commit cf93bc8df9f5f94a0aedc8b52bad0ad4e036737c by Alan Coopersmith
(introduced to enable solaris builds) reqires HAVE_LINUX_KD_H to be set
in order for kd.h to get picked up in igt_kms.c. This is not currently
set in the Android makefile so Android builds are broken.
Signed-off-by: Tim Gore <tim.gore@intel.com>
Signed-off-by: Thomas Wood <thomas.wood@intel.com>
intel_gpu_frequency was added in commit 5fb26d1 (intel_gpu_frequency: A
tool to manipulate Intel GPU frequency), but wasn't added to .gitignore.
Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Thomas Wood <thomas.wood@intel.com>
Add a parameter for the size of the act_upon array in the parse function
since its size cannot be calculated with ARRAY_SIZE from just the
pointer.
Cc: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Thomas Wood <thomas.wood@intel.com>
In order to exercise coherency across swapin/swapout of the same bo,
explicitly loop over all bo a couple of times.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
With the introduction of mmap(wc) and its unbound GTT write domain, we
can now hit the warning inside set-cache-level, complaining about the
failure to do correct cpu cache tracking.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Unlike a GTT mmap, a WC mmap does not have a direct reference to the
object, only to the backing storage. If we want to control the domain
correctly for mmap(wc), we have to keep the bo reference around.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=88356
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
WARNING: very minimally tested
In general you should not need this tool. Its primary purpose is for
benchmarking, and for debugging performance issues.
For many kernel releases now sysfs has supported reading and writing the GPU
frequency. Therefore, this tool provides no new functionality. What it does
provide is an easy to package (for distros) tool that handles the most common
scenarios.
v2:
Get rid of -f from the usage message (Jordan)
Add space before [-s (Jordan)
Add a -c/--custom example (Jordan)
Add a setting for resetting to hardware default (Ken)
Replicate examples in commit message in the source code. (me)
v3:
Its not It's (me)
Add --help/-h to usage
Add Version + man page
Rename tool to intel_gpu_frequency, from intel_frequency
Remove "sudo" from the examples
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Jordan Justen <jordan.l.justen@intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Here are some sample usages:
$ intel_gpu_frequency --get=cur,min,max,eff
cur: 200 MHz
min: 200 MHz
RP1: 200 MHz
max: 1200 MHz
$ intel_gpu_frequency -g
cur: 200 MHz
min: 200 MHz
RP1: 200 MHz
max: 1200 MHz
$ intel_gpu_frequency -geff
RP1: 200 MHz
$ intel_gpu_frequency --set min=300
$ intel_gpu_frequency --get min
cur: 300 MHz
min: 300 MHz
RP1: 200 MHz
max: 1200 MHz
$ intel_gpu_frequency --custom max=900
$ intel_gpu_frequency --get max
cur: 300 MHz
min: 300 MHz
RP1: 200 MHz
max: 900 MHz
$ intel_gpu_frequency --max
$ intel_gpu_frequency -g
cur: 1200 MHz
min: 1200 MHz
RP1: 200 MHz
max: 1200 MHz
$ intel_gpu_frequency -e
$ intel_gpu_frequency -g
cur: 200 MHz
min: 200 MHz
RP1: 200 MHz
max: 200 MHz
$ intel_gpu_frequency --max
$ intel_gpu_frequency -g
cur: 1200 MHz
min: 1200 MHz
RP1: 200 MHz
max: 1200 MHz
$ intel_gpu_frequency --min
$ intel_gpu_frequency -g
cur: 200 MHz
min: 200 MHz
RP1: 200 MHz
max: 200 MHz
We should be able to execute batches up to the full GTT size (give or
take fragmentation), so let's try!
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
After setting up the copy operations, add a hanging batch. This should
mean that we complete the copy and the compare then races against the
GEM reset. Hopefully, this will catch driver bugs where the target
object is no longer accessible after the hang.
Note: hang injection is disabled until the required kernel interface is
completed. But there are useful additional tests here...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
On some systems (ok, most systems!) we may need to enlarge the allowed
number of open files in order to create enough fd to fill the aperture.
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=87572
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Raw system calls aren't portable to other kernels.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Check for the sysconf value used here, not the one used in the
previous function.
Signed-off-by: Alan Coopersmith <alan.coopersmith@oracle.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
A new subtest added to validate the new version of gem_mmap ioctl,
for creating the wc mappings, on yet to be supported flags.
v2: Removed the flags checking for older kernels (Daniel)
Signed-off-by: Akash Goel <akash.goel@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This exercises both the wc mmappings and the extended get_tiling ioctl.
Userspace cannot handle bit17 swizzling through wc mmaps (because bit17
requires swizzling based on the actual physical address of the page -
which is unknown to userspace) and so we need an extended get_tiling
ioctl to report the actual as well as the logical swizzling on an
object. We then check that the contents of the object are tiled and
swizzled correctly when viewed through a wc mmap.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
intel-gpu-tools ships a #!/bin/sh script that has bash-specific syntax
(which breaks on distros such as Debian and Ubuntu where /bin/sh is a
symlink to something other than bash).
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=87888
The kworker threads are used for flip handling and other non-userspace
driver tasks. They are non-blocking and so do not impact upon how
userspace performs, but they do obscure that information in the
overview.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Don't use a negative index into the array if the desired element is
negative, just wrap around properly into the ring for the chart.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
We already allocate enough objects to thrash the ppGTT VMs, so allow us
to reuse the batch buffers for some efficiency gains and through the
contention more towards the ctx->vm.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Play nice, especially with the subtest, and wait for the children to
exit before finishing the test. If we don't we end up with a fork bomb
for some unknown reason...
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Now that we are no longer busy-spinning inside random(), we can spend
more time exercising i915.ko
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
random() being a good multithread-safe RNG is too slow to be used in
stress tests, especially for a seemingly trivial task of randomising the
order of an array.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
This reverts the effect of two commits
06fb6c233dd82aac766aa9206644f6eff668ca99
264e1ac10ac14a098a78cc9f96c4e7cabb124ee5
Both of these were to stop demos/intel_sprite_on
from being built, but the first was just broken.
So this commit re-enables building intel_sprite_on.
However, intel_sprite_on will not build in recent
Android trees. To overcome this the version
of IGT kept in the Android repository will carry a patch
to intel_sprite_on, and the automatic build test of
IGT on android will patch the freedesktop code on the fly.
Signed-off-by: Tim Gore <tim.gore@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Just to make sure this keeps working since a patch to WARN_ON using
dumb buffers in execbuf was accidentally merged.
Signed-off-by: Daniel Vetter <daniel.vetter@intel.com>
intel_sprite_on wont build on Android. Previous
attempt to disable was just wrong!
Signed-off-by: Tim Gore <tim.gore@intel.com>
Signed-off-by: Thomas Wood <thomas.wood@intel.com>
