Signed-off-by: NZhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

We now attempt to free "active" objects following a GPU hang as either
the GPU will be reset or the hang is permenant. In either case, the GPU
writes will not be flushed to main memory and it should be safe to
return that memory back to the system.

The BUG_ON(active) is thus overkill and can erroneously fire after a
EIO.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

For the shared paths on the next generation chipsets.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Like on Sandybridge, disabling the DAC here when doing CRT load detect
avoids forever hangs waiting on the hardware.

test procedure on HP 2740p:
boot with no VGA plugged in, start X,
plug in VGA monitor (1280x1024)
chvt 3
machine hangs waiting forever.
Signed-off-by: NDave Airlie <airlied@redhat.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

Waiting for a hard coded 20ms isn't always enough to make sure a vblank
period has actually occurred, so add code to make sure we really have
passed through a vblank period (or that the pipe is off when disabling).

This prevents problems with mode setting and link training, and seems to
fix a bug like https://bugs.freedesktop.org/show_bug.cgi?id=29278, but
on an HP 8440p instead.  Hopefully also fixes
https://bugs.freedesktop.org/show_bug.cgi?id=29141.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NEric Anholt <eric@anholt.net>

We need to make sure the eDP PLL is enabled before the pipes or planes,
so do it as part of the DP prepare mode set function.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

We need to use I/O port instructions to access VGA registers on
Ironlake+, and it doesn't hurt on other platforms, so switch the VGA
plane disable function over to using them.  Move it to init time as well
while we're at it, no need to repeatedly disable the VGA plane with
every mode set and DPMS event.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

We should disable the panel first when shutting down an eDP link.  And
when turning one on, the panel needs to be enabled before link training
or eDP I/O won't be enabled.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

Ironlake requires that we clear the reset panel bit during power
sequences and restore it afterwards.  Uncondtionally add code to do that
since it should be harmless on SNB+.
Signed-off-by: NJesse Barnes <jbarnes@virtuousgeek.org>

RC6 allows the GPU to enter a lower power state when the GPU is idle.
Signed-off-by: NZou Nan hai <nanhai.zou@intel.com>
[anholt: Fixed the !renderctx error path to actually not enable RC6.]
Signed-off-by: NEric Anholt <eric@anholt.net>

If we have failed to ascertain the fixed mode for the LVDS panel, then
trust the pixel clock ranges reported for the connection when determing
valid modes. This makes intel_sdvo_mode_valid() consistent with
intel_lvds_mode_valid() which is also a no-op is there is no fixed mode
defined. (Since the mode is both validated by SDVO and LVDS, why are
checking against an LVDS fixed mode in SDVO...)

By only defining is_lvds to be true when we actually have an LVDS output
with a fixed mode, we avoid various potential NULL deferences where the
assumption is made that all LVDS outputs have a fixed mode.

References:

  Bug 29449 - [Q35] failure to read EDID/vbios for LVDS, no mode => no output
  https://bugs.freedesktop.org/show_bug.cgi?id=29449

The primary failure in this bug is not finding the EDID and determining
the correct fixed panel mode. However, this patch should fix the
secondary issue of not enabling any of the standard modes for the panel
either.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

RC6 power state requires a logical render context in place for saving
render context.
Signed-off-by: NZou Nan hai <nanhai.zou@intel.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

This is required by the spec, and without this some 3D programs will
hang after resume from RC6 we enable that.
Signed-off-by: NZou Nan hai <nanhai.zou@intel.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

If our watchdog fires and we see that the GPU is idle, but that we
are still waiting on an interrupt, forcibly wake-up the waiter.

i915_do_wait_request() should not be racy, yet there are persistent
reports that 945GM hangs whilst the GPU is idle. This implies that the
hardware is not quite as coherent as the documentation claims - a write
followed by a flush is supposed to be coherent in main memory before the
flush is retired and the irq is emitted. This seems to be a sensible and
elegant guard to force the wait to timeout.

v2: Daniel Vetter pointed out that a warning would be useful to explain
why the machine appeared to stall.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

i830 requires 32bpp cursors to be aligned to 16KB, so we have to expose
the alignment parameter to i915_gem_attach_phys_object().
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

The i845 and i865 have a peculiarlity in that CURBASE is not the trigger
for the vsync update of the cursor registers but instead the
modification of that register is prohibited whilst the cursor is
enabled. Reorder the write sequence for CURPOS, CURCNTR and CURBASE on
i845 to i865 to match.

v2: Remove the checks for i845/i865 from within i9xx_cursor_update()
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

The FBC is dependent upon a few details of the framebuffer so it is
required to be updated within set_base(), so remove the redundant call
from mode_set().
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Add a new macro, wait_for, to simplify the act of waiting on a register
to change state. wait_for() takes three arguments, the condition to
inspect on every loop, the maximum amount of time to wait and whether to
yield the cpu for a length of time after each check.

v2: Upgrade failure messages to DRM_ERROR on the suggestion of
Eric Anholt. We do not expect to hit these conditions as they reflect
programming errors, so if we do we want to be notified.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

The cleanup path for early abort failed to nullify the gem_buffer. The
likely consequence of this is zero, since a failure here should mean
aborting the module load.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Previously, we only remembered to update the watermarks for i9xx, and
incorrectly assumed that the crtc->enabled flag was valid at that point
in the dpms cycle.

Note that on my x201s this makes a SR bug on pipe 1 much easier to hit.
(Since before this patch when disabling pipe 0, we either didn't update
the watermarks at all, or when we did we still thought we had two pipes
enabled and so disabled SR.)

References:

  Bug 28969 - [Arrandale] Screen flickers, suspect Self-Refresh
  https://bugs.freedesktop.org/show_bug.cgi?id=28969Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Within i915_opregion.c there are two blocks of semantically identical
ASLE response codes defined. Only one of those matches the ACPI IGD
OpRegion Specification 0.1, use those.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Acked-by: NMatthew Garrett <mjg59@srcf.ucam.org>
Signed-off-by: NEric Anholt <eric@anholt.net>

Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

shmfs doesn't actually implement i_ops->truncate() so we were not
immedatiately releasing the backing pages when shrinking the gfx cache
under OOM. Instead use a combination of truncate_inode_pages() and
i_ops->truncate_range() as is used by shmem_delete_inode().
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Writing to the DSPBASE register triggers the double-buffered update to
all the control registers, so always write it last in the update
sequence.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

v2: Hook in DP paths to keep FULLSCREEN panel fitting on eDP.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NZhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: NEric Anholt <eric@anholt.net>

Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Directly read the GTT mapping for the contents of the batch buffers
rather than relying on possibly stale CPU caches. Also for completeness
scan the flushing/inactive lists for the current buffers - we are
collecting error state after all.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

In order to reduce the penalty of fallbacks under memory pressure and to
avoid a potential immediate ping-pong of evicting a mmaped buffer, we
move the object to the tail of the inactive list when a page is freshly
faulted or the object is moved into the CPU domain.

We choose not to protect the CPU objects from casual eviction,
preferring to keep the GPU active for as long as possible.

v2: Daniel Vetter found a bug where I forgot that pinned objects are
kept off the inactive list.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

Based in a large part upon Daniel Vetter's implementation and adapted
for handling multiple rings in a single pass.

This should lead to better gtt usage and fixes the page-fault-of-doom
triggered. The fairness is provided by scanning through the GTT space
amalgamating space in rendering order. As soon as we have a contiguous
space in the GTT large enough for the new object (and its alignment),
evict any object which lies within that space. This should keep more
objects resident in the GTT.

Doing throughput testing on a PineView machine with cairo-perf-trace
indicates that there is very little difference with the new LRU scan,
perhaps a small improvement... Except oddly for the poppler trace.

Reference:

  Bug 15911 - Intermittent X crash (freeze)
  https://bugzilla.kernel.org/show_bug.cgi?id=15911

  Bug 20152 - cannot view JPG in firefox when running UXA
  https://bugs.freedesktop.org/show_bug.cgi?id=20152

  Bug 24369 - Hang when scrolling firefox page with window in front
  https://bugs.freedesktop.org/show_bug.cgi?id=24369

  Bug 28478 - Intermittent graphics lockups due to overflow/loop
  https://bugs.freedesktop.org/show_bug.cgi?id=28478

v2: Attempt to clarify the logic and order of eviction through the use
of comments and macros.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NDaniel Vetter <daniel@ffwll.ch>
Signed-off-by: NEric Anholt <eric@anholt.net>

The eviction code is the gnarly underbelly of memory management, and is
clearer if kept separated from the normal domain management in GEM.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

This will be used by the eviction logic to maintain fairness between the
rings.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

This does two little changes:

- Add an alignment parameter for evict_something. It's not really great to
  whack a carefully sized hole into the gtt with the wrong alignment.
  Especially since the fallback path is a full evict.

- With the inactive scan stuff we need to evict more that one object, so
  move the unbind call into the helper function that scans for the object
  to be evicted, too.  And adjust its name.

No functional changes in this patch, just preparation.
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

In order to properly track bound objects, they need to exist on one of
the inactive/active lists or be pinned. As this is a requirement, do the
work inside i915_gem_bind_to_gtt() rather than dotted around the
callsites.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

v2: Add the interrupt status and address.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NEric Anholt <eric@anholt.net>

Using dev_priv__ avoids sparse complaining about shadowed variables in
the *LP_RING() macros.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

As the function has been reduced to a store plus increment, the body is
now smaller than the call so inline it.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

As we check that the ringbuffer will not wrap upon emission, we do not
need to check that incrementing the tail wrapped every time. However, we
do upon advancing just in case the tail is now pointing at the very end
of the ring.

Likewise we can account for the space used during emission in begin()
and avoid decrementing it for every emit.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>

The tail is quadword aligned, so we can add two MI_NOOP as a time.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NEric Anholt <eric@anholt.net>