e7c5233e06bc67, restoring
commit 301ad44cdf1b868b1ab89096721da91fa8541fdc
Author: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Date: Thu Mar 2 10:37:11 2017 +0100
lib: Open debugfs files for the given DRM device
with fixes.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
This reverts commit 301ad44cdf1b868b1ab89096721da91fa8541fdc.
When a render-only device is opened and gem_quiescent_gpu is called, we
need to use the debugfs dir for the master device instead.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
|
|
When opening a DRM debugfs file, locate the right path based on the
given DRM device FD.
This is needed so, in setups with more than one DRM device, any
operations on debugfs files affect the expected DRM device.
v2: - rebased and fixed new API additions
v3: - updated chamelium test, which was missed previously
- use the minor of the device for the debugfs path, not the major
- have a proper exit handler for calling igt_hpd_storm_reset with the
right device fd.
Signed-off-by: Tomeu Vizoso <tomeu.vizoso@collabora.com>
Reviewed-by: Robert Foss <robert.foss@collabora.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
VLV is a sloth and so doing anything extra between the first and last
get_vblank() is likely to increase the chance of failing the test.
Let's move the make_busy() stuff to happen before the first get_vblank()
to reduce the amount of work done there.
These tests are still failing quite often on my VLV, but that is mostly
caused by igt_kms's tendency to still re-read all the connector properties
and whatnot as part of the commit operation. I suspect that fixing that
would eliminate most of the spurious failures.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
|
|
Add changes reflecting the new support for dynamic number of planes per pipe.
Signed-off-by: Robert Foss <robert.foss@collabora.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
This is required for ILK style watermark programming, which currently
inserts an extra vblank wait. This allows the tests to pass on those
platforms.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
This forces the cursor fastpath to be used. While at it, add -atomic
versions of the test to compare legacy pageflip vs atomic.
The nonblocking modeset test is changed slightly, due to cdclk changes
the page flip may be stalled so it's not useful to test it. Just test
that a queued page flip + nonblocking modeset works as intended and
there is no kernel spew.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Trying to run as many cursor updates during a nonblocking modeset won't
work, because both crtc's might be involved in the modeset. Instead
try to run as many cursor updates on both crtc's at the same time,
while trying to queue page flips in between.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
This fixes the test on older platforms that don't have ARGB support.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Nail the cursor update thread to a single cpu, and run a idle busy loop
on the same cpu. This will force it to the highest cpu speed, which will
eliminate cpu speed variations and allows the test to pass.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Kept in the previous commit by accident.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Because in the ideal case we currently wait 50% of the time for every frame,
on some machines cpu scaling gets in the way when we idle a lot and will
cause the machine to throttle to a lower cpu speed. This causes a
failure because we may end up missing vblanks.
Work around this by only running for 1/4th of the time at max speed,
even on low cpu speeds it will be less likely to run into issues then.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
This was running for 150 seconds before, reduce it to 25.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Some tests may wait a vblank on master. Instead of
failing the tests, allow them to succeed if 1 vblank
passed.
Also no longer warn if a page flip is missed, change it to
info instead. As long as no more than 25% of the vblanks are
missed the tests will pass.
Also fix nonblocking modeset vs cursor, which needs an array
of cursor argument. This was a nice buffer overrun before.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
cursor hiding
This way the tests will pass with the legacy cursor hack.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
In case of a nonblocking modeset, the cursor update should block for
the modeset to finish.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
It's more readable like this, and with the ifs it was a completely
separate test anyway.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
make_busy() already existed!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
This is a test for the issue exposed by 97888, we perform a busy
bo update followed by a cursor update.
The page flip and cursor update should complete without hanging,
and the new fb should not be flipped yet as a result of the cursor
update.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Suggested-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
The essence of the basic test is that neither the cursor nor the
nonblocking flip stall the application of the next.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
On skylake there's a add_all_affected_planes call that will make
atomic commit use the old cursor state. Add a test that first does
a page flip then cursor update to expose the issue. The test currently
fails on the vblank wait, but that's a common issue affecting all the
tests.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Since we are experiencing too much noise in BAT from what just looks
like scheduling delays in inspecting the vblank, reduce the basic test
to the fundamentla: check that the cursor ioctl following the nonblocking
flip/modeset occurs within the same vblank. Hopefully, CI + debug builds
are fast enough to do get-vblank; flip; cursor; get-vblank within a
single vblank period.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
When vrefresh is 30 the tests run for twice as long, set it to
vrefresh to make the tests just as long anywhere
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
|
This probably won't pass by design on platforms that need to reallocate
global resources like watermarks on gen9+, but it's a good test anyway.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
|
