created framebuffers were uselessly painted black which just wasted
time from ci.

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

Shrink subtest can end up with oom killing it. Create subprocess
and run it from there so it will at least get reported like:

dynamic child 0 pid:70254 died with signal 9, Killed
Subtest shrink: FAIL (23.906s)

dmesg reports:

[103335.337309] Out of memory: Killed process 70254 (i915_suspend)

Cc: Riana Tauro <riana.tauro@intel.com>
Cc: Anshuman Gupta <anshuman.gupta@intel.com>
Signed-off-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>
Reviewed-by: Sai Gowtham Ch <sai.gowtham.ch@intel.com>

If any of *-without-i915 subtests fails or skips for any reason, it may
leave the i915 module unloaded while keeping our device list populated
with initially collected data.  In a follow up igt_fixture section we then
try to reopen the device.  If the test has been executed with a device
filter specified, an attempt to open the device finds a matching entry
that belongs to the no longer existing device in that initially collected
device list, fails to stat() it, concludes that's because of the device
having been already open, and returns an error.  While that error,
triggered after subtests completion, doesn't affect results of the
subtest, reported by CI togethger with those results it is confusing to
users reviewing those reports.

Fix this issue by refreshing the potentially outdated device list before
continuing with drm_open_driver() if we've been called with a device
filter specified.

While being at it, add a comment that explains why we call
igt_devices_scan() from __igt_device_card_match() but don't force device
rescan, and emit a debug message if we fail in _is_already_opened() on
unsuccessful device stat().

v2: don't free the device list -- we can't tell if it has been populated,
    and igt_devices_free() fails if it hasn't,
  - commit message updated, description improved.

Subtest basic-s3-without-i915: FAIL (9.572s)
(i915_suspend:9050) drmtest-WARNING: card maching filter 0 is already opened
(i915_suspend:9050) drmtest-CRITICAL: Test abort in function drm_open_driver, file ../lib/drmtest.c:639:
(i915_suspend:9050) drmtest-CRITICAL: abort condition: fd < 0
(i915_suspend:9050) drmtest-CRITICAL: Last errno: 2, No such file or directory
(i915_suspend:9050) drmtest-CRITICAL: No known gpu found for chipset flags 0x1 (intel)
Test i915_suspend failed.
**** DEBUG ****
(i915_suspend:9050) drmtest-DEBUG: Looking for devices to open using filter 0: pci:vendor=intel,device=dg2
(i915_suspend:9050) drmtest-DEBUG: Filter matched /dev/dri/card0 | /dev/dri/renderD128
(i915_suspend:9050) drmtest-WARNING: card maching filter 0 is already opened
(i915_suspend:9050) drmtest-CRITICAL: Test abort in function drm_open_driver, file ../lib/drmtest.c:639:
(i915_suspend:9050) drmtest-CRITICAL: abort condition: fd < 0
(i915_suspend:9050) drmtest-CRITICAL: Last errno: 2, No such file or directory
(i915_suspend:9050) drmtest-CRITICAL: No known gpu found for chipset flags 0x1 (intel)
(i915_suspend:9050) igt_core-INFO: Stack trace:
(i915_suspend:9050) igt_core-INFO:   #0 ../lib/igt_core.c:2066 __igt_abort()
(i915_suspend:9050) igt_core-INFO:   #1 ../lib/drmtest.c:573 drm_open_driver()
(i915_suspend:9050) igt_core-INFO:   #2 ../tests/i915/i915_suspend.c:258 __igt_unique____real_main245()
(i915_suspend:9050) igt_core-INFO:   #3 ../tests/i915/i915_suspend.c:245 main()
(i915_suspend:9050) igt_core-INFO:   #4 ../sysdeps/nptl/libc_start_call_main.h:58 __libc_start_call_main()
(i915_suspend:9050) igt_core-INFO:   #5 ../csu/libc-start.c:128 __libc_start_main@@GLIBC_2.34()
(i915_suspend:9050) igt_core-INFO:   #6 [_start+0x2a]
****  END  ****

Fixes: f7aff600 ("tests/i915/i915_suspend: Disable d3cold_allowed for basic-s2idle-without-i915")
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Cc: Riana Tauro <riana.tauro@intel.com>
Link: https://patchwork.freedesktop.org/patch/522501/


Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Remove local 64-bit hash and use common one defined in igt_map.

Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab@kernel.org>

Reduce code duplication by using common hash helpers from igt_map.

Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab@kernel.org>

Creating igt_map requires hashing and comparing function. Common case
is using 32-bit (like fd, etc.) or 64-bit keys (offsets) so adding
such helpers reduces code duplication.

Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Reviewed-by: Mauro Carvalho Chehab <mchehab@kernel.org>

For separate execution there is no risk we may reserve offset which is
still in use by some job. But for running bunch of allocator tests
which may be still on gpu such risk exists. To avoid failures on
grouped execution ensure for 'nopin' subtests gpu is idle first.

Signed-off-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

This reverts commit e6bb684a.

Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>

After introducing the dynamic subtests, tests are getting timedout
on CI. This patch will partially reverts dynamic subtests to avoid
such timeouts.

Fixes: c02f0d48 (tests/kms_cursor_legacy: Convert tests to dynamic)
Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>

Negative test is added to ensure HDR is not enabled on
a non-HDR panel. Commit/modeset won't fail. However, in dmesg
we can find

[drm:drm_hdmi_infoframe_set_hdr_metadata [drm_display_helper]]
EOTF Not Supported

v2: -fixed nitpicks (Kamil)

Signed-off-by: Swati Sharma <swati2.sharma@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Acked-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Negative test is added in which HDR output metadata prop is
set with invalid sizes.

v2: -Fixed nitpicks (Kamil)
    -Add cleanup (Kamil)

Signed-off-by: Swati Sharma <swati2.sharma@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Acked-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

This test can work not only on render engines, as its current code
suggests, but also on compute class engines.  Add them to the set of
engine classes supported by the test, in case a future platform has only
those and no RCS available.

Suggested-by: Chris Wilson <chris.p.wilson@linux.intel.com>
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

This also adds coverage for codepaths related to reading back the
devcore file from sysfs, to help catch issues like
drm/msm#20



Signed-off-by: Rob Clark <robdclark@chromium.org>

Adds a test to various subtests to stress shrinker/eviction.  Various
subtests also add mmap and dma-buf mmap into the mix (the latter because
it uncovered a memory corruption bug due to page mappings not being
correctly shot down).

Signed-off-by: Rob Clark <robdclark@chromium.org>

Get rid of some copy/pasta and open coding of a common pattern, waiting
for GPU commands to complete.

Signed-off-by: Rob Clark <robdclark@chromium.org>

Some qcom devices support CTM but not gamma/degamma LUT.

Note, the remaining failures on msm are due to CTM being supported on
one CRTC but not the other.  But the property is exposed on both.

Signed-off-by: Rob Clark <robdclark@chromium.org>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>

Tests need to be documented, as otherwise its goal will be
lost with time. Keeping documentation out of the sources is also
not such a good idea, as they tend to bitrot.

So, add a script to allow keeping the documentation inlined, and
add tools to verify if the documentation has gaps.

Signed-off-by: Mauro Carvalho Chehab <mchehab@kernel.org>
Acked-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>

Zack Rusin authored 2 years ago and Kamil Konieczny committed 2 years ago

Because igt always tries to load i915 driver first for DRIVER_ANY every
test using DRIVER_ANY on anything but i915 starts by printing a warning
that i915 couldn't be loaded.
Lets switch it to a general info line because it makes it hard to spot
actual warnings when every subtest starts by warning about not being
able to load i915.

v1: changed commit title to lib/igt_kmod (Kamil)

Signed-off-by: Zack Rusin <zackr@vmware.com>
Acked-by: Petri Latvala <petri.latvala@intel.com>

Christopher Snowhill authored 2 years ago and Kamil Konieczny committed 2 years ago

Add label and abbreviated display name for the compute engine class.

Signed-off-by: Christopher Snowhill <kode54@gmail.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Users reported oopses on list corruptions when using i915 perf with a
number of concurrently running graphics applications.  That indicates we
are currently missing some important tests for such scenarios.  Cover
that gap.

Root cause analysis pointed out to an issue in barrier processing code and
its interaction with perf replacing kernel contexts' active barriers with
its own requests.

Add a new test intended for exercising intentionally racy barrier tasks
list processing and its interaction with other i915 subsystems.  As a
first subtest, add one that exercises the interaction of remote requests
with barrier tasks list handling, especially barrier preallocate / acquire
operations performed during context first pin / last unpin.

The code is partially inspired by Chris Wilson's igt@perf@open-race
subtest, which I was not able to get an Ack for from upstream.

v4: fix typo in test description and make it generic so it will not need
    to be changed soon (Kamil),
  - rename workload functions instead of providing name wrappers (Kamil),
  - no need for all physical engines to be tested (Kamil).
v3: don't add the new subtest to gem_ctx_exec which occurred blocklisted,
    create a new test hosting the new subtest, update commit descripion,
  - prepare parameters for perf open still in the main thread to avoid
    test failures on platforms with no perf support (will skip now),
  - call perf open with OA buffer reports disabled, this will make sure
    that the perf API doesn't unnecessarily enable the OA unit, while the
    test still runs the targeted code (Umesh),
  - replace additional code for OA exponent calculations with a reasonable
    hardcoded value (Umesh).
v2: convert to a separate subtest, not a variant of another one (that has
    been dropped from the series),
  - move the subtest out of tests/i915/perf.c (Ashutosh), add it to
    tests/i915/gem_ctx_exec.c,
  - don't touch lib/i915/perf.c (Umesh, Ashutosh), duplicate reused code
    from tests/i915/perf.c in tests/i915/gem_ctx_exec.c.

References: https://gitlab.freedesktop.org/drm/intel/-/issues/6333
Cc: Chris Wilson <chris.p.wilson@intel.com>
Cc: Ashutosh Dixit <ashutosh.dixit@intel.com>
Cc: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Link: https://patchwork.freedesktop.org/patch/522916/


[janusz: close fd on exit from intel_context_first_pin_last_unpin_loop]
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>
Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@linux.intel.com>

In newer hardware versions (i.e. display version >= 14), the second
scaler doesn't support downscaling. Current driver design in the
case of 2 plane scaling scenario is if plane1-US and plane2-DS,
it's reject for now. That's why new tests are added for plane1-DS
and plane2-US, so that different DS+US combinations can be validated.

v2: -minor fix
v3: -change from if-else ladder to switch (JP)

Signed-off-by: Swati Sharma <swati2.sharma@intel.com>
Reviewed-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>

Maíra Canal authored 2 years ago and Kamil Konieczny committed 2 years ago

Add three subtests that combine CL jobs and CSD jobs, to assure the
proper synchronization of different queues, especially the
independence between them. Moreover, tests the relationship between
single syncobjs and multisync using mixed jobs as well.

Reviewed-by: Melissa Wen <mwen@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>

Maíra Canal authored 2 years ago and Kamil Konieczny committed 2 years ago

Add subtests for the DRM_IOCTL_V3D_SUBMIT_CSD, which ensures
that improper parameters return an errno and test multisync and
single sync abilities. For most of the subtests, the empty shader is the
base of the submission, as it is one of the simplest jobs possible,
allowing it to test the synchronization abilities of the V3D.

Reviewed-by: Melissa Wen <mwen@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>

Maíra Canal authored 2 years ago and Kamil Konieczny committed 2 years ago

In order to submit a compute dispatch job, a BO must contain the
assembly shader that corresponds to the job. Therefore, create a helper
to encapsulate a simple compute dispatch job. This helper sets the
configurations (cfg) needed for the job and has the assembled instructions
necessary to process an empty shader, just like the following one:

    #version 310 es
    layout (local_size_x = 1) in;
    void main (void) {}

Reviewed-by: Melissa Wen <mwen@igalia.com>
Signed-off-by: Maíra Canal <mcanal@igalia.com>

Use a helper that checks blitter command flags instead of directly
checking Flat CCS compression support. Delete the function that checks
it. Check for extended block-copy command when initializing the
copy object in block_copy and block_multicopy tests.

Signed-off-by: Karolina Stolarek <karolina.stolarek@intel.com>
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>

Add a predicate that checks if a command has a specific property.
Introduce helpers that check if a platform uses an extended
version of the block-copy command or supports compression. Add
a getter for blt_cmd_info to intel_cmds_info library. Group
functions together in i915_blt header file.

Signed-off-by: Karolina Stolarek <karolina.stolarek@intel.com>
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>

Blitter commands may have different properties depending on the
platform. Add a new field to blt_cmd_info struct to describe
their properties. Update block-copy definitions for DG2 and MTL
to use that field.

Signed-off-by: Karolina Stolarek <karolina.stolarek@intel.com>
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>

The structs here describe the properties of copy commands, not
just tiling formats they support. Make the library name more
generic. Rename blt_tiling_info struct to blt_cmd_info to match
this change.

Signed-off-by: Karolina Stolarek <karolina.stolarek@intel.com>
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>

The struct stores information on more than just one blitter cmd,
so the current name is slightly confusing. Also, we might want
to store information on other copy commands, like the one provided
by render. Change it to intel_cmds_info to emphasize that it
describes available commands on the platform. Change the field
name in intel_device_info and update related definitions to refer
to commands, not tiling.

Signed-off-by: Karolina Stolarek <karolina.stolarek@intel.com>
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>

Nidhi gupta authored 2 years ago and Bhanuprakash Modem committed 2 years ago

To optimize the test execution time on hardware and simulation,
limit the execution to two (first & last) pipes and 2 planes
(first & last).

This patch will also provide an option (command line flag '-e') to
execute on all pipes and planes.

Example: ./kms_plane_alpha_blend -e --run-subtest alpha-7efc

V2: Edited commit message (Bhanu)
V3: New function for simulation constraints (Kamil)
    Update commit message (Bhanu)
v4: Restrict the execution only on 2 pipes for
    real hardware aswell (Ankit)
v5: Provide extended option for plane also (Bhanu)
    Move pipe declaration inside fixture (Ankit)

Signed-off-by: Nidhi Gupta <nidhi1.gupta@intel.com>
Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Reviewed-by: Ankit Nautiyal <ankit.k.nautiyal@intel.com>

Instead of clubbing all draw methods, have a separate subtest for
each draw method.

Fixes: 9e71c27c (tests/i915/kms_draw_crc: Convert tests to dynamic)
Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>

To optimize the test execution time, limit the execution to
two (first & last) pipes. This patch will also provide an
option (command line flag '--extended') to execute on all
pipes.

Example:
./kms_cursor_edge_walk --extended --run-subtest 64x64-left-edge

Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Signed-off-by: Nidhi Gupta <nidhi1.gupta@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>

Instead of clubbing all dynamic subtests, let's have an individual
subtest for each size & type.

Fixes: 54ff3199 (tests/kms_cursor_edge_walk: Convert tests to dynamic)
Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>

As we are using same function for cleanup, no need to do other
stuff, just return after unsetting the output to pipe.

V2: - New helper to unset output to pipe (Karthik)

Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>

Due to the more number of iterations, "modest-transition" tests are
getting timed out on pre-si CI. Reducing the number of iterations to
half can save a lot of execution time.

V2: - Separate patch for cleanup

Signed-off-by: Bhanuprakash Modem <bhanuprakash.modem@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>

Make the colored bars full height so that we can actually see
the async flip activity on every scanline, as opposed to just
being able to see it in the middle of the screen.

Cc: Arun R Murthy <arun.r.murthy@intel.com>
Reviewed-by: Karthik B S <karthik.b.s@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

Remeber to call igt_put_cairo_ctx() after rendering to make
sure the results actually end up in the framebuffer (in case
the rendering happens to a temporary buffer and there needs
to be a copy back at the end).

Cc: Arun R Murthy <arun.r.murthy@intel.com>
Reviewed-by: Arun R Murthy <arun.r.murthy@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>

Add skeleton for adding invalid parameter tests and add
two tests which are expected to return -EINVAL or -ERANGE

Signed-off-by: Juha-Pekka Heikkila <juhapekka.heikkila@gmail.com>
Reviewed-by: Luca Coelho <luciano.coelho@intel.com>

Riana Tauro authored 2 years ago and Ashutosh Dixit committed 2 years ago

add a test that verifies the i915 pmu directory and
reads all the events

v2 : use igt_assert_eq instead of igt_assert
     check errno after loop (Umesh)

Signed-off-by: Riana Tauro <riana.tauro@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>

Add CSV output mode.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Caleb Callaway <caleb.callaway@intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>