When dealing with a working set larger than the GATT, or even the
mappable aperture when touching through the GTT, we end up with evicting
objects only to rebind them at a new offset again later. Moving an
object into and out of the GTT requires clflushing the pages, thus
causing a double-clflush penalty for rebinding.

To avoid having to clflush on rebinding, we can track the pages as they
are evicted from the GTT and only relinquish those pages on memory
pressure.

As usual, if it were not for the handling of out-of-memory condition and
having to manually shrink our own bo caches, it would be a net reduction
of code. Alas.

Note: The patch also contains a few changes to the last-hope
evict_everything logic in i916_gem_execbuffer.c - we no longer try to
only evict the purgeable stuff in a first try (since that's superflous
and only helps in OOM corner-cases, not fragmented-gtt trashing
situations).

Also, the extraction of the get_pages retry loop from bind_to_gtt (and
other callsites) to get_pages should imo have been a separate patch.

v2: Ditch the newly added put_pages (for unbound objects only) in
i915_gem_reset. A quick irc discussion hasn't revealed any important
reason for this, so if we need this, I'd like to have a git blame'able
explanation for it.

v3: Undo the s/drm_malloc_ab/kmalloc/ in get_pages that Chris noticed.
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: Split out code movements and rant a bit in the commit message
with a few Notes. Done v2]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Prep work to make Chris Wilson's unbound tracking patch a bit easier
to read. Alas, I'd have preferred that moving the page allocation
retry loop from bind to get_pages would have been a separate patch,
too. But that looks like real work ;-)
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Added new haswell_write_eld() to initialize Haswell HDMI audio registers
to generate an unsolicited response to the audio controller driver to
indicate that the controller sequence should start.
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Wang Xingchao <xingchao.wang@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

We need to call these before we transfer the damaged areas to the device
not before/after we setup the long lived vmaps.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

In order for udl vmap to work properly, we need to push the object
into the CPU domain before we start copying the data to the USB device.

This along with the udl change avoids userspace explicit mapping to
be used.

v2: add a flag for userspace to query to know if Intel kernel driver can
deal with the vmap flushing properly. In theory udl would need a flag also,
but I intend to push the patches very close to each other and other drivers
should do the right thing from the start.

I've added a test to my intel-gpu-tools prime branch, however testing
this is a bit messy since the only way to get udl to vmap is to rendering
something. I've tested this with real code as well to make sure it works.
Signed-off-by: Dave Airlie <airlied@redhat.com>
[danvet: resolved conflict, which required reallocating the PARAM
number to 21.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

This is left over from the old PLL sharing code and isn't useful now
that PLLs are shared when possible.
Signed-off-by: Keith Packard <keithp@keithp.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

New-ish devices have 3 pipes, so let's not just hardcode 2 but use the
for_each_pipe() macro and make struct intel_display_error_state is big
enough.

V2: Also add the number of pipes emitted (Chris Wilson)
Signed-off-by: Damien Lespiau <damien.lespiau@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Yet again the too close relationship between the fb helper and the
crtc helper code strikes. This time around the fb helper resets all
encoder->crtc pointers to NULL before starting to set up it's own
mode. Which is total bullocks, because this will clobber the existing
output routing, which the new drm/i915 code depends upon to be
absolutely correct.

The crtc helper itself doesn't really care about that, since it
disables unused encoders in a rather roundabout way anyway.

Two places call drm_setup_crts:

- For the initial fb config. I've auditted all current drivers, and
  they all allocate their encoders with kzalloc. So there's no need to
  clear encoder->crtc once more.

- When processing hotplug events while showing the fb console. This
  one is a bit more tricky, but both the crtc helper code and the new
  drm/i915 modeset code disable encoders if their crtc is changed and
  that encoder isn't part of the new config. Also, both disable any
  disconnected outputs, too.

  Which only leaves encoders that are on, connected, but for some odd
  reason the fb helper code doesn't want to use. That would be a bug
  in the fb configuration selector, since it tries to light up as many
  outputs as possible.

v2: Kill the now unused encoders variable.
Acked-by: Dave Airlie <airlied@gmail.com>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Use _PIPE macro to get correct register definition for IBX/CPT, discard
old variable "i" way.
Signed-off-by: Wang Xingchao <xingchao.wang@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
[danvet: Added the DIP_PORT_SEL #define from a preceeding patch in the
series that needs more work.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

HDMI audio related registers will be configured in write_eld callback.
Signed-off-by: Wang Xingchao <xingchao.wang@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Add hsw audio registers definition
Signed-off-by: Wang Xingchao <xingchao.wang@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Simply to make the ilk+ crtc disable path clearer and more symmetric
with the enable function.

Also switch to intel_crtc for the enable function.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

... and move a few others only used by i915_dma.c into the dri1
dungeon.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

All dvo drivers only support 2 dpms states, and our dvo driver
even switches of the dvo port for anything else than DPMS_ON. Hence
ditch this complexity and simply use bool enable.

While reading through this code I've noticed that the mode_set
function of ch7017 is a bit peculiar - it disable the lvds again, even
though the crtc helper code should have done that ... This might be to
work around an issue at driver load, we pretty much ignore the hw
state when taking over.

v2: Also do the conversion for the new ns2501 driver.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Since it's redundant - we can get the attached encoder in the
functions themselves.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

A few things need adjustement:
- Change the dpms state by calling the dpms connector function and
  not some crtc helper internal callbacks. Otherwise this will break
  once we switch to our own dpms handling.
- Instead of tracking and restoring intel_crtc->dpms_mode use the
  connector's dpms variable - the former relies on the dpms compuation
  rules used by the crtc helper. And it would break when the encoder
  is cloned and the other output has a different dpms state. But luckily
  no one is crazy enough for that.
- Properly clear the connector -> encoder -> crtc linking, even when
  failing (note that the crtc helper removes the encoder -> crtc link
  in disabled_unused_functions for us).
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Now that all affected i830M systems have the pipe A quirk set,
we don't need to do any special dances in the overlay code any
longer. And reading through the code I'm rather dubios that it
actually does what it claims to do ...

As a nice benefit this rips out a users of the crtc helper dpms
callback.

v2: As suggested by Chris Wilson, replace the code by an appropriate
WARN to ensure that the pipe A is indeed running.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

With the pipe A quirk properly fixed up for i830M, this shouldn't be
required any longer.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-Off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

For some odd reason we've missed i830 and a i855 variant. Also
kill the two now redundant i830 entries.

v2: Don't add the missing 855 id to the pipe A quirk list, we seem to
lack justification for it.
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Wrong order of parameters passed-in when calling hdmi/adpa
/lvds_pipe_enabled(), 2nd and 3rd parameters are reversed.

This bug was indroduced by

commit 1519b995
Author: Keith Packard <keithp@keithp.com>
Date:   Sat Aug 6 10:35:34 2011 -0700

    drm/i915: Fix PCH port pipe select in CPT disable paths

The reachable tag for this commit is v3.1-rc1-3-g1519b995Signed-off-by: Anhua Xu <anhua.xu@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Backmerge Linux 3.6-rc2 to resolve a few funny conflicts before we put
even more madness on top:

- drivers/gpu/drm/i915/i915_irq.c: Just a spurious WARN removed in
  -fixes, that has been changed in a variable-rename in -next, too.

- drivers/gpu/drm/i915/intel_ringbuffer.c: -next remove scratch_addr
  (since all their users have been extracted in another fucntion),
  -fixes added another user for a hw workaroudn.
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Following a report of a crash during an automount expire I found that
the locking in fs/autofs4/expire.c:get_next_positive_subdir() was wrong.
Not only is the locking wrong but the function is more complex than it
needs to be.

The function is meant to calculate (and dget) the next entry in the list
of directories contained in the root of an autofs mount point (an autofs
indirect mount to be precise). The main problem was that the d_lock of
the owner of the list was not being taken when walking the list, which
lead to list corruption under load. The only other lock that needs to
be taken is against the next dentry candidate so it can be checked for
usability.
Signed-off-by: Ian Kent <raven@themaw.net>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull VFIO fix from Alex Williamson:
 "Just a trivial patch to include vfio.h in the installed headers so we
  can complete userspace integration into QEMU."

* tag 'vfio-for-v3.6-rc1' of git://github.com/awilliam/linux-vfio:
  vfio: Include vfio.h in installed headers

Pull fuse updates from Miklos Szeredi.

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/mszeredi/fuse:
  fuse: verify all ioctl retry iov elements
  fuse: add missing INIT flag descriptions
  fuse: add missing INIT flags
  fuse: update attributes on aio_read
  fuse: invalidate inode mapping if mtime changes
  fuse: add FUSE_AUTO_INVAL_DATA init flag

Merge tag 'stable/for-linus-3.6-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen

Pull Xen fix from Konrad Rzeszutek Wilk:
 "Way back in v3.5 we added a mechanism to populate back pages that were
  released (they overlapped with MMIO regions), but neglected to reserve
  the proper amount of virtual space for extend_brk to work properly.

  Coincidentally some other commit aligned the _brk space to larger area
  so I didn't trigger this until it was run on a machine with more than
  2GB of MMIO space."

 * On machines with large MMIO/PCI E820 spaces we fail to boot b/c
   we failed to pre-allocate large enough virtual space for extend_brk.

* tag 'stable/for-linus-3.6-rc1-tag' of git://git.kernel.org/pub/scm/linux/kernel/git/konrad/xen:
  xen/p2m: Reserve 8MB of _brk space for P2M leafs when populating back.

Pull SuperH fixes from Paul Mundt.

* tag 'sh-for-linus' of git://github.com/pmundt/linux-sh:
  sh: intc: Handle domain association for sparseirq pre-allocated vectors.
  sh: sh7269: Fix LCD pinmux
  sh: dma: fix request_irq usage

Moved to djbw@fb.com

Cc: Dave Jiang <dave.jiang@intel.com>
Cc: Vinod Koul <vinod.koul@linux.intel.com>
Signed-off-by: Dan Williams <djbw@fb.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

When dumping "Code: " sections from an oops, the trapping instruction
%rip points to can be a string copy

  2b:*  f3 a5                   rep movsl %ds:(%rsi),%es:(%rdi)

and the line contain a bunch of ":".  Current "cut" selects only the and
the second field output looks funnily overlaid this:

  2b:*  f3 a5                   rep movsl %ds     <-- trapping instruction:(%rsi),%es:(%rdi

Fix this by selecting the remaining fields too.

Cc: Andrew Morton <akpm@linux-foundation.org>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: linux-kbuild@vger.kernel.org
Signed-off-by: Borislav Petkov <borislav.petkov@amd.com>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>

Pull two slave-dmaengine fixes from Vinod Koul:
 "One fixes the correct use of clock API in imx driver and the other
  enables clock for tegra driver, which is used for other tegra driver
  conversion to dmanegine in -next."

* 'fixes' of git://git.infradead.org/users/vkoul/slave-dma:
  dma: tegra: enable/disable dma clock
  dma: imx-dma: Fix kernel crash due to missing clock conversion

Pull more drm fixes from Dave Airlie:
 "Just some intel and nouveau ones this time, intel has more edp panel
  fixes for macbooks and nouveau has a suspend/resume regression fix in
  there."

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux:
  drm/i915: Apply post-sync write for pipe control invalidates
  drm/i915: reorder edp disabling to fix ivb MacBook Air
  drm/nv86/fifo: suspend fix
  drm/nouveau: disable copy engine on NVAF
  nouveau: fixup scanout enable in nvc0_pm
  drm/nouveau/aux: mask off higher bits of auxch index in i2c table entry
  drm/nvd0/disp: mask off high 16 bit of negative cursor x-coordinate
  drm/i915: ensure i2c adapter is all set before adding it
  drm/i915: ignore eDP bpc settings from vbt
  drm/i915: Fix blank panel at reopening lid
  drm/nve0/fifo: add support for the flip completion swmthd

Pull two sparc fixes from David S. Miller.

* git://git.kernel.org/pub/scm/linux/kernel/git/davem/sparc:
  sparc64: Be less verbose during vmemmap population.
  sparc64: do not clobber personality flags in sys_sparc64_personality()

* 'drm-nouveau-fixes' of git://git.freedesktop.org/git/nouveau/linux-2.6:
  drm/nv86/fifo: suspend fix
  drm/nouveau: disable copy engine on NVAF
  nouveau: fixup scanout enable in nvc0_pm
  drm/nouveau/aux: mask off higher bits of auxch index in i2c table entry
  drm/nvd0/disp: mask off high 16 bit of negative cursor x-coordinate
  drm/nve0/fifo: add support for the flip completion swmthd

Daniel Vetter writes:

"A few important fixers:
- fix various lvds backlight issues, regressed in 3.6 (Takashi Iwai)
- make the retina mbp work (ignore bogus edp bpc value in vbt)
- fix a gmbus regression introduced in (iirc) 3.4 (Jani Nikula)
- fix an edp panel power sequence regression, fixes the new macbook air
- apply the tlb invalidate w/a

Otherwise we still have another gmbus regression (patches are awaiting
tested-bys) and there's something odd going with some rare systems not
entering rc6 often enough (and hence blowing through too much power).  It
seems to be a timing-related issue and can be mitigated by frobbing the
magic tuning parameters. We're still working on that one. Also, we still
have some fallout from the hw context support, but you can only hit that
with mesa master."

* 'drm-intel-fixes' of git://people.freedesktop.org/~danvet/drm-intel:
  drm/i915: Apply post-sync write for pipe control invalidates
  drm/i915: reorder edp disabling to fix ivb MacBook Air
  drm/i915: ensure i2c adapter is all set before adding it
  drm/i915: ignore eDP bpc settings from vbt
  drm/i915: Fix blank panel at reopening lid

On a 2-node machine with 256GB of ram we get 512 lines of
console output, which is just too much.

This mimicks Yinghai Lu's x86 commit c2b91e2e
(x86_64/mm: check and print vmemmap allocation continuous) except that
we aren't ever going to get contiguous block pointers in between calls
so just print when the virtual address or node changes.

This decreases the output by an order of 16.

Also demote this to KERN_DEBUG.
Signed-off-by: David S. Miller <davem@davemloft.net>

When invalidating the TLBs it is documentated as requiring a post-sync
write. Failure to do so seems to result in a GPU hang.

Exposure to this hang on IVB seems to be a result of removing the extra
stalls required for SNB pipecontrol workarounds:

commit 6c6cf5aa
Author: Chris Wilson <chris@chris-wilson.co.uk>
Date:   Fri Jul 20 18:02:28 2012 +0100

    drm/i915: Only apply the SNB pipe control w/a to gen6

Note: Manually switch the pipe_control cmd to 4 dwords to avoid a
(silent) functional conflict with -next. This way will get a loud (but
conflict with next (since the scratch_addr has been deleted there).

Reported-and-tested-by: yex.tian@intel.com
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=53322Acked-by: Ben Widawsky <ben@bwidawsk.net>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
[danvet: added note about merge conflict with -next.]
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

eDP is tons of fun. It turns out that at least the new MacBook Air 5,1
model absolutely doesn't like the new force vdd dance we've introduced
in

commit 6cb49835
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Sun May 20 17:14:50 2012 +0200

    drm/i915: enable vdd when switching off the eDP panel

But that patch also tried to fix some neat edp sequence issue with the
force_vdd timings. Closer inspection reveals that we've raised
force_vdd only to do the aux channel communication dp_sink_dpms. If we
move the edp_panel_off below that, we don't need any force_vdd for the
disable sequence, which makes the Air happy.

Unfortunately the reporter of the original bug that the above commit
fixed is travelling, so we can't test whether this regresses things.
But my theory is that since we don't check for any power-off ->
force_vdd-on delays in edp_panel_vdd_on, this was the actual
root-cause of this failure. With that force_vdd dance completely
eliminated, I'm hopeful the original bug stays fixed, too.

For reference the old bug, which hopefully doesn't get broken by this:

https://bugzilla.kernel.org/show_bug.cgi?id=43163

In any case, regression fixers win over plain bugfixes, so this needs
to go in asap.

v2: The crucial pieces seems to be to clear the force_vdd flag
uncoditionally, too, in edp_panel_off. Looks like this is left behind
by the firmware somehow.

v3: The Apple firmware seems to switch off the panel on it's own, hence
we still need to keep force_vdd on, but properly clear it when switching
the panel off.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=45671Tested-by: Roberto Romer <sildurin@gmail.com>
Tested-by: Daniel Wagner <wagi@monom.org>
Tested-by: Keith Packard <keithp@keithp.com>
Cc: stable@vger.kernel.org
Cc: Keith Packard <keithp@keithp.com>
Signed-off-by: Daniel Vetter <daniel.vetter@ffwll.ch>

Pull s390 patches from Martin Schwidefsky:
 "Included are bug fixes and a patch to enable system call filtering
  with BPF."

* 'for-linus' of git://git.kernel.org/pub/scm/linux/kernel/git/s390/linux:
  s390/compat: fix mmap compat system calls
  s390/compat: fix compat wrappers for process_vm system calls
  s390: do not clobber personality flags in sys_32_personality()
  s390/seccomp: add support for system call filtering using BPF
  s390/sclp_sdias: Add missing break and "fall through"
  s390/mm: remove MAX_PHYSADDR_BITS define

Pull drm fixes from Dave Airlie:
 "Radeon and intel fixes mostly, one fix to the mgag200 driver to not
  hang on certain server variants."

* 'drm-fixes' of git://people.freedesktop.org/~airlied/linux: (32 commits)
  drm/radeon: fix typo in function header comment
  drm/radeon/kms: implement timestamp userspace query (v2)
  drm/radeon/kms: add MSAA texture support for r600-evergreen
  drm/radeon/kms: reorder code in r600_check_texture_resource
  drm/radeon: fence virtual address and free it once idle v4
  drm/radeon: fix some missing parens in asic macros
  drm/radeon: add some new SI pci ids
  drm/radeon: fix ordering in pll picking on dce4+
  drm/radeon: do not reenable crtc after moving vram start address
  drm/radeon: fix bank tiling parameters on cayman
  drm/radeon: fix bank tiling parameters on evergreen
  drm/radeon: fix bank tiling parameters on SI
  drm/radeon: properly handle crtc powergating
  drm/radeon: properly handle SS overrides on TN (v2)
  drm/radeon/dce4+: set a more reasonable cursor watermark
  drm/radeon: fix handling for ddc type 5 on combios
  drm/mgag200: fix G200ER pll picking algorithm
  drm/edid: Fix potential memory leak in edid_load()
  drm/udl: Use ERR_CAST inlined function instead of ERR_PTR(PTR_ERR(.. [1]
  drm/radeon/kms: allow "invalid" DB formats as a means to disable DB
  ...

Commit 72121572 ("GPIO: gpio-pxa: fix devicetree functions") added an
"xlate" function pointer to the irq_domain_ops, but this function is nor
declared or defined anywhere when CONFIG_OF is disabled, causing the
build error:

  drivers/gpio/gpio-pxa.c:532:11: error: 'irq_domain_xlate_twocell' undeclared here (not in a function)

Extending the DT-only code section to cover the irq_domain_ops and the
pxa_gpio_dt_ids solves this problem and makes it clearer which code is
actually used without DT.
Signed-off-by: Arnd Bergmann <arnd@arndb.de>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>