The dithering introduced in

commit 3b5c78a3
Author: Adam Jackson <ajax@redhat.com>
Date:   Tue Dec 13 15:41:00 2011 -0800

    drm/i915/dp: Dither down to 6bpc if it makes the mode fit

stores the INTEL_MODE_DP_FORCE_6BPC flag in the private_flags of the
adjusted mode, while i9xx_crtc_mode_set() and ironlake_crtc_mode_set() use
the original mode, without the flag, so it would never have any
effect. However, the BPC was clamped by VBT settings, making things work by
coincidence, until that part was removed in

commit 4344b813
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Fri Aug 10 11:10:20 2012 +0200

Use adjusted_mode instead of mode when checking for
INTEL_MODE_DP_FORCE_6BPC to make the flag have effect.

v2: Don't forget to fix this in i9xx_crtc_mode_set() also, pointed out by
Daniel both before and after sending the first patch.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=47621
CC: Adam Jackson <ajax@redhat.com>
CC: stable@vger.kernel.org
Signed-off-by: NJani Nikula <jani.nikula@intel.com>
Reviewed-by: NAdam Jackson <ajax@redhat.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Finishes commit 02d71956
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu Aug 9 16:44:54 2012 +0200

    drm/i915: properly guard ilk ips state

The core functions were annotated with their locking requirements, but
we overlooked that they were exported, without any control over the
locking, to debugfs. So in order to enable debugfs to read the registers
without triggering sanity checks, we change the exported entry points to
properly take the required locks before calling the core routines.
Reported-by: Nyangguang <guang.a.yang@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=55304Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

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

... even if the actual infoframe is smaller than the maximum possible
size.

If we don't write all the 32 DIP data bytes the InfoFrame ECC may not
be correctly calculated in some cases (e.g., when changing the port),
and this will lead to black screens on HDMI monitors. The ECC value is
generated by the hardware.

I don't see how this should break anything since we're writing 0 and
that should be the correct value, so this patch should be safe.

Notice that on IVB and older we actually have 64 bytes available for
VIDEO_DIP_DATA, but only bytes 0-31 actually store infoframe data: the
others are either read-only ECC values or marked as "reserved". On HSW
we only have 32 bytes, and the ECC value is stored on its own separate
read-only register. See BSpec.

This patch fixes bug #46761, which is marked as a regression
introduced by commit 4e89ee17:
    drm/i915: set the DIP port on ibx_write_infoframe

Before commit 4e89 we were just failing to send AVI infoframes when we
needed to change the port, which can lead to black screens in some
cases. After commit 4e89 we started sending infoframes, but with a
possibly wrong ECC value. After this patch I hope we start sending
correct infoframes.

Version 2:
  - Improve commit message
  - Try to make the code more clear

Cc: stable@vger.kernel.org
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=46761Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: NRodrigo Vivi <rodrigo.vivi@gmail.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This should never happen, but the silent "return" makes me wonder
every time I try to debug InfoFrame bugs, so promote this to BUG() to
make sure people will complain if we ever break this.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This fixes the gpio reset problem so the Retina MBP works, but avoids
breaking the Dell systems. Ben will work on a better solution for 3.7.

Tested by me on retina MBP.
Signed-off-by: NDave Airlie <airlied@redhat.com>

Cards typically have 5-7 scratch registers; one of these is reserved for
rdev->rptr_save_reg. Unfortunately the reservation is done in function
r100_cp_init, which is called by all drivers except r600 - and this
function is also invoked on resume from suspend. After several resumes,
no scratch registers are free and graphics acceleration is disabled.

Dmesg then reports either:
   *ERROR* radeon: cp failed to get scratch reg (-22).
   *ERROR* radeon: cp isn't working(-22).
   radeon 0000:01:00.0: failed initializing CP (-22).
or:
   *ERROR* radeon: failed to get scratch reg (-22).
   *ERROR* radeon: failed testing IB on GFX ring (-22).
   *ERROR* ib ring test failed (-22).

The chain of calls on boot for all except r600 is:
radeon_init -> ... -> (rXXX_init) -> rXXX_startup -> r100_cp_init

The chain of calls on resume for all except r600 is:
rXXX_resume -> rXXX_startup -> r100_cp_init.

R600 correctly allocates rptr_save_reg in r600_init (ie once only, not
in resume). However moving the code into the init functions for all
drivers means touching 4 drivers. So instead, this patch just adds a
test in r100_cp_init to avoid reallocating on resume. As the rdev
structure is allocated via kzalloc in radeon_driver_load_kms, and zero
is not a valid registerid, zero safely implies not-yet-allocated.

This issue appears to have been introduced in c7eff978 (3.6.0-rcN)
Signed-off-by: NSimon Kitching <skitching@vonos.net>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Somehow this hunk got dropped from my last patch. We do not have the
rc6_attrs when there is no CONFIG_PM so this causes a compilation error.
Reported-by: NFengguang Wu <fengguang.wu@intel.com>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

As we make the simplification of using a power-of-two size for the
execbuffer handle-to-object TLB, we should validate that this is actually
true and so clarify that premise.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Clear Audio Enable bit to trigger unsolicated event to notify Audio
Driver part the HDMI hot plug change. The patch fixed the bug when
remove HDMI cable the bit was not cleared correctly.

In intel_enable_hdmi(), if intel_hdmi->has_audio been true, the "Audio enable bit" will
be set to trigger unsolicated event to notify Alsa driver the change.

intel_hdmi->has_audio will be reset to false from intel_hdmi_detect() after
remove the hdmi cable, here's debug log:

[  187.494153] [drm:output_poll_execute], [CONNECTOR:17:HDMI-A-1] status updated from 1 to 2
[  187.525349] [drm:intel_hdmi_detect], HDMI: has_audio = 0

so when comes back to intel_disable_hdmi(), the "Audio enable bit" will not be cleared. And this
cause the eld infomation and pin presence doesnot update accordingly in alsa driver side.

This patch will also trigger unsolicated event to alsa driver to notify the hot plug event:

[  187.853159] ALSA sound/pci/hda/patch_hdmi.c:772 HDMI hot plug event: Codec=3 Pin=5 Presence_Detect=0 ELD_Valid=1
[  187.853268] ALSA sound/pci/hda/patch_hdmi.c:990 HDMI status: Codec=3 Pin=5 Presence_Detect=0 ELD_Valid=0
Signed-off-by: NWang Xingchao <xingchao.wang@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

By providing a callback for when we need to bind the pages, and then
release them again later, we can shorten the amount of time we hold the
foreign pages mapped and pinned, and importantly the dmabuf objects then
behave as any other normal object with respect to the shrinker and
memory management.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Hopefully this makes userspace slightly less confused about us
frobbing the dpms state behind its back. Yeah, it would be better
to be more careful with not changing the dpms state, but that is
quite more invasive.
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

... because our current set_mode implementation doesn't bother to adjust
for the dpms state, we just forcefully update it. So stop pretending that
we're better than we are and rip out this extranous call.

Note that this totally confuses userspace, because the exposed connector
property isn't actually updated ...
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Because they should have been disabled when shutting down the display
pipe previously. To ensure that this is the case, add a few assserts
instead of unconditionally disabling the fdi link.
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Even with the old crtc helper code we should have disabled all
encoders on that pipe by now, and with the new code this would
definitely paper over a bug. We already have the necessary checks
in place in intel_disable_transcoder, so if we accidentally leave
a pch port on, this will be caught.

Hence just rip this all out.

Note that up to the patch in this giant modeset series that removes
the LVDS special case to avoid disabling LVDS in the encoder->prepare
callback ("drm/i915/lvds: ditch ->prepare special case"), this was not
the case for all outputs.

Also note that in

commit 1b3c7a47
Author: Zhenyu Wang <zhenyuw@linux.intel.com>
Date:   Wed Nov 25 13:09:38 2009 +0800

    drm/i915: Fix LVDS stability issue on Ironlake

this was already discovered independently and worked around. How I
bloody hate this entire mess of cludges piled on top of other cludges.
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

In the future we may like to experiment with using a WC map of the GTT
portion. However, that will conflict with i915.ko mapping the entire bar
as UC in order to access the GPU registers. Instead we can shrink the
register ioremap to only map the register block.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Tested-by (IVB): Ben Widawsky <ben@bwidawsk.net>
Acked-by: NBen Widawsky <ben@bwidawsk.net>
[danvet: Squashed-in follow-up fix for gen2/3 registers file size from
Chris Wilson.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This is useful for userspace utilities which wish to use the previous
interface, specifically for micromanaging the increase/decrease steps by
setting min == max.
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Provide a standardized sysfs interface for setting min, and max
frequencies.  The code which reads the limits were lifted from the
debugfs files. As a brief explanation, the limits are similar to the CPU
p-states. We have 3 states:

RP0 - ie. max frequency
RP1 - ie. "preferred min" frequency
RPn - seriously lowest frequency

Initially Daniel asked me to clamp the writes to supported values, but
in conforming to the way the cpufreq drivers seem to work, instead
return -EINVAL (noticed by Jesse in discussion).
The values can be used by userspace wishing to control the limits of the
GPU (see the CC list for people who care).

v4: Make exceeding the soft limits return -EINVAL as well (Daniel)

v3: bug fix (Ben) -  was passing the MHz value to gen6_set_rps instead of
the step value. To fix, deal only with step values by doing the divide
at the top.

v2: add the dropped mutex_unlock in error cases (Chris)
EINVAL on both too min, or too max (Daniel)

v2 Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
CC: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This has been tons of fun to figure out with git blame. The first
notion of this code block goes back to the original cpu edp enabling
for ilk in

commit 32f9d658
Author: Zhenyu Wang <zhenyuw@linux.intel.com>
Date:   Fri Jul 24 01:00:32 2009 +0800

    drm/i915: Add eDP support on IGDNG mobile chip

Two things are notable in this commit wrt to the this edp special
case:
- The IS_eDP check _only_ fires for DP A, i.e. cpu edp ports.
- The cpu edp port is disabled at the top of the dp_link_down function.

My theory is that these hacks was added to work around the completely
different modeset sequence for cpu edp ports compared to pch edp
ports. With the cpu edp confusion on ilk (and snb/ivb) now fixed up,
this shouldn't be required any more.

The really interesting question is how this special cases survived
this long in the code. The first step is declaring the pch port D as
eDP if it's used for an internal panel:

commit b329530c
Author: Adam Jackson <ajax@redhat.com>
Date:   Fri Jul 16 14:46:28 2010 -0400

    drm/i915/dp: Correctly report eDP in the core connector type

This commit unfortunately failed to notice that not all edp ports are
created equal. Then follow a flurry of refactorings, culminating in a
patch from Keith Packard which resulted in the current logic (by
making it "correct" for all platforms that have edp):

commit 417e822d
Author: Keith Packard <keithp@keithp.com>
Date:   Tue Nov 1 19:54:11 2011 -0700

    drm/i915: Treat PCH eDP like DP in most places

None of these cleanups or refactorings supply any reason why we need
this code, they've simply carried it on as-is.

Hence presume it might be harmful with the current code and rip it
out. We do rewrite the link training bits completely anyway when
re-training the link.
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The exec_list is of type drm_i915_gem_exec_object2 and so casting it to
a drm_i915_gem_relocation_entry is very confusing!
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJani Nikula <jani.nikula@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

This is rather a hack to fix brightness hotkeys on a Clevo laptop. CADL is not
used anywhere in the driver code at the moment, but it could be used in BIOS as
is the case with the Clevo laptop.

The Clevo B7130 requires the CADL field to contain at least the ID of
the LCD device. If this field is empty, the ACPI methods that are called
on pressing brightness / display switching hotkeys will not trigger a
notification. As a result, it appears as no hotkey has been pressed.

Reference: https://bugs.freedesktop.org/show_bug.cgi?id=45452Tested-by: NPeter Wu <lekensteyn@gmail.com>
Signed-off-by: NPeter Wu <lekensteyn@gmail.com>
Acked-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

See bspec, Vol3 Part2, Section 1.1.3 "Display Mode Set Sequence". This
applies to all platforms where we currently support eDP on, i.e. ilk,
snb & ivb.

Without this change we fail to light up the eDP port on previously
unused crtcs (likely because something is stuck on the old pipe), and
we also fail to properly disable the old pipe (i.e. bit 30 in the
PIPECONF register is stuck as set until the next reboot).

v2: Rebased on top of the edp panel off sequence changes in 3.6-rc2.
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=44001Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

These have been added because dp links are fiddle things and don't
like it when we try to re-train an enabled output (or disable a
disabled output harder). And because the crtc helper code is
ridiculously bad add tracking the modeset state.

But with the new code in place it is simply a bug to disable a disabled
encoder or to enable an enabled encoder again. Hence convert these to
WARNs (and bail out for safety), but flatten all conditionals in the
code itself.
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

With the previous patch to clean up where exactly these two functions
are getting called, this patch can tackle the enable/disable code
itself:

- WARN if the port enable bit is in the wrong state or if the edp pll
  bit is in the wrong state, just for paranoia's sake.
- Don't disable the edp pll harder in the modeset functions just for
  fun.
- Don't set the edp pll enable flag in intel_dp->DP in modeset, do
  that while changing the actual hw state. We do the same with the
  actual port enable bit, so this is a bit more consistent.
- Track the current DP register value when setting things up and add
  some comments how intel_dp->DP is used in the disable code.

v2: Be more careful with resetting intel_dp->DP - otherwise dpms
off->on will fail spectacularly, becuase we enable the eDP port when
we should only enable the eDP pll.
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

By using the new pre_enable/post_disable functions.

To ensure that we only frob the cpu edp pll while the pipe is off add
the relevant asserts. Thanks to the new output state staging, this is
now really easy.

With this fixed we can now finally rip out the special-case handling
in the dp dpms code and replace it by the common intel_connector_dpms.
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The cpu eDP encoder has some horrible hacks to set up the DP pll at
the right time. To be able to move them to the right place, add some
more encoder callbacks so that this can happen at the right time.

LVDS has some similar funky hacks, but that would require more work
(we need to move around the pll setup a bit). Hence for now only
wire these new callbacks up for ilk+ - we only have cpu eDP on these
platforms.

v2: Bikeshed the vtable ordering, requested by Chris Wilson.
Reviewed-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

It's bogus.

If I've followed the history of this piece of code correctly, i.e. the
initial register write with the following vblank wait, this goes all
the way back to the original enabling of DP support in

commit a4fc5ed6
Author: Keith Packard <keithp@keithp.com>
Date:   Tue Apr 7 16:16:42 2009 -0700

    drm/i915: Add Display Port support

Unfortunately it seems to be nothing more than glorified duct-tape and
sometimes actively harmful. Adam Jackson noticed this for CPT
platforms with

commit e8519464
Author: Adam Jackson <ajax@redhat.com>
Date:   Thu Jul 21 17:48:38 2011 -0400

    drm/i915/dp: Don't turn CPT DP ports on too early

Unfortunately this kept the code around for ilk and gm45.

The specific failure case I'm seeing here is that after a dpms off/on
cycle we have the bits from the last link training (hopefully
successful link training) set in intel_dp->DP. This is requiered so
that complete_link_train can enable the port with the right tuning
values.

Unfortunately writing these again to the disabled port at dpms on time
kills the port somehow until it's disabled - dp link training fails in
an endless loop without this patch on my mobile ilk and gm45.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: NChris Wilson <chris@chris-wilson.co.uk>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=51493Signed-Off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

With the new "standardized" sysfs interfaces we need to be a bit more
careful about setting the RPS values.

Because the sysfs code and the rps workqueue can run at the same time,
if the sysfs setter wins the race to the mutex, the workqueue can come
in and set a value which is out of range (ie. we're no longer protecting
by RPINTLIM).

I was not able to actually make this error occur in testing.
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

In order to keep our cached values in sync with the hardware, we need a
posting read here.

CC: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Userspace applications such as PowerTOP are interesting in being able to
read the current GPU frequency. The patch itself sets up a generic array
for gen6 attributes so we can easily add other items in the future (and
it also happens to be just about the cleanest way to do this).

The patch is a nice addition to
commit 1ac02185dff3afac146d745ba220dc6672d1d162
Author: Daniel Vetter <daniel.vetter@ffwll.ch>
Date:   Thu Aug 30 13:26:48 2012 +0200

    drm/i915: add a tracepoint for gpu frequency changes

Reading the GPU frequncy can be done by reading a file like:
/sys/class/drm/card0/render_frequency_mhz
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Magic numbers are bad mmmkay. In this case in particular the value is
especially weird because the docs say multiple things. We'll need this
value for sysfs, so extracting it is useful for that as well.
Reviewed-by: NJesse Barnes <jbarnes@virtuousgeek.org>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Name variables a bit better for copy-pasters. This got turned up as part
of review for upcoming sysfs patches.
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
[danvet: resolved conflicts due to missing some earlier patches.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Because declaring a variable in the beginning of the function, then
initializing it 100 lines later, then using it 100 lines later does
not make our code look good IMHO.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Because ironlake_crtc_mode_set is a giant function that used to have
404 lines. Let's try to make it less complex/confusing.
Signed-off-by: NPaulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

Rather than have multiple data structures for describing our page layout
in conjunction with the array of pages, we can migrate all users over to
a scatterlist.

One major advantage, other than unifying the page tracking structures,
this offers is that we replace the vmalloc'ed array (which can be up to
a megabyte in size) with a chain of individual pages which helps reduce
memory pressure.

The disadvantage is that we then do not have a simple array to iterate,
or to access randomly. The common case for this is in the relocation
processing, which will typically fit within a single scatterlist page
and so be almost the same cost as the simple array. For iterating over
the array, the extra function call could be optimised away, but in
reality is an insignificant cost of either binding the pages, or
performing the pwrite/pread.

v2: Fix drm_clflush_sg() to not invoke wbinvd as well! And fix the
trivial compile error from rebasing.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

By using the recently introduced pinning of pages, we can safely drop
the mutex in the knowledge that the pages are not going to disappear
beneath us, and so we can simplify the code for iterating over the pages.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

By using the recently introduced pinning of pages, we can safely drop
the mutex in the knowledge that the pages are not going to disappear
jeneath us, and so we can simplify the code for iterating over the pages.

Note: The old code had such complicated page refcounting since it used
obj->pages as a micro-optimization if it's there, but that could
(before this patch) disappear when we drop the dev->struct_mutex.
Hence some manual page refcounting was required for the slow path,
complicated by the fact that pages returned by shmem_read_mapping_page
already have a pageref, which needs to be dropped again.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
[danvet: Added note to explain the question Ben raised in review.]
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

We need to refcount our pages in order to prevent reaping them at
inopportune times, such as when they currently vmapped or exported to
another driver. However, we also wish to keep the lazy deallocation of
our pages so we need to take a pin/unpinned approach rather than a
simple refcount.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

In order to specialise functions depending upon the type of object, we
can attach vfuncs to each object via a new ->ops pointer.

For instance, this will be used in future patches to only bind pages from
a dma-buf for the duration that the object is used by the GPU - and so
prevent them from pinning those pages for the entire of the object.

v2: Bonus comments.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NBen Widawsky <ben@bwidawsk.net>
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>