| Age | Commit message (Collapse) | Author |
|---|
|
Data types are defined differently on 32-bit systems, causing gcc to
complain about printf format specifiers not matching the size of the
variables passed in. Use PRIu64 and %zu where appropriate.
Reviewed-by: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Thierry Reding <treding@nvidia.com>
|
|
We need to check that the kernel rejects attempts to pass in
unknown flags.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
Instead of:
(debugfs_test:1499) DEBUG: Reading file "data"
(debugfs_test:1499) DEBUG: Could not open file "data" with error: Input/output error
Print:
(debugfs_test:1360) DEBUG: Entering subdir crtc-2
(debugfs_test:1360) DEBUG: Entering subdir crc
(debugfs_test:1360) DEBUG: Reading file "data"
(debugfs_test:1360) DEBUG: Could not open file "data" with error: Input/output error
(debugfs_test:1360) DEBUG: Reading file "control"
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
|
|
For measuring the cost of preemption, inject a low priority spinner
between the two measurements; the difference between the preemption and
the normal dispatch includes both the cost of the spinner dispatch and
of preempting it. Close enough for us to estimate the cost of
preemption, though we don't measure the cost of preemption on the local
ring!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
CC: Michał Winiarski <michal.winiarski@intel.com>
|
|
If the system has bsd2, we do not know which ring the kernel will alias
I915_EXEC_BSD onto and so we do not what the matching string should be.
Skip the unknown.
v2: Deny the aliased I915_EXEC_BSD exists at all; be specific!
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103324
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
CC: Michał Winiarski <michal.winiarski@intel.com>
|
|
Since the headless subtest wants to compare execution latency of a
headless mode vs a single head, it needs to be able to do a modeset and
ergo it requires DRM_MASTER.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
__gem_create() doesn't touch the outparam *handle on failure, so we can
no longer assert that it zero. This is reasonable to remove as it is just
testing the library itself and not the kernel, so no loss in coverage.
We already had to remove the false assertion that gem_create() must fail
following fd exhaustion (as we can not prevent the kernel from freeing
VFS fdspace in between calls).
The last remaining change is that we do no need to rely on an external
path for open() as dup() will do the job of exhausting the fdtable.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103365
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
Emphasize that we want to test synchronisation using the dmabuf API
(prime_sync_start, prime_sync_end) and so drop the manual
synchronisation using the GEM API (gem_sync).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
|
|
Since the driver's guc submission method requires the breadcrumbs irq
for feeding requests to the guc, we cannot simply simulate a missing irq
by disabling the interrupts.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
kms_atomic_transition was updating already disabled planes and committing
them nonblockingly. This results in sporadic -EBUSY failures because
planes that are unbound have their own timeline.
The solution is not unbinding already unbound planes, making the test
pass. There was also a related kernel bug causing failures in the same
way, but that is now fixed.
While at it, only check for fd completion in the nonblocking case,
in the blocking case it's theoretically possible the commit completes
before we check the that the commit didn't complete, most probably when
the test is run in a debugger.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=102671
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Petri Latvala <petri.latvala@intel.com>
|
|
kmstest_get_crtc was skipping because at that point the crtc was not
active yet, instead we should only use igt_assert_plane_visible
directly. Unexport kmstest_get_crtc, since nothing here should need it.
While at it fix a small leak in igt_assert_plane_visible, the only
remaining user.
Additionally, it' s not allowed to obtain crc on a disabled pipe, so
wait until the pipe is enabled before allocating the crc.
Changes since v1:
- Move igt_pipe_crc_new until after pipe is enabled.
- Move test_init() into test_setup(), it's redundant..
- Remove pipe_crc from data, no need to have it there.
Changes since v2:
- Increment fb pointer for igt_plane_set_fb, in order to set a different
fb on each plane correctly.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
igt_pipe_refresh was added for the forked helper, we could just call it
from the fork to prevent races against MODE_ID.
The static 1024x768 mode is allocated on the stack and returned, which will
result in a garbage mode. Make the fallback mode global to prevent this.
igt_output_set_pipe is unneeded, override_mode is enough to force the
right mode.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103342
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
Now that we can set individual properties through the igt_kms api,
we no longer need to duplicate functionality from igt_kms. Set invalid
properties directly, and rewrite kms_atomic.c to use igt_display.
This will allow us to remove a lot of code in kms_atomic.c,
and benefit from how igt_kms can set up a valid configuration,
instead of having to inherit it from fbcon.
Changes since v1:
- Fix test failure when atomic_invalid_params is run standalone.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Daniel Stone <daniels@collabora.com>
[mlankhorst: Use kmstest_set_connector_dpms (mkahola)]
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
IGT has an api that can find a matching pipe for a given connector,
so use that.
igt_output_override_mode already forces a modeset on an affected pipe
since an earlier commit, so the second call to igt_output_set_pipe
can be removed.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
igt_pipe_get_property is about to be removed, so use
igt_pipe_obj_get_prop instead. This requires adding 2 more properties
to the crtc property list. Also get rid of the Broadcast RGB call,
this is already handled in igt_kms.
Change order for DEGAMMA_LUT and GAMMA_LUT around, else this test will
fail if legacy gamma is set. In the legacy path, this will update
GAMMA_LUT to the new size before DEGAMMA_LUT is set.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
This can be handled by generic properties.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
This allows test to test whether a property is supported, in
a nice and clean way. It removes the need for special case
functions like igt_plane_supports_rotation.
Convert the users of igt_plane_supports_rotation and remove the
extra check in drm_plane_commit, this is already checked below
when setting plane properties.
Changes since v1:
- Use the correct has_prop in kms_crtc_background_color.c
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
Setting broadcast rgb was only used by chamelium, but is now handled
in igt_display by default. This means that chamelium doesn't need to
set it, and this can be hidden. The broadcast RGB property is intel
connector specific, so rename the enum to intel_broadcast_rgb_mode.
Keep the property and enum public in case someone wants to test the
property later.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
With the replace_prop_blob functions we can safely replace the blob for
any property, without having to care about error handling ourselves.
This will for example allow overriding color management blobs, or for
kms_atomic set invalid mode blobs.
The color management blob functions are removed, they can now be
replaced by direct calls to replace the properties.
Changes since v1:
- Fix chamelium tests.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
In the future I want to allow tests to commit more properties,
but for this to work I have to fix all properties to work better
with atomic commit. Instead of special casing each
property make a bitmask for all property changed flags, and try to
commit all properties.
This has been the most involved one, since legacy pipe commit still
handles a lot of the properties differently from the rest.
Changes since v1:
- Dump all changed properties on commit.
- Fix bug in igt_pipe_refresh().
Changes since v2:
- Set pipe ACTIVE property changed flag on init.
Changes since v3:
- Add a missing igt_pipe_refresh() to kms_atomic_interruptible.
Changes since v4:
- Perform error handling when setting custom crtc properties.
Changes since v5:
- Only attempt to commit changes properties.
Changes since v6:
- Clear OUT_FENCE_PTR on succesful commit.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
In the future I want to allow tests to commit more properties,
but for this to work I have to fix all properties to work better
with atomic commit. Instead of special casing each
property make a bitmask for all property changed flags, and try to
commit all properties.
Changes since v1:
- Remove special dumping of src and crtc coordinates.
- Dump all modified coordinates.
Changes since v2:
- Move igt_plane_set_prop_changed up slightly.
Changes since v3:
- Fix wrong ordering of set_position in kms_plane_lowres causing a test failure.
Changes since v4:
- Back out resetting crtc position in igt_plane_set_fb() and
document it during init. Tests appear to rely on it being preserved.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
In the future I want to allow tests to commit more properties,
but for this to work I have to fix all properties to work better
with atomic commit. Instead of special casing each
property make a bitmask for all property changed flags, and try to
commit all properties.
Changs since v1:
- Mention which properties we set to what.
- Assert the property to be set is valid.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Mika Kahola <mika.kahola@intel.com>
|
|
Moving code out of the boost function will allow its usage
in other/new test scenarios.
Signed-off-by: Radoslaw Szwichtenberg <radoslaw.szwichtenberg@intel.com>
Reviewed-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
|
|
Since I accidentally broke the build for some, by putting the pretty
printer for submission inside ifdef HAVE_PROCPS, it's time to move the
whole thing into lib/i915 while fixing this mistake.
Let's also rename the pretty printer and add a doc to it as well as the
section.
Fixes: f6dfe556659f ("lib: Extract helpers for determining submission method")
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Cc: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
We never declared that we were about to read from the mmap after copying
into it using the BLT (a missed call to prime_sync_start); leaving its
coherency ill-defined. For completeness, add the missing
prime_sync_end() as well.
v2: Another prime_sync_start/end around the is-zero check in
test_read_flush()
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103168
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
|
|
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>
|
|
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>
|
|
This was the predecessor test to kms_frontbuffer_tracking, which not
just tests fbc, and also in a much more through-out fashion.
Both tests authored by Paulo, but looking at git history there's a
clear switch with the introduction of kms_frontbuffer_tracking. After
that only maintenance commits to kms_fbc_crc. Not point having
redundant tests, so lets remove the old one.
This will cut a bit more than 3 minutes of test-time from kbl-shard.
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Acked-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Acked-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
|
|
As we use rounding when converting from hw clocks to time, we may report
a residency greater than wallclock (by a small margin), so allow us to
overshoot by 5% (just due to our inaccuracy).
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
|
|
Ignore the unexpected success when the CPU cache is randomly flushed
that makes !llc appear to work without sync. It happens, the cpu cache
is a fickle beast that we do not have sole control over. Instead limit
the test to detect failures when the API is being adhered to.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103168
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
|
|
Another example of something that is used across different tests, and
should be moved to lib.
v2: Break the trend of expanding ioctl_wrappers
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Cc: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Katarzyna Dec <katarzyna.dec@intel.com>
Cc: Petri Latvala <petri.latvala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Katarzyna Dec <katarzyna.dec@intel.com>
Acked-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
|
|
Couple of tests are using either determining scheduler capabilities or
pretty printing. Let's move those to helpers in lib. We can also keep
the value obtained from getparam static.
v2: Break the trend of expanding ioctl_wrappers
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Cc: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Katarzyna Dec <katarzyna.dec@intel.com>
Cc: Petri Latvala <petri.latvala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Katarzyna Dec <katarzyna.dec@intel.com>
Acked-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
|
|
Couple of tests are using either determining submission method, or
pretty printing. Let's move those to helpers in lib.
v2: s/igt_show/gem_show
Signed-off-by: Michał Winiarski <michal.winiarski@intel.com>
Cc: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Katarzyna Dec <katarzyna.dec@intel.com>
Cc: Petri Latvala <petri.latvala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Katarzyna Dec <katarzyna.dec@intel.com>
Acked-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
|
|
For per-engine checks, check the engine actually exists (important for
bsd aliasing).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=101905
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
The first active pipe might not be pipe A, and we shouldn't rely on
fbcon for a working crtc. Use igt_kms to set up something basic, and
use the kmstest helper to get an event from the correct pipe.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
When we suspend, we do not want to be interrupted by a pending reset, or
failure of that reset. After the resume, the errors should be noted as
usual.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Michał Winiarski <michal.winiarski@intel.com>
|
|
If the connected output is disabled before this test is run, then the test
will fail because actual_brightness stays at 0 instead of the value
of the desired brightness. Fix this by reading intel_backlight/device
for the output name, and enable that output at start of this test.
Changes since v1:
- Enable the connected output, instead of enabling all outputs.
Changes since v2:
- Set a dummy fb, just in case.
Changes since v3:
- Use igt_output_get_plane_type, not get_plane.
Sample failure:
IGT-Version: 1.20-g01c550a27986 (x86_64) (Linux: 4.14.0-rc3-fbconhax+ x86_64)
(pm_backlight:1203) CRITICAL: Test assertion failure function test_and_verify, file pm_backlight.c:111:
(pm_backlight:1203) CRITICAL: Failed assertion: ({ typeof(0) _a = (0); typeof(val - tolerance) _b = (val - tolerance); _a > _b ? _a : _b; }) <= result
(pm_backlight:1203) CRITICAL: error: 91200 > 0
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
|
|
fails, v3.
Make sure read_all_entries has all outputs possible enabled, but also
add a test that runs with all outputs disabled.
This will maximize coverage of debugfs reading, and allows the test not
to be dependent on fbcon for setup.
Changes since v1:
- Put framebuffers on all primary planes, to get a crc from i915_sink_src_eDP1.
Changes since v2:
- Use igt_output_get_plane_type, not igt_output_get_plane.
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
[mlankhorst: Add comment suggested by ickle.]
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
Include igt_syncobj.c in libigt so that tests/syncobj_basic actually
builds, and let's also build tests/syncobj_wait.
v2: Also update lib_headers (Petri)
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Petri Latvala <petri.latvala@intel.com>
|
|
Most of the gem_bad_length code is compiled out because creating
a zero-length object is not allowed anymore by i915 and thus it is
not possible to execute it. The remaining part checks that creation of
a zero-length object does indeed fail, which is also checked by
gem_create/create-invalid-size.
The only difference between the 2 tests is that gem_bad_length checks
specifically for EINVAL while create-invalid-size only looks for
failure, so after adding the EINVAL check to create-invalid-size it is
safe to drop gem_bad_length.
v2 (Joonas):
- Clarify header line, split the patch, fix Meson build and add Cc:
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
Replace the local recursive batch execbuf with the equivalent
igt_spin_t.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Arkadiusz Hiler <arkadiusz.hiler@intel.com>
|
|
Closing the sw_sync timeline now signals any remaining fences upon it;
but test_wait_snapshot requires the fence to continue to be busy so that
the __syncobj_wait() will return with -ETIME.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103188
References: ea4d5a270b57 ("dma-buf/sw_sync: force signal all unsignaled fences on dying timeline")
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
|
|
Closing the sw_sync timeline now signals any remaining fences upon it;
but test_wait_interrupted requires the fence to be busy for the
__syncobj_wait() not to immediately return with -ETIME (and so be
interrupted instead).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=103187
References: ea4d5a270b57 ("dma-buf/sw_sync: force signal all unsignaled fences on dying timeline")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jason Ekstrand <jason@jlekstrand.net>
Cc: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
|
|
Remove the bare igt_ioctl(SYNCOBJ_WAIT) in favour of __syncobj_wait()
for tidier error reporting.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jason Ekstrand <jason@jlekstrand.net>
Cc: Dave Airlie <airlied@redhat.com>
Reviewed-by: Jason Ekstrand <jason@jlekstrand.net>
|
|
Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
|
|
This adds both trivial error-checking tests as well as more complex
tests which actually test whether or not waits do what they're supposed
to do. They only currently work on i915 but it should be simple to hook
them up for other drivers by simply implementing the little function
pointer hook provided at the top for triggering a syncobj.
v2:
- Actually add the reset tests.
v3:
- Only do one execbuf for trigger
- Use do_ioctl and do_ioctl_err
- Better check for syncobj support
- Add local_/LOCAL_ defines of things
- Use a timer instead of a pthread
v4:
- Use ioctl wrappers
- Use VGEM instead of i915
- Combine a bunch of the simple tests into one function
v5:
- Combinatorially generate basic tests
- Use sw_sync instead of using vgem directly
- Add even more tests
v6:
- Rebase on the new SYNCOBJ_RESET API
- Add tests for SYNCOBJ_SIGNAL
v7:
- Improve the signal and reset bad pad tests
Reviewed-by: Dave Airlie <airlied@redhat.com>
[airlied: fix build]
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
|
|
Signed-off-by: Jason Ekstrand <jason@jlekstrand.net>
Reviewed-by: Dave Airlie <airlied@redhat.com>
|
|
10bpc formats aren't supported on all platforms, so skip the test when
we can't create the framebuffer.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Petri Latvala <petri.latvala@intel.com>
|
|
At the moment, the verify tests use an extremely brutal write-read of
every dword, degrading performance to UC. If we break those up into
cachelines, we can do a wcb write/read at a time instead, roughly 8x
faster. We lose the accuracy of the forced wcb flushes around every dword,
but we are retaining the overall behaviour of checking reads following
writes instead. To compensate, we do check that a single dword write/read
before using wcb aligned accesses.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
|
|
Not a test that inspects strict ordering of execution, but one that makes
sure that we can survive a small bit of stress. From each cpu we submit a
small number of batches at different priorities to different engines,
with the expectation that they pass through unscathed.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski@intel.com
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
|
