tests/gem_exec_bad_domains: enable conflicting write domains test

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

lib: fix the fix for gen5 workaround emmision

Fix the regression introduced in

commit bfbe813f8fb587017c4e1d73c51395c2837eb395
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Tue May 29 22:14:06 2012 +0200

    lib: fix gen5 workaround emission

Signed-off-by: Imre Deak <imre.deak@intel.com>

lib/intel_batchbuffer: remove code w/o effect

Introduced when refactoring the patch in

commit c1ee0bb53269ded7b79966d081518d689639bac7
Author: Imre Deak <imre.deak@intel.com>
Date:   Mon Jul 29 16:43:31 2013 +0300

    intel_batchbuffer: add support for non-32bit blt copies

No functional change.

Signed-off-by: Imre Deak <imre.deak@intel.com>

tests/gem_pipe_control_store_loop: Add subtest for reused buffers

This exercises the slightly faulty kernel w/a that Eric fixed in

commit e844b990b1df9242bb91b7d490552f3198946838
Author: Eric Anholt <eric@anholt.net>
Date:   Tue Jul 31 15:35:01 2012 -0700

    drm/i915: Don't forget to apply SNB PIPE_CONTROL GTT workaround.

    If a buffer that was the target of a PIPE_CONTROL from userland was a
    reused one that hadn't been evicted which had not previously had this
    workaround applied, then the early return for a correct
    presumed_offset in this function meant we would not bind it into the
    GTT and the write would land somewhere else.

    Fixes reproducible failures with GL_EXT_timer_query usage in apitrace,
    and I also expect it to fix the intermittent OQ issues on snb that
    danvet's been working on.

    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=48019
    Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=52932
Signed-off-by: Eric Anholt <eric@anholt.net>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Carl Worth <cworth@cworth.org>
Tested-by: Carl Worth <cworth@cworth.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests: add missing igt_exit() calls

Forgotten while converting to subtests, then copy&pasted.

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

lib/drmtest: add igt_skip_on macro

I've fumbled the logic inversion when converting to igt_require way
too often, so lets add something for dummies like me ;-)

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

tests: Update .gitignore

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

tests: add gem_persisten_relocs

This reproduces the 3.7 relocation regression ...

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

intel_gtt: Raw PTE dumper mode

./tools/intel_gtt -d | head
GTT offset |                 PTEs
--------------------------------------------------------
  0x000000 | 0xe4005015 0xe2854015 0xe283e015 0xe283f015
  0x004000 | 0xe28ba015 0xe28bb015 0xe28b6015 0xe28b7015
  0x008000 | 0xe2828015 0xe2829015 0xe282a015 0xe282b015
  0x00c000 | 0xe2928015 0xe2929015 0xe292a015 0xe292b015
  0x010000 | 0xe2918015 0xe2919015 0xe291a015 0xe291b015
  0x014000 | 0xe291c015 0xe291d015 0xe291e015 0xe291f015
  0x018000 | 0xe2920015 0xe2921015 0xe2922015 0xe2923015
  0x01c000 | 0xe2924015 0xe2925015 0xe2926015 0xe2927015

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

intel_gtt: Properly support gen6+ GTT PTEs

This finishes the objective in the last patch which was to actually deal
with physical addresses, and not the PTEs.

GEN6+ Provided support for physical addresses above 4GB. I'm not
actually sure what Ironlake supported, and don't feel like firing up the
timemachine.

v2: Add support for gen4, gen5, and haswell.

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

intel_gtt: Use function to get the physical address

The GTT PTEs that the tool is trying to compare is really about
addresses, and not the PTE itself. To accomplish this, make which
calculates the physical address we actually want.

This commit itself doesn't change any functionality; just the wording in
the code.

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

gem_vmap_blits: Demote warning to note

The warning that vmap isn't supported is useful, but it shouldn't get in
the way of developers (or distros) being able to use -Werror.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>

intel_reg_dumper: Silence GCC for uninitialized clock

GCC 4.8.1 seems to think clock may be uninitialized.

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

tests/gem_reloc_vs_gpu: add thrashing tests

Using the i915_gem_drop_caches debugfs interface to thrash without
really thrashing.

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

tests/gem_reloc_vs_gpu: add forked versions

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

lib/drmtest: use igt_require/assert in the prefault helpers

tests/gem_reloc_vs_gpu: Add faulting reloc tests

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

lib/drmtest: include sys/mman.h from drmtest.h

We need it for mmapping to get at PROT_READ|WRITE anyway.

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

lib/drmtest: extract gem_execbuf helper

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

tests/gem_reloc_vs_gpu: add interruptible version

Exercise a bug where we've failed to propagate the error code
correctly.

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

kms_flip: Fix use of fb_width for PAN subtests

We need to be careful to remember that fb-width is not always the same
as hdisplay, since for panning we allocate a larger framebuffer. So fix
up the printfs to use hdisplay/vsisplay since that should be uniform
across the array.

Regression from
commit 919d68901187fa797a9b648fcf87c838fae22fa3
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Thu Aug 29 15:33:53 2013 +0100

    kms_flips: Operate on an array of crtc

Buzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68832
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>

lib/drmtest: Improve output when igt_waitchildren fails

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

test/gem_concurrent_blt: remove hack for testing igt_fork

Oops, this shouldn't have been committed ...

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68830
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_tiled_swapping: fix igt_require conversion

I seem to be incompetent at logic ...

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68831
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

tests/gem_mmap_gtt: fix access checks

Reading manpages advisable ;-)

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

tests/gem_mmap_gtt: clarify access check checks a bit

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

lib/drmtest: Print info when children died due to signals

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

tests/gem_evict_everything: add swapping and forked subtests

Much better at hitting the list corruption here on my machines
than what we have thus far.

Note that somehow I just can't reproduce the bug any more. No idea
why. But I guess it's time to simply push this pile out.

v2: Limit threads and rounds to something reasonable.

v3: Use igt_permute_array to avoid EINVAL due to duplicated bo.

v4:
- Add a variant of the forked tests with multiple drm fds.
- Tune the swapped forked tests a bit to complete in a reasonable
  amount of time.

v5: Add some memory pressure from the cpu by using cpu mmaps (which
directly hit shmem, so bypass gem completely).

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

tests/gem_evict_*: Fix leak in copy

This turned out to be the reason one one of my tests was hitting
the list corruption bug - we need a good deal more memory pressure.
So I'll now add a new testcase for that.

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

tests/gem_tiled_swapping: Use igt_require

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

lib/drmtest: extract igt_fork from gem_concurrent_blt

Making sure that we correctly collect the exit codes from all children
is a bit a hassle. So add another magic igt codeblock for easy forking
and joining.

Note that children are (currently at least) not allowed to call igt_skip.
Add an assert to enforce this.

v2:
- Properly propagate the exit code.
- Fix the segfault.
- Add a child int and num_children paramter to the magic codeblock as
  suggested by Chris Wilson.
- Don't dump noise into stdout when a child thread fails, the parent
  will do that for us already.

v3: Now with some docs.

v4: Fixup igt_waitchildren to properly reset state so it can be used
again.

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

kms_flip: Remove debugging leftovers

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

kms_flip: Set everything to zero to disable a CRTC

Just setting fb=0 is not enough as the kernel thinks userspace is
insane.

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

kms_flip: Use the first mode if we find no matching modes for the crtc pair

We will check that we can set the mode on both crtcs before use, so
hopefully this will work...

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

kms_flip: Hook up primary events for page-flips

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

kms_flip: Run on pairs of connected outputs.

The goal is to flip the same framebuffer on a pair of CRTCs (clone mode)
and check that (a) the flip works in all combinations of workloads, and
(b) that we can hit the desired refresh rate under the simplest, most
ideal of conditions.

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

kms_flips: Operate on an array of crtc

This should be no functional change as we operate on an array of
crtc[1]. Later we shall test clone mode across a number of crtc.

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

kms_flip: Exercise flip-vs-render

For machine with split BCS/RCS rings, we also need to test whether we
correctly wait upon outstanding render work before flipping and changing
modes. This should also serve to exercise the ring selection code for
flips.

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

overlay: Increase idle timeout to 30s

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

overlay: Hide idle processes

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

overlay: Autohide Flip counter after a period of inactivity

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

overlay: Monitor per-ring context switch rate

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

overlay: Make it easier to renice

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

overlay: Generate unique name for snapshots

Since we no longer increment the counter every frame, we need to use a
timestamp instead.

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

overlay: Add number of running processes to CPU display

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

overlay: Include the core count in the cpu info line

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

overlay: Scale the CPU busy value by the numer of active cores

Having 100% always mean one processor fully utilised is much easier to
understand (than having to remember how many cores are on each machine
and scale appropriately).

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

overlay: Tidy presentation of ring busy values

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

overlay: Restore the correct default font size

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

overlay: read rc6 status from perf

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

overlay: Add support for multi-monitor positioning

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

overlay: Set the text color in case we do not have RC6

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

overlay: improve alignment of some labels

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

overlay: Read interrupts from perf

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

overlay: Read frequency from perf

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

overlay: Read power from perf_events

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

overlay: Align the process names in the memory table

And use a shade of gray to distinguish from the total instead.

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

lib/drmtest: restore in_fixture assert

I didn't really want to revert those, too ...

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

Revert "tests/gem_concurrent_blit: Fix segmentation fault"

This reverts commit 912a7d855600aadb937517ec5bab26bfd9b8953d.

Now fixed for real!

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

lib/drmtest: don't complete fixtures with a longjmp

Longjmp creates havoc with stack variables of the current stackframe.
And since fixtures should be used to set up such variables creating
havoc isn't a great idea. With this I can revert a bunch of
bogus patches I've done to paper over this by moving stack variables
to be global.

The same issue is actually a feature for subtest blocks since subtests
should be independant anyway.

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

tests/gem_mmap_gtt: Add testcase for the vma access manager

Currently fails since the patches aren't merged yet.

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

tests/gem_mmap: Convert to subtests

I wanted to write a testcase for gtt mmap access restriction checking
and only noticed after conversion to subtests that this here is for
cpu mmaps ;-)

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

tests/kms_setmode: don't save/restore vt graphics for dry-run

Signed-off-by: Imre Deak <imre.deak@intel.com>

lib/drmtest: fix handling of -h --help argument parsing

So far we handled -h or --help arguments properly only if the test
called igt_subtest_init_parse_opts(). Fix this for igt_subtest_init()
callers too. Make sure we still don't exit for any other
unknown options, which the caller may parse with a second getopt scan.

Signed-off-by: Imre Deak <imre.deak@intel.com>

overlay: Trim the padding slightly

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

overlay: Count number of semaphores used by each process

This required me to contract the per-process information considerably,
hopefully readability is not sacrificed too much.

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

overlay: Allow simple positioning and resizing

Using window.size=<width>x<height> or window.size=<scale>% in
the config file, or --size=<scale>% or --size=<width>x<height>

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

overlay: Monitor interrupts

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

overlay: Detach from terminal

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

overlay: Add a raw kms output

For when you don't have any display server, use brute force.

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

overlay: Draw the cpu busy line on top of the waits

By drawing it behind the waits, it is obscured by the translucent waits.

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

overlay: simplify types for object counts for 32/64bit porting

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

overlay: Expose sampling period/frequency as a user parameter

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

tests/gem_evict_*: Add interruptible modes

Also I've stumbled over igt_fixture again. Somehow gcc loves to wreak
havoc with stack variables set up in igt_fixtures ... Duct-tape in the
form of moving fd out to global scope applied.

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

overlay: Fix power reading from debugfs

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

overlay: Distinguish the root debugfs path and our dri node

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

overlay: Search for right debugfs path

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

tests/sysfs_rps: Fix igt_require conversion

Another fumble caught by QA.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68461
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

overlay: Replace debug ticker with hostname

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

overlay: Close DRI2 device after authentication failure

So that we don't leave a master fd open when operating remotely.

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

overlay: Double buffer the x11 window

For pleasant remote transport.

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

overlay: Make the config parser more lenient

Allow a dot-notation outside of the [ini] style to make commandline
parameters easier to add.

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

tests/pc8: don't work around the eDP I2C bug

On the current Kernels, i2c operations on disabled eDP panels don't
work because we don't enable the panel before doing the i2c
transactions. Whenever we try these transactions we print an error
message on dmesg and fail. The current test suite was trying to work
around this problem: we count how many eDP outputs we have and add
them to the final result.

Daniel asked me to not work around the bug, so now the i2c test will
fail until we actually fix the Kernel bug (or stop using eDP panels).

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>

tests/pc8: fix function return value

Looks like I forgot to push this chunk... The test was passing in
cases where you only have eDP monitors due to the fact that we
consider i2c won't work on eDP monitors.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>

overlay: Rudiments of config files and option parsing

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

overlay: Fill the blank charts with error messages

Instead of showing nothing, show a "space left intentionally blank"
message.

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

overlay: Remove busy estimation

Too inaccurate to be useful.

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

tests: ddx_intel_after_fbdev loads intel ddx after fbdev was loaded.

v2: Don't add to kernel tests.

Signed-off-by: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

overlay: Fix underflow for clamping initial values

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

overlay: i915 pmu doesn't require a specific frequency

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

overlay: Put a faint outline around each chart

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

overlay: Fade the graphs behind the text

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

tests: Add gem_evict_alignment

Clone gem_evict_everything and use it to exercise the misalignment
paths.

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

lib/drmtest: fix logic fumble in gem_context_create

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=68363
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

overlay: Fix old mmio gpu busyness probing

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

overlay: Fix sampling of i915 perf event

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

overlay: Fix display of rc6 states

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

overlay: Add copyright statements

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

overlay: Use the new i915 PMU to query GPU busyness

And so avoid having to hold forcewake indefinitely.

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

overlay: Take a snapshot on SIGUSR1

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