This is meant to test the sysfs entry for showing rc6 residency in
milliseconds. Remember, sysfs is a permanent interface.
v2: use new get_card interface to try "all" devices
check rc6p and rc6pp in addition to rc6
v3: rename rc6_residency.c to sysfs_rc6_residency.c
print better error messages
skip test if rc6 isn't enabled
v4: update to new sysfs names
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Designed to exercise this patch to i915.ko:
diff --git a/drivers/gpu/drm/i915/i915_gem.c b/drivers/gpu/drm/i915/i915_gem.c
index fbf1118..57ae1f2 100644
--- a/drivers/gpu/drm/i915/i915_gem.c
+++ b/drivers/gpu/drm/i915/i915_gem.c
@@ -3181,9 +3181,11 @@ i915_gem_object_set_to_cpu_domain(struct drm_i915_gem_obj
if (ret)
return ret;
- ret = i915_gem_object_wait_rendering(obj);
- if (ret)
- return ret;
+ if (write || obj->pending_gpu_write) {
+ ret = i915_gem_object_wait_rendering(obj);
+ if (ret)
+ return ret;
+ }
i915_gem_object_flush_gtt_write_domain(obj);
By exercising the conditions whereby should either of the checks be missed
an error is detected.
Used to check that i915_error_state works. Unfortunately this
kills the gpu because wedging is permanent.
Base on a patch by Ben Widawsky.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
This removes cut and pasted code and uses a more central source.
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Currently all we see is gem_read: ret == 0 failed, where it would help
to see the errno and/or the ret.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
MI_MEM_VIRTUAL actually means use global gtt now, not setting the bit
means use ppgtt. On previous gens, not setting the bits ment 'use
physical memory'. So what, the usual confusion.
Note that for some odd reasong this is broken on gen6, but only on the
bsd ring. Unexpected.
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
... but actually run the test with reduced memory requirements,
as the messages claims. Additional print it to stdout, stderr
seems to imply FAIL in our QA's testing rig.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Without enabling largefile support the prototype for mmap64() is broken
on Linux/x86_64 with the result being a 32-bit integer sign-extended
to fill a 64-bit pointer. Hilarity ensues.
This allows to specify '-d' parameter which will decode individual bits in
each register being read.
The register bits are printed horizontally for space reasons. This
requires more than 80x25 terminal to see them all. An alternative solution
would be to print them vertically, but this will become much more
difficult to read when printing multiple registers at the same time.
v2: fix spacing to get us a bit closer to the code nirvana.
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
The registers must be passed on the command line and will be read
sequentially, one at a time.
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
This will allow us to pass more options to it in the future.
v2: fix whitespacing issues and improve scary warning text as suggested by
Paul Menzel.
Reviewed-by: Paul Menzel <paulepanter@users.sourceforge.net>
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Astonishing how dense I sometimes am ...
- increase the bo size so that we don't have any round-up to next tile
size areas (which the 2d blit go over, but the 1d pread/pwrite calls
will notice).
- correctly set tiling bits when copying back to a linear buffer.
- read back through a tiled bo to avoid messing with swizzling.
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Oops, my mistake for copying a compile fix from another machine...
If we can't detect how much RAM we have, 0 is not a suitable default, so
keep the error for the time being and only downgrade the error for
unknown swap.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Double-include ftl and local variable shadowing. While fixing the
later I've noticed that we mix up width and height in the blt copy
function.
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
On SandyBridge, the BLT commands were split from the RENDER commands as
well as the BSD split inherited from Ironlake. So we need to make sure
we do exercise each ring, and in order to do so we also need to make
sure each batch takes longer to execute than it takes for us to
submit it.
v2: Exercise each ring sequentially.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
I was interested in finding why my IVB system is not getting GPU turbo
after suspend/resume. The piece that looks weird to me is that
INTERRUPT_THRESHOLD is sitting at 0, whereas pre-suspend it's
0x12000000.
The tool allows you to change the panel fitter settings, so you can
change the size of the screen being displayed on your monitor without
changing the real pixel size of your desktop. The biggest use case for
this tool is to work around overscan done by TVs and some monitors in
interlaced mode.
v2: reviews by Ben, Chris and Rodrigo
- don't install
- use intel_register_access_init
- check for maximum X and Y values
- add a disclaimer saying this is not the real solution
- print less when pf is disabled and option '-l' is used
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Before this patch, handling dump files was wrong:
- when HAS_PCH_SPLIT was specified, intel_reg_dumper segfaulted inside
intel_check_pch()
- the "devid" variable was used but not set
- there was no way to specify the device id of the machine used to
generate the dump file
This patch fixes this behavior with the following changes:
- the HAS_PCH_SPLIT variable is gone
- there is now a '-d' argument that can be used to specify the device id
used to interpret the results
- when a dump file is used but the '-d' argument is not provided, an
Ironlake machine is assumed
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
We need to use _INSTRUCTION as the reloc domain because otherwise the
ppgtt pipe_control w/a for snb won't kick in and the test fails.
The storedw tests for blt and bsd are still disabled because the
corresponding patch to flag ppgtt support isn't merged upstream yet.
Without ppgtt these hang my snb here.
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
