Release 1.1

testdisplay: shut up compiler

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: move drm_fourcc.h include under the #ifdef protection

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/testdisplay.c: make the 3th pipe finish all the mode setting.

If test all modes rather than preferred mode, remove framebuffer and
set CRTC to zero after each connector mode setting.

Signed-off-by: Sun Yi <yi.sun@intel.com>
Reviewed-by: Eugeni Dodonov <eugeni.dodonov@intel.com>

intel-decode: fix flush dword post sync parse

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>

tests/gem_double_irq_loop: silence the compiler

gem_double_irq_loop.c: In function ‘dummy_reloc_loop’:
gem_double_irq_loop.c:62:9: warning: unused variable ‘j’ [-Wunused-variable]

tests/gem_exec_blt: silence the compiler by failing on error

gem_exec_blt.c: In function ‘gem_exec’:
gem_exec_blt.c:174:6: warning: variable ‘ret’ set but not used [-Wunused-but-set-variable]

Propagate the failure and exit(1).

tests/gem_exec_nop: silence the compiler by failing on error

gem_exec_nop.c: In function ‘exec’:
gem_exec_nop.c:101:6: warning: variable ‘ret’ set but not used [-Wunused-but-set-variable]

Propagate the failure and exit(1).

tests/gem_stress: silence the compiler by using '%zu' for size_t

gem_stress.c: In function ‘main’:
gem_stress.c:980:3: warning: format ‘%lu’ expects argument of type ‘long unsigned int’, but argument 4 has type ‘unsigned int’ [-Wformat]

tests/gem_largeobject: silence the compiler

gem_largeobject.c: In function ‘test_large_object’:
gem_largeobject.c:95:20: warning: cast from pointer to integer of different size [-Wpointer-to-int-cast]

testdisplay: don't test planes if i915 color key ioctl is not available

scripts/who: Add the process id to the output

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

scripts: Add a trivial script to show which programs are using the GPU

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

decode: Fix interpretation of i915 8bit formats

2/3 are reserved with the correct values for A8/I8 being 4/5
respectively.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

testdisplay: update for final plane & fb ABI

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>

tests/gem_partial_pwrite_pread: don't trash gtt unnecessarily

On chips that don't have a unmappable gtt part it's utterly pointless.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add gem_partial_pwrite_pread

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_reloc_vs_gpu: fixup this mess

- fixup pitch mess noticed by Chris Wilson.
- we need to diable bo reuse otherwise the kernel won't do the relocs
  (we might get an already gtt-bound batch from the dummy load).

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_reloc_vs_gpu: emit special batch on blt ring on gen6+

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_reloc_vs_gpu: switch to COLOR_BLT instead of PIPE_CONTROL

Less fancy command that works everywhere. Suggested by Chris Wilons.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_tiled_pread_pwrite: kill set-but-unsed warning

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add gem_reloc_vs_gpu

Tests whether the kernel properly waits for the gpu before
applying a reloc.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Revert some accidental things

... somehow slipped in.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add gem_exec_faulting_reloc

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add gem_exec_bad_domains

To check whether the kernel properly rejects non-gpu domains in
relocs.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add gem_tiled_pread_pwrite

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

improve README

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_gtt_speed: Add modes that force that to the gtt domain

No sane userspace doesn't upload data to the gpu without actually
using it there ...

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Add missing executables to .gitignore.

tests: implement snb+ XY_SETUP_CLIP_BLT workaround

Better safe than sorry.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/sysfs_edit_reader: posixify

Debian's default sh complains ...

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: properly terminate tests when values mismatch

Not much use running them in the test rig otherwise.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_hang: fix copy-paste problem

I suspect that we should do different things for different pipes..

Spotted by Thomas Jarosh on #intel-gfx freenode.

Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>

tests: add a test for checking edid reading delays

With base on EDID timing testing, when we take more than 1s to run
xrandr command, something is wrong.. So add this test for testing the time
we take to read the status of all the connectors from sysfs. It should do
us an average picture of how long we'd take to run xrandr (roughtly 2x
that value).

Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>

tests: add test for kernel segmentation fault

As asked by Daniel Vetter, this is a tech which checks if we can cause
division by zero in kernel by reading the i915_emon_status debugfs
entry repeatably.

Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>

intel_gpu_top: use debugfs interface for forcewake

Using debugfs facilities for forcewake-related stuff.

Acked-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>

tests: disable pipe_control on i965g

That hw is broken.

Reported-by: Dave Airlie <airlied@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Call Arrandale gen5 not gen4.

testdisplay: Round tiled allocations up to pot stride and fence size

Be simple and use the strictest requirements from gen2/3 with old
kernels so that this simply works everywhere.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42585

testdisplay: find planes for each pipe

Fixes a bug where we were masking against the wrong value.

testdisplay: update to new addfb2 definition

cargo-cult the build system into shape

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

always require libdrm

... and also add the missing files to lib/Makefile.am

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_tiled_pread: fixup detiler for gen2

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_largeobject: accomodate for puny apertures

... like on my i855gm.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

i-g-t/debugger - make the debugger work with distcheck

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>

i-g-t/debugger: Only allow the shader debugger if the user has specified a gen4asm

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>

tests: mark gem_bad_address as HANG

Kills my little i855gm.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/debugfs_reader: fight bashism

debian's dash doesn't like this.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: actually add the drm client check script

Ooops.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: check for other drm clients

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_bad_length: disable broken batchbuffer

Submitting random crap hangs the gpu. Suprise!

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: don't run kernel check from make check

Add a new target test for that.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tools: script to gather bug report data

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

make distcheck fixes

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add test to read all debugfs files

Catches a bunch of forcewake warnings on snb+

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: compile fix for !DRM_PLANES

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add gem_unref_active_buffers

The crucial trick to reproduce the bug is that we need to have
a decent pile of active bos to retire. Because we unref the bo
after having it moved off the active list, our recursion depth
in fdo bug #42180 is limited by the number of active objects that
can retire at the same time.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=42180
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: allow overlay plane to be moved & scaled at runtime

For additional testing.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>

intel_reg_dumper: handle 3 pipe configs when dumping HDMI config

Could be on pipe A, B, or C.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>

testdisplay: add stub disable_planes() function

To fix the compile with non-plane libdrms.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>

tools/decode: decode XY_SCANLINES_BLT

tools/decode: decode XY_SETUP_* cmds

And extract decode code for br00 and br01.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: add sprite scaling and location support

Allow testdisplay to locate the sprite on the screen and potentially
scale it using different crtc width/height vs the source width/height
(determined by the resolution on the target pipe).

Also fix exit, making sure we properly disable all the planes.

Signed-off-by: Jesse Barnes <jbarnes@virtuougseek.org>

tools/intel_error_decode: decode gen4+ fences

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tools/decode: properly decode gen6+ PIPE_CONTROL flags

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

top: Avoid FPE by removing meaningless metric

intel_reg_checker: New tool to accumulate checks for HW workarounds.

The goal of this tool is to be usable as a test for whether something
(suspend/resume, GPU reset, bugs, whatever) may have lost various
required hardware setup specified in the docs.

Reviewed-By: Eugeni Dodonov <eugeni.dodonov@intel.com>

intel/decode: print out chipset gen

... instead of i965+ for almost everything that Intel is still selling.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tools/decode: don't move around the display register section

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tools/decode: don't forget to print the name of the last ring

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Allow intel_dump_decode to read data from stdin.

Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>

tools/intel_dump_decode: make devid parsing more useful

We use hex for these!

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tools/decode: retain the ring name

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add gem_hangcheck_forcewake

Provokes the forcewake warning when the hangcheck runs and no
one waits for the gpu (and hence holds the dev->struct_mutex).

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_fence_trash: some retuning

We actually want to test lru behaviour, so do a bit of work with
the fence before yielding to the next thread (we use twice as many
fences as there are, so yielding always is pretty bad, no matter how
clever our fence stealing).

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_*_blits: reduce buffer count to not trash swap

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Allow to output statistics to stdout.

This allows intel_gpu_top to run both in statistics-collecting mode
(collecting the per-ring statistics in gnuplot-friendly format) and
ncurses top-like mode at the same time.

It also allows to output the statistics directly to stdout, by using "-o
-", so the results can be parsed directly via a popen() parsing.

If you are using intel_gpu_top as previously (without any command-line
arguments), it should change nothing for you. If you were using its
logging facilities (e.g., the '-o file'), note that the logging will keep
running, but the detailed top-like interface will be on the screen at the
same time.

Signed-off-by: Eugeni Dodonov <eugeni.dodonov@intel.com>

bios_reader: Allow forcing the device ID with DEVICE=0xNNNN

As a pleasant side effect you can now feed it an opregion dump and it'll
decode without crashing.

Signed-off-by: Adam Jackson <ajax@redhat.com>

intel_reg_dumper: use none instead of NULL if no port is assigned to TRANS_DP

Makes the output a little more readable.

Treat PPT like CPT as we do in the kernel

Set the pch type to CPT if we detect a PPT as well.

tests: disable storedw tests on !render

According to docs, this should work with ppgtt and just convert to
noops otherwise. In reality, it hangs the machine.

So just disable these for the moment.

Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tools/decode: add missing shift to MI_SEM_MBOX decoder

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: return 0 when successful

We use the "ret" variable to check function return values, and these
values are not necessarily 0 on the successful cases, so we need to
reset "ret" after we finish using it.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

getstats: remove useless assertion

getstats.c:48:2: warning: comparison of unsigned expression >= 0 is
always true

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm_open_any: explain why we are abort()ing

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: pimped error msg]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: remove wrong return statement

It seems that the kernel patches required by the TEST_PLANES feature are
still not upstream, so I didn't test this patch.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: don't enter the main loop if dump_info or test_all_modes

If calls inside update_display fail, the function returns and we don't
quit the program if dump_info or test_all_modes. So we enter the main
loop and keep waiting for user input, even on cases where we are not
supposed to require user input. To fix this, we move the check to
outside the update_display function.  As a side effect, we also do
cleanup instead of just exit(0) and we return 1 in case update_display
failed.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: test all modes if no option is provided

Previously, when called without any arguments, the application required
user input to finish.  However, testdisplay is ran by "make check", and
it is not a good idea to run a program that requires user input in "make
check". So we change the default behavior to something that doesn't
require user input. You can still access the previous behavior using the
"-m" argument.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Add missing files to .gitignore

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

testdisplay: fix compiler warnings

testdisplay.c:117:1: warning: comparison between signed and unsigned
integer expressions
testdisplay.c:125:1: warning: comparison between signed and unsigned
integer expressions
testdisplay.c:145:1: warning: comparison between signed and unsigned
integer expressions
testdisplay.c:1060:16: warning: comparison between signed and unsigned
integer expressions

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_mmap_gtt: also test gtt pwrite paths

This needs a properly pre-faulted dst bo.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: build all tests by default

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: separate testcases in gem_bad_length

Creating a zero-length bo should fail, so check that. Still run
the minimal batchbuffer, but without the zero-length reloc.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: disable gem_bad_blit

Hangs gen3 and simply writes garbage into the unmappable part of
gtt on gen4+, which might cause issues later on.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_bad_length: dont throw garbage at the gpu

It doesn't like it. Really.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

intel: we don't do distcheck for the debugger

intel_gpu_dump: really kill it

This is getting a bit hilarious ...

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

intel_gpu_dump: also kill manpage

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tools: remove intel_gpu_dump

Superseded by the kernel's i915_error_state dumping facility.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>