drm/i915: Mark CPU cache as dirty when used for rendering

On LLC, or even snooped, machines rendering via the GPU ends up in the CPU
cache. This cacheline dirt also needs to be flushed to main memory when
moving to an incoherent domain, such as the display's scanout engine.
Mostly, this happens because either the object is marked as dirty from
its first use or is avoided by setting the object into the display
domain from the start.

v2: Treat WT as not requiring a clflush prior to use on the display
engine as well.

Fixes: 0f71979ab7fb ("drm/i915: Performed deferred clflush inside set-cache-level")
References: https://bugs.freedesktop.org/show_bug.cgi?id=95414
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <stable@vger.kernel.org> # v4.0+
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161107165204.7008-1-chris@chris-wilson.co.uk

drm/i915: Add assert for no pending GPU requests during suspend/resume in LR mode

During resume we will reset the SW/HW tracking for each ring head/tail
pointers and so are not prepared to replay any pending requests (as
opposed to GPU reset time). Add an assert for this both to the suspend
and the resume code.

v2:
- Check for ELSP port idle already during suspend and check !gt.awake
  during resume. (Chris)
v3:
- Move the !gt.awake check to i915_gem_resume().
v4:
- s/intel_lr_engines_idle/intel_execlists_idle/ (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1478510405-11799-4-git-send-email-imre.deak@intel.com

drm/i915: Make sure engines are idle during GPU idling in LR mode

We assume that the GPU is idle once receiving the seqno via the last
request's user interrupt. In execlist mode the corresponding context
completed interrupt can be delayed though and until this latter
interrupt arrives we consider the request to be pending on the ELSP
submit port. This can cause a problem during system suspend where this
last request will be seen by the resume code as still pending. Such
pending requests are normally replayed after a GPU reset, but during
resume we reset both SW and HW tracking of the ring head/tail pointers,
so replaying the pending request with its stale tail pointer will leave
the ring in an inconsistent state. A subsequent request submission can
lead then to the GPU executing from uninitialized area in the ring
behind the above stale tail pointer.

Fix this by making sure any pending request on the ELSP port is
completed before suspending. I used a polling wait since the completion
time I measured was <1ms and since normally we only need to wait during
system suspend. GPU idling during runtime suspend is scheduled with a
delay (currently 50-100ms) after the retirement of the last request at
which point the context completed interrupt must have arrived already.

The chance of this bug was increased by

commit 1c777c5d1dcdf8fa0223fcff35fb387b5bb9517a
Author: Imre Deak <imre.deak@intel.com>
Date:   Wed Oct 12 17:46:37 2016 +0300

    drm/i915/hsw: Fix GPU hang during resume from S3-devices state

but it could happen even without the explicit GPU reset, since we
disable interrupts afterwards during the suspend sequence.

v2:
- Do an unlocked poll-wait first. (Chris)
v3-4:
- s/intel_lr_engines_idle/intel_execlists_idle/ and move
  i915.enable_execlists check to the new helper. (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98470
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1478510405-11799-3-git-send-email-imre.deak@intel.com

drm/i915: Avoid early GPU idling due to race with new request

There is a small race where a new request can be submitted and retired
after the idle worker started to run which leads to idling the GPU too
early. Fix this by deferring the idling to the pending instance of the
worker.

This scenario was pointed out by Chris.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1478510405-11799-2-git-send-email-imre.deak@intel.com

drm/i915: Avoid early GPU idling due to already pending idle work

Atm, in case an idle work handler is already pending but haven't yet
started to run, retiring a new request will not extend the active period
as required, rather simply leaves the pending idle work to be scheduled
at the original expiration time. This may lead to idling the GPU too
early. Fix this by using the delayed-work scheduler alternative which
makes sure the handler's expiration time is extended in this case.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Requested-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1478510405-11799-1-git-send-email-imre.deak@intel.com

drm/i915: Limit Valleyview and earlier to only using mappable scanout

Valleyview appears to be limited to only scanning out from the first 512MiB
of the Global GTT. Lets presume that this behaviour was inherited from the
display block copied from g4x (not Ironlake) and all earlier generations
are similarly affected, though testing suggests different symptoms. For
simplicity, impose that these platforms must scanout from the mappable
region. (For extra simplicity, use HAS_GMCH_DISPLAY even though this
catches Cherryview which does not appear to be limited to the low
aperture for its scanout.)

v2: Use HAS_GMCH_DISPLAY() to more clearly convey my intent about
limiting this workaround to the old style of display engine.

v3: Update changelog to reflect testing by Ville Syrjälä
v4: Include the changes to the comments as well

Reported-by: Luis Botello <luis.botello.ortega@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98036
Fixes: 2efb813d5388 ("drm/i915: Fallback to using unmappable memory for scanout")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Akash Goel <akash.goel@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.9-rc1+
Link: http://patchwork.freedesktop.org/patch/msgid/20161107110128.28762-1-chris@chris-wilson.co.uk
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com

drm/i915: Round tile chunks up for constructing partial VMAs

When we split a large object up into chunks for GTT faulting (because we
can't fit the whole object into the aperture) we have to align our cuts
with the fence registers. Each partial VMA must cover a complete set of
tile rows or the offset into each partial VMA is not aligned with the
whole image. Currently we enforce a minimum size on each partial VMA,
but this minimum size itself was not aligned to the tile row causing
distortion.

Reported-by: Andreas Reis <andreas.reis@gmail.com>
Reported-by: Chris Clayton <chris2553@googlemail.com>
Reported-by: Norbert Preining <preining@logic.at>
Tested-by: Norbert Preining <preining@logic.at>
Tested-by: Chris Clayton <chris2553@googlemail.com>
Fixes: 03af84fe7f48 ("drm/i915: Choose partial chunksize based on tile row size")
Fixes: a61007a83a46 ("drm/i915: Fix partial GGTT faulting") # enabling patch
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98402
Testcase: igt/gem_mmap_gtt/medium-copy-odd
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.9-rc1+
Link: http://patchwork.freedesktop.org/patch/msgid/20161107105443.27855-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915: Remove the vma from the object list upon close

Currently, the vma is being unlink from the object lookup on destroy.
However, we are meant to be decoupling it upon close so that the user
cannot access the closed vma whilst it remains active on the GPU.

[   34.074858] kernel BUG at drivers/gpu/drm/i915/i915_gem_gtt.c:3561!
[   34.074875] invalid opcode: 0000 [#1] PREEMPT SMP
[   34.074888] Modules linked in: snd_hda_intel i915 x86_pkg_temp_thermal coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel lpc_ich mei_me mei snd_hda_codec_realtek snd_hda_codec_generic snd_hda_codec_hdmi snd_hda_codec snd_hwdep snd_hda_core i2c_designware_platform i2c_designware_core snd_pcm e1000e ptp pps_core sdhci_acpi sdhci mmc_core i2c_hid [last unloaded: i915]
[   34.075010] CPU: 1 PID: 6224 Comm: gem_close_race Tainted: G     U          4.9.0-rc3-CI-CI_DRM_1800+ #1
[   34.075034] Hardware name:                  /NUC5i7RYB, BIOS RYBDWi35.86A.0355.2016.0224.1501 02/24/2016
[   34.075057] task: ffff8802459a8040 task.stack: ffffc90000524000
[   34.075074] RIP: 0010:[<ffffffffa0392cbc>]  [<ffffffffa0392cbc>] i915_gem_obj_lookup_or_create_vma+0x8c/0xc0 [i915]
[   34.075118] RSP: 0018:ffffc90000527b68  EFLAGS: 00010202
[   34.075135] RAX: ffff8802426c5e40 RBX: 0000000000000000 RCX: ffff8802447fc2a8
[   34.075158] RDX: 0000000000000000 RSI: ffff8802447fc2a8 RDI: ffff880248a4a880
[   34.075181] RBP: ffffc90000527b88 R08: 0000000000000008 R09: 0000000000000000
[   34.075203] R10: 0000000000000001 R11: 0000000000000000 R12: ffff880248a4a880
[   34.075225] R13: ffff8802447fc2a8 R14: ffff880243e9afa8 R15: ffff880248a4a9c8
[   34.075248] FS:  00007f9b43e59740(0000) GS:ffff880256c80000(0000) knlGS:0000000000000000
[   34.075273] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[   34.075292] CR2: 00007f9b43419140 CR3: 000000024455d000 CR4: 00000000003406e0
[   34.075314] Stack:
[   34.075323]  0000000000000000 ffffc90000527bd0 ffff880243cb8008 ffff880243e9afa8
[   34.075353]  ffffc90000527c08 ffffffffa03874c7 ffffc90000527bb8 ffff880243e9afa8
[   34.075383]  ffff880243e9afb0 ffffc90000527e10 ffff8802447fc2a8 ffff880243cb8040
[   34.075414] Call Trace:
[   34.075435]  [<ffffffffa03874c7>] eb_lookup_vmas.isra.7+0x247/0x330 [i915]
[   34.075468]  [<ffffffffa0388c34>] i915_gem_do_execbuffer.isra.15+0x604/0x1a10 [i915]
[   34.075507]  [<ffffffffa039c957>] ? i915_gem_object_get_sg+0x347/0x380 [i915]
[   34.075532]  [<ffffffff811a69ce>] ? __might_fault+0x3e/0x90
[   34.075562]  [<ffffffffa038a430>] i915_gem_execbuffer2+0xc0/0x250 [i915]
[   34.075585]  [<ffffffff81552926>] drm_ioctl+0x1f6/0x480
[   34.075604]  [<ffffffff8100107a>] ? trace_hardirqs_on_thunk+0x1a/0x1c
[   34.075635]  [<ffffffffa038a370>] ? i915_gem_execbuffer+0x330/0x330 [i915]
[   34.075658]  [<ffffffff81202d2e>] do_vfs_ioctl+0x8e/0x690
[   34.075677]  [<ffffffff8181582d>] ? _raw_spin_unlock_irqrestore+0x3d/0x60
[   34.075700]  [<ffffffff810fcd51>] ? SyS_timer_settime+0x141/0x1e0
[   34.075721]  [<ffffffff810d6de2>] ? trace_hardirqs_on_caller+0x122/0x1b0
[   34.075742]  [<ffffffff8120336c>] SyS_ioctl+0x3c/0x70
[   34.075760]  [<ffffffff8181602e>] entry_SYSCALL_64_fastpath+0x1c/0xb1
[   34.075781] Code: 44 a0 48 c7 c2 9a 7e 43 a0 be e0 0d 00 00 48 c7 c7 a0 45 44 a0 e8 55 b8 ce e0 48 85 db 74 a3 49 83 bd f8 03 00 00 00 74 99 0f 0b <0f> 0b 48 89 da 4c 89 ee 4c 89 e7 e8 04 a9 ff ff 48 89 da 49 89
[   34.075955] RIP  [<ffffffffa0392cbc>] i915_gem_obj_lookup_or_create_vma+0x8c/0xc0 [i915]
[   34.075994]  RSP <ffffc90000527b68>

Testcase: igt/gem_close_race/basic-threads
Fixes: db6c2b4151f2 ("drm/i915: Store the vma in an rbtree...")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161104161241.25871-1-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

drm/i915: Reinit polling before hpd when resuming

Now that we don't run the connector reprobing from i915_drm_resume(), we
need to make it so we don't have to wait for reprobing to finish so that
we actually speed things up. In order to do this, we need to make sure
that i915_drm_resume() doesn't get blocked by i915_hpd_poll_init_work()
while trying to acquire the mode_config lock that
drm_kms_helper_poll_enable() needs to acquire.

The easiest way to do this is to just enable polling before hpd. This
shouldn't break anything since at that point we have everything else we
need for polling enabled.

As well, this should result in a rather significant improvement in how
quickly we can resume the system.

Signed-off-by: Lyude <lyude@redhat.com>
Tested-by: David Weinehall <david.weinehall@linux.intel.com>
Reviewed-by: David Weinehall <david.weinehall@linux.intel.com>
Testcase: analyze_suspend.py -config config/suspend-callgraph.cfg -filter i915

drm/i915: Remove redundant reprobe in i915_drm_resume

Weine's investigation on benchmarking the suspend/resume process pointed
out a lot of the time in suspend/resume is being spent reprobing. While
the reprobing process is a lengthy one for good reason, we don't need to
hold up the entire suspend/resume process while we wait for it to
finish. Luckily as it turns out, we already trigger a full connector
reprobe in i915_hpd_poll_init_work(), so we can just ditch reprobing in
i915_drm_resume() entirely.

This won't lead to less time spent resuming just yet since now the
bottleneck will be waiting for the mode_config lock in
drm_kms_helper_poll_enable(), since that will be held as long as
i915_hpd_poll_init_work() is reprobing all of the connectors. But we'll
address that in the next patch.

Signed-off-by: Lyude <lyude@redhat.com>
Tested-by: David Weinehall <david.weinehall@linux.intel.com>
Reviewed-by: David Weinehall <david.weinehall@linux.intel.com>
Testcase: analyze_suspend.py -config config/suspend-callgraph.cfg -filter i915

drm/i915/dp: Extend BDW DP audio workaround to GEN9 platforms

According to BSpec, cdclk for BDW has to be not less than 432 MHz with DP
audio enabled, port width x4, and link rate HBR2 (5.4 GHz). With cdclk less
than 432 MHz, enabling audio leads to pipe FIFO underruns and displays
cycling on/off.

Let's apply this work around to GEN9 platforms too, as it fixes the same
issue.

v2: Move drm_device to drm_i915_private conversion

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=97907
Cc: stable@vger.kernel.org
Cc: Libin Yang <libin.yang@linux.intel.com>
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1478117601-19122-1-git-send-email-dhinakaran.pandiyan@intel.com

drm/i915/dp: BDW cdclk fix for DP audio

According to BSpec, cdclk for BDW has to be not less than 432 MHz with DP
audio enabled, port width x4, and link rate HBR2 (5.4 GHz). With cdclk less
than 432 MHz, enabling audio leads to pipe FIFO underruns and displays
cycling on/off.

From BSpec:
"Display» BDW-SKL» dpr» [Register] DP_TP_CTL [BDW+,EXCLUDE(CHV)]
Workaround : Do not use DisplayPort with CDCLK less than 432 MHz, audio
enabled, port width x4, and link rate HBR2 (5.4 GHz), or else there may
be audio corruption or screen corruption."

Since, some DP configurations (e.g., MST) use port width x4 and HBR2
link rate, let's increase the cdclk to >= 432 MHz to enable audio for those
cases.

v4: Changed commit message
v3: Combine BDW pixel rate adjustments into a function (Jani)
v2: Restrict fix to BDW
    Retain the set cdclk across modesets (Ville)
Cc: stable@vger.kernel.org
Signed-off-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Jani Nikula <jani.nikula@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1478026080-2925-1-git-send-email-dhinakaran.pandiyan@intel.com

drm/i915: Fix pages pin counting around swizzle quirk

commit bc0629a76726 ("drm/i915: Track pages pinned due to swizzling
quirk") fixed one problem, but revealed a whole lot more. The root cause
of the pin count mismatch for the swizzle quirk (for L-shaped memory on
gen3/4) was that we were incrementing the pages_pin_count upon getting
the backing pages but then overwriting the pages_pin_count to set it to
1 afterwards. With a little bit of adjustment to satisfy the GEM_BUG_ON
sanitychecks, the fix is to replace the explicit atomic_set with an
atomic_inc.

v2: Consistently use atomics (not mix atomics and helpers) within the
lowlevel get_pages routines. This makes the atomic operations much
clearer.

Fixes: 1233e2db199d ("drm/i915: Move object backing storage manipulation")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161104103001.27643-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915: Fix test on inputs for vma_compare()

When supplying a view to vma_compare() it is required that the supplied
i915_address_space is the global GTT. I tested the VMA instead (which is
the current position in the rbtree and maybe from any address space).

Reported-by: Matthew Auld <matthew.auld@intel.com>
Tested-by: Matthew Auld <matthew.auld@intel.com>
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98579
Fixes: db6c2b4151f2 ("drm/i915: Store the vma in an rbtree...")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161103200852.23431-1-chris@chris-wilson.co.uk
Reviewed-by: Matthew Auld <matthew.auld@intel.com>

drm/i915/guc: Cache the client mapping

Use i915_gem_object_pin_map() for the guc client's lifetime to replace
the peristent kmap + frequent kmap_atomic with a permanent vmapping.
This avoids taking the obj->mm.lock mutex whilst inside irq context
later.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98571
Fixes: 96d776345277 ("drm/i915: Use a radixtree for random access...");
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20161102175051.29163-9-chris@chris-wilson.co.uk
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>

drm/i915: Tidy slab cache allocations

We can use the preferred KMEM_CACHE helper for brevity.

Also simplifiy error unwind by only setting the ENOMEM
error code once.

v2: Add forgotten changes. (Joonas Lahtinen)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> (v1)
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1478099699-28652-1-git-send-email-tvrtko.ursulin@linux.intel.com

drm/i915: Introduce HAS_64BIT_RELOC

Move has_64bit_reloc into dev_priv->info. This will make it visible
in the feature listing debug output.

v2:
- Keep the struct member to keep GCC fragile but happy (Chris)
v3:
- More detailed commit message (Chris)
- Include forgotten CHV and BXT (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1478162386-5018-1-git-send-email-joonas.lahtinen@linux.intel.com

drm/i915: Show the execlist queue in debugfs/i915_engine_info

When looking at freezes whilst working on execlists, knowing the order
of the pending requests in the driver is useful.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20161027000348.4641-2-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915: Unify global_list into global_link

$ sed -i -r 's/\bglobal_list\b/global_link/g' *.c *.h

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/1478081764-8058-1-git-send-email-joonas.lahtinen@linux.intel.com

drm/i915: Move hangcheck code out from i915_irq.c

Create new file for hangcheck specific code, intel_hangcheck.c,
and move all related code in it.

v2: s/intel_engine_hangcheck/intel_engine (Chris)

No functional changes.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1478018583-5816-1-git-send-email-mika.kuoppala@intel.com

drm/i915/gtt: Mark tlbs dirty on clear

Now when clearing ptes can modify upper level pdp's,
we need to mark them dirty so that they will be flushed
correctly.

Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1478006856-8313-1-git-send-email-mika.kuoppala@intel.com

drm/i915/gtt: Fix pte clear range

Comparing pte index to a number of entries is wrong
when clearing a range of pte entries. Use end marker
of 'one past' to correctly point adequate number of
ptes to the scratch page.

v2: assert early instead of warning late (Chris)
v3: removed consts (Joonas)

Fixes: d209b9c3cd28 ("drm/i915/gtt: Split gen8_ppgtt_clear_pte_range")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98282
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Michel Thierry <michel.thierry@intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Reported-by: Mike Lothian <mike@fireburn.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Tested-by: Mike Lothian <mike@fireburn.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@intel.com>

drm/i915/bxt: Don't set OCL2_LDOFUSE_PWR_DIS bit in phy init sequence

Hardware engineers confirmed that writing to it has no effect, as implied
by the FIXME comment.

v2: Also remove comment from bxt_ddi_phy_verify_state(). (Imre)
Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Ander Conselvan de Oliveira <ander.conselvan.de.oliveira@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1478069096-11209-1-git-send-email-ander.conselvan.de.oliveira@intel.com

drm/i915: Allow shrinking of userptr objects once again

Commit 1bec9b0bda3d ("drm/i915/shrinker: Only shmemfs objects
are backed by swap") stopped considering the userptr objects
in shrinker callbacks.

Restore that so idle userptr objects can be discarded in order
to free up memory.

One change further to what was introduced in 1bec9b0bda3d is
to start considering userptr objects in oom but that should
also be a correct thing to do.

v2: Introduce I915_GEM_OBJECT_IS_SHRINKABLE. (Chris Wilson)

Signed-off-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Fixes: 1bec9b0bda3d ("drm/i915/shrinker: Only shmemfs objects are backed by swap")
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: <stable@vger.kernel.org>
Link: http://patchwork.freedesktop.org/patch/msgid/1478011450-6634-1-git-send-email-tvrtko.ursulin@linux.intel.com

drm/i915: Pass dev_priv to intel_init_pm()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-27-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to ilk_setup_wm_latency() & co.

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-26-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to intel_suspend_hw()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-25-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to init_clock_gating

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-24-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to single_enabled_crtc()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-23-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to rest of IS_FOO() macros for the old platforms

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-22-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to IS_BROADWATER/IS_CRESTLINE

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-21-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to HAS_FW_BLC

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-20-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to .get_fifo_size()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-19-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to i915_pineview_get_mem_freq() and i915_ironlake_get_mem_freq()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-18-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to IS_PINEVIEW()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-17-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to IS_MOBILE()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-16-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to .get_display_clock_speed()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-15-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to cdclk update funcs

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-14-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to intel_crtc_init()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-13-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Always use intel_get_crtc_for_pipe()

Replace the open coded dev_priv->pipe_to_crtc_mapping[] usage with
intel_get_crtc_for_pipe().

Mostly done with coccinelle, with a few manual tweaks

@@
expression E1, E2;
@@
(
- E1->pipe_to_crtc_mapping[E2]
+ intel_get_crtc_for_pipe(E1, E2)
|
- E1->plane_to_crtc_mapping[E2]
+ intel_get_crtc_for_plane(E1, E2)
)

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-12-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to intel_get_crtc_for_pipe()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-11-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to g4x wm functions

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-10-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to vlv force pll functions

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-9-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to intel_wait_for_vblank()

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-8-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Store struct intel_crtc * in {pipe,plane}_to_crtc_mapping[]

A lot of users of the {pipe,plane}_to_crtc_mapping[] will end up
casting the result to intel_crtc, so let's just store the intel_crtc
pointer in the first place.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-7-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Use struct intel_crtc in legacy platform wm code

Unify our approach to things by using intel_crtc instead of drm_crtc.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-6-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass intel_crtc to update_wm functions

Unify our approach to things by passing around intel_crtc instead of
drm_crtc.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-5-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass intel_crtc to intel_crtc_active()

Unify our approach to things by passing around intel_crtc instead of
drm_crtc.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-4-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to skl_init_scalers()

Unify our approach to things by passing around dev_priv instead of dev.
While at it let's do some house cleaning: s/intel_foo/foo/ and move
things into tighter scope.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-3-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Pass dev_priv to plane constructors

Unify our approach to things by passing around dev_priv instead of dev.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477946245-14134-2-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>

drm/i915: Export a function to flush the context upon pinning

For legacy contexts we employ an optimisation to only flush the context
when binding into the global GTT. This avoids stalling on the GPU when
reloading an active context. Wrap this detail up into a helper and
export it for a potential third user. (Longer term, context pinning
needs to be reworked as the current handling of switch context pins too
late and so risks eviction and corrupting the request. Plans, plans,
plans.)

v2: Expand the comment explaining the optimisation for avoiding the
stall on active contexts.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20161030132820.32163-1-chris@chris-wilson.co.uk
Reviewed-by: Matthew Auld <matthew.auld@intel.com>

drm/i915/gen9+: Use the watermarks from crtc_state for everything, v2.

There's no need to keep a duplicate skl_pipe_wm around any more,
everything can be discovered from crtc_state, which we pass around
correctly now even in case of plane disable.

The copy in intel_crtc->wm.skl.active is equal to
crtc_state->wm.skl.optimal after the atomic commit completes.
It's useful for two-step watermark programming, but not required for
gen9+ which does it in a single step. We can pull the old allocation
from old_crtc_state.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477489299-25777-9-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>

drm/i915/skl+: Clean up minimum allocations, v2.

Move calculating minimum allocations to a helper, which cleans up the
code some more. The cursor is still allocated in advance because it
doesn't count towards data rate and should always be reserved.

changes since v1:
- Change comment to have a extra opening line. (Matt)
- Rebase to remove unused plane->pipe == pipe, handled by the iterator
  now. (Paulo)

Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477489299-25777-7-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>

drm/i915/skl+: Remove minimum block allocation from crtc state.

This is not required any more now that we get fresh state from
drm_atomic_crtc_state_for_each_plane_state. Zero all state
in advance.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477489299-25777-6-git-send-email-maarten.lankhorst@linux.intel.com

drm/i915/skl+: Remove data_rate from watermark struct, v2.

It's only used in one function, and can be calculated without caching it
in the global struct by using drm_atomic_crtc_state_for_each_plane_state.

There are loops over all planes, including planes that don't exist.
This is harmless, because data_rate will always be 0 for them and we
never program them when updating watermarks.

Changes since v1:
- Rename rate back to data_rate, and change array name to
  plane_data_rate. (Matt)
- Remove whitespace. (Paulo)

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477489299-25777-5-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>

drm/i915/gen9+: Use for_each_intel_plane_on_crtc in skl_print_wm_changes, v2.

Using for_each_intel_plane_on_crtc will allow us to find all allocations
that may have changed, not just the one added by the atomic state.

This will print changes to plane allocations for crtc's when some
planes are not added to the atomic state.

Changes since v1:
- Rephrase commit message. (Ville)
- Use plane->base.id and plane->name to kill off cursor special
  case. (Ville)
- Add intel_crtc to prevent a line wrap. (Paulo)
- Line wrap debug messages.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/c9f7dc1a-d23a-7c16-b2b7-1c23dd07ed35@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>

drm/i915/gen9+: Use cstate plane mask instead of crtc->state.

I'm planning on getting rid of all obj->state dereferences,
and replace thhem with accessor functions.
Remove this one early, they're equivalent because removed
planes are already part of the state, else they could not
have been removed.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477489299-25777-3-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>

drm/i915/skl+: Prepare for removing data rate from skl watermark state, v2.

Caching is not required, drm_atomic_crtc_state_for_each_plane_state can
be used to inspect the states of all planes assigned to the CRTC even
if they are not part of _state, so we can just recalculate every time.

Changes since v1:
- Remove plane->pipe checks, they're implied by the macros.
- Split unrelated changes to a separate commit.

Signed-off-by: Maarten Lankhorst <maarten.lankhorst@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477489299-25777-2-git-send-email-maarten.lankhorst@linux.intel.com
Reviewed-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Matt Roper <matthew.d.roper@intel.com>

drm/i915: Improve lockdep tracking for obj->mm.lock

The shrinker may appear to recurse into obj->mm.lock as the shrinker may
be called from a direct reclaim path whilst handling get_pages. We
filter out recursing on the same obj->mm.lock by inspecting
obj->mm.pages, but we do want to take the lock on a second object in
order to reap their pages. lockdep spots the recursion on the same
lockclass and needs annotation to avoid a false positive. To keep the
two paths distinct, create an enum to indicate which subclass of
obj->mm.lock we are using. This removes the false positive and avoids
masking real bugs.

Suggested-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101121134.27504-1-chris@chris-wilson.co.uk
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915: Store the vma in an rbtree under the object

With full-ppgtt one of the main bottlenecks is the lookup of the VMA
underneath the object. For execbuf there is merit in having a very fast
direct lookup of ctx:handle to the vma using a hashtree, but that still
leaves a large number of other lookups. One way to speed up the lookup
would be to use a rhashtable, but that requires extra allocations and
may exhibit poor worse case behaviour. An alternative is to use an
embedded rbtree, i.e. no extra allocations and deterministic behaviour,
but at the slight cost of O(lgN) lookups (instead of O(1) for
rhashtable). The major of such tree will be very shallow and so not much
slower, and still scales much, much better than the current unsorted
list.

v2: Bump vma_compare() to return a long, as we return the result of
comparing two pointers.

References: https://bugs.freedesktop.org/show_bug.cgi?id=87726
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101115400.15647-1-chris@chris-wilson.co.uk

drm/i915: Track pages pinned due to swizzling quirk

If we have a tiled object and an unknown CPU swizzle pattern, we pin the
pages to prevent the object from being swapped out (and us corrupting
the contents as we do not know the access pattern and so cannot convert
it to linear and back to tiled on reuse). This requires us to remember
to drop the extra pinning when freeing the object, or else we trigger
warnings about the pin leak. In commit fbbd37b36fa5 ("drm/i915: Move
object release to a freelist + worker"), the object free path was
deferred to a worker, but the unpinning of the quirk, along with marking
the object as reclaimable, was left on the immediate path (so that if
required we could reclaim the pages under memory pressure as early as
possible). However, this split introduced a bug where the pages were no
longer being unpinned if they were marked as unneeded.

[  231.800401] WARNING: CPU: 1 PID: 90 at drivers/gpu/drm/i915/i915_gem.c:4275 __i915_gem_free_objects+0x326/0x3c0 [i915]
[  231.800403] WARN_ON(i915_gem_object_has_pinned_pages(obj))
[  231.800405] Modules linked in:
[  231.800406]  snd_hda_intel i915 snd_hda_codec_generic mei_me snd_hda_codec coretemp snd_hwdep mei lpc_ich snd_hda_core snd_pcm e1000e ptp pps_core [last unloaded: i915]
[  231.800426] CPU: 1 PID: 90 Comm: kworker/1:4 Tainted: G     U          4.9.0-rc2-CI-CI_DRM_1780+ #1
[  231.800428] Hardware name: LENOVO 7465CTO/7465CTO, BIOS 6DET44WW (2.08 ) 04/22/2009
[  231.800456] Workqueue: events __i915_gem_free_work [i915]
[  231.800459]  ffffc9000034fc80 ffffffff8142dd65 ffffc9000034fcd0 0000000000000000
[  231.800465]  ffffc9000034fcc0 ffffffff8107e4e6 000010b300000001 0000000000001000
[  231.800469]  ffff88011d3db740 ffff880130ef0000 0000000000000000 ffff880130ef5ea0
[  231.800474] Call Trace:
[  231.800479]  [<ffffffff8142dd65>] dump_stack+0x67/0x92
[  231.800484]  [<ffffffff8107e4e6>] __warn+0xc6/0xe0
[  231.800487]  [<ffffffff8107e54a>] warn_slowpath_fmt+0x4a/0x50
[  231.800491]  [<ffffffff811d12ac>] ? kmem_cache_free+0x2dc/0x340
[  231.800520]  [<ffffffffa009ef36>] __i915_gem_free_objects+0x326/0x3c0 [i915]
[  231.800548]  [<ffffffffa009effe>] __i915_gem_free_work+0x2e/0x50 [i915]
[  231.800552]  [<ffffffff8109c27c>] process_one_work+0x1ec/0x6b0
[  231.800555]  [<ffffffff8109c1f6>] ? process_one_work+0x166/0x6b0
[  231.800558]  [<ffffffff8109c789>] worker_thread+0x49/0x490
[  231.800561]  [<ffffffff8109c740>] ? process_one_work+0x6b0/0x6b0
[  231.800563]  [<ffffffff8109c740>] ? process_one_work+0x6b0/0x6b0
[  231.800566]  [<ffffffff810a2aab>] kthread+0xeb/0x110
[  231.800569]  [<ffffffff810a29c0>] ? kthread_park+0x60/0x60
[  231.800573]  [<ffffffff818164a7>] ret_from_fork+0x27/0x40

Moving to a separate flag for tracking the quirked pin is overkill for
the bug (since we only have to interchange the two tests in
i915_gem_free_object) but it does reduce a complicated test on all
objects and provide a sanitycheck for uncommon code paths.

Fixes: fbbd37b36fa5 ("drm/i915: Move object release to a freelist + worker")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101100317.11129-2-chris@chris-wilson.co.uk

drm/i915: Avoid accessing request->timeline outside of its lifetime

Whilst waiting on a request, we may do so without holding any locks or
any guards beyond a reference to the request. In order to avoid taking
locks within request deallocation, we drop references to its timeline
(via the context and ppgtt) upon retirement. We should avoid chasing
such pointers outside of their control, in particular we inspect the
request->timeline to see if we may restore the RPS waitboost for a
client. If we instead look at the engine->timeline, we will have similar
behaviour on both full-ppgtt and !full-ppgtt systems and reduce the
amount of reward we give towards stalling clients (i.e. only if the
client stalls and the GPU is uncontended does it reclaim its boost).
This restores behaviour back to pre-timelines, whilst fixing:

[  645.078485] BUG: KASAN: use-after-free in i915_gem_object_wait_fence+0x1ee/0x2e0 at addr ffff8802335643a0
[  645.078577] Read of size 4 by task gem_exec_schedu/28408
[  645.078638] CPU: 1 PID: 28408 Comm: gem_exec_schedu Not tainted 4.9.0-rc2+ #64
[  645.078724] Hardware name:                  /        , BIOS PYBSWCEL.86A.0027.2015.0507.1758 05/07/2015
[  645.078816]  ffff88022daef9a0 ffffffff8143d059 ffff880235402a80 ffff880233564200
[  645.078998]  ffff88022daef9c8 ffffffff81229c5c ffff88022daefa48 ffff880233564200
[  645.079172]  ffff880235402a80 ffff88022daefa38 ffffffff81229ef0 000000008110a796
[  645.079345] Call Trace:
[  645.079404]  [<ffffffff8143d059>] dump_stack+0x68/0x9f
[  645.079467]  [<ffffffff81229c5c>] kasan_object_err+0x1c/0x70
[  645.079534]  [<ffffffff81229ef0>] kasan_report_error+0x1f0/0x4b0
[  645.079601]  [<ffffffff8122a244>] kasan_report+0x34/0x40
[  645.079676]  [<ffffffff81634f5e>] ? i915_gem_object_wait_fence+0x1ee/0x2e0
[  645.079741]  [<ffffffff81229951>] __asan_load4+0x61/0x80
[  645.079807]  [<ffffffff81634f5e>] i915_gem_object_wait_fence+0x1ee/0x2e0
[  645.079876]  [<ffffffff816364bf>] i915_gem_object_wait+0x19f/0x590
[  645.079944]  [<ffffffff81636320>] ? i915_gem_object_wait_priority+0x500/0x500
[  645.080016]  [<ffffffff8110fb30>] ? debug_show_all_locks+0x1e0/0x1e0
[  645.080084]  [<ffffffff8110abdc>] ? check_chain_key+0x14c/0x210
[  645.080157]  [<ffffffff8110a796>] ? __lock_is_held+0x46/0xc0
[  645.080226]  [<ffffffff8163bc61>] ? i915_gem_set_domain_ioctl+0x141/0x690
[  645.080296]  [<ffffffff8163bcc2>] i915_gem_set_domain_ioctl+0x1a2/0x690
[  645.080366]  [<ffffffff811f8f85>] ? __might_fault+0x75/0xe0
[  645.080433]  [<ffffffff815a55f7>] drm_ioctl+0x327/0x640
[  645.080508]  [<ffffffff8163bb20>] ? i915_gem_obj_prepare_shmem_write+0x3a0/0x3a0
[  645.080603]  [<ffffffff815a52d0>] ? drm_ioctl_permit+0x120/0x120
[  645.080670]  [<ffffffff8110abdc>] ? check_chain_key+0x14c/0x210
[  645.080738]  [<ffffffff81275717>] do_vfs_ioctl+0x127/0xa20
[  645.080804]  [<ffffffff8120268c>] ? do_mmap+0x47c/0x580
[  645.080871]  [<ffffffff811da567>] ? vm_mmap_pgoff+0x117/0x140
[  645.080938]  [<ffffffff812755f0>] ? ioctl_preallocate+0x150/0x150
[  645.081011]  [<ffffffff81108c53>] ? up_write+0x23/0x50
[  645.081078]  [<ffffffff811da567>] ? vm_mmap_pgoff+0x117/0x140
[  645.081145]  [<ffffffff811da450>] ? vma_is_stack_for_current+0x90/0x90
[  645.081214]  [<ffffffff8110d853>] ? mark_held_locks+0x23/0xc0
[  645.082030]  [<ffffffff81288408>] ? __fget+0x168/0x250
[  645.082106]  [<ffffffff819ad517>] ? entry_SYSCALL_64_fastpath+0x5/0xb1
[  645.082176]  [<ffffffff81288592>] ? __fget_light+0xa2/0xc0
[  645.082242]  [<ffffffff8127604c>] SyS_ioctl+0x3c/0x70
[  645.082309]  [<ffffffff819ad52e>] entry_SYSCALL_64_fastpath+0x1c/0xb1
[  645.082374] Object at ffff880233564200, in cache kmalloc-8192 size: 8192
[  645.082431] Allocated:
[  645.082480] PID = 28408
[  645.082535]  [  645.082566] [<ffffffff8103ae66>] save_stack_trace+0x16/0x20
[  645.082623]  [  645.082656] [<ffffffff81228b06>] save_stack+0x46/0xd0
[  645.082716]  [  645.082756] [<ffffffff812292fd>] kasan_kmalloc+0xad/0xe0
[  645.082817]  [  645.082848] [<ffffffff81631752>] i915_ppgtt_create+0x52/0x220
[  645.082908]  [  645.082941] [<ffffffff8161db96>] i915_gem_create_context+0x396/0x560
[  645.083027]  [  645.083059] [<ffffffff8161f857>] i915_gem_context_create_ioctl+0x97/0xf0
[  645.083152]  [  645.083183] [<ffffffff815a55f7>] drm_ioctl+0x327/0x640
[  645.083243]  [  645.083274] [<ffffffff81275717>] do_vfs_ioctl+0x127/0xa20
[  645.083334]  [  645.083372] [<ffffffff8127604c>] SyS_ioctl+0x3c/0x70
[  645.083432]  [  645.083464] [<ffffffff819ad52e>] entry_SYSCALL_64_fastpath+0x1c/0xb1
[  645.083551] Freed:
[  645.083599] PID = 27629
[  645.083648]  [  645.083676] [<ffffffff8103ae66>] save_stack_trace+0x16/0x20
[  645.083738]  [  645.083770] [<ffffffff81228b06>] save_stack+0x46/0xd0
[  645.083830]  [  645.083862] [<ffffffff81229203>] kasan_slab_free+0x73/0xc0
[  645.083922]  [  645.083961] [<ffffffff812279c9>] kfree+0xa9/0x170
[  645.084021]  [  645.084053] [<ffffffff81629f60>] i915_ppgtt_release+0x100/0x180
[  645.084139]  [  645.084171] [<ffffffff8161d414>] i915_gem_context_free+0x1b4/0x230
[  645.084257]  [  645.084288] [<ffffffff816537b2>] intel_lr_context_unpin+0x192/0x230
[  645.084380]  [  645.084413] [<ffffffff81645250>] i915_gem_request_retire+0x620/0x630
[  645.084500]  [  645.085226] [<ffffffff816473d1>] i915_gem_retire_requests+0x181/0x280
[  645.085313]  [  645.085352] [<ffffffff816352ba>] i915_gem_retire_work_handler+0xca/0xe0
[  645.085440]  [  645.085471] [<ffffffff810c725b>] process_one_work+0x4fb/0x920
[  645.085532]  [  645.085562] [<ffffffff810c770d>] worker_thread+0x8d/0x840
[  645.085622]  [  645.085653] [<ffffffff810d21e5>] kthread+0x185/0x1b0
[  645.085718]  [  645.085750] [<ffffffff819ad7a7>] ret_from_fork+0x27/0x40
[  645.085811] Memory state around the buggy address:
[  645.085869]  ffff880233564280: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  645.085956]  ffff880233564300: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  645.086053] >ffff880233564380: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  645.086138]                                ^
[  645.086193]  ffff880233564400: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb
[  645.086283]  ffff880233564480: fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb fb

v2: Add a comment to document the hint like nature of
 intel_engine_last_submit()

Fixes: 73cb97010d4f ("drm/i915: Combine seqno + tracking into a global timeline struct")
Fixes: 80b204bce8f2 ("drm/i915: Enable multiple timelines")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101100317.11129-1-chris@chris-wilson.co.uk

drm/i915: Move the recently scanned objects to the tail after shrinking

During shrinking, we walk over the list of objects searching for
victims. Any that are not removed are put back into the global list.
Currently, they are put back in order (at the front) which means they
will be first to be scanned again. If we instead move them to the rear
of the list, we will scan new potential victims on the next pass and
waste less time rescanning unshrinkable objects. Normally the lists are
kept in rough order to shrinking (with object least frequently used at
the start), by moving just scanned objects to the rear we are
acknowledging that they are still in use.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101084843.3961-3-chris@chris-wilson.co.uk

drm/i915: Discard objects from mm global_list after being shrunk

In the shrinker, we can safely remove an empty object (obj->mm.pages ==
NULL) after having discarded the pages because we are holding the
struct_mutex.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101084843.3961-2-chris@chris-wilson.co.uk

drm/i915: Use the full hammer when shutting down the rcu tasks

To flush all call_rcu() tasks (here from i915_gem_free_object()) we need
to call rcu_barrier() (not synchronize_rcu()). If we don't then we may
still have objects being freed as we continue to teardown the driver -
in particular, the recently released rings may race with the memory
manager shutdown resulting in sporadic:

[  142.217186] WARNING: CPU: 7 PID: 6185 at drivers/gpu/drm/drm_mm.c:932 drm_mm_takedown+0x2e/0x40
[  142.217187] Memory manager not clean during takedown.
[  142.217187] Modules linked in: i915(-) x86_pkg_temp_thermal intel_powerclamp coretemp crct10dif_pclmul crc32_pclmul ghash_clmulni_intel lpc_ich snd_hda_codec_realtek snd_hda_codec_generic mei_me mei snd_hda_codec_hdmi snd_hda_codec snd_hwdep snd_hda_core snd_pcm e1000e ptp pps_core [last unloaded: snd_hda_intel]
[  142.217199] CPU: 7 PID: 6185 Comm: rmmod Not tainted 4.9.0-rc2-CI-Trybot_242+ #1
[  142.217199] Hardware name: LENOVO 10AGS00601/SHARKBAY, BIOS FBKT34AUS 04/24/2013
[  142.217200]  ffffc90002ecfce0 ffffffff8142dd65 ffffc90002ecfd30 0000000000000000
[  142.217202]  ffffc90002ecfd20 ffffffff8107e4e6 000003a40778c2a8 ffff880401355c48
[  142.217204]  ffff88040778c2a8 ffffffffa040f3c0 ffffffffa040f4a0 00005621fbf8b1f0
[  142.217206] Call Trace:
[  142.217209]  [<ffffffff8142dd65>] dump_stack+0x67/0x92
[  142.217211]  [<ffffffff8107e4e6>] __warn+0xc6/0xe0
[  142.217213]  [<ffffffff8107e54a>] warn_slowpath_fmt+0x4a/0x50
[  142.217214]  [<ffffffff81559e3e>] drm_mm_takedown+0x2e/0x40
[  142.217236]  [<ffffffffa035c02a>] i915_gem_cleanup_stolen+0x1a/0x20 [i915]
[  142.217246]  [<ffffffffa034c581>] i915_ggtt_cleanup_hw+0x31/0xb0 [i915]
[  142.217253]  [<ffffffffa0310311>] i915_driver_cleanup_hw+0x31/0x40 [i915]
[  142.217260]  [<ffffffffa0312001>] i915_driver_unload+0x141/0x1a0 [i915]
[  142.217268]  [<ffffffffa031c2c4>] i915_pci_remove+0x14/0x20 [i915]
[  142.217269]  [<ffffffff8147d214>] pci_device_remove+0x34/0xb0
[  142.217271]  [<ffffffff8157b14c>] __device_release_driver+0x9c/0x150
[  142.217272]  [<ffffffff8157bcc6>] driver_detach+0xb6/0xc0
[  142.217273]  [<ffffffff8157abe3>] bus_remove_driver+0x53/0xd0
[  142.217274]  [<ffffffff8157c787>] driver_unregister+0x27/0x50
[  142.217276]  [<ffffffff8147c265>] pci_unregister_driver+0x25/0x70
[  142.217287]  [<ffffffffa03d764c>] i915_exit+0x1a/0x71 [i915]
[  142.217289]  [<ffffffff811136b3>] SyS_delete_module+0x193/0x1e0
[  142.217291]  [<ffffffff818174ae>] entry_SYSCALL_64_fastpath+0x1c/0xb1
[  142.217292] ---[ end trace 6fd164859c154772 ]---
[  142.217505] [drm:show_leaks] *ERROR* node [6b6b6b6b6b6b6b6b + 6b6b6b6b6b6b6b6b]: inserted at
                [<ffffffff81559ff3>] save_stack.isra.1+0x53/0xa0
                [<ffffffff8155a98d>] drm_mm_insert_node_in_range_generic+0x2ad/0x360
                [<ffffffffa035bf23>] i915_gem_stolen_insert_node_in_range+0x93/0xe0 [i915]
                [<ffffffffa035c855>] i915_gem_object_create_stolen+0x75/0xb0 [i915]
                [<ffffffffa036a51a>] intel_engine_create_ring+0x9a/0x140 [i915]
                [<ffffffffa036a921>] intel_init_ring_buffer+0xf1/0x440 [i915]
                [<ffffffffa036be1b>] intel_init_render_ring_buffer+0xab/0x1b0 [i915]
                [<ffffffffa0363d08>] intel_engines_init+0xc8/0x210 [i915]
                [<ffffffffa0355d7c>] i915_gem_init+0xac/0xf0 [i915]
                [<ffffffffa0311454>] i915_driver_load+0x9c4/0x1430 [i915]
                [<ffffffffa031c2f8>] i915_pci_probe+0x28/0x40 [i915]
                [<ffffffff8147d315>] pci_device_probe+0x85/0xf0
                [<ffffffff8157b7ff>] driver_probe_device+0x21f/0x430
                [<ffffffff8157baee>] __driver_attach+0xde/0xe0

In particular note that the node was being poisoned as we inspected the
list, a  clear indication that the object is being freed as we make the
assertion.

v2: Don't loop, just assert that we do all the work required as that
will be better at detecting further errors.

Fixes: fbbd37b36fa5 ("drm/i915: Move object release to a freelist + worker")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161101084843.3961-1-chris@chris-wilson.co.uk

drm/i915: Reorganize sprite init

Kill the switch statement from the sprite init code and replace with a
more straightforward if ladder. Now each significant evolution of the
sprite hardware is in its own neat box.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477411083-19255-5-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Bail if plane/crtc init fails

Due to the plane->index not getting readjusted in drm_plane_cleanup(),
we can't continue initialization of some plane/crtc init fails.
Well, we sort of could I suppose if we left all initialized planes on
the list, but that would expose those planes to userspace as well.

But for crtcs the situation is even worse since we assume that
pipe==crtc index occasionally, so we can't really deal with a partially
initialize set of crtcs.

So seems safest to just abort the entire thing if anything goes wrong.
All the failure paths here are kmalloc()s anyway, so it seems unlikely
we'd get very far if these start failing.

v2: Add (enum plane) case to silence gcc

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477411083-19255-4-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Initialize planes in a reasonable order

The zpos magic sorting uses the object ID to solve conflicting zpos
values. Let's initialize our planes in an order that makes the object
IDs agree with the normal primary->sprites->cursor z order.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477411083-19255-3-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Don't try to initialize sprite planes on pre-ilk

We don't currently implement support for sprite planes on pre-ilk
platforms, so let's leave num_sprites at 0 so that we don't get
spurious errors during driver init.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477411083-19255-2-git-send-email-ville.syrjala@linux.intel.com
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>

drm/i915: Mark up obj->mm.lock for shrinker

As we may allocate from within the obj->mm.lock we may enter the
shrinker for direct reclaim. Operating on the current object is
prevented by checking for obj->mm.pages (which is only set as the last
operation in the allocation path). However, we need to identify the
single recursion of accessing another object's obj->mm.lock as the two
locks have identical class and so appear to be the same to lockdep,
convincing it that a deadlock is possible. Use mutex_lock_nested() to
remove the false positive.

[ 2165.945734] =================================
[ 2165.945749] [ INFO: inconsistent lock state ]
[ 2165.945765] 4.9.0-rc2+ #2 Tainted: G        W
[ 2165.945781] ---------------------------------
[ 2165.945796] inconsistent {RECLAIM_FS-ON-W} -> {IN-RECLAIM_FS-W} usage.
[ 2165.945816] kswapd0/62 [HC0[0]:SC0[0]:HE1:SE1] takes: (&obj->mm.lock){+.+.?.}, at: [<ffffffffc0289a1f>] i915_gem_shrink+0x29f/0x500 [i915]
[ 2165.945904] {RECLAIM_FS-ON-W} state was registered at:
[ 2165.945931] [<ffffffffb10bd50f>] mark_held_locks+0x6f/0xa0
[ 2165.945956] [<ffffffffb10bf889>] lockdep_trace_alloc+0x69/0xc0
[ 2165.945982] [<ffffffffb11eea53>] kmem_cache_alloc_trace+0x33/0x2a0
[ 2165.946019] [<ffffffffc028a28a>] i915_gem_object_get_pages_stolen+0x6a/0xd0 [i915]
[ 2165.946060] [<ffffffffc027e1d0>] ____i915_gem_object_get_pages+0x20/0x60 [i915]
[ 2165.946098] [<ffffffffc027e268>] __i915_gem_object_get_pages+0x58/0x70 [i915]
[ 2165.946138] [<ffffffffc028a3dc>] _i915_gem_object_create_stolen+0xec/0x120 [i915]
[ 2165.946177] [<ffffffffc028af73>] i915_gem_object_create_stolen_for_preallocated+0xf3/0x3f0 [i915]
[ 2165.946222] [<ffffffffc02bae43>] intel_alloc_initial_plane_obj.isra.125+0xd3/0x200 [i915]
[ 2165.946266] [<ffffffffc02cb1c1>] intel_modeset_init+0x931/0x1530 [i915]
[ 2165.946301] [<ffffffffc023d584>] i915_driver_load+0xa14/0x14a0 [i915]
[ 2165.946335] [<ffffffffc0248aff>] i915_pci_probe+0x4f/0x70 [i915]
[ 2165.946362] [<ffffffffb13cc452>] local_pci_probe+0x42/0xa0
[ 2165.946386] [<ffffffffb13cd903>] pci_device_probe+0x103/0x150
[ 2165.946411] [<ffffffffb14adeb3>] driver_probe_device+0x223/0x430
[ 2165.946436] [<ffffffffb14ae1a3>] __driver_attach+0xe3/0xf0
[ 2165.946461] [<ffffffffb14ab943>] bus_for_each_dev+0x73/0xc0
[ 2165.946485] [<ffffffffb14ad5ee>] driver_attach+0x1e/0x20
[ 2165.946508] [<ffffffffb14ad003>] bus_add_driver+0x173/0x270
[ 2165.946533] [<ffffffffb14aee70>] driver_register+0x60/0xe0
[ 2165.946557] [<ffffffffb13cbd6d>] __pci_register_driver+0x5d/0x60
[ 2165.946606] [<ffffffffc0378057>] soundcore_open+0x17/0x230 [soundcore]
[ 2165.946636] [<ffffffffb1000450>] do_one_initcall+0x50/0x180
[ 2165.946661] [<ffffffffb117fd2d>] do_init_module+0x5f/0x1f1
[ 2165.946685] [<ffffffffb1108964>] load_module+0x2174/0x2a80
[ 2165.946709] [<ffffffffb11094df>] SYSC_finit_module+0xdf/0x110
[ 2165.946734] [<ffffffffb110952e>] SyS_finit_module+0xe/0x10
[ 2165.946758] [<ffffffffb1742aea>] entry_SYSCALL_64_fastpath+0x18/0xad
[ 2165.946776] irq event stamp: 90871
[ 2165.946788] hardirqs last  enabled at (90871):
[ 2165.946805] [<ffffffffb173e9da>] __mutex_unlock_slowpath+0x11a/0x1c0
[ 2165.946823] hardirqs last disabled at (90870):
[ 2165.946839] [<ffffffffb173e91b>] __mutex_unlock_slowpath+0x5b/0x1c0
[ 2165.946856] softirqs last  enabled at (90858):
[ 2165.946872] [<ffffffffb174581a>] __do_softirq+0x39a/0x4c6
[ 2165.946887] softirqs last disabled at (90671):
[ 2165.946902] [<ffffffffb1066cea>] irq_exit+0xea/0xf0
[ 2165.946916] other info that might help us debug this:
[ 2165.946936]  Possible unsafe locking scenario:
[ 2165.946955]        CPU0
[ 2165.946965]        ----
[ 2165.946975]   lock(&obj->mm.lock);
[ 2165.947000]   <Interrupt>
[ 2165.947010]     lock(&obj->mm.lock);
[ 2165.947035] *** DEADLOCK ***
[ 2165.947054] 2 locks held by kswapd0/62:
[ 2165.947067]  #0: (shrinker_rwsem){++++..}, at: [<ffffffffb119a20e>] shrink_slab.part.40+0x5e/0x5d0
[ 2165.947120]  #1: (&dev->struct_mutex){+.+.+.}, at: [<ffffffffc028954b>] i915_gem_shrinker_lock+0x1b/0x60 [i915]
[ 2165.948909] stack backtrace:
[ 2165.950650] CPU: 2 PID: 62 Comm: kswapd0 Tainted: G        W 4.9.0-rc2+ #2
[ 2165.951587] Hardware name: LENOVO 80MX/Lenovo E31-80, BIOS DCCN34WW(V2.03) 12/01/2015
[ 2165.952484]  ffffc90000b5f8c8 ffffffffb137f645 ffff88016c5a2700 ffffffffb25f20a0
[ 2165.953395]  ffffc90000b5f918 ffffffffb10bcecd 0000000000000000 ffff880100000001
[ 2165.954305]  0000000000000001 000000000000000a ffff88016c5a2fd0 ffff88016c5a2700
[ 2165.955240] Call Trace:
[ 2165.956170]  [<ffffffffb137f645>] dump_stack+0x68/0x93
[ 2165.957071]  [<ffffffffb10bcecd>] print_usage_bug+0x1dd/0x1f0
[ 2165.957979]  [<ffffffffb10bd439>] mark_lock+0x559/0x5c0
[ 2165.958875]  [<ffffffffb10bc3f0>] ?  print_shortest_lock_dependencies+0x1b0/0x1b0
[ 2165.959829]  [<ffffffffb10be04d>] __lock_acquire+0x66d/0x12a0
[ 2165.960729]  [<ffffffffb11ef541>] ? __slab_free+0xa1/0x340
[ 2165.961625]  [<ffffffffb10dba5d>] ?  debug_lockdep_rcu_enabled+0x1d/0x20
[ 2165.962530]  [<ffffffffb10bd50f>] ? mark_held_locks+0x6f/0xa0
[ 2165.963457]  [<ffffffffb10bf0b0>] lock_acquire+0xf0/0x1f0
[ 2165.964368]  [<ffffffffc0289a1f>] ? i915_gem_shrink+0x29f/0x500 [i915]
[ 2165.965269]  [<ffffffffc0289a1f>] ? i915_gem_shrink+0x29f/0x500 [i915]
[ 2165.966150]  [<ffffffffb173d837>] mutex_lock_nested+0x77/0x420
[ 2165.967030]  [<ffffffffc0289a1f>] ? i915_gem_shrink+0x29f/0x500 [i915]
[ 2165.967952]  [<ffffffffc027c7a1>] ?  __i915_gem_object_put_pages.part.58+0x161/0x1b0 [i915]
[ 2165.968835]  [<ffffffffc0289a1f>] i915_gem_shrink+0x29f/0x500 [i915]
[ 2165.969712]  [<ffffffffc0289e40>] i915_gem_shrinker_scan+0x70/0xb0 [i915]
[ 2165.970591]  [<ffffffffb119a3ae>] shrink_slab.part.40+0x1fe/0x5d0
[ 2165.971504]  [<ffffffffb119f19c>] shrink_node+0x22c/0x320
[ 2165.972371]  [<ffffffffb11a05fb>] kswapd+0x38b/0x9b0
[ 2165.973238]  [<ffffffffb11a0270>] ?  mem_cgroup_shrink_node+0x330/0x330
[ 2165.974068]  [<ffffffffb108630f>] kthread+0xff/0x120
[ 2165.974929]  [<ffffffffb1086210>] ? kthread_park+0x60/0x60
[ 2165.975847]  [<ffffffffb1742d57>] ret_from_fork+0x27/0x40

Reported-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Fixes: 1233e2db199d ("drm/i915: Move object backing storage manipulation...")
Testcase: igt/gem_ctx_create/maximum-swap
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161031124048.30355-1-chris@chris-wilson.co.uk

MAINTAINERS: drop dri-devel list for i915

In practice, none of the i915 developers Cc dri-devel for strictly i915
specific patches. Make MAINTAINERS reflect reality, and reduce random
i915 specific noise on dri-devel.

Also, we have a fairly large crowd reading and responding on intel-gfx,
and we're pretty good at involving dri-devel when that is appropriate.

Cc: dri-devel@lists.freedesktop.org
Acked-by: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477498292-9808-1-git-send-email-jani.nikula@intel.com

drm/i915/vlv: Prevent enabling hpd polling in late suspend

One of the CI machines began to run into issues with the hpd poller
suddenly waking up in the midst of the late suspend phase. It looks like
this is getting caused by the fact we now deinitialize power wells in
late suspend, which means that intel_hpd_poll_init() gets called in late
suspend causing polling to get re-enabled. So, when deinitializing power
wells on valleyview we now refrain from enabling polling in the midst of
suspend.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98040
Fixes: 19625e85c6ec ("drm/i915: Enable polling when we don't have hpd")
Signed-off-by: Lyude <lyude@redhat.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Jani Saarinen <jani.saarinen@intel.com>
Cc: Petry Latvala <petri.latvala@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/1477499769-1966-1-git-send-email-lyude@redhat.com

drm/i915: Enable multiple timelines

With the infrastructure converted over to tracking multiple timelines in
the GEM API whilst preserving the efficiency of using a single execution
timeline internally, we can now assign a separate timeline to every
context with full-ppgtt.

v2: Add a comment to indicate the xfer between timelines upon submission.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-35-chris@chris-wilson.co.uk

drm/i915: Defer setting of global seqno on request to submission

Defer the assignment of the global seqno on a request to its submission.
In the next patch, we will only allocate the global seqno at that time,
here we are just enabling the wait-for-submission before wait-for-seqno
paths.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-34-chris@chris-wilson.co.uk

drm/i915: Reserve space in the global seqno during request allocation

A restriction on our global seqno is that they cannot wrap, and that we
cannot use the value 0. This allows us to detect when a request has not
yet been submitted, its global seqno is still 0, and ensures that
hardware semaphores are monotonic as required by older hardware. To
meet these restrictions when we defer the assignment of the global
seqno, we must check that we have an available slot in the global seqno
space during request construction. If that test fails, we wait for all
requests to be completed and reset the hardware back to 0.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-33-chris@chris-wilson.co.uk

drm/i915: Convert breadcrumbs spinlock to be irqsafe

The breadcrumbs are about to be used from within IRQ context sections
(e.g. nouveau signals a fence from an interrupt handler causing us to
submit a new request) and/or from bottom-half tasklets (i.e.
intel_lrc_irq_handler), therefore we need to employ the irqsafe spinlock
variants.

For example, deferring the request submission to the
intel_lrc_irq_handler generates this trace:

[   66.388639] =================================
[   66.388650] [ INFO: inconsistent lock state ]
[   66.388663] 4.9.0-rc2+ #56 Not tainted
[   66.388672] ---------------------------------
[   66.388682] inconsistent {SOFTIRQ-ON-W} -> {IN-SOFTIRQ-W} usage.
[   66.388695] swapper/1/0 [HC0[0]:SC1[1]:HE0:SE0] takes:
[   66.388706]  (&(&b->lock)->rlock){+.?...} , at: [<ffffffff81401c88>] intel_engine_enable_signaling+0x78/0x150
[   66.388761] {SOFTIRQ-ON-W} state was registered at:
[   66.388772]   [   66.388783] [<ffffffff810bd842>] __lock_acquire+0x682/0x1870
[   66.388795]   [   66.388803] [<ffffffff810bedbc>] lock_acquire+0x6c/0xb0
[   66.388814]   [   66.388824] [<ffffffff8161753a>] _raw_spin_lock+0x2a/0x40
[   66.388835]   [   66.388845] [<ffffffff81401e41>] intel_engine_reset_breadcrumbs+0x21/0xb0
[   66.388857]   [   66.388866] [<ffffffff81403ae7>] gen8_init_common_ring+0x67/0x100
[   66.388878]   [   66.388887] [<ffffffff81403b92>] gen8_init_render_ring+0x12/0x60
[   66.388903]   [   66.388912] [<ffffffff813f8707>] i915_gem_init_hw+0xf7/0x2a0
[   66.388927]   [   66.388936] [<ffffffff813f899b>] i915_gem_init+0xbb/0xf0
[   66.388950]   [   66.388959] [<ffffffff813b4980>] i915_driver_load+0x7e0/0x1330
[   66.388978]   [   66.388988] [<ffffffff813c09d8>] i915_pci_probe+0x28/0x40
[   66.389003]   [   66.389013] [<ffffffff812fa0db>] pci_device_probe+0x8b/0xf0
[   66.389028]   [   66.389037] [<ffffffff8147737e>] driver_probe_device+0x21e/0x430
[   66.389056]   [   66.389065] [<ffffffff8147766e>] __driver_attach+0xde/0xe0
[   66.389080]   [   66.389090] [<ffffffff814751ad>] bus_for_each_dev+0x5d/0x90
[   66.389105]   [   66.389113] [<ffffffff81477799>] driver_attach+0x19/0x20
[   66.389134]   [   66.389144] [<ffffffff81475ced>] bus_add_driver+0x15d/0x260
[   66.389159]   [   66.389168] [<ffffffff81477e3b>] driver_register+0x5b/0xd0
[   66.389183]   [   66.389281] [<ffffffff812fa19b>] __pci_register_driver+0x5b/0x60
[   66.389301]   [   66.389312] [<ffffffff81aed333>] i915_init+0x3e/0x45
[   66.389326]   [   66.389336] [<ffffffff81ac2ffa>] do_one_initcall+0x8b/0x118
[   66.389350]   [   66.389359] [<ffffffff81ac323a>] kernel_init_freeable+0x1b3/0x23b
[   66.389378]   [   66.389387] [<ffffffff8160fc39>] kernel_init+0x9/0x100
[   66.389402]   [   66.389411] [<ffffffff816180e7>] ret_from_fork+0x27/0x40
[   66.389426] irq event stamp: 315865
[   66.389438] hardirqs last  enabled at (315864): [<ffffffff816178f1>] _raw_spin_unlock_irqrestore+0x31/0x50
[   66.389469] hardirqs last disabled at (315865): [<ffffffff816176b3>] _raw_spin_lock_irqsave+0x13/0x50
[   66.389499] softirqs last  enabled at (315818): [<ffffffff8107a04c>] _local_bh_enable+0x1c/0x50
[   66.389530] softirqs last disabled at (315819): [<ffffffff8107a50e>] irq_exit+0xbe/0xd0
[   66.389559]
[   66.389559] other info that might help us debug this:
[   66.389580]  Possible unsafe locking scenario:
[   66.389580]
[   66.389598]        CPU0
[   66.389609]        ----
[   66.389620]   lock(&(&b->lock)->rlock);
[   66.389650]   <Interrupt>
[   66.389661]     lock(&(&b->lock)->rlock);
[   66.389690]
[   66.389690]  *** DEADLOCK ***
[   66.389690]
[   66.389715] 2 locks held by swapper/1/0:
[   66.389728]  #0: (&(&tl->lock)->rlock){..-...}, at: [<ffffffff81403e01>] intel_lrc_irq_handler+0x201/0x3c0
[   66.389785]  #1: (&(&req->lock)->rlock/1){..-...}, at: [<ffffffff813fc0af>] __i915_gem_request_submit+0x8f/0x170
[   66.389854]
[   66.389854] stack backtrace:
[   66.389959] CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.9.0-rc2+ #56
[   66.389976] Hardware name:                  /        , BIOS PYBSWCEL.86A.0027.2015.0507.1758 05/07/2015
[   66.389999]  ffff88027fd03c58 ffffffff812beae5 ffff88027696e680 ffffffff822afe20
[   66.390036]  ffff88027fd03ca8 ffffffff810bb420 0000000000000001 0000000000000000
[   66.390070]  0000000000000000 0000000000000006 0000000000000004 ffff88027696ee10
[   66.390104] Call Trace:
[   66.390117]  <IRQ>
[   66.390128]  [<ffffffff812beae5>] dump_stack+0x68/0x93
[   66.390147]  [<ffffffff810bb420>] print_usage_bug+0x1d0/0x1e0
[   66.390164]  [<ffffffff810bb8a0>] mark_lock+0x470/0x4f0
[   66.390181]  [<ffffffff810ba9d0>] ? print_shortest_lock_dependencies+0x1b0/0x1b0
[   66.390203]  [<ffffffff810bd75d>] __lock_acquire+0x59d/0x1870
[   66.390221]  [<ffffffff810bedbc>] lock_acquire+0x6c/0xb0
[   66.390237]  [<ffffffff810bedbc>] ? lock_acquire+0x6c/0xb0
[   66.390255]  [<ffffffff81401c88>] ? intel_engine_enable_signaling+0x78/0x150
[   66.390273]  [<ffffffff8161753a>] _raw_spin_lock+0x2a/0x40
[   66.390291]  [<ffffffff81401c88>] ? intel_engine_enable_signaling+0x78/0x150
[   66.390309]  [<ffffffff81401c88>] intel_engine_enable_signaling+0x78/0x150
[   66.390327]  [<ffffffff813fc170>] __i915_gem_request_submit+0x150/0x170
[   66.390345]  [<ffffffff81403e8b>] intel_lrc_irq_handler+0x28b/0x3c0
[   66.390363]  [<ffffffff81079d97>] tasklet_action+0x57/0xc0
[   66.390380]  [<ffffffff8107a249>] __do_softirq+0x119/0x240
[   66.390396]  [<ffffffff8107a50e>] irq_exit+0xbe/0xd0
[   66.390414]  [<ffffffff8101afd5>] do_IRQ+0x65/0x110
[   66.390431]  [<ffffffff81618806>] common_interrupt+0x86/0x86
[   66.390446]  <EOI>
[   66.390457]  [<ffffffff814ec6d1>] ? cpuidle_enter_state+0x151/0x200
[   66.390480]  [<ffffffff814ec7a2>] cpuidle_enter+0x12/0x20
[   66.390498]  [<ffffffff810b639e>] call_cpuidle+0x1e/0x40
[   66.390516]  [<ffffffff810b65ae>] cpu_startup_entry+0x10e/0x1f0
[   66.390534]  [<ffffffff81036133>] start_secondary+0x103/0x130

(This is split out of the defer global seqno allocation patch due to
realisation that we need a more complete conversion if we want to defer
request submission even further.)

v2: lockdep was warning about mixed SOFTIRQ contexts not HARDIRQ
contexts so we only need to use spin_lock_bh and not disable interrupts.

v3: We need full irq protection as we may be called from a third party
interrupt handler (via fences).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-32-chris@chris-wilson.co.uk

drm/i915: Create a unique name for the context

This will be used for communicating issues with this context to
userspace, so we want to identify the parent process and the individual
context. Note that the name isn't quite unique, it makes the presumption
of there only being a single device fd per process.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-31-chris@chris-wilson.co.uk

drm/i915: Move the global sync optimisation to the timeline

Currently we try to reduce the number of synchronisations (now the
number of requests we need to wait upon) by noting that if we have
earlier waited upon a request, all subsequent requests in the timeline
will be after the wait. This only applies to requests in this timeline,
as other timelines will not be ordered by that waiter.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-30-chris@chris-wilson.co.uk

drm/i915: Defer breadcrumb emission

Move the actual emission of the breadcrumb for closing the request from
i915_add_request() to the submit callback. (It can be moved later when
required.) This allows us to defer the allocation of the global_seqno
from request construction to actual submission, allowing us to emit the
requests out of order (wrt to the order of their construction, they
still will only be executed one all of their dependencies are resolved
including that all earlier requests on their timeline have been
submitted.) We have to specialise how we then emit the request in order
to write into the preallocated space, rather than at the tail of the
ringbuffer (which will have been advanced by the addition of new
requests).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-29-chris@chris-wilson.co.uk

drm/i915: Record space required for breadcrumb emission

In the next patch, we will use deferred breadcrumb emission. That requires
reserving sufficient space in the ringbuffer to emit the breadcrumb, which
first requires us to know how large the breadcrumb is.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-28-chris@chris-wilson.co.uk

drm/i915: Rename ->emit_request to ->emit_breadcrumb

Now that the emission of the request tail and its submission to hardware
are two separate steps, engine->emit_request() is confusing.
engine->emit_request() is called to emit the breadcrumb commands for the
request into the ring, name it such (engine->emit_breadcrumb).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-27-chris@chris-wilson.co.uk

drm/i915: Introduce a global_seqno for each request

Though we will have multiple timelines, we still have a single timeline
of execution. This we can use to provide an execution and retirement order
of requests. This keeps tracking execution of requests simple, and vital
for preserving a single waiter (i.e. so that we can order the waiters so
that only the earliest to wakeup need be woken). To accomplish this we
distinguish the seqno used to order requests per-context (external) and
that used internally for execution.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-26-chris@chris-wilson.co.uk

drm/i915: Wait first for submission, before waiting for request completion

In future patches, we will no longer be able to wait on a static global
seqno and instead have to break our wait up into phases. First we wait
for the global seqno assignment (upon submission to hardware), and once
submitted we wait for the hardware to complete.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-25-chris@chris-wilson.co.uk

drm/i915: Queue the idling context switch after all other timelines

Before suspend, we wait for the switch to the kernel context. In order
for all the other context images to be complete upon suspend, that
switch must be the last operation by the GPU (i.e. this idling request
must not overtake any pending requests). To make this request execute last,
we make it depend on every other inflight request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-24-chris@chris-wilson.co.uk

drm/i915: Combine seqno + tracking into a global timeline struct

Our timelines are more than just a seqno. They also provide an ordered
list of requests to be executed. Due to the restriction of handling
individual address spaces, we are limited to a timeline per address
space but we use a fence context per engine within.

Our first step to introducing independent timelines per context (i.e. to
allow each context to have a queue of requests to execute that have a
defined set of dependencies on other requests) is to provide a timeline
abstraction for the global execution queue.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-23-chris@chris-wilson.co.uk

drm/i915: Restore nonblocking awaits for modesetting

After combining the dma-buf reservation object and the GEM reservation
object, we lost the ability to do a nonblocking wait on the i915 request
(as we blocked upon the reservation object during prepare_fb). We can
instead convert the reservation object into a fence upon which we can
asynchronously wait (including a forced timeout in case the DMA fence is
never signaled).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-22-chris@chris-wilson.co.uk

drm/i915: Move GEM activity tracking into a common struct reservation_object

In preparation to support many distinct timelines, we need to expand the
activity tracking on the GEM object to handle more than just a request
per engine. We already use the struct reservation_object on the dma-buf
to handle many fence contexts, so integrating that into the GEM object
itself is the preferred solution. (For example, we can now share the same
reservation_object between every consumer/producer using this buffer and
skip the manual import/export via dma-buf.)

v2: Reimplement busy-ioctl (by walking the reservation object), postpone
the ABI change for another day. Similarly use the reservation object to
find the last_write request (if active and from i915) for choosing
display CS flips.

Caveats:

 * busy-ioctl: busy-ioctl only reports on the native fences, it will not
warn of stalls (in set-domain-ioctl, pread/pwrite etc) if the object is
being rendered to by external fences. It also will not report the same
busy state as wait-ioctl (or polling on the dma-buf) in the same
circumstances. On the plus side, it does retain reporting of which
*i915* engines are engaged with this object.

 * non-blocking atomic modesets take a step backwards as the wait for
render completion blocks the ioctl. This is fixed in a subsequent
patch to use a fence instead for awaiting on the rendering, see
"drm/i915: Restore nonblocking awaits for modesetting"

 * dynamic array manipulation for shared-fences in reservation is slower
than the previous lockless static assignment (e.g. gem_exec_lut_handle
runtime on ivb goes from 42s to 66s), mainly due to atomic operations
(maintaining the fence refcounts).

 * loss of object-level retirement callbacks, emulated by VMA retirement
tracking.

 * minor loss of object-level last activity information from debugfs,
could be replaced with per-vma information if desired

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-21-chris@chris-wilson.co.uk

drm/i915: Use lockless object free

Having moved the locked phase of freeing an object to a separate worker,
we can now declare to the core that we only need the unlocked variant of
driver->gem_free_object, and can use the simple unreference internally.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-20-chris@chris-wilson.co.uk

drm/i915: Move object release to a freelist + worker

We want to hide the latency of releasing objects and their backing
storage from the submission, so we move the actual free to a worker.
This allows us to switch to struct_mutex freeing of the object in the
next patch.

Furthermore, if we know that the object we are dereferencing remains valid
for the duration of our access, we can forgo the usual synchronisation
barriers and atomic reference counting. To ensure this we defer freeing
an object til after an RCU grace period, such that any lookup of the
object within an RCU read critical section will remain valid until
after we exit that critical section. We also employ this delay for
rate-limiting the serialisation on reallocation - we have to slow down
object creation in order to prevent resource starvation (in particular,
files).

v2: Return early in i915_gem_tiling() ioctl to skip over superfluous
work on error.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-19-chris@chris-wilson.co.uk

drm/i915: Acquire the backing storage outside of struct_mutex in set-domain

As we can locklessly (well struct_mutex-lessly) acquire the backing
storage, do so in set-domain-ioctl to reduce the contention on the
struct_mutex.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-18-chris@chris-wilson.co.uk

drm/i915: Implement pwrite without struct-mutex

We only need struct_mutex within pwrite for a brief window where we need
to serialise with rendering and control our cache domains. Elsewhere we
can rely on the backing storage being pinned, and forgive userspace any
races against us.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-17-chris@chris-wilson.co.uk

drm/i915: Implement pread without struct-mutex

We only need struct_mutex within pread for a brief window where we need
to serialise with rendering and control our cache domains. Elsewhere we
can rely on the backing storage being pinned, and forgive userspace any
races against us.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-16-chris@chris-wilson.co.uk

drm/i915/dmabuf: Acquire the backing storage outside of struct_mutex

Use the per-object mm.lock to allocate the backing storage (and hold a
reference to it across the dmabuf access) without resorting to
struct_mutex.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-15-chris@chris-wilson.co.uk

drm/i915: Move object backing storage manipulation to its own locking

Break the allocation of the backing storage away from struct_mutex into
a per-object lock. This allows parallel page allocation, provided we can
do so outside of struct_mutex (i.e. set-domain-ioctl, pwrite, GTT
fault), i.e. before execbuf! The increased cost of the atomic counters
are hidden behind i915_vma_pin() for the typical case of execbuf, i.e.
as the object is typically bound between execbufs, the page_pin_count is
static. The cost will be felt around set-domain and pwrite, but offset
by the improvement from reduced struct_mutex contention.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-14-chris@chris-wilson.co.uk

drm/i915: Pass around sg_table to get_pages/put_pages backend

The plan is to move obj->pages out from under the struct_mutex into its
own per-object lock. We need to prune any assumption of the struct_mutex
from the get_pages/put_pages backends, and to make it easier we pass
around the sg_table to operate on rather than indirectly via the obj.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-13-chris@chris-wilson.co.uk

drm/i915: Refactor object page API

The plan is to make obtaining the backing storage for the object avoid
struct_mutex (i.e. use its own locking). The first step is to update the
API so that normal users only call pin/unpin whilst working on the
backing storage.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-12-chris@chris-wilson.co.uk

drm/i915: Use radixtree to jump start intel_partial_pages()

We can use the radixtree index of the obj->pages to find the start
position of the desired partial range.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-11-chris@chris-wilson.co.uk

drm/i915: Use a radixtree for random access to the object's backing storage

A while ago we switched from a contiguous array of pages into an sglist,
for that was both more convenient for mapping to hardware and avoided
the requirement for a vmalloc array of pages on every object. However,
certain GEM API calls (like pwrite, pread as well as performing
relocations) do desire access to individual struct pages. A quick hack
was to introduce a cache of the last access such that finding the
following page was quick - this works so long as the caller desired
sequential access. Walking backwards, or multiple callers, still hits a
slow linear search for each page. One solution is to store each
successful lookup in a radix tree.

v2: Rewrite building the radixtree for clarity, hopefully.

v3: Rearrange execbuf to avoid calling i915_gem_object_get_sg() from
within an atomic section and so relax the allocation context to a simple
GFP_KERNEL and mutex.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-10-chris@chris-wilson.co.uk

drm/i915: Markup GEM API with lockdep asserts

Add lockdep_assert_held(struct_mutex) to the API preamble of the
internal GEM interfaces.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-9-chris@chris-wilson.co.uk

drm/i915: Reuse the active golden render state batch

The golden render state is constant, but we recreate the batch setting
it up for every new context. If we keep that batch in a volatile cache
we can safely reuse it whenever we need to initialise a new context. We
mark the pages as purgeable and use the shrinker to recover pages from
the batch whenever we face memory pressues, recreating that batch afresh
on the next new context.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Joonas Lahtinen <joonas.lahtien@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20161028125858.23563-8-chris@chris-wilson.co.uk