A significant cost in setting up a wait is the overhead of enabling the
interrupt. As we disable the interrupt whenever the queue of waiters is
empty, if we are frequently waiting on alternating batches, we end up
re-enabling the interrupt on a frequent basis. We do want to disable the
interrupt during normal operations as under high load it may add several
thousand interrupts/s - we have been known in the past to occupy whole
cores with our interrupt handler after accidentally leaving user
interrupts enabled. As a compromise, leave the interrupt enabled until
the next IRQ, or the system is idle. This gives a small window for a
waiter to keep the interrupt active and not be delayed by having to
re-enable the interrupt.

v2: Restore hangcheck/missed-irq detection for continuations
v3: Be more careful restoring the hangcheck timer after reset
v4: Be more careful restoring the fake irq after reset (if required!)
v5: Redo changes to intel_engine_wakeup()
v6: Factor out __intel_engine_wakeup()
v7: Improve commentary for declaring a missed wakeup
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170227205850.2828-4-chris@chris-wilson.co.uk

This patch makes the I915_PARAM_HAS_EXEC_CONSTANTS getparam return 0
(indicating the optional feature is not supported), and makes execbuf
always return -EINVAL if the flags are used.

Apparently, no userspace ever shipped which used this optional feature:
I checked the git history of Mesa, xf86-video-intel, libva, and Beignet,
and there were zero commits showing a use of these flags.  Kernel commit
72bfa19c apparently introduced the feature prematurely.  According
to Chris, the intention was to use this in cairo-drm, but "the use was
broken for gen6", so I don't think it ever happened.

'relative_constants_mode' has always been tracked per-device, but this
has actually been wrong ever since hardware contexts were introduced, as
the INSTPM register is saved (and automatically restored) as part of the
render ring context. The software per-device value could therefore get
out of sync with the hardware per-context value.  This meant that using
them is actually unsafe: a client which tried to use them could damage
the state of other clients, causing the GPU to interpret their BO
offsets as absolute pointers, leading to bogus memory reads.

These flags were also never ported to execlist mode, making them no-ops
on Gen9+ (which requires execlists), and Gen8 in the default mode.

On Gen8+, userspace can write these registers directly, achieving the
same effect.  On Gen6-7.5, it likely makes sense to extend the command
parser to support them.  I don't think anyone wants this on Gen4-5.

Based on a patch by Dave Gordon.

v3: Return -ENODEV for the getparam, as this is what we do for other
    obsolete features.  Suggested by Chris Wilson.

Cc: stable@vger.kernel.org
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=92448Signed-off-by: NKenneth Graunke <kenneth@whitecape.org>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170215093446.21291-1-kenneth@whitecape.orgAcked-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>

A request is assigned a global seqno only when it is on the hardware
execution queue. The global seqno can be used to maintain a list of
requests on the same engine in retirement order, for example for
constructing a priority queue for waiting. Prior to its execution, or
if it is subsequently removed in the event of preemption, its global
seqno is zero. As both insertion and removal from the execution queue
may operate in IRQ context, it is not guarded by the usual struct_mutex
BKL. Instead those relying on the global seqno must be prepared for its
value to change between reads. Only when the request is complete can
the global seqno be stable (due to the memory barriers on submitting
the commands to the hardware to write the breadcrumb, if the HWS shows
that it has passed the global seqno and the global seqno is unchanged
after the read, it is indeed complete).
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170223074422.4125-9-chris@chris-wilson.co.uk

Setting retire=true is identical to using origin=ORIGIN_CS, so make the
same simplification to intel_fb_obj_flush() as already employed for
intel_fb_obj_invalidate().
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170222114049.28456-6-chris@chris-wilson.co.uk

Flushing the cachelines for an object is slow, can be as much as 100ms
for a large framebuffer. We currently do this under the struct_mutex BKL
on execution or on pageflip. But now with the ability to add fences to
obj->resv for both flips and execbuf (and we naturally wait on the fence
before CPU access), we can move the clflush operation to a workqueue and
signal a fence for completion, thereby doing the work asynchronously and
not blocking the driver or its clients.

v2: Introduce i915_gem_clflush.h and use a new name, split out some
extras into separate patches.
Suggested-by: NAkash Goel <akash.goel@intel.com>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170222114049.28456-5-chris@chris-wilson.co.uk

Generalise the skip for physical and stolen objects by skipping anything
we do not have a valid address for inside the sg.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NMatthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170222114049.28456-4-chris@chris-wilson.co.uk

We have three different paths by which userspace wants to flush the
display plane (i.e. objects with obj->pin_display). Use a common helper
to identify those paths and to simplify a later change.

v2: Include the conditional in the name, i915_gem_object_flush_if_display
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170222114049.28456-3-chris@chris-wilson.co.uk

For use in the next patch, take the current is-coherent helper and add
it to i915_gem_object.h
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170222114049.28456-2-chris@chris-wilson.co.uk

The change_domain tracepoint has been inaccurate for a few years - it
doesn't fully capture the domains, especially with userspace bypassing
them. It is defunct, misleading and time to be removed.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170222114049.28456-1-chris@chris-wilson.co.uk

If we wait upon the full (i.e. all shared fences, or upon an exclusive
fence) reservation object successfully, we know that all fences beneath
it have been signaled, so long as no new fences were added whilst we
slept. If the reservation_object remains the same, as detected by its
seqcount, we can then reap all the fences upon completion.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Reviewed-by: NMatthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170217151304.16665-6-chris@chris-wilson.co.uk

The physical object is treated as permanently pinned. If we fail to take
this initial pin during i915_gem_object_attach_phys() we need to revert
it back to an ordinary shmemfs object before reporting the failure.

v2: git-add
Reported-by: NMika Kuoppala <mika.kuoppala@intel.com>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170215163900.11606-1-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>

We wait upon jiffies, but report the time elapsed using a
high-resolution timer. This discrepancy can lead to us timing out the
wait prior to us reporting the elapsed time as complete.

This restores the squelching lost in commit e95433c7 ("drm/i915:
Rearrange i915_wait_request() accounting with callers").

Fixes: e95433c7 ("drm/i915: Rearrange i915_wait_request() accounting with callers")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.10-rc1+
Cc: stable@vger.kernel.org
Link: http://patchwork.freedesktop.org/patch/msgid/20170216125441.30923-1-chris@chris-wilson.co.ukReviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>

The i915_gem_object_wait_fence() uses an incoming timeout=0 to query
whether the current fence is busy or idle, without waiting. This can be
used by the wait-ioctl to implement a busy query.

Fixes: e95433c7 ("drm/i915: Rearrange i915_wait_request() accounting with callers")
Testcase: igt/gem_wait/basic-busy-write-all
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.10-rc1+
Cc: stable@vger.kernel.org
Link: http://patchwork.freedesktop.org/patch/msgid/20170212215344.16600-1-chris@chris-wilson.co.ukReviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
(cherry picked from commit d892e939)
Signed-off-by: NJani Nikula <jani.nikula@intel.com>

Following a reset, the context and page directory registers are lost.
However, the queue of requests that we resubmit after the reset may
depend upon them - the registers are restored from a context image, but
that restore may be inhibited and may simply be absent from the request
if it was in the middle of a sequence using the same context. If we
prime the CCID/PD registers with the first request in the queue (even
for the hung request), we prevent invalid memory access for the
following requests (and continually hung engines).

v2: Magic BIT(8), reserved for future use but still appears unused.
v3: Some commentary on handling innocent vs guilty requests
v4: Add a wait for PD_BASE fetch. The reload appears to be instant on my
Ivybridge, but this bit probably exists for a reason.

Fixes: 821ed7df ("drm/i915: Update reset path to fix incomplete requests")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170207152437.4252-1-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>
(cherry picked from commit c0dcb203)
Signed-off-by: NJani Nikula <jani.nikula@intel.com>

The i915_gem_object_wait_fence() uses an incoming timeout=0 to query
whether the current fence is busy or idle, without waiting. This can be
used by the wait-ioctl to implement a busy query.

Fixes: e95433c7 ("drm/i915: Rearrange i915_wait_request() accounting with callers")
Testcase: igt/gem_wait/basic-busy-write-all
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.10-rc1+
Cc: stable@vger.kernel.org
Link: http://patchwork.freedesktop.org/patch/msgid/20170212215344.16600-1-chris@chris-wilson.co.ukReviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>

Write into an object using WB, WC, GTT, and GPU paths and make sure that
our internal API is sufficient to ensure coherent reads and writes.

v2: Avoid invalid free upon allocation error
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NMatthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170213171558.20942-21-chris@chris-wilson.co.uk

The phys object is a rarely used device (only very old machines require
a chunk of physically contiguous pages for a few hardware interactions).
As such, it is not exercised by CI and to combat that we want to add a
test that exercises the phys object on all platforms.

v2: Always set err on error paths and not rely on inheriting the err.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170213171558.20942-17-chris@chris-wilson.co.uk

We would like to be able to exercise huge allocations even on memory
constrained devices. To do this we create an object that allocates only
a few pages and remaps them across its whole range - each page is reused
multiple times. We can therefore pretend we are rendering into a much
larger object.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NMatthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170213171558.20942-9-chris@chris-wilson.co.uk

A simulacrum of drm_i915_private to let us pretend interactions with the
device.

v2: Tidy init error paths
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NMatthew Auld <matthew.auld@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170213171558.20942-6-chris@chris-wilson.co.uk

Start exercising the scattergather lists, especially looking at
iteration after coalescing.

v2: Comment on the peculiarity of table construction (i.e. why this
sg_table might be interesting).
v3: Added one __func__ to identify expect_pfn_sg()
v4: Loop until we have crossed the chain boundary (forcing sg_table to
do multiple allocations) before squelching a potential ENOMEM from oom.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170213171558.20942-2-chris@chris-wilson.co.uk

Following a hang and reset, we know that the engine is idle and all
context state has been saved or lost. Consequently, we know that the
engine is no longer referencing the last context and we can relinquish
our tracking.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Tvrtko Ursulin <tursulin@ursulin.net>
Link: http://patchwork.freedesktop.org/patch/msgid/20170212172002.23072-5-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>

The signal threads may be running concurrently with the GPU reset. The
completion from the GPU run asynchronous with the reset and two threads
may see different snapshots of the state, and the signaler may mark a
request as complete as we try to reset it. We don't tolerate 2 different
views of the same state and complain if we try to mark a request as
failed if it is already complete. Disable the signal threads during
reset to prevent this conflict (even though the conflict implies that
the state we resetting to is invalid, we have already made our
decision!).

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=99733
References: https://bugs.freedesktop.org/show_bug.cgi?id=99671Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170212172002.23072-4-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>

Disabling the tasklet leaves it if scheduled on the ready to run list
until it is re-enabled. This will leave the ksoftird thread spinning
until satisfied. To prevent this situation on starting the GPU reset, we
want to kill the tasklet first and then disable. The same problem will
arise when a tasklet is scheduled from another device, so a better
solution is required for the general case.
Reported-by: NTvrtko Ursulin <tvrtko.ursulin@intel.com>
Fixes: 1f7b847d ("drm/i915: Disable engine->irq_tasklet around resets")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170212172002.23072-3-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>

As the request is not complete, it should not be signaled. Assert that
this is true before we process the request for a reset.

References: https://bugs.freedesktop.org/show_bug.cgi?id=99671Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170212172002.23072-1-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>

When updating the bookkeeping following the reset, we need the seqno to
be coherent on the CPU prior to trusting its result for deciding whether
any request is completed. We need the irq_barrier before we start making
these decisions, i.e. in reset_prepare.

References: https://bugs.freedesktop.org/show_bug.cgi?id=99733Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170210185214.23463-1-chris@chris-wilson.co.ukReviewed-by: NMichel Thierry <michel.thierry@intel.com>

As dmabufs may live beyond the PCI device removal, we need to flush the
freed object worker on device release, and include a warning in case
there is a leak.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170210163523.17533-3-chris@chris-wilson.co.uk

When we restart the engines, and we have active requests, a request on
the first engine may complete and queue a request to the second engine
before we try to restart the second engine. That queueing of the
request may race with the engine to restart, and so may corrupt the
current state. Disabling the engine->irq_tasklet prevents the two paths
from writing into ELSP simultaneously (and modifyin the execlists_port[]
at the same time).

Fixes: 821ed7df ("drm/i915: Update reset path to fix incomplete requests")
Testcase: igt/gem_exec_fence/await-hang
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170208143033.11651-3-chris@chris-wilson.co.uk

Currently we do a reset prepare/finish around the call to reset the GPU,
but it looks like we need a later stage after the hw has been
reinitialised to allow GEM to restart itself. Start by splitting the 2
GEM phases into 3:

  prepare - before the reset, check if GEM recovered, then stop GEM

  reset - after the reset, update GEM bookkeeping

  finish - after the re-initialisation following the reset, restart GEM
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170208143033.11651-2-chris@chris-wilson.co.uk

Just a simple refactor to move a loop and some support code out of
i915_gem_init_hw(). This is in preparation for avoiding a race between
the tasklet writing to ELSP whilst simultaneously being written by
engine->init_hw() following a GPU reset.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Reviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170208143033.11651-1-chris@chris-wilson.co.uk

Since to unbind an object, we may need a powered up device to access the
GTT entries, we only shrink bound objects if awake. Callers to
i915_gem_shrink_all() had to take this into account and take the rpm
wakeref, but we can move this wakeref into the shrink_all itself for
convenience and making the function live up to its name.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170208104710.18089-1-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>

The goal of the WARN was to catch when we are still actively using the
fence as we go into the runtime suspend. However, the reg->pin_count is
too coarse as it does not distinguish between exclusive ownership of the
fence register from activity.

I've not improved on the WARN, nor have we captured this WARN in an
exact igt, but it is showing up regularly in the wild:

[ 1915.935332] WARNING: CPU: 1 PID: 10861 at drivers/gpu/drm/i915/i915_gem.c:2022 i915_gem_runtime_suspend+0x116/0x130 [i915]
[ 1915.935383] WARN_ON(reg->pin_count)[ 1915.935399] Modules linked in:
 snd_hda_intel i915 drm_kms_helper vgem netconsole scsi_transport_iscsi fuse vfat fat x86_pkg_temp_thermal coretemp intel_cstate intel_uncore snd_hda_codec_hdmi snd_hda_codec_generic snd_hda_codec snd_hwdep snd_hda_core snd_pcm snd_timer snd mei_me mei serio_raw intel_rapl_perf intel_pch_thermal soundcore wmi acpi_pad i2c_algo_bit syscopyarea sysfillrect sysimgblt fb_sys_fops drm r8169 mii video [last unloaded: drm_kms_helper]
[ 1915.935785] CPU: 1 PID: 10861 Comm: kworker/1:0 Tainted: G     U  W       4.9.0-rc5+ #170
[ 1915.935799] Hardware name: LENOVO 80MX/Lenovo E31-80, BIOS DCCN34WW(V2.03) 12/01/2015
[ 1915.935822] Workqueue: pm pm_runtime_work
[ 1915.935845]  ffffc900044fbbf0 ffffffffac3220bc ffffc900044fbc40 0000000000000000
[ 1915.935890]  ffffc900044fbc30 ffffffffac059bcb 000007e6044fbc60 ffff8801626e3198
[ 1915.935937]  ffff8801626e0000 0000000000000002 ffffffffc05e5d4e 0000000000000000
[ 1915.935985] Call Trace:
[ 1915.936013]  [<ffffffffac3220bc>] dump_stack+0x4f/0x73
[ 1915.936038]  [<ffffffffac059bcb>] __warn+0xcb/0xf0
[ 1915.936060]  [<ffffffffac059c4f>] warn_slowpath_fmt+0x5f/0x80
[ 1915.936158]  [<ffffffffc052d916>] i915_gem_runtime_suspend+0x116/0x130 [i915]
[ 1915.936251]  [<ffffffffc04f1c74>] intel_runtime_suspend+0x64/0x280 [i915]
[ 1915.936277]  [<ffffffffac0926f1>] ? dequeue_entity+0x241/0xbc0
[ 1915.936298]  [<ffffffffac36bb85>] pci_pm_runtime_suspend+0x55/0x180
[ 1915.936317]  [<ffffffffac36bb30>] ? pci_pm_runtime_resume+0xa0/0xa0
[ 1915.936339]  [<ffffffffac4514e2>] __rpm_callback+0x32/0x70
[ 1915.936356]  [<ffffffffac451544>] rpm_callback+0x24/0x80
[ 1915.936375]  [<ffffffffac36bb30>] ? pci_pm_runtime_resume+0xa0/0xa0
[ 1915.936392]  [<ffffffffac45222d>] rpm_suspend+0x12d/0x680
[ 1915.936415]  [<ffffffffac69f6d7>] ? _raw_spin_unlock_irq+0x17/0x30
[ 1915.936435]  [<ffffffffac0810b8>] ? finish_task_switch+0x88/0x220
[ 1915.936455]  [<ffffffffac4534bf>] pm_runtime_work+0x6f/0xb0
[ 1915.936477]  [<ffffffffac074353>] process_one_work+0x1f3/0x4d0
[ 1915.936501]  [<ffffffffac074678>] worker_thread+0x48/0x4e0
[ 1915.936523]  [<ffffffffac074630>] ? process_one_work+0x4d0/0x4d0
[ 1915.936542]  [<ffffffffac074630>] ? process_one_work+0x4d0/0x4d0
[ 1915.936559]  [<ffffffffac07a2c9>] kthread+0xd9/0xf0
[ 1915.936580]  [<ffffffffac07a1f0>] ? kthread_park+0x60/0x60
[ 1915.936600]  [<ffffffffac69fe62>] ret_from_fork+0x22/0x30

In the case the register is pinned, it should be present and we will
need to invalidate them to be restored upon resume as we cannot expect
the owner of the pin to call get_fence prior to use after resume.

Fixes: 7c108fd8 ("drm/i915: Move fence cancellation to runtime suspend")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98804Reported-by: NLionel Landwerlin <lionel.g.landwerlin@linux.intel.com>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Imre Deak <imre.deak@linux.intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.10-rc1+
Link: http://patchwork.freedesktop.org/patch/msgid/20170203125717.8431-1-chris@chris-wilson.co.ukReviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
(cherry picked from commit e0ec3ec6)
Signed-off-by: NJani Nikula <jani.nikula@intel.com>

On a non-llc system, the objects are created with .cache_level =
CACHE_NONE and so the transition to uncached for scanout is a no-op.
However, if the object was never written to, it will still be in the CPU
domain (having been zeroed out by shmemfs). Those cachelines need to be
flushed prior to display.

Reported-and-tested-by: Vito Caputo
Fixes: a6a7cc4b ("drm/i915: Always flush the dirty CPU cache when pinning the scanout")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.10-rc1+
Link: http://patchwork.freedesktop.org/patch/msgid/20170109111932.6342-1-chris@chris-wilson.co.ukReviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
(cherry picked from commit 69aeafea)
Signed-off-by: NJani Nikula <jani.nikula@intel.com>

Following a reset, the context and page directory registers are lost.
However, the queue of requests that we resubmit after the reset may
depend upon them - the registers are restored from a context image, but
that restore may be inhibited and may simply be absent from the request
if it was in the middle of a sequence using the same context. If we
prime the CCID/PD registers with the first request in the queue (even
for the hung request), we prevent invalid memory access for the
following requests (and continually hung engines).

v2: Magic BIT(8), reserved for future use but still appears unused.
v3: Some commentary on handling innocent vs guilty requests
v4: Add a wait for PD_BASE fetch. The reload appears to be instant on my
Ivybridge, but this bit probably exists for a reason.

Fixes: 821ed7df ("drm/i915: Update reset path to fix incomplete requests")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170207152437.4252-1-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>

The goal of the WARN was to catch when we are still actively using the
fence as we go into the runtime suspend. However, the reg->pin_count is
too coarse as it does not distinguish between exclusive ownership of the
fence register from activity.

I've not improved on the WARN, nor have we captured this WARN in an
exact igt, but it is showing up regularly in the wild:

[ 1915.935332] WARNING: CPU: 1 PID: 10861 at drivers/gpu/drm/i915/i915_gem.c:2022 i915_gem_runtime_suspend+0x116/0x130 [i915]
[ 1915.935383] WARN_ON(reg->pin_count)[ 1915.935399] Modules linked in:
 snd_hda_intel i915 drm_kms_helper vgem netconsole scsi_transport_iscsi fuse vfat fat x86_pkg_temp_thermal coretemp intel_cstate intel_uncore snd_hda_codec_hdmi snd_hda_codec_generic snd_hda_codec snd_hwdep snd_hda_core snd_pcm snd_timer snd mei_me mei serio_raw intel_rapl_perf intel_pch_thermal soundcore wmi acpi_pad i2c_algo_bit syscopyarea sysfillrect sysimgblt fb_sys_fops drm r8169 mii video [last unloaded: drm_kms_helper]
[ 1915.935785] CPU: 1 PID: 10861 Comm: kworker/1:0 Tainted: G     U  W       4.9.0-rc5+ #170
[ 1915.935799] Hardware name: LENOVO 80MX/Lenovo E31-80, BIOS DCCN34WW(V2.03) 12/01/2015
[ 1915.935822] Workqueue: pm pm_runtime_work
[ 1915.935845]  ffffc900044fbbf0 ffffffffac3220bc ffffc900044fbc40 0000000000000000
[ 1915.935890]  ffffc900044fbc30 ffffffffac059bcb 000007e6044fbc60 ffff8801626e3198
[ 1915.935937]  ffff8801626e0000 0000000000000002 ffffffffc05e5d4e 0000000000000000
[ 1915.935985] Call Trace:
[ 1915.936013]  [<ffffffffac3220bc>] dump_stack+0x4f/0x73
[ 1915.936038]  [<ffffffffac059bcb>] __warn+0xcb/0xf0
[ 1915.936060]  [<ffffffffac059c4f>] warn_slowpath_fmt+0x5f/0x80
[ 1915.936158]  [<ffffffffc052d916>] i915_gem_runtime_suspend+0x116/0x130 [i915]
[ 1915.936251]  [<ffffffffc04f1c74>] intel_runtime_suspend+0x64/0x280 [i915]
[ 1915.936277]  [<ffffffffac0926f1>] ? dequeue_entity+0x241/0xbc0
[ 1915.936298]  [<ffffffffac36bb85>] pci_pm_runtime_suspend+0x55/0x180
[ 1915.936317]  [<ffffffffac36bb30>] ? pci_pm_runtime_resume+0xa0/0xa0
[ 1915.936339]  [<ffffffffac4514e2>] __rpm_callback+0x32/0x70
[ 1915.936356]  [<ffffffffac451544>] rpm_callback+0x24/0x80
[ 1915.936375]  [<ffffffffac36bb30>] ? pci_pm_runtime_resume+0xa0/0xa0
[ 1915.936392]  [<ffffffffac45222d>] rpm_suspend+0x12d/0x680
[ 1915.936415]  [<ffffffffac69f6d7>] ? _raw_spin_unlock_irq+0x17/0x30
[ 1915.936435]  [<ffffffffac0810b8>] ? finish_task_switch+0x88/0x220
[ 1915.936455]  [<ffffffffac4534bf>] pm_runtime_work+0x6f/0xb0
[ 1915.936477]  [<ffffffffac074353>] process_one_work+0x1f3/0x4d0
[ 1915.936501]  [<ffffffffac074678>] worker_thread+0x48/0x4e0
[ 1915.936523]  [<ffffffffac074630>] ? process_one_work+0x4d0/0x4d0
[ 1915.936542]  [<ffffffffac074630>] ? process_one_work+0x4d0/0x4d0
[ 1915.936559]  [<ffffffffac07a2c9>] kthread+0xd9/0xf0
[ 1915.936580]  [<ffffffffac07a1f0>] ? kthread_park+0x60/0x60
[ 1915.936600]  [<ffffffffac69fe62>] ret_from_fork+0x22/0x30

In the case the register is pinned, it should be present and we will
need to invalidate them to be restored upon resume as we cannot expect
the owner of the pin to call get_fence prior to use after resume.

Fixes: 7c108fd8 ("drm/i915: Move fence cancellation to runtime suspend")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=98804Reported-by: NLionel Landwerlin <lionel.g.landwerlin@linux.intel.com>
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Imre Deak <imre.deak@linux.intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.10-rc1+
Link: http://patchwork.freedesktop.org/patch/msgid/20170203125717.8431-1-chris@chris-wilson.co.ukReviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>

The drm_mm range manager claimed to support top-down insertion, but it
was neither searching for the top-most hole that could fit the
allocation request nor fitting the request to the hole correctly.

In order to search the range efficiently, we create a secondary index
for the holes using either their size or their address. This index
allows us to find the smallest hole or the hole at the bottom or top of
the range efficiently, whilst keeping the hole stack to rapidly service
evictions.

v2: Search for holes both high and low. Rename flags to mode.
v3: Discover rb_entry_safe() and use it!
v4: Kerneldoc for enum drm_mm_insert_mode.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Cc: "Christian König" <christian.koenig@amd.com>
Cc: David Airlie <airlied@linux.ie>
Cc: Russell King <rmk+kernel@armlinux.org.uk>
Cc: Daniel Vetter <daniel.vetter@intel.com>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Cc: Sean Paul <seanpaul@chromium.org>
Cc: Lucas Stach <l.stach@pengutronix.de>
Cc: Christian Gmeiner <christian.gmeiner@gmail.com>
Cc: Rob Clark <robdclark@gmail.com>
Cc: Thierry Reding <thierry.reding@gmail.com>
Cc: Stephen Warren <swarren@wwwdotorg.org>
Cc: Alexandre Courbot <gnurou@gmail.com>
Cc: Eric Anholt <eric@anholt.net>
Cc: Sinclair Yeh <syeh@vmware.com>
Cc: Thomas Hellstrom <thellstrom@vmware.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Reviewed-by: Sinclair Yeh <syeh@vmware.com> # vmwgfx
Reviewed-by: Lucas Stach <l.stach@pengutronix.de> #etnaviv
Signed-off-by: NDaniel Vetter <daniel.vetter@ffwll.ch>
Link: http://patchwork.freedesktop.org/patch/msgid/20170202210438.28702-1-chris@chris-wilson.co.uk

On a non-llc system, the objects are created with .cache_level =
CACHE_NONE and so the transition to uncached for scanout is a no-op.
However, if the object was never written to, it will still be in the CPU
domain (having been zeroed out by shmemfs). Those cachelines need to be
flushed prior to display.

Reported-and-tested-by: Vito Caputo
Fixes: a6a7cc4b ("drm/i915: Always flush the dirty CPU cache when pinning the scanout")
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Cc: <drm-intel-fixes@lists.freedesktop.org> # v4.10-rc1+
Link: http://patchwork.freedesktop.org/patch/msgid/20170109111932.6342-1-chris@chris-wilson.co.ukReviewed-by: NDaniel Vetter <daniel.vetter@ffwll.ch>

The GPU may be in an unknown state following resume and module load. The
previous occupant may have left contexts loaded, or other dangerous
state, which can cause an immediate GPU hang for us. The only save
course of action is to reset the GPU prior to using it - similarly to
how we reset the GPU prior to unload (before a second user may be
affected by our leftover state).

We need to reset the GPU very early in our load/resume sequence so that
any stale HW pointers are revoked prior to any resource allocations we
make (that may conflict).

A reset should only be a couple of milliseconds on a slow device, a cost
we should easily be able to absorb into our initialisation times.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>
Link: http://patchwork.freedesktop.org/patch/msgid/20170124110135.6418-2-chris@chris-wilson.co.uk

When pinning into the global GTT, an error from creating the VMA is
unlikely, so mark it so.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170119192659.31789-4-chris@chris-wilson.co.ukReviewed-by: NJoonas Lahtinen <joonas.lahtinen@linux.intel.com>

Now that i915_gem_object_bump_inactive_ggtt() exists, also make use of
it for the LRU bumping from i915_gem_object_unpin_from_display()
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170119192659.31789-2-chris@chris-wilson.co.ukReviewed-by: NJonas Lahtinen <joonas.lahtinen@linux.intel.com>

Since a change in cache level is likely to trigger an unbind, avoid
waiting under the mutex by preemptively doing an unlocked wait.
Signed-off-by: NChris Wilson <chris@chris-wilson.co.uk>
Link: http://patchwork.freedesktop.org/patch/msgid/20170119082211.21257-1-chris@chris-wilson.co.ukReviewed-by: NMika Kuoppala <mika.kuoppala@intel.com>