Accidentally left in the hack to run the
"overwrite-source-interruptible" for only one loop, used whilst testing.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Hiding the initial set_bo() required for the "overwrite-source" tests
lead to a nice bit of hilarity as I missed repeating the initialisation
for the multiple loops of the interruptible version of
"overwrite-source".
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Daniel preferred to keep the old tests intact lest we accidentally break
them, and to add the new interruptible tests as new subtests.
In the process also make sure the GPU is idle before starting each loop.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
In order to exercise the bug behind:
commit 22fd5ca947b58901927d100d2b1aa0f1672b3435
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date: Fri Jun 28 16:54:08 2013 +0100
drm/i915: Only clear write-domains after a successful wait-seqno
we need to check for concurrent access with the potential to be
interrupted by a signal. The framework for doing so is already in place,
so just enable it and repeat the tests for longer to give better chance
of being interrupted at just the wrong moment.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
I'd been working under the falsehood that we would always generate an
error for an invalid reloc->target_handle before reserving any object.
However, only the execlist is checked up front for validity before
reservation so ENOSPC is a genuine error condition raised by the test.
Fix it so that we stop reporting that limit as a test failure.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=65391
Limit the broken handles to UINT32_MAX-4096 so that we can be sure that
they do not alias with a valid handle.
References: https://bugs.freedesktop.org/show_bug.cgi?id=65391
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We may fail to set a mode if it fails some hidden constraints, such as
bandwidth on the third pipe. This is expected, so skip testing such
modes.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=66111
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Certain modes may not be supported by certain combinations of pipes.
This is impossible to determine upfront, and we await an atomic
modesetting query operation. In the meantime, if we fail to set a mode,
just skip that test.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=66000
Add a test going through all connectors/crtcs/modes/formats painting to
a front FB with CPU or painting to a back FB with CPU and blitting it
to the front FB.
Only formats understood by cairo are supported for now.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Factor out parts that will be used by an upcoming patch adding
kmstest_create_fb2.
Also call the fb paint functions directly, there is not much
point in passing a function pointer for that.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
This is used by multiple test cases, so make it shared.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
These are used by multiple test cases, so make them shared.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
What this test is interested in is the handling of the LUT for very
large arrays, irrespective of whether such batch are actually
executable. So adjust the pass/fail checks to be explicit in the error
they are looking for, so that we do not conflate memory/aperture
pressure as a failure in the LUT API.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=65391
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Add a double buffer and a single buffer version of the above sequence,
to check if the modeset does a DPMS ON.
Tested on IVB, with and without the relevant kernel fix, got the
expected results.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Currently when exiting with error, we'll get stuck in a DPMS OFF state
if the error happens while we have DPMS OFF set in the test sequence.
This happens even though we switch back to text mode at exit. This might
be a bug in itself to be fixed later, but in any case we want a working
console, so do an explicit DPMS ON.
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Let's not use the 2K variants of the frequencies it does not help
in anything here and the explanations are hopefuly more understandable
this way.
On top of that, I noticed that we can just compute the desired min/max
boundaries for r2 and n2, so use that instead of the two tests to
discard out of range values.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Assume that the budget for those frequencies were tuned after the
reference table was created.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Because the reference frequencies were stored in KHz, we've lost
precision for 37762500Hz and the test is failing. Let's express the
reference frequencies in Hz then.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Oops. This has been broken since:
commit d798ab9c2a092a08e14c6f2dfb1777376ba407a8
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date: Tue Dec 18 22:55:58 2012 +0100
tests: add testcase to check igt runtime enviroment
This is one reason for some of the sporadic kms_flip failures.
One such is https://bugs.freedesktop.org/show_bug.cgi?id=59834.
v2:
- use unsigned long for KDSETMODE/KDGETMODE
- fix passing the parameter to KDGETMODE as it should be by value
- actually testing that it works..
v3:
- don't do an explicit DPMS_ON, only switch to graphics mode.
v4:
- use the newly added drmtest_set_vt_graphics_mode(), which will also
take care of restoring the original mode. (Daniel)
Signed-off-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
-q will prevent grep from writing to stdout and print "root" when make
test is running as root.
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Aside from adding VEBOX support, this test allows allows the "mixed"
test case variant to work on platforms that have 2 rings which was
previously not the case. To the best of my memory, this might therefore
impact G45, and Ironlake.
Signed-off-by: Zhong Li <zhong.li@intel.com>
[Ben: added commit message, remove num_rings = 1, s/LOCACL/LOCAL]
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
1. add functions check kernel enable a ring or not.
2. add function gem_get_num_rings() to check how many rings kernel has
enable.
3. gem_ring_sync_loop.c will call gem_get_num_rings() directly instead
of original static fucntion get_number_rings().
Signed-off-by: Zhong Li <zhong.li@intel.com>
[Ben: Wrapped commit message + whitespace fixes]
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
The code is surround by a #ifdef...#endif to avoid to break compiling against
the current libdrm release
v2 (Ben): Use VEBOX get param. Thankfully Daniel let us carve this out
way back when.
Spacing cleanups
Signed-off-by: Xiang, Haihao <haihao.xiang@intel.com>
Signed-off-by: Zhong Li <zhong.li@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Also add a subtest for the fd=handle_to_fd(), fd2=dup(fd), close(fd)
case (idea from Kristian Høgsberg).
Signed-off-by: Imre Deak <imre.deak@intel.com>
v2:
- add a new subtest instead of modifying the original test (Daniel)
- add a new subtest for testing dup (Kristian)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Make sure the kernel returns EDEADLK when the number of fences is
exceeded for gen2-3. For gen4+ the test makes sure the kernel ignores
the EXEC_OBJECT_NEEDS_FENCE flag.
Note that I changed the code not to round the num_fences to an even
number. Not sure why that was there, and if there's a reason for it,
we need to add it back.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Make sure that debugfs entry works as expected by reading
back the sequence number that was written.
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
gem_tiling_max_stride writes a data pattern to an X-tiled buffer using
the maximum supported stride, reads the data back as linear, and
verifies that the data didn't get scrambled on the way.
The test also checks that some invalid stride values are rejected
properly.
v2: Check invalid strides
v3: Check invalid stride with Y-tiling
Include a few more invalid stride values
Fix gen3 X-tile size
v4: A few more invalid strides :)
Drop the useless memset()
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
lib/drmtest.c provides gem_available_fences(). Use it where
appropriate.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
IVB+ supports 32 fence registers, bump the maximum in the test.
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
With this at least the y-tiled test reliably fails on my machines, but
x-tiled still passes on some. More ideas to tune this highly welcome.
v2: Fill cpu caches with data for each newly allocated bo. This seems
to do the trick on my snb here _really_ reliably. So apparently the
backsnoop for llc gtt writes is the crucial ingredient here to make
the test fail.
While at it, also stop leaking mmap space.
v3: Fixup commit message.
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
