Create the RCS and BCS batches once and reuse them to avoid the
rewriting and relocation overhead on every submission - the goal is to
focus on the interring synchronisation overhead.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
intel_batchbuffer.c: In function 'fill_object':
intel_batchbuffer.c:589:20: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast]
obj->relocs_ptr = (uint64_t)relocs;
^
intel_batchbuffer.c: In function 'exec_blit':
intel_batchbuffer.c:598:21: error: cast from pointer to integer of different size [-Werror=pointer-to-int-cast]
exec.buffers_ptr = (uint64_t)objs;
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Since we are considering implementing read-read optimisations for mixed
engine workloads, make sure that we at least have a few tests that check
for coherency when doing so.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The choice is to either move the igt_require from the buffer allocation
(and allow the allocation to fail) inside the igt_fixture, or move the
buffer allocation to the subtest. Moving it to the subtest has the
advantage of ensuring that every test has the same initial state (no
chance of leaking state across tests) and speeding up running a single
subtest.
Note this is necessary in order to run igt/gem_concurrent_blit on older
kernels, otherwise the requirement for mmap(wc) causes us to skip the
majority of tests.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
New test core_getparams consists of 2 subtests, each one testing
the ability of userspace to query the correct value of a GT config
attribute: subslice total or EU total. drm/i915 implementation of
these queries is required for Cherryview and Gen9+ devices (non-
simulated).
v2: Duplicate small amount of new libdrm functionality to avoid
bumping libdrm version requirement (Daniel). Convert some
igt_asserts to the appropriate comparison variants. Add a
test description.
v3: Actually use the LOCAL GETPARAM defines. Otherwise can't build
against older libdrm as intended by v2.
For: VIZ-4636
Signed-off-by: Jeff McGee <jeff.mcgee@intel.com>
Reviewed-by: Zhigang Gong <zhigang.gong@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
igt_interactive_debug should be defined in igt_core.c, rather than the
header, to avoid it being defined more than once.
Reported-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Thomas Wood <thomas.wood@intel.com>
Fix the following warning:
igt_fb.c: In function 'igt_create_fb_with_bo_size':
igt_fb.c:414:2: warning: format '%llx' expects argument of type
'long long unsigned int', but argument 9 has type 'uint64_t' [-Wformat=]
igt_debug("%s(width=%d, height=%d, format=0x%x [bpp=%d], tiling=%llx, size=%d\n",
introduced by commit:
commit e36091d1c7010e825897dc4487f9985ab353973b
Author: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Date: Tue Mar 3 14:11:01 2015 +0000
tiling: Convert framebuffer helpers to use fb modifiers
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
There's no fencing for those tiling layouts, so we create a linear bo
for cairo to play with, and when cairo is finished with it, we do a fast
copy blit to the fb BO with its final tiling.
v2: Move to correct domain after CPU is done with the object (-EINVAL). (Tvrtko Ursulin)
Correct arguments passed in to framebuffer creation (segfault). (Tvrtko Ursulin)
Pass zero stride to kernel as it expects for Yf&Ys. (Tvrtko Ursulin)
v3: Rebase for gem_mmap__cpu changes. (Tvrtko Ursulin)
v4: Rebase for addfb2.5. (Tvrtko Ursulin)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
This converts the IGT API only, underneath legacy set_tiling is still used.
v2: One got away in kms_flip.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
So we can use it with bare kernel types, without going through libdrm
bos.
v2: Don't forget the object handle. (Tvrtko)
Correct surface pitch calculation. (Tvrtko)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
v2: Adjust for BB handling changes. (Tvrtko Ursulin)
Correct XY_FAST_COPY_DST_TILING_Yf. (Tvrtko Ursulin)
v3: New tiling modes are not defined in the kernel any more. (Tvrtko Ursulin)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Just a few basic tests to make sure fb modifiers can be used and
behave sanely when mixed with the old set_tiling API.
v2:
* Review feedback from Daniel Vetter:
1. Move cap detection into the subtest so skipping works.
2. Added some gtkdoc comments.
3. Two more test cases.
4. Removed unused parts for now.
v3:
* Removed two tests which do not make sense any more after the
fb modifier rewrite.
v4:
* Moved gtkdoc comments into .c file.
* Moved all initialization into fixtures.
* Rebased for fb modifier changes.
v5:
* Added bad modifier subtest.
Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Change the formatting asserts into requires and add the contents into
the debug log.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=89525
The integer comparison macros give us better error output by including
the actual values that failed the comparison.
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
i915 was using the main atomic 'disable plane' to turn off sprite planes
during a CRTC disable. This was problematic because it modified the
plane state, preventing us from recovering the original state later.
One such case was that during a DPMS OFF followed by a DPMS ON, any
sprite planes would not be restored properly.
Let's add a test that toggles DPMS off and on and ensures that the CRC
remains the same (i.e., planes are successfully restored unchanged).
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
I missed the quiet "Log start" between the test failure and the debug
output (and so was very confused by the repetition). You have to shout
at me!
v2: Thomas suggested I make the end of the logging clear as well. Since
I completely missed that there was a end marker, he must be right!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Thomas Wood <thomas.wood@intel.com>
The user should be able to specify a negative timeout to indefinitely
wait upon a bo becoming idle.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
So convert from uint64_t to int64_t. The distinction becomes important
when you realise what test we were missing...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
The gem_render_tiled_blits test tends to get oom killed
on low memory (< 4GB) Android systems. This is because the
test tries to allocate (sysinfo.totalram * 9 / 10) in
buffer objects and the remaining 10% of memory is not
always enough for the Android system.
A similar issue with gem_render_linear_blits was resolved
by creating several subtests. A "basic" subtest that uses
minimal memory buffers to test the basic operation, and
two stress tests which are skipped if there is insufficient
memory. The first stress test uses more memory than the
graphics apperture and the second uses enough to ensure
that swap space is used (if present).
This patch makes the same changes to gem_render_tiled_blits.
v2: Following comments from Daniel Vetter:
a) Use igt_main macro instead of "open coding", and
b) cull some leading spaces
Signed-off-by: Tim Gore <tim.gore@intel.com>
The gem_render_linear_blits test tends to get oom killed
on low memory (< 4GB) Android systems. This is because the
test tries to allocate (sysinfo.totalram * 9 / 10) in
buffer objects and the remaining 10% of memory is not
always enough for the Android system.
After a discussion with Chris Wilson I have split this
test into a "basic" and an "apperture-thrash" subtest,
in the same way as gem_linear_blits. The basic test
uses just two buffer objects and the apperture-thrash
test is skipped if there is insuffiecient memory.
v2: Following comment from Chris Wilson:
a) Remove the command line option for count.
b) Add a third subtest to ensure swap is tested
v3: Replace some leading spaces with tabs
v4: Follwing comment from Daniel Vetter:
a) Use igt_main macro instead of "open coding", and
b) cull some more leading spaces
Signed-off-by: Tim Gore <tim.gore@intel.com>
Add 64bit ptes and 8MB mmiobar offset for gen8
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Acked-by: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
The only thing the kernel can do is pin the buffers, which essentially
means no swapped tiled objects.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
We fail to detect some VGA monitors using our I2C method, leading to
bug reports from QA. If you look at the dmesg of these cases, you'll
see the Kernel complaining about EDID reading mostly FFs and then
disabling bit-mangling. Since we don't want to reimplement everything
the Kernel does, let's just accept the fact that some VGA outputs
won't be properly detected.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=84273
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
igt_kms extensively uses line continuation when dumping state updates
at the debug level. They got badly mangled with the recent changes to
for the log handling functions. Two separate fixes:
- Don't prepend domain and other metainformation when it's just a
continuation line.
- Dont add newlines when dumping the log recorder.
If someone interleaves different log level messages this will go awry,
but really just don't do that.
Cc: Thomas Wood <thomas.wood@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
This way the debug output in case of failures is nicer since we dump
the entire test condition.
Also replace one open-coded igt_assert_eq.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
We've been leaking the framebuffers that get created inside the
legacy -> universal cursor compatibility layer and nobody noticed. Add
an i-g-t test to check debugfs and ensure we end up the same number of
framebuffers we started with after performing cursor operations.
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
The bare "Test requirement: modes" message is too cryptic, I had to go and
read the source code to understand the missing requirement.
Signed-off-by: Marc Herbert <marc.herbert@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Added new media_rc6_residency_subtest for chv & vlv.
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With current code we are not considering the RC6 residency during sysfs
read. This is causing test to fail due to incorrect residency_accuracy check
This patch consider code time spent for accuracy check
Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
