drm/i915: use __wait_seqno for ring throttle

It turns out throttle had an almost identical  bit of code to do the
wait. Now we can call the new helper directly.  This is just a bonus,
and not needed for the overall series.

v2: remove irq_get/put which is now in __wait_seqno (Ben)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: remove polled wait from throttle

It's about to go away anyway. Just here to help bisection.

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: extract __wait_seqno from i915_wait_request

i915_wait_request is actually a fairly large function encapsulating
quite a few different operations. Because being able to wait on seqnos
in various conditions is useful, extracting that bit of code to a helper
function seems useful

v2: pull the irq_get/put as well (Ben)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: drop polled waits from i915_wait_request

The only time irq_get should fail is during unload or suspend. Both of
these points should try to quiesce the GPU before disabling interrupts
and so the atomic polling should never occur.

This was recommended by Chris Wilson as a way of reducing added
complexity to the polled wait which I introduced in an RFC patch.

09:57 < ickle_> it's only there as a fudge for waiting after irqs
after uninstalled during s&r, we aren't actually meant to hit it
09:57 < ickle_> so maybe we should just kill the code there and fix the breakage

v2: return -ENODEV instead of -EBUSY when irq_get fails

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: kill waiting_seqno

The waiting_seqno is not terribly useful, and as such we can remove it
so that we'll be able to extract lockless code.

v2: Keep the information for error_state (Chris)
Check if ring is initialized in hangcheck (Chris)
Capture the waiting ring (Chris)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: add some bikeshed to clarify a comment.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: move vbetool invoked ier stuff

This extra bit of interrupt enabling code doesn't belong in the wait
seqno function. If anything we should pull it out to a helper so the
throttle code can also use it. The history is a bit vague, but I am
going to attempt to just dump it, unless someone can argue otherwise.

Removing this allows for a shared lock free wait seqno function. To keep
tabs on this issue though, the IER value is stored on error capture
(recommended by Chris Wilson)

v2: fixed typo EIR->IER (Ben)
Fix some white space (Ben)
Move IER capture to globally instead of per ring (Ben)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
[danvet: ier is a 16 bit reg on gen2!]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: remove do_retire from i915_wait_request

This originates from a hack by me to quickly fix a bug in an earlier
patch where we needed control over whether or not waiting on a seqno
actually did any retire list processing. Since the two operations aren't
clearly related, we should pull the parameter out of the wait function,
and make the caller responsible for retiring if the action is desired.

The only function call site which did not get an explicit retire_request call
(on purpose) is i915_gem_inactive_shrink(). That code was already calling
retire_request a second time.

v2: don't modify any behavior excepit i915_gem_inactive_shrink(Daniel)

Signed-off-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: use the new masked bit macro some more

I've missed this one.

v2: Chris Wilson noticed another register.
v3: Color choice improvements.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove unused and unloved vblank macros

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Inline I915_INTERRUPT_ENABLE_FIX

Since there is only one remaining user of I915_INTERRUPT_ENABLE_FIX,
expand it at the callsite. Quoting Jesse Barnes:

"I'd really like to get rid of these defines at the top of i915_irq.c.
Some are unused and the others just make you check for the right bits
everytime your read the code."

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
[danvet: Add bikeshed suggested by Jesse.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Handle PendingFlip on gen3 robustly

We appear to allow too many pending pageflips as evidenced by an
apparent pin-leak. So borrow the pageflip completion logic from i8xx for
handling PendingFlip in a robust manner.

v2: Address Jesse's reminders about the nuances of gen3 IRQ handling.

References: https://bugzilla.kernel.org/show_bug.cgi?id=41882
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Cleanup gen3 irq uninstall

Bring the for-each-pipe loops together so that the code is easier on the
eyes.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: HWSTAM is only 16-bit on gen3

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove gen4 irq code from gen3 irq routine

A couple of miscellaneous cleanups as well to move per-loop condition
variables within the scope of the loop and the update of the DRI1
breadcrumb to the tail of the function.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove gen3 irq code from gen4 irq routine

And a couple of miscellaneous cleanups to the main body of the IRQ loop;
move per-loop condition variables within the scope of the loop and move
the old DRI1 breadcrumb to the tail of the function and so only execute
it once.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Clear FlipDone semantics change for pageflipping on gen3

On later gen3, you are able to select the meaning of the FlipPending
status bit in IIR and change it to FlipDone. This was sometimes done by
the BIOS leading to confusion on just how pageflipping worked on gen3.
Simplify the implementation by using the legacy meaning for all gen3
machines.

Note: this makes all gen3 machines equally broken...

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Duplicate and split the gen3/4 irq handler

In preparation for rewriting the gen3 irq handler.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: pending_flip_is_done is gen3, name it so

And remove the cargo-culted copy from the valleyview irq handler.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove redundant initialisation of per-ring IRQ waitqueues

The waitqueues are already initialised during ring initialisation so
kill the redundant and duplicated code to do so in each generations IRQ
installer.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Unconditionally initialise the interrupt workers

Rather than duplicate similar code across the IRQ installers, perform
the initialisation of the workers upfront. This will lead to simpler
teardown and quiescent code as we can assume that the workers have
been initialised.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: check for kms in dri1 ioctls

Calling these when gem assumes full control of the hw won't end
in anything else than tears. So be a bit more paranoid here.

Just serves as documentation.

v2: Bail out with ENODEV as suggested by Chris Wilson.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: rip out dev_priv->has_gem

Always true these days. It has been added originally to work
around some issues with the agp layer in 2.6.29:

commit ac5c4e76180a74c7f922f6fa71ace0cef45fa433
Author: Dave Airlie <airlied@redhat.com>
Date:   Fri Dec 19 15:38:34 2008 +1000

    drm/i915: GEM on PAE has problems - disable it for now.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: rip out GEM drm feature checks

We always set it so there's no point in checking. We could
instead add a bit that tells us whether gem is actually
initialized (i.e. either kms or gem_init_ioctl called), but
that's imho not worth it.

So just rip it out.

There's a little change in the wait_ring timeout, but we've never
run with anything else than the 60 second timeout, even on dri1
userspace.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: disallow gem ums init ioctl for kms

This ioctl used in a kms driver is only useful to create massive
havoc.

v2: Bail out with -ENODEV as suggested by Chris Wilson.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: properly check for MODESET for kms driver ioctls

Also ditch the cargo-culted dev_priv checks - either we have a
giant hole in our setup code or this is useless. Plainly bogus
to check for it in either case.

v2: Chris Wilson noticed that I've missed one bogus dev_priv check.

v3: The check in the overlay code is redundant (Chris)

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: CR clock gating is recommend to be set on PineView

The specs recommend that this bit be set on PineView. No reason is
given, but it sounds like a powersaving bit that we should expect the
BIOS to be setting...

v2: Rebase on top of _MASKED_ENABLE_BIT

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Move GEM initialisation from i915_dma.c to i915_gem.c

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Split the stolen handling for GEM out of i915_dma.c

We slightly modify the initialisation sequence to move the
initialisation of the memory managers earlier and in particular before
probing outputs and detecting any existing output configuration. This is
essential if we wish to track preallocated objects and preserve them
whilst initialising GEM.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove the deferred-free list

The use of the mm_list by deferred-free breaks the following patches to
extend the range of objects tracked. We can simplify things if we just
make the unbind during free uninterrutible.

Note that unbinding should never fail, because we hold an additional
reference on every active object. Only the ilk vt-d workaround breaks
this, but already takes care of not failing by waiting for the gpu to
quiescent non-interruptible. But the existence of the deferred free
list casted some doubts on this theory, hence WARN if the unbind fails
and only then retry non-interruptible.

We can kill this additional code after a release in case the theory is
indeed right and no one has hit that WARN.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove the list of pinned inactive objects

Simplify object tracking by removing the inactive but pinned list. The
only place where this was used is for counting the available memory,
which is just as easy performed by checking all objects on the rare
occasions it is required (application startup). For ease of debugging,
we keep the reporting of pinned objects through the error-state and
debugfs.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove i915_gem_evict_inactive()

This was only used by one external caller who would just be as happy
with evict-everything, so perform the replacement and make the function
private.

In the process we note that unbinding the inactive list should not fail,
and make it a warning instead.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Bump the inactive LRU on set-to-GTT-domain

Currently, we only bump the inactive LRU of an object when we bind
into the GTT for a page-fault. As the object may be used many times
before its mapping is zapped, we do not mark it as active as
frequently as we should. Userspace should be calling set-to-GTT-domain
before each pointer deference (for synchronous access) and so is a
good place to mark the buffer as active.

Marking the buffer as recently used places it at the end of the
inactive eviction queue, though still before anything with outstanding
rendering. This reduces the likelihood of evicting a buffer that is
going to be used again by the CPU in the near future. This way we can
hopefully avoid to kick out upload buffers right before we use them on
the gpu.

Note that we need to check that the object is not active or pinned,
for otherwise we create havoc on the active/pinned lists, which also
use obj->mm_list.

The active lists are sorted by and evicted in last GPU rendering
order, access by the CPU to a still active buffer therefore does not
affect its eviction ordering. Pinned objects are currently excluded
from eviction, therefore the only list that we need to bump for GTT
access by the CPU is the inactive list.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Added further explanations to the commit message as discussed
on irc.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove too early plane enable on pre-PCH hardware

Enabling the plane before we have assigned valid address means that it
will access random PTE (often with conflicting memory types) and cause
GPU lockups. However, enabling the plane too early appears to workaround
a number of bugs in our modesetting code.

Cc: Franz Melchior <melchior.franz@gmail.com>
References: https://bugs.freedesktop.org/show_bug.cgi?id=39947
References: https://bugs.freedesktop.org/show_bug.cgi?id=41091
References: https://bugs.freedesktop.org/show_bug.cgi?id=49041
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Use a global lock for modifying global irq flags

We were attempting to use a per-ring spinlock whilst modifying global
IRQ flags. A recipe for rare missed interrupts.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: implement Disable4x2SubspanOptimization w/a for ivb, too

Copy&pasted from the vlv setup code. According to docs, we need that
on ivb, too.

v2: Use new masked bit handling macros.

Reviewed-by: Ben Widawsky <ben@bwidawsk.net>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: create macros to handle masked bits

... and put them to so good use.

Note that there's functional change in vlv clock gating code, we now
no longer spuriously read back the current value of the bit. According
to Bspec the high bits should always read zero, so ORing this in
should have no effect.

Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: manage PCH PLLs separately from pipes

PCH PLLs aren't required for outputs on the CPU, so we shouldn't just
treat them as part of the pipe.

So split the code out and manage PCH PLLs separately, allocating them
when needed or trying to re-use existing PCH PLL setups when the timings
match.

v2: add num_pch_pll field to dev_priv (Daniel)
    don't NULL the pch_pll pointer in disable or DPMS will fail (Jesse)
    put register offsets in pll struct (Chris)

v3: Decouple enable/disable of PLLs from get/put.
v4: Track temporary PLL disabling during modeset
v5: Tidy PLL initialisation by only checking for num_pch_pll == 0 (Eugeni)
v6: Avoid mishandling allocation failure by embedding the small array of
    PLLs into the device struct

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=44309
Signed-off-by: Jesse Barnes <jbarnes@virtuousgeek.org> (up to v2)
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk> (v3+)
Reviewed-by: Eugeni Dodonov <eugeni.dodonov@intel.com>
Tested-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: i8xx interrupt handler

gen2 hardware has some significant differences from the other interrupt
routines that were glossed over and then forgotten about in the
transition to KMS. Such as

- 16bit IIR
- PendingFlip status bit

This patch reintroduces a handler specifically for gen2 for the purpose
of handling pageflips correctly, simplifying code in the process.

v2: Also fixup ring get/put irq to only access 16bit registers (Daniel)

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=24202
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=41793
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: use posting_read16 in intel_ringbuffer.c and kill _driver
from the function names.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Only the zap the VMA after updating the tiling parameters

If we fail to unbind and so abort the change in tiling, we will have
removed the VMA for the object for no reason. The likelihood of unbind
failing is slim (other than ERESTARTSYS which will cause userspace to
try again), so the change is mostly for the principle.

Also improve the slightly stale comment.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Clarify the semantics of tiling_changed

Rename obj->tiling_changed to obj->fence_dirty so that it is clear that
it flags when the parameters for an active fence (including the
no-fence) register are changed.

Also, do not set this flag when the object does not have a fence
register allocated currently and the gpu does not depend upon the
unfence. This case works exactly like when a tiled object lost its
fence and hence does not need additional handling for the tiling
change in the code.

v2: Use fence_dirty to better express what the flag tracks and add a few
more details to the comments to serve as a reminder of how the GPU also
uses the unfenced register slot.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Add some bikeshed to the commit message about the stricter
use of fence_dirty.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: fixup tv load-detect on enabled but not active crtc

When fixing up the crt load detect code I've failed to notice the same
problem in the tv load detect code. Again, unconditionally use the
load detect pipe infrastructure, it gets things right.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: re-add static qualifier to get_cxsr_latency

This got lost in the intel_pm.c move.

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

drm/i915: [sparse] __iomem fixes for gem

As with one of the earlier patches in the series, we're forced to cast
for copy_[to|from]_user. Again because of the nature of the GEN x86
exclusivity, this should be safe.

Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
[danvet: Added some bikeshed.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: [sparse] __iomem fixes for debugfs

These were mostly straight forward. No forced casting needed.

Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
[danvet: fix conflict with ringbuffer_data removal and drop the hunk
about the status page - that needs more care to fix up.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: [sparse] __iomem fixes for overlay

With the exception of a forced cast for phys_obj stuff (a problem in
other patches as well) all of these are fairly simple __iomem compliance
fixes.

As with other patches, yank/paste errors may exist.

Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
[danvet: Added comment to explain the __iomem cast.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: [sparse] __iomem fixes for opregion

Almost all of the errors related __iomem problems.

Most of the changes here are trivial, however there is plenty of chance
for yank/paste errors.

Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Remove i915_gem_ringbuffer_(data|info) from debugfs

This was originally used as an attempt to diagnose GPU hangs, but was
never very reliable and superseded by the i915_error_state capture on
hangcheck. It now lies languishing unused and unwanted.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915/sprite: Avoid incurring extra vblank stall when updating plane on IVB

IvyBridge requires an extra frame between disabling the low power
watermarks and enabling scaling on the sprite plane. If the scaling
is already enabled, then we have already disabled the low power
watermarks and need not incur an extra wait.

Similarly, as we disable the scaling when turning off the sprite plane,
we can update the scaling enabled flag and restore the low power
watermarks.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Jesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: print computed bpp in dp link configuration

Pretty useful to debug our DP bandwidth woes.

v2: Also print out the required and available link bandwidth,
suggested by Chris Wilson.

v3: Also print out the input parameters so that diagnosing failures to
find a valid dp link configuration is possible.

v4: s/Display port/DP/ to shorten the output.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Merge tag 'drm-intel-next-2012-04-23' of git://people.freedesktop.org/~danvet/drm-intel into drm-core-next

Daniel Vetter writes:

A new drm-intel-next pull. Highlights:
- More gmbus patches from Daniel Kurtz, I think gmbus is now ready, all
 known issues fixed.
- Fencing cleanup and pipelined fencing removal from Chris.
- rc6 residency interface from Ben, useful for powertop.
- Cleanups and code reorg around the ringbuffer code (Ben&me).
- Use hw semaphores in the pageflip code from Ben.
- More vlv stuff from Jesse, unfortunately his vlv cpu is doa, so less
 merged than I've hoped for - we still have the unused function warning :(
- More hsw patches from Eugeni, again, not yet enabled fully.
- intel_pm.c refactoring from Eugeni.
- Ironlake sprite support from Chris.
- And various smaller improvements/fixes all over the place.

Note that this pull request also contains a backmerge of -rc3 to sort out
a few things in -next. I've also had to frob the shortlog a bit to exclude
anything that -rc3 brings in with this pull.

Regression wise we have a few strange bugs going on, but for all of them
closer inspection revealed that they've been pre-existing, just now
slightly more likely to be hit. And for most of them we have a patch
already. Otherwise QA has not reported any regressions, and I'm also not
aware of anything bad happening in 3.4.

* tag 'drm-intel-next-2012-04-23' of git://people.freedesktop.org/~danvet/drm-intel: (420 commits)
  drm/i915: rc6 residency (fix the fix)
  drm/i915/tv: fix open-coded ARRAY_SIZE.
  drm/i915: invalidate render cache on gen2
  drm/i915: Silence the change of LVDS sync polarity
  drm/i915: add generic power management initialization
  drm/i915: move clock gating functionality into intel_pm module
  drm/i915: move emon functionality into intel_pm module
  drm/i915: move drps, rps and rc6-related functions to intel_pm
  drm/i915: fix line breaks in intel_pm
  drm/i915: move watermarks settings into intel_pm module
  drm/i915: move fbc-related functionality into intel_pm module
  drm/i915: Refactor get_fence() to use the common fence writing routine
  drm/i915: Refactor fence clearing to use the common fence writing routine
  drm/i915: Refactor put_fence() to use the common fence writing routine
  drm/i915: Prepare to consolidate fence writing
  drm/i915: Remove the unsightly "optimisation" from flush_fence()
  drm/i915: Simplify fence finding
  drm/i915: Discard the unused obj->last_fenced_ring
  drm/i915: Remove unused ring->setup_seqno
  drm/i915: Remove fence pipelining
  ...

drm/radeon/kms/hdmi: use relative offsets, official regs

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Tested-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: keep HDMI state in separated variable

If we want hdmi_offset to be relative to the first block, zero value can
be used also for enabled block.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Tested-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: get rid of r600_hdmi_find_free_block

R6xx has routable blocks, but there's nothing wrong in assignment based
on dig_encoder. We didn't really need that algorithm.

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Tested-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: get rid of hdmi_config_offset

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Tested-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: move audio params to separated struct

Signed-off-by: Rafał Miłecki <zajec5@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Tested-by: Christian König <deathsimple@vodafone.de>
Reviewed-by: Christian König <deathsimple@vodafone.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: fix collision between two patches breaking build

this fixes a report that the new load code needed to be updated for
ajax's validity changes.

Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: fixed: Add dfixed_frac

This helper macro retrieves the fractional part of a fixed20_12 20.12
fixed-point number.

Signed-off-by: Robert Morell <rmorell@nvidia.com>
Signed-off-by: Olof Johansson <olofj@chromium.org>
Signed-off-by: Thierry Reding <thierry.reding@avionic-design.de>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: Set the mapping mask

Some boards such as the Intel D2700MUD allow you to have over 4GB of RAM.
The GTT on the PVR based devices is 32bit however. Hugh Dickins points out
that we should therefore be setting the mapping gfp mask.

This is not the whole fix for the problem. Some further shmem patches will
be needed to deal with the corner cases.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: Add the base elements of CDV hotplug support

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: Add ops for hotplug support.

This provides the needed callback hooks to add hotplug display support to
the GMA36x0 devices.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

cdv: continue synching up with updated reference code

In particular clean up the errata handling and correct the crtc masks. We do
this a bit differently using our device abstraction for neatness.

This doesn't address the ACPI opregion and hotplug plumbing, nor the IRQ related
changes that will need. It touches on backlight init but the full backlight
support is not in this change set.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: Clean up weirdness in the cdv mode test code

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: support 1080p

The problem in console mode is lack of linear memory. We can solve that by
dropping to 16bpp. The mode setting X server will allocate its own GEM
framebuffer in 32bpp and all will be well.

We could just do 16bpp anyway but that would be a regression on the lower
modes as many distributions don't yet ship the generic mode setting KMS
drivers.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: panel presence check

Introduce a panel presence check for Cedartrail. Non netbook devices don't
necessarily have a panel attached.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: intel_bios updates

Pull in various i915 bits that we will need to begin tackling the LVDS detect
and ACPI events. We try and drift towards the i915 version of the code with
the long term goal that at least some of it can one day be unified.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: mark framebuffer pages write combining

We don't want them uncached, combining will do nicely and fixes the performance
problem with the generic modesetting X server.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: Update the Cedarview clock handling

Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: read the PLL bits

We need to pull more stuff from the VBT in order to configure the clocking
correctly in all cases. Add the relevant bits from the other CDV driver work.

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

gma500: Fix leak of uncached page

This was reported a long time ago (and I apologize to whoever it was that
reported it as I've lost the original report).

Signed-off-by: Alan Cox <alan@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Store vendor IDs directly in the EDID quirk structure

EDID vendor IDs are always 3 characters long (4 with the terminating
0).  It doesn't make any sense to have a (possibly 8-byte) pointer
to the ID string in the quirk structure.

Signed-off-by: Ian Pilcher <arequipeno@gmail.com>
Reviewed-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Try harder to fix up base EDID blocks

Requiring the first byte of the EDID base block header to be 0 means we
don't fix up as many transfer errors as we could.  Instead have the
callers specify whether it's meant to be block 0 or not, and
conditionally run header fixup based on that.

Bugzilla: https://bugzilla.redhat.com/812890
Signed-off-by: Adam Jackson <ajax@redhat.com>
Reviewed-by: Alex Deucher <alexdeucher@gmail.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Dave Airlie <airlied@redhat.com>

vga: fix build when fbdev is a module

This fixes the build breakage reported by Stephen in -next
when merging the drm-next tree.

Signed-off-by: Dave Airlie <airlied@redhat.com>

drm: Unify and fix idr error handling

The error handling code w.r.t. idr usage looks inconsistent.

In the case of drm_mode_object_get() and drm_ctxbitmap_next() the error
handling is also incomplete.

Unify the code to follow the same pattern always.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

vga-switcheroo: select VGA arbitration.

Since Matthew's changes we have to select arbitration.

Reported-by: devh on #radeon
Signed-off-by: Dave Airlie <airlied@redhat.com>

efifb: Implement vga_default_device() (v2)

EFI doesn't typically make use of the legacy VGA ROM, but it may still be
configured to pass that through to a given video device. This may lead to
an inaccurate choice of default video device. Add support to efifb to pick
out the correct active video device.

v2: fix if->ifdef

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Acked-by: hpa@zytor.com
Cc: matt.fleming@intel.com
Signed-off-by: Dave Airlie <airlied@redhat.com>

x86: Use vga_default_device() when determining whether an fb is primary

IORESOURCE_ROM_SHADOW is not necessarily an indication that the hardware
is the primary device. Add support for using the vgaarb functions and
fall back if nothing's set them.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Cc: mingo@redhat.com
Acked-by: hpa@zytor.com
Signed-off-by: Dave Airlie <airlied@redhat.com>

vga-switcheroo: Use vga_default_device()

vga-switcheroo currently changes the default VGA device by fiddling with
the IORESOURCE_ROM_SHADOW flag on the device. This isn't strictly accurate,
since there's no guarantee that switching also changes the ROM decoding.
Switch over to using the vgaarb functions for this.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

vgaarb: Add support for setting the default video device (v2)

The default VGA device is a somewhat fluid concept on platforms with
multiple GPUs. Add support for setting it so switching code can update
things appropriately, and make sure that the sysfs code returns the right
device if it's changed.

v2: Updated to fix builds when __ARCH_HAS_VGA_DEFAULT_DEVICE is false.

Signed-off-by: Matthew Garrett <mjg@redhat.com>
Acked-by: H. Peter Anvin <hpa@zytor.com>
Acked-by: benh@kernel.crashing.org
Cc: airlied@redhat.com
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: fix up audio interrupt handling

- add support for rs6xx
- add support for DCE4/5
- fixup 6xx/7xx

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: add register definitions for audio

This adds register definitions for HDMI/DP audio on
DCE2/3/4/5 hardware.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/radeon/kms: improve bpc handling (v2)

Improve handling of bpc (bits per color) in radeon.
In most cases we want 8 except for HDMI, DP, LVDS, and eDP.

v2: handle DP better.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Tested-by: Lennert Buytenhek <buytenh@wantstofly.org>
Signed-off-by: Dave Airlie <airlied@redhat.com>

mm: fixup compilation error due to an asm write through a const pointer

This regression has been introduced in

commit f56f821feb7b36223f309e0ec05986bb137ce418
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Sun Mar 25 19:47:41 2012 +0200

    mm: extend prefault helpers to fault in more than PAGE_SIZE

I have failed to notice this because x86 asm seems to happily compile
things as-is.

Reported-by: Geert Uytterhoeven <geert@linux-m68k.org
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Add packed attribute to new gtf2 and cvt structs

The new structs added in struct detailed_data_monitor_range must be
marked with packed attribute although the outer struct itself is
already marked as packed.  Otherwise these 7-bytes structs may be
aligned, and give the wrong position and size for the data.

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Acked-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Add a workaround for 1366x768 HD panel

HD panel (1366x768) found most commonly on laptops can't be represented
exactly in CVT/DMT expression, which leads to 1368x768 instead, because
1366 can't be divided by 8.

Add a hack to convert to 1366x768 manually as an exception.

Signed-off-by: Takashi Iwai <tiwai@suse.de>
Acked-by: Adam Jackson <ajax@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/i915: rc6 residency (fix the fix)

Chris' fix for my 32b breakage was incorrect. do_div returns a
remainder. Go back to a divide macro which is more 32b friendly.

Tested on x86-64.

This has only been compile tested on 32b systems.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=48756
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Sincere-apologies: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Ben Widawsky <benjamin.widawsky@intel.com>
[danvet: fixup 32bit compile-fail.]
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915/tv: fix open-coded ARRAY_SIZE.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm: replace open-coded ARRAY_SIZE with macro

[airlied: fixed one more new one added since]

Signed-off-by: Jim Cromie <jim.cromie@gmail.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: add missing NULL checks.

Reviewed-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Generate modes from extra_modes for range descriptors

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Add extra_modes

Some common sizes that don't show up in DMT.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Give the est3 mode struct a real name

We want the same type for extra modes inferred from ranges.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Update range descriptor struct for EDID 1.4

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Do drm_dmt_modes_for_range() for all range descriptor types

EDID 1.4 retcons the meaning of the "GTF feature" bit to mean "is
continuous frequency", and moves the set of supported timing formulas
into the range descriptor itself.  In any event, the range descriptor
can act as a filter on the DMT list without regard to a specific timing
formula.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Fix some comment typos in the DMT mode list

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Add the reduced blanking DMT modes to the DMT list

Copied from the list in xserver.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: s/drm_gtf_modes_for_range/drm_dmt_modes_for_range/

Slightly more honest naming.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Remove a misleading comment

mode_in_range() handles what this was warning about.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Allow drm_mode_find_dmt to hunt for reduced-blanking modes

It won't find any, yet.  Fix up callers to match: standard mode codes
will look prefer r-b modes for a given size if present, EST3 mode codes
will look for exactly the r-b-ness mentioned in the mode code.  This
might mean fewer modes matched for EST3 mode codes between now and when
the DMT mode list regrows the r-b modes, but practically speaking EST3
codes don't exist in the wild.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Rewrite drm_mode_find_dmt search loop

No functional change, but will make an upcoming change clearer.

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>

drm/edid: Document drm_mode_find_dmt

Signed-off-by: Adam Jackson <ajax@redhat.com>
Tested-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>