GitHub/LineageOS/G12/android_kernel_amlogic_linux-4.9.git
10 years agodrm/i915: Try harder to get FBC
Ben Widawsky [Mon, 30 Jun 2014 17:41:24 +0000 (10:41 -0700)]
drm/i915: Try harder to get FBC

The GEN FBC unit provides the ability to set a low pass on frames it
attempts to compress. If a frame is less than a certain amount
compressibility (2:1, 4:1) it will not bother. This allows the driver to
reduce the size it requests out of stolen memory.

Unluckily, a few months ago, Ville actually began using this feature for
framebuffers that are 16bpp (not sure why not 8bpp). In those cases, we
are already using this mechanism for a different purpose, and so we can
only achieve one further level of compression (2:1 -> 4:1)

FBC GEN1, ie. pre-G45 is ignored.

The cleverness of the patch is Art's. The bugs are mine.

v2: Update message and including missing threshold case 3 (Spotted by Arthur).

Cc: Art Runyan <arthur.j.runyan@intel.com>
Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
10 years agodrm/i915: Extract CFB threshold calculation
Ben Widawsky [Thu, 19 Jun 2014 19:06:11 +0000 (12:06 -0700)]
drm/i915: Extract CFB threshold calculation

Right now, there is no threshold (0 means fail, 1 means 1:1 compression
limit). This is to split the function/non-functional change of the next
patch.

The next patch will start to attempt to reduce the amount of CFB space
we need for dire situations. It will be contained within this function.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
10 years agodrm/i915: Move compressed_fb to static allocation
Ben Widawsky [Thu, 19 Jun 2014 19:06:10 +0000 (12:06 -0700)]
drm/i915: Move compressed_fb to static allocation

We are already using the size to determine whether or not to free the
object, so there is no functional change there. Almost everything else
has changed to static allocations of the drm_mm_node too.

Aside from bringing this inline with much of our other code, this makes
error paths slightly simpler, which benefits the look of an upcoming
patch.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
10 years agodrm/i915: replace ALIGN(PAGE_SIZE) by PAGE_ALIGN
Fabian Frederick [Tue, 1 Jul 2014 18:39:41 +0000 (20:39 +0200)]
drm/i915: replace ALIGN(PAGE_SIZE) by PAGE_ALIGN

use mm.h definition

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: intel-gfx@lists.freedesktop.org
Cc: dri-devel@lists.freedesktop.org
Signed-off-by: Fabian Frederick <fabf@skynet.be>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
10 years agodrm/i915/bdw: Use timeout mode for RC6 on bdw
Tom O'Rourke [Wed, 9 Apr 2014 18:44:06 +0000 (11:44 -0700)]
drm/i915/bdw: Use timeout mode for RC6 on bdw

Higher RC6 residency is observed using timeout mode
instead of EI mode.  This applies to Broadwell only.
The difference is particularly noticeable with video
playback.

Signed-off-by: Tom O'Rourke <Tom.O'Rourke@intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
10 years agodrm/i915: flush delayed_resume_work when suspending
Paulo Zanoni [Fri, 27 Jun 2014 21:51:51 +0000 (18:51 -0300)]
drm/i915: flush delayed_resume_work when suspending

It is possible that, by the time we run i915_drm_freeze(),
delayed_resume_work was already queued but did not run yet. If it
still didn't run after intel_runtime_pm_disable_interrupts(), by the
time it runs it will try to change the interrupt registers with the
interrupts already disabled, which will trigger a WARN. We can
reliably reproduce this with the pm_rpm system-suspend test case.

In order to avoid the problem, we have to flush the work before
disabling the interrupts. We could also cancel the work instead of
flushing it, but that would require us to put a runtime PM reference -
and any other resource we may need in the future - in case the work
was already queued, so I believe flushing the work is more
future-proof, although less efficient. But I can also change this part
if someone requests.

Another thing I tried was to move the intel_suspend_gt_powersave()
call to before intel_runtime_pm_disable_interrupts(), but since that
function needs to be called after the interrupts are already disabled,
due to dev_priv->rps.work, this strategy didn't work.

Testcase: igt/pm_rpm/system-suspend
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=80517
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
10 years agodrm/i915: Don't try to look up object for non-existent fb
Matt Roper [Wed, 25 Jun 2014 00:05:02 +0000 (17:05 -0700)]
drm/i915: Don't try to look up object for non-existent fb

crtc->primary->fb may be NULL upon entry to intel_pipe_set_base() if the
primary plane has previously been disabled via the universal plane
interface.  We need to check for NULL before trying to reference
old_fb's obj.

This fixes a regression introduced in

        commit a071fa00647bc9a3c53f917b236fff9aea175e3a
        Author: Daniel Vetter <daniel.vetter@ffwll.ch>
        Date:   Wed Jun 18 23:28:09 2014 +0200

        drm/i915: Introduce accurate frontbuffer tracking

Testcase: igt/kms_universal_plane
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
10 years agodrivers/i915: Fix unnoticed failure of init_ring_common()
Konrad Zapalowicz [Thu, 19 Jun 2014 17:07:15 +0000 (19:07 +0200)]
drivers/i915: Fix unnoticed failure of init_ring_common()

This commit add check for return value of init_ring_common() in the
init_render_ring(). Now, when failure is detected the error code is
propagated to the caller instead of being ignored.

Signed-off-by: Konrad Zapalowicz <bergo.torino@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Track frontbuffer invalidation/flushing
Daniel Vetter [Thu, 19 Jun 2014 14:01:59 +0000 (16:01 +0200)]
drm/i915: Track frontbuffer invalidation/flushing

So these are the guts of the new beast. This tracks when a frontbuffer
gets invalidated (due to frontbuffer rendering) and hence should be
constantly scaned out, and when it's flushed again and can be
compressed/one-shot-upload.

Rules for flushing are simple: The frontbuffer needs one more full
upload starting from the next vblank. Which means that the flushing
can _only_ be called once the frontbuffer update has been latched.

But this poses a problem for pageflips: We can't just delay the
flushing until the pageflip is latched, since that would pose the risk
that we override frontbuffer rendering that has been scheduled
in-between the pageflip ioctl and the actual latching.

To handle this track asynchronous invalidations (and also pageflip)
state per-ring and delay any in-between flushing until the rendering
has completed. And also cancel any delayed flushing if we get a new
invalidation request (whether delayed or not).

Also call intel_mark_fb_busy in both cases in all cases to make sure
that we keep the screen at the highest refresh rate both on flips,
synchronous plane updates and for frontbuffer rendering.

v2: Lots of improvements

Suggestions from Chris:
- Move invalidate/flush in flush_*_domain and set_to_*_domain.
- Drop the flush in busy_ioctl since it's redundant. Was a leftover
  from an earlier concept to track flips/delayed flushes.
- Don't forget about the initial modeset enable/final disable.
  Suggested by Chris.

Track flips accurately, too. Since flips complete independently of
rendering we need to track pending flips in a separate mask. Again if
an invalidate happens we need to cancel the evenutal flush to avoid
races.

v3:
Provide correct header declarations for flip functions. Currently not
needed outside of intel_display.c, but part of the proper interface.

v4: Add proper domain management to fbcon so that the fbcon buffer is
also tracked correctly.

v5: Fixup locking around the fbcon set_to_gtt_domain call.

v6: More comments from Chris:
- Split out fbcon changes.
- Drop superflous checks for potential scanout before calling intel_fb
  functions - we can micro-optimize this later.
- s/intel_fb_/intel_fb_obj_/ to make it clear that this deals in gem
  object. We already have precedence for fb_obj in the pin_and_fence
  functions.

v7: Clarify the semantics of the flip flush handling by renaming
things a bit:
- Don't go through a gem object but take the relevant frontbuffer bits
  directly. These functions center on the plane, the actual object is
  irrelevant - even a flip to the same object as already active should
  cause a flush.
- Add a new intel_frontbuffer_flip for synchronous plane updates. It
  currently just calls intel_frontbuffer_flush since the implemenation
  differs.

This way we achieve a clear split between one-shot update events on
one side and frontbuffer rendering with potentially a very long delay
between the invalidate and flush.

Chris and I also had some discussions about mark_busy and whether it
is appropriate to call from flush. But mark busy is a state which
should be derived from the 3 events (invalidate, flush, flip) we now
have by the users, like psr does by tracking relevant information in
psr.busy_frontbuffer_bits. DRRS (the only real use of mark_busy for
frontbuffer) needs to have similar logic. With that the overall
mark_busy in the core could be removed.

v8: Only when retiring gpu buffers only flush frontbuffer bits we
actually invalidated in a batch. Just for safety since before any
additional usage/invalidate we should always retire current rendering.
Suggested by Chris Wilson.

v9: Actually use intel_frontbuffer_flip in all appropriate places.
Spotted by Chris.

v10: Address more comments from Chris:
- Don't call _flip in set_base when the crtc is inactive, avoids redunancy
  in the modeset case with the initial enabling of all planes.
- Add comments explaining that the initial/final plane enable/disable
  still has work left to do before it's fully generic.

v11: Only invalidate for gtt/cpu access when writing. Spotted by Chris.

v12: s/_flush/_flip/ in intel_overlay.c per Chris' comment.

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use new frontbuffer bits to increase pll clock
Daniel Vetter [Wed, 18 Jun 2014 11:59:13 +0000 (13:59 +0200)]
drm/i915: Use new frontbuffer bits to increase pll clock

The downclocking checks a few more things, so not that simple to
convert. Also, this should get unified with the drrs handling and also
use the locking of that. Otoh the drrs locking is about as hapzardous
as no locking, at least on first sight.

For easier conversion ditch the upclocking on unload - we'll turn off
everything anyway.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't take runtime PM reference around freeze/thaw
Jesse Barnes [Wed, 18 Jun 2014 16:52:56 +0000 (09:52 -0700)]
drm/i915: don't take runtime PM reference around freeze/thaw

We should be taking the right power well refs these days, so this
shouldn't be necessary.  It also gets in the way of re-using these
routines for S0iX states, as those need all the power saving features
enabled.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: use runtime irq suspend/resume in freeze/thaw
Jesse Barnes [Wed, 18 Jun 2014 16:52:55 +0000 (09:52 -0700)]
drm/i915: use runtime irq suspend/resume in freeze/thaw

We don't need to uninstall the full handler, simply disabling interrupts
ought to be enough.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Properly track domain of the fbcon fb
Daniel Vetter [Wed, 18 Jun 2014 13:05:19 +0000 (15:05 +0200)]
drm/i915: Properly track domain of the fbcon fb

X could end up putting the fbcon fb into other domains, e.g.
for smooth take-overs. Also we want this for accurate frontbuffer
tracking: The set_config is an implicit flush and will re-enable
psr and similar features, so we need to bring the bo back into
the gtt domain.

v2: Add FIXME comment about fbcon locking fun in atomic context,
requested by Chris.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Print obj->frontbuffer_bits in debugfs output
Daniel Vetter [Wed, 18 Jun 2014 12:46:49 +0000 (14:46 +0200)]
drm/i915: Print obj->frontbuffer_bits in debugfs output

Can be useful to figure out imbalances and bugs in the frontbuffer
tracking.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Introduce accurate frontbuffer tracking
Daniel Vetter [Wed, 18 Jun 2014 21:28:09 +0000 (23:28 +0200)]
drm/i915: Introduce accurate frontbuffer tracking

So from just a quick look we seem to have enough information to
accurately figure out whether a given gem bo is used as a frontbuffer
and where exactly: We have obj->pin_count as a first check with no
false negatives and only negligible false positives. And then we can
just walk the modeset objects and figure out where exactly a buffer is
used as scanout.

Except that we can't due to locking order: If we already hold
dev->struct_mutex we can't acquire any modeset locks, so could
potential chase freed pointers and other evil stuff.

So we need something else. For that introduce a new set of bits
obj->frontbuffer_bits to track where a buffer object is used. That we
can then chase without grabbing any modeset locks.

Of course the consumers of this (DRRS, PSR, FBC, ...) still need to be
able to do their magic both when called from modeset and from gem
code. But that can be easily achieved by adding locks for these
specific subsystems which always nest within either kms or gem
locking.

This patch just adds the relevant update code to all places.

Note that if we ever support multi-planar scanout targets then we need
one frontbuffer tracking bit per attachment point that we expose to
userspace.

v2:
- Fix more oopsen. Oops.
- WARN if we leak obj->frontbuffer_bits when freeing a gem buffer. Fix
  the bugs this brought to light.
- s/update_frontbuffer_bits/update_fb_bits/. More consistent with the
  fb tracking functions (fb for gem object, frontbuffer for raw bits).
  And the function name was way too long.

v3: Size obj->frontbuffer_bits correctly so that all pipes fit in.

v4: Don't update fb bits in set_base on failure. Noticed by Chris.

v5: s/i915_gem_update_fb_bits/i915_gem_track_fb/ Also remove a few
local enum pipe variables which are now no longer needed to make the
function arguments no drop over the 80 char limit.

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Drop schedule_back from psr_exit
Daniel Vetter [Wed, 18 Jun 2014 11:59:05 +0000 (13:59 +0200)]
drm/i915: Drop schedule_back from psr_exit

It doesn't make sense to never again schedule the work, since by the
time we might want to re-enable psr the world might have changed and
we can do it again.

The only exception is when we shut down the pipe, but that's an
entirely different thing and needs to be handled in psr_disable.

Note that later patch will again split psr_exit into psr_invalidate
and psr_flush. But the split is different and this simplification
helps with the transition.

v2: Improve the commit message a bit.

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Ditch intel_edp_psr_update
Daniel Vetter [Wed, 18 Jun 2014 11:59:03 +0000 (13:59 +0200)]
drm/i915: Ditch intel_edp_psr_update

We have _enable/_disable interfaces now for the modeset sequence and
intel_edp_psr_exit for workarounds.

The callsites in intel_display.c are all redundant with the modeset
sequence enable/disable calls in intel_ddi.c. The one in
intel_sprite.c is real and needs to be switched to psr_exit.

If this breaks anything then we need to augment the enable/disable
functions accordingly.

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Drop unecessary complexity from psr_inactivate
Daniel Vetter [Wed, 18 Jun 2014 11:59:02 +0000 (13:59 +0200)]
drm/i915: Drop unecessary complexity from psr_inactivate

It's not needed and further more will get in the way of a sane
locking scheme - psr_exit _can't_ take modeset locks due to lock
inversion, and at least once dp mst hits the connector list
is no longer static.

But since we track all state in dev_priv->psr there is no need
at all.

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Remove ctx->last_ring
Oscar Mateo [Wed, 18 Jun 2014 16:16:03 +0000 (17:16 +0100)]
drm/i915: Remove ctx->last_ring

The original comment that introduced it said:

commit 0009e46cd54324c4af20b0b52b89973b1b914167
Author: Ben Widawsky <ben@bwidawsk.net>
Date:   Fri Dec 6 14:11:02 2013 -0800

    drm/i915: Track which ring a context ran on

    Previously we dropped the association of a context to a ring. It is
    however very important to know which ring a context ran on (we could
    have reused the other member, but I was nitpicky).

    This is very important when we switch address spaces, which unlike
    context objects, do change per ring.

    As an example, if we have:

            RCS   BCS
    ctx            A
    ctx      A
    ctx      B
    ctx            B

    Without tracking the last ring B ran on, we wouldn't know to switch the
    address space on BCS in the last row.

But this is not really true, because we are already checking to != from (with
"from" being = ring->last_context) and that should be enough to make sure we
switch to the right address space.

We would have a problem if we switched the context object for every ring (since
then we would fail to do it in some situations) but we only switch it for the
render ring, so we don't care.

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoMerge branch 'topic/soix' into drm-intel-next-queued
Daniel Vetter [Wed, 18 Jun 2014 09:44:05 +0000 (11:44 +0200)]
Merge branch 'topic/soix' into drm-intel-next-queued

Jesse's SOix work required some patches from acpi-next, so pull it in
through a topic barnch.

Conflicts:
drivers/gpu/drm/i915/i915_drv.c
drivers/gpu/drm/i915/intel_pm.c

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Ack interrupts before handling them (CHV)
Oscar Mateo [Mon, 16 Jun 2014 15:11:00 +0000 (16:11 +0100)]
drm/i915/chv: Ack interrupts before handling them (CHV)

Otherwise, we might receive a new interrupt before we have time to
ack the first one, eventually missing it.

Without an atomic XCHG operation with mmio space, this patch merely
reduces the window in which we can miss an interrupt (especially when
you consider how heavyweight the I915_READ/I915_WRITE operations are).

Notice that, before clearing a port-sourced interrupt in the IIR, the
corresponding interrupt source status in the PORT_HOTPLUG_STAT must be
cleared.

Spotted by Bob Beckett <robert.beckett@intel.com>.

v2:
- Add warning to commit message and comments to the code as per Chris
  Wilson's request.
- Imre Deak pointed out that the pipe underrun flag might not be signaled
  in IIR, so do not make valleyview_pipestat_irq_handler depend on it.

v3: Improve the source code comment.

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: Ack interrupts before handling them (GEN8)
Oscar Mateo [Mon, 16 Jun 2014 15:10:59 +0000 (16:10 +0100)]
drm/i915/bdw: Ack interrupts before handling them (GEN8)

Otherwise, we might receive a new interrupt before we have time to
ack the first one, eventually missing it.

The right order should be:

1 - Disable Master Interrupt Control.
2 - Find the category of interrupt that is pending.
3 - Find the source(s) of the interrupt and clear the Interrupt Identity bits (IIR)
4 - Process the interrupt(s) that had bits set in the IIRs.
5 - Re-enable Master Interrupt Control.

Without an atomic XCHG operation with mmio space, the above merely reduces the window
in which we can miss an interrupt (especially when you consider how heavyweight the
I915_READ/I915_WRITE operations are).

Spotted by Bob Beckett <robert.beckett@intel.com>.

v2: Add warning to commit message and comments to the code as per Chris Wilson's request.

v3: Improve the source code comment.

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/vlv: Ack interrupts before handling them (VLV)
Oscar Mateo [Mon, 16 Jun 2014 15:10:58 +0000 (16:10 +0100)]
drm/i915/vlv: Ack interrupts before handling them (VLV)

Otherwise, we might receive a new interrupt before we have time to
ack the first one, eventually missing it.

Without an atomic XCHG operation with mmio space, this patch merely
reduces the window in which we can miss an interrupt (especially when
you consider how heavyweight the I915_READ/I915_WRITE operations are).

Notice that, before clearing a port-sourced interrupt in the IIR, the
corresponding interrupt source status in the PORT_HOTPLUG_STAT must be
cleared.

Spotted by Bob Beckett <robert.beckett@intel.com>.

v2:
- Reorder the IIR clearing to reduce the window even further.
- Add warning to commit message and comments to the code as per Chris
  Wilson's request.
- Imre Deak pointed out that the pipe underrun flag might not be signaled
  in IIR, so do not make valleyview_pipestat_irq_handler depend on it.

v3: Improve the source code comment.

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Ack interrupts before handling them (GEN5 - GEN7)
Oscar Mateo [Mon, 16 Jun 2014 15:10:57 +0000 (16:10 +0100)]
drm/i915: Ack interrupts before handling them (GEN5 - GEN7)

Otherwise, we might receive a new interrupt before we have time to ack the first
one, eventually missing it.

According to BSPec, the right order should be:

1 - Disable Master Interrupt Control.
2 - Find the source(s) of the interrupt.
3 - Clear the Interrupt Identity bits (IIR).
4 - Process the interrupt(s) that had bits set in the IIRs.
5 - Re-enable Master Interrupt Control.

Without an atomic XCHG operation with mmio space, the above merely reduces the window
in which we can miss an interrupt (especially when you consider how heavyweight the
I915_READ/I915_WRITE operations are).

We maintain the "disable SDE interrupts when handling" hack since apparently it works.

Spotted by Bob Beckett <robert.beckett@intel.com>.

v2: Add warning to commit message and comments to the code as per Chris Wilson's request.
v3: Improve the source comments.

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't BUG_ON in i915_gem_obj_offset
Daniel Vetter [Tue, 17 Jun 2014 20:34:38 +0000 (22:34 +0200)]
drm/i915: Don't BUG_ON in i915_gem_obj_offset

A WARN_ON is perfectly fine.

The BUG in here seems to be the cause behind hard-hangs when I cat the
i915_gem_pageflip debugfs file (which calls this from an irq
spinlock). But only while running a full igt run after a while. I
still need to root cause the underlying issue.

I'll also start reject patches which add new BUG_ON but don't come
with a really good justification for it. The general rule really
should be to just WARN and hope the driver survives for long enough.

v2: Make the WARN a bit more useful per Chris' suggestion.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Grab dev->struct_mutex in i915_gem_pageflip_info
Daniel Vetter [Tue, 17 Jun 2014 20:34:37 +0000 (22:34 +0200)]
drm/i915: Grab dev->struct_mutex in i915_gem_pageflip_info

We could walk of a bad list otherwise when someone concurrently
unbinds stuff for fun.

I've suspected this as the root-cause behind seemingly inconsistent
state, but alas it's not.

Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add some L3 registers to the parser whitelist
Brad Volkin [Tue, 17 Jun 2014 21:10:34 +0000 (14:10 -0700)]
drm/i915: Add some L3 registers to the parser whitelist

Beignet needs these in order to program the L3 cache config for
OpenCL workloads, particularly when using SLM.

Signed-off-by: Brad Volkin <bradley.d.volkin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't prefault the entire obj if the vma is smaller
Ville Syrjälä [Tue, 17 Jun 2014 18:03:00 +0000 (21:03 +0300)]
drm/i915: Don't prefault the entire obj if the vma is smaller

Take the minimum of the object size and the vma size and prefault
only that much. Avoids a SIGBUS when mmapping only a portion of the
object.

Prefaulting was introduced here:
 commit b90b91d87038f6b257b40a02b42ed4f9705e06f0
 Author: Chris Wilson <chris@chris-wilson.co.uk>
 Date:   Tue Jun 10 12:14:40 2014 +0100

    drm/i915: Prefault the entire object on first page fault

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Testcase: igt/gem_mmap/short-mmap
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Replaced Blitter ring based flips with MMIO flips
Sourab Gupta [Mon, 2 Jun 2014 11:17:17 +0000 (16:47 +0530)]
drm/i915: Replaced Blitter ring based flips with MMIO flips

This patch enables the framework for using MMIO based flip calls,
in contrast with the CS based flip calls which are being used currently.

MMIO based flip calls can be enabled on architectures where
Render and Blitter engines reside in different power wells. The
decision to use MMIO flips can be made based on workloads to give
100% residency for Media power well.

v2: The MMIO flips now use the interrupt driven mechanism for issuing the
flips when target seqno is reached. (Incorporating Ville's idea)

v3: Rebasing on latest code. Code restructuring after incorporating
Damien's comments

v4: Addressing Ville's review comments
    -general cleanup
    -updating only base addr instead of calling update_primary_plane
    -extending patch for gen5+ platforms

v5: Addressed Ville's review comments
    -Making mmio flip vs cs flip selection based on module parameter
    -Adding check for DRIVER_MODESET feature in notify_ring before calling
     notify mmio flip.
    -Other changes mostly in function arguments

v6: -Having a seperate function to check condition for using mmio flips (Ville)
    -propogating error code from i915_gem_check_olr (Ville)

v7: -Adding __must_check with i915_gem_check_olr (Chris)
    -Renaming mmio_flip_data to mmio_flip (Chris)
    -Rebasing on latest nightly

v8: -Rebasing on latest code
    -squash 3rd patch in series(mmio setbase vs page flip race) with this patch
    -Added new tiling mode update in intel_do_mmio_flip (Chris)

v9: -check for obj->last_write_seqno being 0 instead of obj->ring being NULL in
intel_postpone_flip, as this is a more restrictive condition (Chris)

v10: -Applied Chris's suggestions for squashing patches 2,3 into this patch.
These patches make the selection of CS vs MMIO flip at the page flip time, and
make the module parameter for using mmio flips as tristate, the states being
'force CS flips', 'force mmio flips', 'driver discretion'.
Changed the logic for driver discretion (Chris)

v11: Minor code cleanup(better readability, fixing whitespace errors, using
lockdep to check mutex locked status in postpone_flip, removal of __must_check
in function definition) (Chris)

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Sourab Gupta <sourab.gupta@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Tested-by: Chris Wilson <chris@chris-wilson.co.uk> # snb, ivb
[danvet: Fix up parameter alignement checkpatch spotted.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add missing statics to recent psr functions
Daniel Vetter [Mon, 16 Jun 2014 17:51:21 +0000 (19:51 +0200)]
drm/i915: Add missing statics to recent psr functions

Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Broaden FBC resolution limit to 4096*4096
Daisy Sun [Mon, 16 Jun 2014 22:48:18 +0000 (15:48 -0700)]
drm/i915: Broaden FBC resolution limit to 4096*4096

Staring from HSW, the resolution limit of FBC has increased to
4096*4096

Issue: VIZ-2813
Change-Id: I842f64e3cf2c0d18d29ef1bcfef3b9bb1f1764ac
Signed-off-by: Daisy Sun <daisy.sun@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Added write-enable pte bit supportt
Akash Goel [Tue, 17 Jun 2014 05:29:42 +0000 (10:59 +0530)]
drm/i915: Added write-enable pte bit supportt

This adds support for a write-enable bit in the entry of GTT.
This is handled via a read-only flag in the GEM buffer object which
is then used to see how to set the bit when writing the GTT entries.
Currently by default the Batch buffer & Ring buffers are marked as read only.

v2: Moved the pte override code for read-only bit to 'byt_pte_encode'. (Chris)
    Fixed the issue of leaving 'gt_old_ro' as unused. (Chris)

v3: Removed the 'gt_old_ro' field, now setting RO bit only for Ring Buffers(Daniel).

v4: Added a new 'flags' parameter to all the pte(gen6) encode & insert_entries functions,
    in lieu of overloading the cache_level enum (Daniel).

v5: Removed the superfluous VLV check & changed the definition location of PTE_READ_ONLY flag (Imre)

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Akash Goel <akash.goel@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Handle disabled primary plane in debugfs i915_display_info (v2)
Matt Roper [Mon, 16 Jun 2014 17:12:55 +0000 (10:12 -0700)]
drm/i915: Handle disabled primary plane in debugfs i915_display_info (v2)

Now that the primary plane can be disabled independently of the CRTC,
the debugfs code needs to be updated to recognize when the primary plane
is disabled and not try to return information about the primary plane's
framebuffer.

This change prevents a NULL dereference when reading i915_display_info
with a disabled primary plane.

v2: Replace a seq_printf() with seq_puts() (suggested by Damien)

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix comment about our plane remapping on gen2/3
Daniel Vetter [Mon, 16 Jun 2014 00:08:26 +0000 (02:08 +0200)]
drm/i915: Fix comment about our plane remapping on gen2/3

Spotted while crawling around in the area.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: update intel_dp_voltage_max comment
Paulo Zanoni [Fri, 13 Jun 2014 21:45:41 +0000 (18:45 -0300)]
drm/i915: update intel_dp_voltage_max comment

Any comment containing "current Intel hardware supports" quickly
becomes obsolete, so remove it and let people discover the information
by looking at the function implementation.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: update BDW DDI buffer translations
Paulo Zanoni [Fri, 13 Jun 2014 21:45:40 +0000 (18:45 -0300)]
drm/i915: update BDW DDI buffer translations

Two BSpec updates changed the recommended values for BDW eDP and DP
DDI buffer translations. Now the signal levels also match the HSW signal
levels, which simplify things a little bit.

It seems some DP sinks don't work properly without voltage level 0 and
pre-emphasis level 3, so this patch may fix some bugs on
panels/monitors that happen on BDW but not on HSW.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Simplify processing of the golden render context state
Chris Wilson [Tue, 10 Jun 2014 10:23:33 +0000 (11:23 +0100)]
drm/i915: Simplify processing of the golden render context state

Rewrite i915_gem_render_state.c for the purposes of clarity and
compactness, in the process we can eliminate some dodgy math that did
not handle 64bit addresses correctly.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Damien Lespiau <damien.lespiau@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Simplify i915_gem_release_all_mmaps()
Chris Wilson [Mon, 16 Jun 2014 07:57:44 +0000 (08:57 +0100)]
drm/i915: Simplify i915_gem_release_all_mmaps()

An object can only have an active gtt mapping if it is currently bound
into the global gtt. Therefore we can simply walk the list of all bound
objects and check the flag upon those for an active gtt mapping.

From commit 48018a57a8f5900e7e53ffaa0adeb784095accfb
Author: Paulo Zanoni <paulo.r.zanoni@intel.com>
Date:   Fri Dec 13 15:22:31 2013 -0200

    drm/i915: release the GTT mmaps when going into D3

Also note that the WARN is inappropriate for this function as GPU
activity is orthogonal to GTT mmap status. Rather it is the caller that
relies upon this condition and so it should assert that the GPU is idle
itself.

References: https://bugs.freedesktop.org/show_bug.cgi?id=80081
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@gmail.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Tested-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix memory leak in intel_dsi_init() error path
Christoph Jaeger [Fri, 13 Jun 2014 19:51:22 +0000 (21:51 +0200)]
drm/i915: Fix memory leak in intel_dsi_init() error path

intel_dsi_init() bails out without freeing the memory 'intel_dsi' and
'intel_connector' point to. Simply bail out before allocating memory.

Picked up by Coverity - CID 1222750.

Signed-off-by: Christoph Jaeger <christophjaeger@linux.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Improve PSR debugfs status.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:45 +0000 (10:16 -0700)]
drm/i915: Improve PSR debugfs status.

Now we have the active/inactive state for exit and this actually changes the
HW enable bit the status was a bit confusing for users. So let's provide
more info.

Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Force PSR exit by inactivating it.
Rodrigo Vivi [Fri, 13 Jun 2014 12:10:03 +0000 (05:10 -0700)]
drm/i915: Force PSR exit by inactivating it.

The perfect solution for psr_exit is the hardware tracking the changes and
doing the psr exit by itself. This scenario works for HSW and BDW with some
environments like Gnome and Wayland.

However there are many other scenarios that this isn't true. Mainly one right
now is KDE users on HSW and BDW with PSR on. User would miss many screen
updates. For instances any key typed could be seen only when mouse cursor is
moved. So this patch introduces the ability of trigger PSR exit on kernel side
on some common cases that.

Most of the cases are coverred by psr_exit at set_domain. The remaining cases
are coverred by triggering it at set_domain, busy_ioctl, sw_finish and
mark_busy.

The downside here might be reducing the residency time on the cases this
already work very wall like Gnome environment. But so far let's get focused
on fixinge issues sio PSR couild be used for everybody and we could even
get it enabled by default. Later we can add some alternatives to choose the
level of PSR efficiency over boot flag of even over crtc property.

v2: remove exit from connector_dpms. Daniel pointed this is the wrong way and
also this isn't needed for BDW and HSW anyway.

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/vlv: disable PPGTT on early revs v3
Jesse Barnes [Fri, 13 Jun 2014 16:28:33 +0000 (09:28 -0700)]
drm/i915/vlv: disable PPGTT on early revs v3

Early revs didn't have PPGTT support, so disable there.

v2: add debug msg when disabling on early stepping
v3: enable on other B3 packages as well (untested) (Ville)

References: https://bugs.freedesktop.org/show_bug.cgi?id=79669
References: https://bugs.freedesktop.org/show_bug.cgi?id=79670
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Acked-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix __user sparse warning
Ville Syrjälä [Fri, 13 Jun 2014 13:42:51 +0000 (16:42 +0300)]
drm/i915: Fix __user sparse warning

CHECK   linux/drivers/gpu/drm/i915/i915_gem_execbuffer.c
linux/drivers/gpu/drm/i915/i915_gem_execbuffer.c:1529:47: warning: incorrect type in initializer (different address spaces)
linux/drivers/gpu/drm/i915/i915_gem_execbuffer.c:1529:47:    expected struct drm_i915_gem_exec_object2 *user_exec_list
linux/drivers/gpu/drm/i915/i915_gem_execbuffer.c:1529:47:    got void [noderef] <asn:1>*
linux/drivers/gpu/drm/i915/i915_gem_execbuffer.c:1533:61: warning: incorrect type in argument 1 (different address spaces)
linux/drivers/gpu/drm/i915/i915_gem_execbuffer.c:1533:61:    expected void [noderef] <asn:1>*dst
linux/drivers/gpu/drm/i915/i915_gem_execbuffer.c:1533:61:    got unsigned long long *<noident>

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Print PCI revision in i915_dump_device_info()
Ville Syrjälä [Fri, 13 Jun 2014 12:39:56 +0000 (15:39 +0300)]
drm/i915: Print PCI revision in i915_dump_device_info()

Knowing the device stepping may be crucial in analyzing problems. Since
we always ask bug reporters for dmegs with drm.debug=0xe (or something)
it would be nice if the PCI revision is already included in the dump.
Avoids having to ask for lspci output as well.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add locking around framebuffer_references--
Ville Syrjälä [Fri, 13 Jun 2014 08:10:53 +0000 (11:10 +0300)]
drm/i915: Add locking around framebuffer_references--

obj->framebuffer_references isn't an atomic_t so the decrement needs to
be protected by some lock. struct_mutex seems like the appropriate lock
here, and we may already take it for the obj unref anyway.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: PSR HSW: update after enabling sprite.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:46 +0000 (10:16 -0700)]
drm/i915: PSR HSW: update after enabling sprite.

On the current structure HSW doesn't support PSR with sprites enabled
but sprites can be enabled after PSR was enabled what would cause
user to miss screen updates.

v2: move it to update_plane.

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: BDW PSR: Remove DDIA limitation for Broadwell.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:44 +0000 (10:16 -0700)]
drm/i915: BDW PSR: Remove DDIA limitation for Broadwell.

Broadwell has a PSR per transcoder, where DDIA supports
link disable and link standby modes while other
transcoders only support link standby.

Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: BDW PSR: Remove limitations that aren't valid for BDW.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:43 +0000 (10:16 -0700)]
drm/i915: BDW PSR: Remove limitations that aren't valid for BDW.

Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: BDW PSR: Add single frame update support.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:42 +0000 (10:16 -0700)]
drm/i915: BDW PSR: Add single frame update support.

When link is in stand by and PSR exit is triggered by a primary or sprite
plane flip this mode allows only one single updated frame to be send to
display than get back to PSR immediately.

Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Do not try to enable PSR when Panel doesn't suport it.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:40 +0000 (10:16 -0700)]
drm/i915: Do not try to enable PSR when Panel doesn't suport it.

Also do not cache aux info. That info could be related to another panel.

Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't let update_psr function actually enable PSR.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:39 +0000 (10:16 -0700)]
drm/i915: Don't let update_psr function actually enable PSR.

Being more conservative by enabling PSR only on psr_enable function.

Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use HAS_PSR to avoid unecessary interactions.
Rodrigo Vivi [Thu, 12 Jun 2014 17:16:38 +0000 (10:16 -0700)]
drm/i915: Use HAS_PSR to avoid unecessary interactions.

Let's be more conservative and protect platforms that don't
support PSR from unecessary interactions.

Reviewed-by: Vijay Purushothaman <vijay.a.purushothaman@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Switch to unified plane cursor handling (v4)
Matt Roper [Tue, 10 Jun 2014 15:28:13 +0000 (08:28 -0700)]
drm/i915: Switch to unified plane cursor handling (v4)

The DRM core will translate calls to legacy cursor ioctls into universal
cursor calls automatically, so there's no need to maintain the legacy
cursor support.  This greatly simplifies the transition since we don't
have to handle reference counting differently depending on which cursor
interface was called.

The aim here is to transition to the universal plane interface with
minimal code change.  There's a lot of cleanup that can be done (e.g.,
using state stored in crtc->cursor->fb rather than intel_crtc) that is
left to future patches.

v4:
 - Drop drm_gem_object_unreference() that is no longer needed now that
   we receive the GEM obj directly rather than looking up the ID.
v3:
 - Pass cursor obj to intel_crtc_cursor_set_obj() if cursor fb changes,
   even if 'visible' is false.  intel_crtc_cursor_set_obj() will notice
   that the cursor isn't visible and disable it properly, but we still
   need to get intel_crtc->cursor_addr set properly so that we behave
   properly if the cursor becomes visible again in the future without
   changing the cursor buffer (noted by Chris Wilson and verified
   via i-g-t kms_cursor_crc).
 - s/drm_plane_init/drm_universal_plane_init/.  Due to type
   compatibility between enum and bool, everything actually works
   correctly with the wrong init call, except for the type of plane that
   gets exposed to userspace (it shows up as type 'primary' rather than
   type 'cursor').
v2:
 - Remove duplicate dimension checks on cursor
 - Drop explicit cursor disable from crtc destroy (fb & plane
   destruction will take care of that now)
 - Use DRM plane helper to check update parameters

Cc: intel-gfx@lists.freedesktop.org
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Pallavi G<pallavi.g@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add intel_crtc_cursor_set_obj() to set cursor buffer (v2)
Matt Roper [Tue, 10 Jun 2014 15:28:12 +0000 (08:28 -0700)]
drm/i915: Add intel_crtc_cursor_set_obj() to set cursor buffer (v2)

Refactor cursor buffer setting such that the code to actually update the
cursor lives in a new function, intel_crtc_cursor_set_obj(), and takes
a GEM object as a parameter.  The existing legacy cursor ioctl handler,
intel_crtc_cursor_set() will now perform the userspace handle lookup and
then call this new function.

This refactoring is in preparation for the universal plane cursor
support where we'll want to update the cursor with an actual GEM buffer
object (obtained via drm_framebuffer) rather than a userspace handle.

v2:  Drop obvious kerneldoc and replace with note about function's
     reference consumption

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Pallavi G<pallavi.g@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Allow drivers to register cursor planes with crtc
Matt Roper [Tue, 10 Jun 2014 15:28:11 +0000 (08:28 -0700)]
drm: Allow drivers to register cursor planes with crtc

Universal plane support had placeholders for cursor planes, but didn't
actually do anything with them.  Save the cursor plane reference inside
the crtc and update the cursor plane parameter from void* to drm_plane.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Pallavi G<pallavi.g@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Avoid NULL deference when disabling a plane from userspace
Chris Wilson [Fri, 13 Jun 2014 14:22:28 +0000 (15:22 +0100)]
drm: Avoid NULL deference when disabling a plane from userspace

To disable a plane, userspace passes in an framebuffer id of 0. This
causes us to pass CRTC == NULL to setplane_internal, who promptly
deferences it to grab the struct drm_device. Oops.

[ 1296.467327] BUG: unable to handle kernel NULL pointer dereference at   (null)
[ 1296.467332] IP: [<c134dc51>] setplane_internal+0x11/0x280
[ 1296.467338] *pde = 00000000
[ 1296.467341] Oops: 0000 [#1] SMP
[ 1296.467344] Modules linked in: ccm bnep bluetooth snd_hda_codec_hdmi snd_hda_codec_idt snd_hda_codec_generic snd_hda_intel arc4 iwldvm snd_hda_controller snd_hda_codec mac80211 snd_hwdep snd_seq snd_seq_device snd_pcm snd_timer iwlwifi sdhci_pci snd cfg80211 x86_pkg_temp_thermal hp_wmi sdhci sparse_keymap mmc_core crc32c_intel rfkill microcode hp_accel lpc_ich lis3lv02d wmi mfd_core serio_raw input_polldev soundcore e1000e ptp pps_core
[ 1296.467367] CPU: 1 PID: 672 Comm: Xorg Tainted: G        W     3.15.0-rc8+ #351
[ 1296.467369] Hardware name: Hewlett-Packard HP ProBook 6360b/1620, BIOS 68SCF Ver. B.42 12/29/2010
[ 1296.467371] task: f423b5c0 ti: c2332000 task.ti: c2332000
[ 1296.467374] EIP: 0060:[<c134dc51>] EFLAGS: 00013286 CPU: 1
[ 1296.467376] EIP is at setplane_internal+0x11/0x280
[ 1296.467378] EAX: 00000000 EBX: c2333e90 ECX: 00000000 EDX: f3165600
[ 1296.467380] ESI: f430f400 EDI: 00000000 EBP: c2333e14 ESP: c2333dd4
[ 1296.467382]  DS: 007b ES: 007b FS: 00d8 GS: 0033 SS: 0068
[ 1296.467384] CR0: 80050033 CR2: 00000000 CR3: 00159000 CR4: 000407d0
[ 1296.467385] Stack:
[ 1296.467387]  000200da 00000002 c2333de8 c15dc4a0 f430f400 c2333e00 c134c54f eeeeeeee
[ 1296.467391]  f430f400 00000007 f416b480 c2333e14 00000000 c2333e90 f430f400 00000000
[ 1296.467396]  c2333e4c c1350aed 00000000 00000000 00000000 00000000 00000000 00000000
[ 1296.467400] Call Trace:
[ 1296.467406]  [<c15dc4a0>] ? mutex_lock+0x10/0x28
[ 1296.467408]  [<c134c54f>] ? _object_find+0x5f/0x90
[ 1296.467413]  [<c1350aed>] drm_mode_setplane+0x10d/0x1f0
[ 1296.467416]  [<c13509e0>] ? drm_mode_getplane+0x100/0x100
[ 1296.467420]  [<c1342e4d>] drm_ioctl+0x1bd/0x4f0
[ 1296.467423]  [<c13509e0>] ? drm_mode_getplane+0x100/0x100
[ 1296.467427]  [<c111c023>] ? handle_mm_fault+0x5d3/0xb30
[ 1296.467431]  [<c1118f31>] ? tlb_finish_mmu+0x11/0x40
[ 1296.467435]  [<c1342c90>] ? drm_ioctl_flags+0x40/0x40
[ 1296.467438]  [<c11593d2>] do_vfs_ioctl+0x2f2/0x4d0
[ 1296.467443]  [<c1226512>] ? inode_has_perm.isra.32+0x32/0x40
[ 1296.467446]  [<c122662f>] ? file_has_perm+0x7f/0x90
[ 1296.467449]  [<c1226fec>] ? selinux_file_ioctl+0x4c/0xf0
[ 1296.467452]  [<c1159610>] SyS_ioctl+0x60/0x90
[ 1296.467456]  [<c15e578c>] sysenter_do_call+0x12/0x22
[ 1296.467457] Code: 3f cf ff eb dd ba 3f 00 00 00 b8 d9 c9 7f c1 e8 e6 3f cf ff eb d9 8d 74 26 00 55 89 e5 57 56 53 83 ec 34 66 66 66 66 90 89 45 f0 <8b> 00 85 c9 89 d6 89 cb 89 45 ec 0f 84 16 01 00 00 8b 45 f0 e8
[ 1296.467485] EIP: [<c134dc51>] setplane_internal+0x11/0x280 SS:ESP 0068:c2

Fixes regression from
commit b02fd7fd8a541c3d590bfdda23365a927b507ceb
Author: Matt Roper <matthew.d.roper@intel.com>
Date:   Tue Jun 10 08:28:10 2014 -0700

    drm: Support legacy cursor ioctls via universal planes when possible (v4)

While at it move the plane parameter to the first position in
setplane_internal since that's the main object we're manipulating.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Pallavi G<pallavi.g@intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
[danvet: Add note about parameter reordering.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Support legacy cursor ioctls via universal planes when possible (v4)
Matt Roper [Tue, 10 Jun 2014 15:28:10 +0000 (08:28 -0700)]
drm: Support legacy cursor ioctls via universal planes when possible (v4)

If drivers support universal planes and have registered a cursor plane
with the DRM core, we should use that universal plane support when
handling legacy cursor ioctls.  Drivers that transition to universal
planes won't have to maintain separate legacy ioctl handling; drivers
that don't transition to universal planes will continue to operate
without any change to behavior.

Note that there's a bit of a mismatch between the legacy cursor ioctls
and the universal plane API's --- legacy ioctl's use driver buffer
handles directly whereas the universal plane API takes drm_framebuffers.
Since there's no way to recover the driver handle from a
drm_framebuffer, we can implement legacy ioctl's in terms of universal
plane interfaces, but cannot implement universal plane interfaces in
terms of legacy ioctls.  Specifically, there's no way to create a
general cursor helper in the way we previously created a primary plane
helper.

It's important to land this patch before any patches that add universal
cursor support to individual drivers so that drivers don't have to worry
about juggling two different styles of reference counting for cursor
buffers when userspace mixes and matches legacy and universal cursor
calls.  With this patch, a driver that switches to universal cursor
support may assume that all cursor buffers are wrapped in a
drm_framebuffer and can rely on framebuffer reference counting for all
cursor operations.

v4:
 - Add comments pointing out setplane_internal's reference-eating
   semantics.
v3:
 - Drop drm_mode_rmfb() call that is no longer needed now that we're
   using setplane_internal(), which takes care of deref'ing the
   appropriate framebuffer.
v2:
 - Use new add_framebuffer_internal() function to create framebuffer
   rather than trying to call directly into the ioctl interface and
   look up the handle returned.
 - Use new setplane_internal() function to update the cursor plane
   rather than calling through the ioctl interface.  Note that since
   we're no longer looking up an fb_id, no extra reference will be
   taken here.
 - Grab extra reference to fb under lock in !BO case to avoid issues
   where racing userspace could cause the fb to be destroyed out from
   under us after we grab the fb pointer.

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Pallavi G<pallavi.g@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Refactor setplane to allow internal use (v3)
Matt Roper [Tue, 10 Jun 2014 15:28:09 +0000 (08:28 -0700)]
drm: Refactor setplane to allow internal use (v3)

Refactor DRM setplane code into a new setplane_internal() function that
takes DRM objects directly as parameters rather than looking them up by
ID.  We'll use this in a future patch when we implement legacy cursor
ioctls on top of the universal plane interface.

v3:
 - Move integer overflow checking from setplane_internal to setplane
   ioctl.  The upcoming legacy cursor support via universal planes needs
   to maintain current cursor ioctl semantics and not return error for
   these extreme values (found via intel-gpu-tools kms_cursor_crc test).
v2:
 - Allow planes to be disabled without a valid crtc again (and add
   mention of this to setplane's kerneldoc, since it doesn't seem to be
   mentioned anywhere else).
 - Reformat some parameter line wrap

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Pallavi G<pallavi.g@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm: Refactor framebuffer creation to allow internal use (v2)
Matt Roper [Tue, 10 Jun 2014 15:28:08 +0000 (08:28 -0700)]
drm: Refactor framebuffer creation to allow internal use (v2)

Refactor DRM framebuffer creation into a new function that returns a
struct drm_framebuffer directly.  The upcoming universal cursor support
will want to create framebuffers internally to wrap cursor buffers, so
we want to be able to share that framebuffer creation with the
drm_mode_addfb2 ioctl handler.

v2: Take struct drm_mode_fb_cmd2 parameter directly rather than void*

Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Pallavi G<pallavi.g@intel.com>
Acked-by: Dave Airlie <airlied@linux.ie>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Fix "drm/i915/chv: Add a bunch of pre production workarounds"
Tom O'Rourke [Tue, 10 Jun 2014 23:26:34 +0000 (16:26 -0700)]
drm/i915/chv: Fix "drm/i915/chv: Add a bunch of pre production workarounds"

Correct a merge mishap in commit e4443e459ccf43f2c139358400365fd6a839d40d.

Wa*:chv belongs in cherryview_enable_rps, not gen8_enable_rps.

Signed-off-by: Tom O'Rourke <Tom.O'Rourke@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: Do not write the Semaphore Sync Registers in GEN8+
Oscar Mateo [Wed, 11 Jun 2014 15:17:16 +0000 (16:17 +0100)]
drm/i915/bdw: Do not write the Semaphore Sync Registers in GEN8+

These do not exist anymore.

Spotted while reading through intel_ringbuffer.c

Signed-off-by: Oscar Mateo <oscar.mateo@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Prefault the entire object on first page fault
Chris Wilson [Tue, 10 Jun 2014 11:14:40 +0000 (12:14 +0100)]
drm/i915: Prefault the entire object on first page fault

Inserting additional PTEs has no side-effect for us as the pfn are fixed
for the entire time the object is resident in the global GTT. The
downside is that we pay the entire cost of faulting the object upon the
first hit, for which we in return receive the benefit of removing the
per-page faulting overhead.

On an Ivybridge i7-3720qm with 1600MHz DDR3, with 32 fences,
Upload rate for 2 linear surfaces: 8127MiB/s -> 8134MiB/s
Upload rate for 2 tiled surfaces: 8607MiB/s -> 8625MiB/s
Upload rate for 4 linear surfaces: 8127MiB/s -> 8127MiB/s
Upload rate for 4 tiled surfaces: 8611MiB/s -> 8602MiB/s
Upload rate for 8 linear surfaces: 8114MiB/s -> 8124MiB/s
Upload rate for 8 tiled surfaces: 8601MiB/s -> 8603MiB/s
Upload rate for 16 linear surfaces: 8110MiB/s -> 8123MiB/s
Upload rate for 16 tiled surfaces: 8595MiB/s -> 8606MiB/s
Upload rate for 32 linear surfaces: 8104MiB/s -> 8121MiB/s
Upload rate for 32 tiled surfaces: 8589MiB/s -> 8605MiB/s
Upload rate for 64 linear surfaces: 8107MiB/s -> 8121MiB/s
Upload rate for 64 tiled surfaces: 2013MiB/s -> 3017MiB/s

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: "Goel, Akash" <akash.goel@intel.com>
Testcasee: igt/gem_fence_upload/performance
Reviewed-by: Brad Volkin <bradley.d.volkin@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix VLV CRC reading.
Rodrigo Vivi [Thu, 5 Jun 2014 21:28:17 +0000 (14:28 -0700)]
drm/i915: Fix VLV CRC reading.

Adding missing Display mmio reg offset.

Credits-to: Laws, Philip <philip.laws@intel.com>
Cc: He, Shuang <shuang.he@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Update DRIVER_DATE to 20140606
Daniel Vetter [Fri, 6 Jun 2014 20:40:42 +0000 (22:40 +0200)]
drm/i915: Update DRIVER_DATE to 20140606

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: preserve user forcewake over system suspend/resume
Imre Deak [Fri, 6 Jun 2014 09:59:39 +0000 (12:59 +0300)]
drm/i915: preserve user forcewake over system suspend/resume

Atm, the forcewake refcount will be incorrectly set to zero during
system suspend if there is any reference held via the
i915_forcewake_user debugfs entry.

Fix this by simply not zeroing the sw counters during suspend and
restoring the original state using them. Note that the only other
places where we zeroed the counters were driver load and unload time,
where it was redundant anyway.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=78059
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: fix possible refcount leak when resetting forcewake
Imre Deak [Fri, 6 Jun 2014 11:04:37 +0000 (14:04 +0300)]
drm/i915: fix possible refcount leak when resetting forcewake

If the timer putting the last forcewake refcount was pending and we
canceled it, we'll leak the corresponding forcewake and RPM references.

v2:
- do the ptr casting at the caller instead of adding a separate helper
  for this (Chris)

Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use %c in a format string for the pipe name
Damien Lespiau [Fri, 6 Jun 2014 17:22:06 +0000 (18:22 +0100)]
drm/i915: Use %c in a format string for the pipe name

pipe_name() returns an ascii character.

Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/bdw: Add Broadwell support for debugfs rps freq info
Tom O'Rourke [Fri, 30 May 2014 23:22:10 +0000 (16:22 -0700)]
drm/i915/bdw: Add Broadwell support for debugfs rps freq info

Add Broadwell support to i915_frequency_info
and extend i915_max|min_freq_get|set to (gen >= 6).

v2: generalized support for i915_max|min_freq_get|set (Daniel).

Signed-off-by: Tom O'Rourke <Tom.O'Rourke@intel.com>
Reviewed-by: Jeff McGee <jeff.mcgee@intel.com>
[danvet: Fix checkpatch fail.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't get hw state from DVO chip unless DVO is enabled
Ville Syrjälä [Thu, 5 Jun 2014 16:15:52 +0000 (19:15 +0300)]
drm/i915: Don't get hw state from DVO chip unless DVO is enabled

Certain DVO chips (ns2501 for example) don't like to be accessed unless
the PLL is running. Simply skip the DVO get_hw_state if the DVO port
is disabled.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use named initializers for gmch wm params
Ville Syrjälä [Thu, 5 Jun 2014 16:15:50 +0000 (19:15 +0300)]
drm/i915: Use named initializers for gmch wm params

Using names initializers when filling out the watermark structs
saves you from having go look up the struct definition every
single time.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use the .release hook to drop the stolen drm_mm tracking
Chris Wilson [Fri, 6 Jun 2014 09:22:54 +0000 (10:22 +0100)]
drm/i915: Use the .release hook to drop the stolen drm_mm tracking

Now that we have a release hook into i915_gem_object_free, we can move
the explicit call to the internal stolen function and hook it up
throught the callback instead.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Only wait one vblank when disabling crc if the pipe is on
Daniel Vetter [Fri, 6 Jun 2014 06:22:08 +0000 (08:22 +0200)]
drm/i915: Only wait one vblank when disabling crc if the pipe is on

Otherwise we incur an unsightly WARNING. The mutex locking is a bit
overkill, but it curbs races and eventially we might grow a locking
check in the vblank wait code to make sure the right crtc lock is
held.

This is fallout from

commit 9393707190194eb8b42e412b444a03331db6862f
Author:     Jesse Barnes <jbarnes@virtuousgeek.org>
AuthorDate: Fri Apr 4 16:12:09 2014 -0700

    drm/i915: warn when a vblank wait times out

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=79612
Tested-by: Guo Jinxian <jinxianx.guo@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
[danvet: Rebase on top of drm core ww locking changes.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: make sure PC8 is enabled on suspend and disabled on resume v4
Kristen Carlson Accardi [Thu, 12 Jun 2014 15:35:48 +0000 (08:35 -0700)]
drm/i915: make sure PC8 is enabled on suspend and disabled on resume v4

This matches the runtime suspend paths and allows the system to enter
the lowest power mode at freeze time.

v2: move disable_pc8 call to thaw_early (Imre)
    move enable_pc8 to freeze_late (Imre/Jesse)
v3: drop spurious hunk from _freeze now that we have freeze_late (Jesse)
v4: move back to suspend_late (Imre was right)

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Kristen Carlson Accardi <kristen@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: send proper opregion notifications on suspend/resume
Jesse Barnes [Thu, 12 Jun 2014 15:35:47 +0000 (08:35 -0700)]
drm/i915: send proper opregion notifications on suspend/resume

This indicates to the firmware that it can power down various other
components or bring them back up, depending on the target system state.

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Kristen Carlson Accardi <kristen@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Unifiy GT powersave suspend logic
Daniel Vetter [Thu, 12 Jun 2014 15:48:52 +0000 (17:48 +0200)]
drm/i915: Unifiy GT powersave suspend logic

Jesse's patch to only quiescent our rps work and Imre's fix to address
a race with runtime pm and the forcewake reference held by the used
diverging means to address the same bug: Jesse's patch uses
flush_delayed_work while (since we want to make sure rps is set up)
while Imre's used a cancel+manuel refcount adjustment.

Unify them again by simply reusing intel_suspend_gt_powersave in
intel_disable_gt_powersave.

Cc: Imre Deak <imre.deak@intel.com>
Cc: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: leave rc6 enabled at suspend time v4
Jesse Barnes [Thu, 12 Jun 2014 15:35:45 +0000 (08:35 -0700)]
drm/i915: leave rc6 enabled at suspend time v4

This allows the system to enter the lowest power mode during system freeze.

v2: delete force wake timer at suspend (Imre)
v3: add GT work suspend function (Imre)
v4: use uncore forcewake reset (Daniel)

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Kristen Carlson Accardi <kristen@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: disable power wells on suspend
Kristen Carlson Accardi [Thu, 12 Jun 2014 15:35:44 +0000 (08:35 -0700)]
drm/i915: disable power wells on suspend

We want to make sure everything is disabled and at its lowest power when
freezing.

Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Kristen Carlson Accardi <kristen@linux.intel.com>
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agoMerge commit 'e81a0e771c10de86fdb52c6baf534ff5fdeec72c' into topic/soix
Daniel Vetter [Thu, 12 Jun 2014 15:46:02 +0000 (17:46 +0200)]
Merge commit 'e81a0e771c10de86fdb52c6baf534ff5fdeec72c' into topic/soix

Need to have a baseline of drm-next + acpi-next to merge new cool SOix
features for 3.17. No conflicts, so won't cause trouble.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Add #defines for short/long pulse on gmch platforms
Daniel Vetter [Thu, 5 Jun 2014 07:36:23 +0000 (09:36 +0200)]
drm/i915: Add #defines for short/long pulse on gmch platforms

For no reason at all the public docs lack them, and Dave needs them
for his hpd interrupt rework.

Cc: Dave Airlie <airlied@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Simplify intel_gpu_reset
Robert Beckett [Tue, 3 Jun 2014 08:08:26 +0000 (10:08 +0200)]
drm/i915: Simplify intel_gpu_reset

Replaced ever growing switch for gen version with chained conditionals.
Futre gen's only need to add a new one if they require something different.

Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Robert Beckett <robert.beckett@intel.com>
[danvet: Picked from internal tree and white-wash commit message.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Drop unused lut tables from intel_plane
Matt Roper [Mon, 2 Jun 2014 17:12:06 +0000 (10:12 -0700)]
drm/i915: Drop unused lut tables from intel_plane

Those LUT where defined in the original sprite patch introducing intel_plane,
but were never used.

  commit b840d907fcf6d5d5ef91af4518b3dab3a5da0f75
  Author: Jesse Barnes <jbarnes@virtuousgeek.org>
  Date:   Tue Dec 13 13:19:38 2011 -0800

    drm/i915: add SNB and IVB video sprite support v6

Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Pimp commit message as suggested by Damien]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: runtime PM support for DPMS
Daniel Vetter [Thu, 24 Apr 2014 21:55:42 +0000 (23:55 +0200)]
drm/i915: runtime PM support for DPMS

Keeping track of the power domains is a bit messy since crtc->active
is currently updated by the platform hooks, but we need to be aware of
which state transition exactly is going on. Maybe we simply need to
shovel all the power domain handling down into platform code to
simplify this. But doing that requires some more auditing since
currently the ->mode_set callbacks still read some random registers
(to e.g. figure out the reference clocks).

Also note that intel_crtc_update_dpms is always call first/last even
for encoders which have their own dpms functions. Hence we really only
need to update this place here.

Being a quick "does it blow up?" run not really tested yet.

v2: Don't do runtime PM in the DPMS hooks for HAS_DDI platforms since
that is stalled. Also add a comment to explain what's going on.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: enable PPGTT on VLV
Jesse Barnes [Thu, 29 May 2014 21:33:21 +0000 (14:33 -0700)]
drm/i915: enable PPGTT on VLV

Working for real this time.  i915_ppgtt_info has all sorts of good stuff
in it and X is running nicely on top.

Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Use transcoder as index to MIPI regs
Shashank Sharma [Mon, 2 Jun 2014 12:37:48 +0000 (18:07 +0530)]
drm/i915: Use transcoder as index to MIPI regs

Conceptually, the MIPI registers are addressed by the MIPI transcoder
index, not the pipe. It doesn't matter right now, because there's a
1:1 relationship between pipes and MIPI transcoders, but that change
allows us to break that link in the future

V1: Created new patch to address Damien's review comment.
Replacing _PIPE calls to _TRANSCODER calls
V2: Re-basing on patch 2
V3: Re-basing on patch 2
V4: Re-basing on patch 2

Signed-off-by: Shashank Sharma <shashank.sharma@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Change Mipi register definitions
Shashank Sharma [Mon, 2 Jun 2014 12:37:47 +0000 (18:07 +0530)]
drm/i915: Change Mipi register definitions

Re-define MIPI register definitions in such a way that most of
the existing DSI code can be re-used for future platforms. Register
definitions are re-written using MMIO offset variable, so that without
changing the existing sequence, same code can be generically applied.

V4: Addressing review comments by Damien and Ville, splitting into two patches
This patch removes all the un-necessary formatting changes from previous patch.
V5: Removed 80 char limit formatting for existing MIPI regs
V6: Removed extra space, change one definition

Signed-off-by: Shashank Sharma <shashank.sharma@intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrivers/gpu/drm/i915/dma: style fixes
Robin Schroer [Mon, 2 Jun 2014 14:59:39 +0000 (16:59 +0200)]
drivers/gpu/drm/i915/dma: style fixes

Fixed several double space pointer notations, and added one newline

Signed-off-by: Robin Schroer <sulamiification@gmail.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: use shmem helpers if possible
David Herrmann [Sun, 25 May 2014 12:34:10 +0000 (14:34 +0200)]
drm/i915: use shmem helpers if possible

Instead of shuffling gfp-masks all the time, use the
shmem_read_mapping_page() helper. Note that __GFP_IO and __GFP_WAIT are
set in mapping_gfp_mask() for i915, so the behavior is still the same.

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Signed-off-by: David Herrmann <dh.herrmann@gmail.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Intel-specific primary plane handling (v8)
Matt Roper [Thu, 29 May 2014 15:06:54 +0000 (08:06 -0700)]
drm/i915: Intel-specific primary plane handling (v8)

Intel hardware allows the primary plane to be disabled independently of
the CRTC.  Provide custom primary plane handling to allow this.

v8:
 - Pin/unpin properly when clipping causes the primary plane to be
   disabled when it has previously been enabled.
 - s/drm_primary_helper_check_update/drm_plane_helper_check_update/
v7:
 - Clip primary plane to invisible when crtc is disabled since
   intel_crtc->config.pipe_src_{w,h} may be garbage otherwise.
 - Unpin old fb before pinning new one in the "just pin and
   return" case that is used when the crtc is disabled.
 - Don't treat implicit disabling of the primary plane (caused by
   clipping) the same way as explicit disabling (caused by fb=0).
   For implicit disables, we should leave the fb set and pinned,
   whereas for explicit disables we need to unpin the fb before
   primary->fb is cleared.
v6:
 - Pass rectangles to primary helper check function and get plane
   visibility back.
 - Wait for pending pageflips on primary plane update/disable.
 - Allow primary plane to be updated while the crtc is disabled (changes
   will take effect when the crtc is re-enabled if modeset passes -1
   for the fb id).
 - Drop WARN() if we try to disable the primary plane when it's
   already been disabled.  This will happen if the crtc gets disabled
   after the primary plane has already been disabled independently.
v5:
 - Use new drm_primary_helper_check_update() helper function to check
   setplane parameter validity.
 - Swap primary plane's pipe for pre-gen4 FBC (caught by Ville Syrjälä)
 - Cleanup primary plane properly on crtc init failure
v4:
 - Don't add a primary_plane field to intel_crtc; that was left over
   from a much earlier iteration of this patch series, but is no longer
   needed/used now that the DRM core primary plane support has been
   merged.
v3:
 - Provide gen-specific primary plane format lists (suggested by Daniel
   Vetter).
 - If the primary plane is already enabled, go ahead and just call the
   primary plane helper to do the update (suggested by Daniel Vetter).
 - Don't try to disable the primary plane on destruction; the DRM layer
   should have already taken care of this for us.
v2:
 - Unpin fb properly on primary plane disable
 - Provide an Intel-specific set of primary plane formats
 - Additional sanity checks on setplane (in line with the checks
   currently being done by the DRM core primary plane helper)

Reviewed-by: Chon Ming Lee <chon.ming.lee@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: don't force full modeset if primary plane is disabled (v2)
Matt Roper [Thu, 29 May 2014 15:06:53 +0000 (08:06 -0700)]
drm/i915: don't force full modeset if primary plane is disabled (v2)

In a future patch, we'll allow the primary plane to be disabled by
userspace via the universal plane API.  If a modeset is requested while
the primary plane is disabled, crtc->primary->fb will be NULL which
generally triggers a full modeset (except in fastboot situations).  If
we detect that the crtc is active, but there's no primary plane fb,
we should still allow a simple plane update rather than a full modeset
if the mode isn't actually changing (after re-enabling the primary plane
of course).

v2:
 - Enable plane after set_base to avoid enabling the plane if set_base
   fails, and to make flip+enable atomic (suggested by Ville)
 - Drop BUG to WARN if we somehow enter the 'fb_changed' modeset case
   with the crtc disabled (suggested by Ville)

Reviewed-by: Chon Ming Lee <chon.ming.lee@intel.com>
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Drop locking around fbdev-fb in debugfs
Daniel Vetter [Thu, 29 May 2014 21:27:00 +0000 (23:27 +0200)]
drm/i915: Drop locking around fbdev-fb in debugfs

All the date we print is invariant for the lifetime of the driver.
And none of it would be protected by the mode_config.mutex anyway.
So drop it.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix context locking in debugfs
Daniel Vetter [Thu, 29 May 2014 21:23:08 +0000 (23:23 +0200)]
drm/i915: Fix context locking in debugfs

This goes all the way back to the introduction of this debugfs file,
even though back then no locking really was required. None of the
intermediate patches fixed this.

Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Check for a NULL shared dpll before dereferencing
Chris Wilson [Wed, 28 May 2014 15:16:42 +0000 (16:16 +0100)]
drm/i915: Check for a NULL shared dpll before dereferencing

This doesn't look possible but a little extra defense against the
improbable is worth it - an oops here could lockup the machine.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Make intel_dsi_init() return void
Damien Lespiau [Wed, 28 May 2014 11:30:56 +0000 (12:30 +0100)]
drm/i915: Make intel_dsi_init() return void

Functions that can't fail are such a bliss to work with, it'd be shame
to miss the occasion. The "failure" mode is the DSI connector not being
created, the rest of the initialization can carry on happily.

We weren't even checking that value anyway.

Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Suggested-by: Shobhit Kumar <shobhit.kumar@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Shobhit Kumar <shobhit.kumar@intel.com>
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Also convert the missed return statement due to other patches
merged meanwhile.]
[danvet2: Squash in fixup from Damien to remove empty return; at the
end of intel_dsi_init.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Fix checkpatch errors
Shobhit Kumar [Tue, 27 May 2014 13:53:46 +0000 (19:23 +0530)]
drm/i915: Fix checkpatch errors

Fix warnings introduced by the following commit -

commit 9c92da2c7c17eea79b6321b37592df0a002d24df
Author: Shobhit Kumar <shobhit.kumar@intel.com>
Date:   Fri May 23 21:35:27 2014 +0530

    drm/i915: Add support for Generic MIPI panel driver

Fixed all except the DRM logging which go beyond line 80

Signed-off-by: Shobhit Kumar <shobhit.kumar@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Handle video DIP registers on CHV
Ville Syrjälä [Wed, 9 Apr 2014 10:29:09 +0000 (13:29 +0300)]
drm/i915/chv: Handle video DIP registers on CHV

The DIP registers are a mess on VLV and CHV. The register block on pipe
A is different than the register block on pipes B and C. In order to
handle that using the pipe offsets, we'd need a new pipe offset per
register, which seems wasteful. So instead just use the _PIPE3() macro
to handle these registers.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915: Don't use pipe_offset stuff for DPLL registers
Ville Syrjälä [Wed, 9 Apr 2014 10:29:08 +0000 (13:29 +0300)]
drm/i915: Don't use pipe_offset stuff for DPLL registers

These are just single registers so wasting space for the pipe offsets
seems a bit pointless. So just use the _PIPE3() macro instead.

Also rewrite the _PIPE3() macro to be more obvious, and protect the
arguments properly.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Damien Lespiau <damien.lespiau@intel.com>
[danvet: Frob conflict.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Force PHY clock buffers off after PLL disable
Ville Syrjälä [Tue, 27 May 2014 13:32:55 +0000 (16:32 +0300)]
drm/i915/chv: Force PHY clock buffers off after PLL disable

Now that we forced the clock buffers on in .pre_pll_enable() we
should probably undo the damage after we've turned the PLL off.

We do the clock buffer force enable in the .pre_pll_enable() hook
as we need to know which port is going to be used, but in the disable
case we don't need the port since we just disable the clock buffers
to both channels. So we can do this in chv_disable_pll() instead
of having to add any kind of .post_pll_disable() hook.

v2: Improve the commit message

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Force clock buffer enables
Ville Syrjälä [Tue, 27 May 2014 13:30:18 +0000 (16:30 +0300)]
drm/i915/chv: Force clock buffer enables

Try to force the PHY clock buffer enables to make the clock routing
work.

v2: Fix the pipe B case to actually enable CH0 clock buffers

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Try to program the PHY used clock channel overrides
Ville Syrjälä [Wed, 9 Apr 2014 10:29:05 +0000 (13:29 +0300)]
drm/i915/chv: Try to program the PHY used clock channel overrides

These should make it possible to feed port C from pipe A or port B from
pipe B. Didn't quite seem to work though.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
10 years agodrm/i915/chv: Freq(opcode) request for CHV.
Deepak S [Fri, 23 May 2014 15:30:21 +0000 (21:00 +0530)]
drm/i915/chv: Freq(opcode) request for CHV.

On CHV, All the freq request should be even. So, we need to make sure we
request the opcode accordingly.

v2: Avoid vairable for freq request (ville)

Signed-off-by: Deepak S <deepak.s@linux.intel.com>
Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>