Peter Senna Tschudin authored 3 weeks ago and Kamil Konieczny committed 3 weeks ago

This patch modifies igt_runner to support runner_kmemleak() calls. By
default, kmemleak scanning is disabled, so new command-line options are
introduced to enable it:
 * -k, -k<option>, --kmemleak, --kmemleak=<option>

The available options are:
 * once: Do single kmemleak scan after last test in the test list
 * each: Perform a kmemleak scan after each test completes

By default, kmemleak scanning is disabled. If any kmemleaks are
detected, they will be saved in the igt_runner results directory under
kmemleak.txt.

Additionally, this patch updates serialize_settings() and
read_settings_from_file() to persist igt_runner settings across runs.
This allows settings to be saved when running igt_runner --dry-run and
later restored when executing igt_resume.

The unit tests for igt_runner have been extended to verify:
 * Kmemleak scans are disabled by default
 * Kmemleak scans can be enabled via command-line arguments
 * The kmemleak setting is correctly saved to and restored from disk

To test the new -k command-line option, this patch appends "--overwrite"
to *argv[] in runner_test.c to expand the argument array. This approach
avoids a major refactor of how *argv[] is defined across the file while
keeping the changes isolated to unit testing. Since this only affects
tests, there is no downstream impact.

Cc: vitaly.prosyak@amd.com
Cc: christian.koenig@amd.com
Cc: alexander.deucher@amd.com
Cc: jesse.zhang@amd.com
Cc: harry.wentland@amd.com
Cc: zbigniew.kempczynski@intel.com
Cc: kamil.konieczny@linux.intel.com
Cc: ryszard.knop@intel.com
Cc: lucas.demarchi@intel.com
Cc: katarzyna.piecielska@intel.com
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Vitaly Prosyak <vitaly.prosyak@amd.com>
Signed-off-by: Peter Senna Tschudin <peter.senna@linux.intel.com>

Peter Senna Tschudin authored 3 weeks ago and Kamil Konieczny committed 3 weeks ago

Adds a simple library for interacting with kmemleak and add
unit testing for the library. There are two modes intended to
integrate with igt_runner:
 - once: collect kmemleaks after all test completed
 - each: collect kmemleaks after eachb test completes

Cc: vitaly.prosyak@amd.com
Cc: christian.koenig@amd.com
Cc: alexander.deucher@amd.com
Cc: jesse.zhang@amd.com
Cc: harry.wentland@amd.com
Cc: zbigniew.kempczynski@intel.com
Cc: kamil.konieczny@linux.intel.com
Cc: ryszard.knop@intel.com
Cc: lucas.demarchi@intel.com
Cc: katarzyna.piecielska@intel.com
Reviewed-by: Zbigniew Kempczyński <zbigniew.kempczynski@intel.com>
Reviewed-by: Jonathan Cavitt <jonathan.cavitt@intel.com>
Reviewed-by: Vitaly Prosyak <vitaly.prosyak@amd.com>
Signed-off-by: Peter Senna Tschudin <peter.senna@linux.intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Vinod Govindapillai authored 3 weeks ago and Karthik B S committed 3 weeks ago

Even if FBC is enabled and test is successfull, using warn level
to print the info will make the CI treat this as a failure. In
case FBC is not enabled, the assert will be triggered and print
the error case.

Signed-off-by: Vinod Govindapillai <vinod.govindapillai@intel.com>
Reviewed-by: Jeevan B <jeevan.b@intel.com>

Sai Teja Pottumuttu authored 3 weeks ago and Ashutosh Dixit committed 3 weeks ago

oa-formats test need not be limited to render engine only. Extend
it to use any engine using __for_one_hwe_in_oag macro instead.

Signed-off-by: Sai Teja Pottumuttu <sai.teja.pottumuttu@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>

Pravalika Gurram authored 3 weeks ago and Kamil Konieczny committed 3 weeks ago

Use the fault injection infrastructure to make targeted internal KMD
functions fail when executing xe_exec_queue_create_ioctl()
so that more code paths are tested, such as error handling and unwinding.
include xe_exec_queue_create_bind only for VM_BIND engine class.
xe_exec_queue_destroy_ioctl() and
xe_exec_queue_get_property_ioctl() are not considered as there is
no unwinding code to test with fault injection.

v2: Added separate case for vm_bind engine class

Signed-off-by: Pravalika Gurram <pravalika.gurram@intel.com>
Reviewed-by: Priyanka Dandamudi <priyanka.dandamudi@intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Marcin Bernatowicz authored 3 weeks ago and Kamil Konieczny committed 3 weeks ago

Commit d7bccece44dd ("tests/intel/kms_dp_link_training: Restrict to Intel
GPUs") missed adding DRIVER_XE. Ensure both DRIVER_INTEL and DRIVER_XE
are set to properly restrict the test to Intel GPUs.

Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz@linux.intel.com>
Cc: Kamil Konieczny <kamil.konieczny@linux.intel.com>
Cc: Kunal Joshi <kunal1.joshi@intel.com>
Cc: Arun R Murthy <arun.r.murthy@intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Marcin Bernatowicz authored 3 weeks ago and Kamil Konieczny committed 3 weeks ago

Use DRIVER_INTEL instead of DRIVER_ANY to avoid assertions on non-Intel
hardware. In few of our CI runs test opened any DRM driver and failed
because it was not Intel.

Cc: Kunal Joshi <kunal1.joshi@intel.com>
Cc: Arun R Murthy <arun.r.murthy@intel.com>
Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz@linux.intel.com>
Reviewed-by: Kunal Joshi <kunal1.joshi@intel.com>

Ville Syrjälä authored 4 weeks ago and Kamil Konieczny committed 3 weeks ago

The test only checks igt_display_has_format_mod() before
deciding to test a specific format. But that doesn't
mean the actual plane supports that format. Also check
for that once we know which plane to use.

Should turn the fp16 failures into skips on IVB because it
doesn't have fp16 support on the primary plane.

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

Simulation runs are slow, so account for that by modifying the scope of
the tests.

- Set parameters specific to simulation during init.
- Skip tests that are heavily reliant on timing and that only test SW
  functionality.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Internally the test is running a loop of 1000 iterations and calling
read in each iteration. The test is expecting a single read call to
return at least two reports. If two reports are not read, the logic
continues to the next iteration. This can easily fail if each read just
returns one report. In essence, many reports are available, but the
logic causes test to fail. Instead, count and use every report read
when waiting for two reports to be available.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

If the device is a Virtual Function, skip certain devcoredump validation
steps that are not applicable in this context.

These changes align with kernel updates introduced in:

Link: https://patchwork.freedesktop.org/patch/msgid/20241022010116.342240-2-zhanjun.dong@intel.com



Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz@linux.intel.com>
Reviewed-by: Zhanjun Dong <zhanjun.dong@intel.com>
Cc: Lukasz Laguna <lukasz.laguna@intel.com>
Cc: Michał Wajdeczko <michal.wajdeczko@intel.com>
Cc: Zhanjun Dong <zhanjun.dong@intel.com>

Introduce intel_is_vf_device helper to determine whether a given device
is a Virtual Function (VF). The function reads the VF_CAPABILITY_REGISTER
and checks if the least significant bit is set to identify VF device.

Signed-off-by: Marcin Bernatowicz <marcin.bernatowicz@linux.intel.com>
Reviewed-by: Lukasz Laguna <lukasz.laguna@intel.com>
Cc: Adam Miszczak <adam.miszczak@linux.intel.com>
Cc: Jakub Kolakowski <jakub1.kolakowski@intel.com>
Cc: Lukasz Laguna <lukasz.laguna@intel.com>
Cc: Michał Wajdeczko <michal.wajdeczko@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Cc: Piotr Piórkowski <piotr.piorkowski@intel.com>
Cc: Satyanarayana K V P <satyanarayana.k.v.p@intel.com>
Cc: Tomasz Lis <tomasz.lis@intel.com>

Eudebug is not supported in PF mode with VFs enabled and in VF mode.
Provide subtests to ensure that:
  1. enabling eudebug is not permitted in PF mode with VFs enabled
  2. eudebug sysfs toggle is not available in VF mode
  3. enabling VFs is not permitted when eudebug is enabled

Reviewed-by: Dominik Karol Piątkowski <dominik.karol.piatkowski@intel.com>
Reviewed-by: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
Link: https://lore.kernel.org/r/20250304163227.49986-3-christoph.manszewski@intel.com


Signed-off-by: Christoph Manszewski <christoph.manszewski@intel.com>

Export __xe_eudebug_enable_getset to make it possible to catch error
code in tests.

Reviewed-by: Dominik Karol Piątkowski <dominik.karol.piatkowski@intel.com>
Link: https://lore.kernel.org/r/20250304163227.49986-2-christoph.manszewski@intel.com


Signed-off-by: Christoph Manszewski <christoph.manszewski@intel.com>

Katarzyna Piecielska authored 1 month ago and Kamil Konieczny committed 3 weeks ago

During making changes in Mega feature <-> Sub-category pairs I've
noticed few places were test documentation formatting is different
than pattern:

	/**
	 * TEST:
	 * Category:
	 * Mega feature:
	 * Sub-category:
	 * Functionality:
	 * Test category:
	 *
	 * SUBTEST:
	 * Functionality:
	 * Description:
	*/

So let's align Xe tests to this order.

Also there is no need to duplicate documentation fields in subtests
if Category/Mega feature/Sub-category/Functionality are the same
for subtest as for whole binary.

Last thing is to make sure test description is properly adjusted.

Signed-off-by: Katarzyna Piecielska <katarzyna.piecielska@intel.com>
Reviewed-by: Aditya Chauhan <aditya.chauhan@intel.com>

Katarzyna Piecielska authored 1 month ago and Kamil Konieczny committed 3 weeks ago

Our documentation scripts need to have unique pairs 'Mega
feature - Sub-category', e.g. example below is causing warnings:

Mega feature: WMTP
Sub-category: CMD submission
&&
Mega feature: General Core features
Sub-category: CMD submission

There is a need to make sure that pairs are unique, so let's make change
in xe_compute* tests.

Signed-off-by: Katarzyna Piecielska <katarzyna.piecielska@intel.com>
Reviewed-by: Aditya Chauhan <aditya.chauhan@intel.com>

Using igt_assert_eq() to compare floats was incorrect and led to
false positives, replace it with the appropriate assert.

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

Define min and max oa exponents for the oa exponent test so that they
can be set differently on slow platforms.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

The MI_STORE command needs modification to set the right amount of
dwords.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

unprivileged-single-ctx-counters was hardcoded for a format size of 256
bytes. Fix it up for generic format size.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Keep it simple and just check if enable/disable is working correctly
using the read call and existing uapi.

v2: Drop mmio and use the value returned in read (Ashutosh)
v3: Checkpatch issues and set polled mode (Ashutosh)

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Introduce oa buffer size test separately. Pick a random valid buffer
size for the test.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Use a default buffer size for non-zero-reason for and strip
out testing buffer size in the same test.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Simplify the buffer-fill test for quicker runs on slow platforms.

We only want to test that the BUFFER OVERFLOW status is set when we do
not read the OA stream data in time. To do so, keeping reading zero
bytes of data until you hit a buffer overflow.

v2:
- Add reason to the commit msg
- Drop 'wraparound' from test description
- s/disable/disabled
- drop unnecessary double-bang for int to bool conversion
- Fix regression in test execution time

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

The polling-small-buf test is specifically testing that POLLIN is still
set after an ENOSPC error. In the buggy code, POLLIN was cleared on
the ENOSPC error even though there was data to be read. The test was
putting an upper bound on the number of bytes read in a fixed duration
of time. While this worked, we had occassional failures that were due to
scheduling and other dependencies outside the scope of this test.

In an attempt to make the test more robust,
- use a more definitive method to wait for few reports
- force an ENOSPC error using a small buffer
- Ensure POLLIN is still set in the subsequent poll() call

v2:
- Add "why" in commit message (Ashutosh)
- Use uapi to wait for N reports

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Use the same width and height for render copy frame to make execution
times uniform across tests.

v2: Update commit message (Ashutosh)

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Some tests are not dependent on the oa_exponent value, so just use the
default exponent for such tests, so that the tests can be tuned to run
in reasonable time on slow platforms.

- oa-tlb-validate
- short-reads
- stress-open-close
- mmio-triggered-reports

v2: Explain "why" in commit message (Ashutosh)

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

On slower platforms, the idea is to set the default oa_exponent to a
smaller value so that test execution is faster. Rename the global
exponent to default so it can be set to platform-specific value.

v2: Add more detail in commit message (Ashutosh)

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Rename the argument in max_oa_exponent_for_period_lte() to period_ns to
specify the units of the period.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Drop unused for helper

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Use static for mmio_data and hwe.

Signed-off-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Harish Chegondi authored 1 month ago and Ashutosh Dixit committed 3 weeks ago

A new hardware feature first introduced in PVC gives capability to
periodically sample EU stall state and record counts for different stall
reasons, on a per IP basis, aggregate across all EUs in a subslice and
record the samples in a buffer in each subslice. Eventually, the aggregated
data is written out to a buffer in the memory. This feature is also
supported in XE2 and later architecture GPUs.

Add tests to test EU stall sampling functionality in the Xe driver.
These tests accept several inputs from the user, enable EU stall counters,
run a given workload on a child process while the parent process reads
the stall data and parses the data. The EU stall counters are disabled
once the workload completes execution.

If the user doesn't provide any input workload, GPGPU fill is used as
the workload. gpgpu_fill() and related functions have been reused from
xe_gpgpu_fill.c.

v2: Addressed review feedback from v1
    User can pass an output file to write the EU stall data
    xe_drm.h file changes have been moved into a separate patch

Signed-off-by: Harish Chegondi <harish.chegondi@intel.com>
Reviewed-by: Ashutosh Dixit <ashutosh.dixit@intel.com>

Dominik Grzegorzek authored 1 month ago and Christoph Manszewski committed 3 weeks ago

Limit number of GRFs used by single thread to utilize all
possible threads per EU.

Signed-off-by: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
Reviewed-by: Christoph Manszewski <christoph.manszewski@intel.com>
Link: https://lore.kernel.org/r/20250226104436.257298-2-dominik.grzegorzek@intel.com

Dominik Grzegorzek authored 1 month ago and Christoph Manszewski committed 3 weeks ago

For eu debug testing purposes we need utilize all possible threads per
eu. This is possible only if we limit number of GRFs per single thread.
Add gpgpu_shader interface which allow us to setup that during pipeline
creation. For now define only only mode with 96 grfs.

Signed-off-by: Dominik Grzegorzek <dominik.grzegorzek@intel.com>
Reviewed-by: Christoph Manszewski <christoph.manszewski@intel.com>
Link: https://lore.kernel.org/r/20250226104436.257298-1-dominik.grzegorzek@intel.com

Align with kernel commit cd5bbb2532f2 ("drm/xe/uapi: Add a device query to
get EU stall sampling information").

Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Tejas Upadhyay authored 1 month ago and Lucas De Marchi committed 4 weeks ago

KMD introduced low latency hint support for GT frequency
to guc. Lets add test to validate the feature support.

KMD patch:https://patchwork.freedesktop.org/patch/639677/



Signed-off-by: Tejas Upadhyay <tejas.upadhyay@intel.com>
Link: https://lore.kernel.org/r/20250227122117.711646-1-tejas.upadhyay@intel.com


Signed-off-by: Lucas De Marchi <lucas.demarchi@intel.com>

MDThasleem authored 2 years ago and SANTHOSH GUDDATI committed 4 weeks ago

This negative test was crated to validate ENOSPC when two 2k-4k or 4k-4k
monitors connected through MST to cover bandwidth issue in MST config.

But it will not helpfult to handle current bandwidth which are
DisplayPort (DP) 1.4 supports a total bandwidth of 4 × 8.1 Gbps.

However, with DP 2.1, the bandwidth allocation strategy has
significantly changed to 4 × 20 Gbps of available bandwidth.
This provides significantly greater display capabilities
w.r.t bandwidth.

v2: Fix compilation issue.
v3: Rebase chagnes.

Signed-off-by: Mohammed Thasleem <mohammed.thasleem@intel.com>
Reviewed-by: Jeevan B <jeevan.b@intel.com>

Introduce a small tool for measing the display enging
memory bandwidth utilization. Generally this is available
on SNB+, except on TGL/derivatives where the relevant
registers weren't updated to cope with the new ABOX layout
in the hardware.

Quite handy for confirming that FBC/CCS/etc. are doing their
job.

Not 100% sure about the required scaling factor because
bspec claims it's only needed for MTL, but my ADL definitely
needs it already.

v2: Fix long option handling

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Vinod Govindapillai <vinod.govindapillai@intel.com>

Tvrtko Ursulin authored 1 month ago and Kamil Konieczny committed 1 month ago

Let Valgrind know which memory is initialised by ioctls so running tests
under it becomes less noisy.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>

Tvrtko Ursulin authored 1 month ago and Kamil Konieczny committed 1 month ago

Explicitly initialise ioctl arguments so Valgrind can know we are not
passing in uninitialised memory to the kernel.

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@igalia.com>
Reviewed-by: Kamil Konieczny <kamil.konieczny@linux.intel.com>