drm/i915: Fix ICL output CSC programming

When I refactored the code into its own function I accidentally
misplaced the <<16 shifts for some of the registers causing us
to lose the blue channel entirely.

We should really find a way to test this...

Cc: Uma Shankar <uma.shankar@intel.com>
Fixes: d2c19b06d6ea ("drm/i915: Clean up ilk/icl pipe/output CSC programming")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190425192419.24931-1-ville.syrjala@linux.intel.com
Reviewed-by: Swati Sharma <swati2.sharma@intel.com>

drm/i915: Enable render context support for gen4 (Broadwater to Cantiga)

Broadwater and the rest of gen4  do support being able to saving and
reloading context specific registers between contexts, providing isolation
of the basic GPU state (as programmable by userspace). This allows
userspace to assume that the GPU retains their state from one batch to the
next, minimising the amount of state it needs to reload and manually save
across batches.

v2: CONSTANT_BUFFER woes

Running through piglit turned up an interesting issue, a GPU hang inside
the context load. The context image includes the CONSTANT_BUFFER command
that loads an address into a on-gpu buffer, and the context load was
executing that immediately. However, since it was reading from the GTT
there is no guarantee that the GTT retains the same configuration as
when the context was saved, resulting in stray reads and a GPU hang.

Having tried issuing a CONSTANT_BUFFER (to disable the command) from the
ring before saving the context to no avail, we resort to patching out
the instruction inside the context image before loading.

This does impose that gen4 always reissues CONSTANT_BUFFER commands on
each batch, but due to the use of a shared GTT that was and will remain
a requirement.

v3: ECOSKPD to the rescue

Ville found the magic bit in the ECOSKPD to disable saving and restoring
the CONSTANT_BUFFER from the context image, thereby completely avoiding
the GPU hangs from chasing invalid pointers. This appears to be the
default behaviour for gen5, and so we just need to tweak gen4 to match.

v4: Fix spelling of ECOSKPD and discover it already exists

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Kenneth Graunke <kenneth@whitecape.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419172720.5462-1-chris@chris-wilson.co.uk

drm/i915: Enable render context support for Ironlake (gen5)

Ironlake does support being able to saving and reloading context specific
registers between contexts, providing isolation of the basic GPU state
(as programmable by userspace). This allows userspace to assume that the
GPU retains their state from one batch to the next, minimising the
amount of state it needs to reload, or manually save and restore.

v2: Fix off-by-one in reading CXT_SIZE, and add a comment that the
CXT_SIZE and context-layout do not match in bspec, but the difference is
irrelevant as we overallocate the full page anyway (Ville).

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Kenneth Graunke <kenneth@whitecape.org>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419111749.3910-2-chris@chris-wilson.co.uk

drm/i915/ringbuffer: EMIT_INVALIDATE *before* switch context

Despite what I think the prm recommends, commit f2253bd9859b
("drm/i915/ringbuffer: EMIT_INVALIDATE after switch context") turned out
to be a huge mistake when enabling Ironlake contexts as the GPU would
hang on either a MI_FLUSH or PIPE_CONTROL immediately following the
MI_SET_CONTEXT of an active mesa context (more vanilla contexts, e.g.
simple rendercopies with igt, do not suffer).

Ville found the following clue,

  "[DevCTG+]: For the invalidate operation of the pipe control, the
   following pointers are affected. The
   invalidate operation affects the restore of these packets. If the pipe
   control invalidate operation is completed
   before the context save, the indirect pointers will not be restored from
   memory.
   1. Pipeline State Pointer
   2. Media State Pointer
   3. Constant Buffer Packet"

which suggests by us emitting the INVALIDATE prior to the MI_SET_CONTEXT,
we prevent the context-restore from chasing the dangling pointers within
the image, and explains why this likely prevents the GPU hang.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419111749.3910-1-chris@chris-wilson.co.uk

drm/i915: Move sandybride pcode access to intel_sideband.c

sandybride_pcode is another sideband, so move it to their new home.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-8-chris@chris-wilson.co.uk

drm/i915: Merge sandybridge_pcode_(read|write)

These routines are identical except in the nature of the value parameter.
For writes it is a pure in-param, but for a read, we need an out-param.
Since they differ in a single line, merge the two routines into one.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-7-chris@chris-wilson.co.uk

drm/i915: Merge sbi read/write into a single accessor

Since intel_sideband_read and intel_sideband_write differ by only a
couple of lines (depending on whether we feed the value in or out),
merge the two into a single common accessor.

v2: Restore vlv_flisdsi_read() lost during rebasing.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-6-chris@chris-wilson.co.uk

drm/i915: Separate sideband declarations to intel_sideband.h

Split the sideback declarations out of the ginormous i915_drv.h

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-5-chris@chris-wilson.co.uk

drm/i915: Replace pcu_lock with sb_lock

We now have two locks for sideband access. The general one covering
sideband access across all generation, sb_lock, and a specific one
covering sideband access via the punit on vlv/chv. After lifting the
sb_lock around the punit into the callers, the pcu_lock is now redudant
and can be separated from its other use to regulate RPS (essentially
giving RPS a lock all of its own).

v2: Extract a couple of minor bug fixes.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-4-chris@chris-wilson.co.uk

drm/i915: Lift sideband locking for vlv_punit_(read|write)

Lift the sideband acquisition for vlv_punit_read and vlv_punit_write
into their callers, so that we can lock the sideband once for a sequence
of operations, rather than perform the heavyweight acquisition on each
request.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-3-chris@chris-wilson.co.uk

drm/i915: Lift acquiring the vlv punit magic to a common sb-get

As we now employ a very heavy pm_qos around the punit access, we want to
minimise the number of synchronous requests by performing one for the
whole punit sequence rather than around individual accesses. The
sideband lock is used for this, so push the pm_qos into the sideband
lock acquisition and release, moving it from the lowlevel punit rw
routine to the callers. In the first step, we move the punit magic into
the common sideband lock so that we can acquire a bunch of ports
simultaneously, and if need be extend the workaround protection later.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-2-chris@chris-wilson.co.uk

drm/i915: Disable preemption and sleeping while using the punit sideband

While we talk to the punit over its sideband, we need to prevent the cpu
from sleeping in order to prevent a potential machine hang.

Note that by itself, it appears that pm_qos_update_request (via
intel_idle) doesn't provide a sufficient barrier to ensure that all core
are indeed awake (out of Cstate) and that the package is awake. To do so,
we need to supplement the pm_qos with a manual ping on_each_cpu.

v2: Restrict the heavy-weight wakeup to just the ISOF_PORT_PUNIT, there
is insufficient evidence to implicate a wider problem atm. Similarly,
restrict the w/a to Valleyview, as Cherryview doesn't have an angry cadre
of users.

The working theory, courtesy of Ville and Hans, is the issue lies within
the power delivery and so is likely to be unit and board specific and
occurs when both the unit/fw require extra power at the same time as the
cpu package is changing its own power state.

References: https://bugzilla.kernel.org/show_bug.cgi?id=109051
References: https://bugs.freedesktop.org/show_bug.cgi?id=102657
References: https://bugzilla.kernel.org/show_bug.cgi?id=195255
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Hans de Goede <hdegoede@redhat.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190426081725.31217-1-chris@chris-wilson.co.uk

drm/i915: Allow multiple user handles to the same VM

It was noted that we made the same mistake for VM_ID as for object
handles, whereby we ensured that we only allocated a single handle for
one ppgtt. This has the unfortunate consequence for userspace that they
need to reference count the handles to avoid destroying an active ID. If
we allow multiple handles to the same ppgtt, userspace can freely
unreference any handle they own without fear of destroying the same
handle in use elsewhere.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190425054333.27299-1-chris@chris-wilson.co.uk

drm/i915: Explicitly pin the logical context for execbuf

In order to separate the reservation phase of building a request from
its emission phase, we need to pull some of the request alloc activities
from deep inside i915_request to the surface, GEM_EXECBUFFER.

v2: Be frivolous, use a local drm_i915_private.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190425050143.811-1-chris@chris-wilson.co.uk

drm/i915: Invert the GEM wakeref hierarchy

In the current scheme, on submitting a request we take a single global
GEM wakeref, which trickles down to wake up all GT power domains. This
is undesirable as we would like to be able to localise our power
management to the available power domains and to remove the global GEM
operations from the heart of the driver. (The intent there is to push
global GEM decisions to the boundary as used by the GEM user interface.)

Now during request construction, each request is responsible via its
logical context to acquire a wakeref on each power domain it intends to
utilize. Currently, each request takes a wakeref on the engine(s) and
the engines themselves take a chipset wakeref. This gives us a
transition on each engine which we can extend if we want to insert more
powermangement control (such as soft rc6). The global GEM operations
that currently require a struct_mutex are reduced to listening to pm
events from the chipset GT wakeref. As we reduce the struct_mutex
requirement, these listeners should evaporate.

Perhaps the biggest immediate change is that this removes the
struct_mutex requirement around GT power management, allowing us greater
flexibility in request construction. Another important knock-on effect,
is that by tracking engine usage, we can insert a switch back to the
kernel context on that engine immediately, avoiding any extra delay or
inserting global synchronisation barriers. This makes tracking when an
engine and its associated contexts are idle much easier -- important for
when we forgo our assumed execution ordering and need idle barriers to
unpin used contexts. In the process, it means we remove a large chunk of
code whose only purpose was to switch back to the kernel context.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424200717.1686-5-chris@chris-wilson.co.uk

drm/i915: Pass intel_context to i915_request_create()

Start acquiring the logical intel_context and using that as our primary
means for request allocation. This is the initial step to allow us to
avoid requiring struct_mutex for request allocation along the
perma-pinned kernel context, but it also provides a foundation for
breaking up the complex request allocation to handle different scenarios
inside execbuf.

For the purpose of emitting a request from inside retirement (see the
next patch for engine power management), we also need to lift control
over the timeline mutex to the caller.

v2: Note that the request carries the active reference upon construction.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424200717.1686-4-chris@chris-wilson.co.uk

drm/i915: Introduce context->enter() and context->exit()

We wish to start segregating the power management into different control
domains, both with respect to the hardware and the user interface. The
first step is that at the lowest level flow of requests, we want to
process a context event (and not a global GEM operation). In this patch,
we introduce the context callbacks that in future patches will be
redirected to per-engine interfaces leading to global operations as
required.

The intent is that this will be guarded by the timeline->mutex, except
that retiring has not quite finished transitioning over from being
guarded by struct_mutex. So at the moment it is protected by
struct_mutex with a reminded to switch.

v2: Rename default handlers to intel_context_enter_engine.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424200717.1686-3-chris@chris-wilson.co.uk

drm/i915: Pull the GEM powermangement coupling into its own file

Split out the powermanagement portion (GT wakeref, suspend/resume) of
GEM from i915_gem.c into its own file.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424200717.1686-2-chris@chris-wilson.co.uk

drm/i915: Introduce struct intel_wakeref

For controlling runtime pm of the GT and engines, we would like to have
a callback to do extra work the first time we wake up and the last time
we drop the wakeref. This first/last access needs serialisation and so
we encompass a mutex with the regular intel_wakeref_t tracker.

v2: Drop the _once naming and report the errors.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc; Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424200717.1686-1-chris@chris-wilson.co.uk

drm/i915: Move GraphicsTechnology files under gt/

Start partitioning off the code that talks to the hardware (GT) from the
uapi layers and move the device facing code under gt/

One casualty is s/intel_ringbuffer.h/intel_engine.h/ with the plan to
subdivide that header and body further (and split out the submission
code from the ringbuffer and logical context handling). This patch aims
to be simple motion so git can fixup inflight patches with little mess.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Acked-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Acked-by: Jani Nikula <jani.nikula@intel.com>
Acked-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424174839.7141-1-chris@chris-wilson.co.uk

drm/i915/selftests: Verify whitelist of context registers

The RING_NONPRIV allows us to add registers to a whitelist that allows
userspace to modify them. Ideally such registers should be safe and
saved within the context such that they do not impact system behaviour
for other users. This selftest verifies that those registers we do add
are (a) then writable by userspace and (b) only affect a single client.

Opens:
- Is GEN9_SLICE_COMMON_ECO_CHICKEN1 really write-only?

v2: Remove the blatant copy-paste.
v3: Emulate userspace register writes via the batch again.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424110941.9869-1-chris@chris-wilson.co.uk

drm/i915: Store the default sseu setup on the engine

As we push for better compartmentalisation, it is more convenient to
copy the default sseu configuration from the engine into the derived
logical context, than it is to dig it out from i915->runtime_info.

v2: Use intel_sseu_from_device_info() to describe the converter

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190424095134.30249-1-chris@chris-wilson.co.uk

drm/i915/icl: Fix clockgating issue when using scalers

Fixes the clock-gating issue when pipe scaling is enabled.
(Lineage #2006604312)

V2: Fix typo in headline(Chris)
    Handle the non double buffered nature of the register(Ville)
V3: Fix checkpatch warning. BAT failure for V2 on gen3 looks unrelated.
V4: Split the icl and skl wa's(Ville)
V5: Split the checks for icl and skl(Ville)
V6: Correct the flipped checks in intel_pre_plane_update(Ville)
V7: Use enum for pipe and extend the WA for plane scalers(Ville)
V8: Eliminate the redundant use of pch_pfit(Ville)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Ville Syrjala <ville.syrjala@linux.intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Clint Taylor <clinton.a.taylor@intel.com>
Cc: Aditya Swarup <aditya.swarup@intel.com>
Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190417185901.14833-1-radhakrishna.sripada@intel.com

drm/i915: Fix skl+ max plane width

The spec has changed since skl_max_plane_width() was written.
Now the SKL limits are lower than what they were initially, and
GLK and ICL have different limits. Update the code to match the
spec.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190418195907.23912-1-ville.syrjala@linux.intel.com
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>

drm/i915/icl: Fix MG_DP_MODE() register programming

Fix the order of lane, port parameters passed to the register macro.

Note that this was already partly fixed by commit
37fc7845df7b6 ("drm/i915: Call MG_DP_MODE() macro with the right parameters order")

While at it simplify things by using the macro directly instead of an
unnecessary redirection via an array.

v2:
- Add a note the commit message about simplifying things. (José)

Fixes: 58106b7d816e1 ("drm/i915: Make MG PHY macros semantically consistent")
Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Aditya Swarup <aditya.swarup@intel.com>
Signed-off-by: Imre Deak <imre.deak@intel.com>
Reviewed-by: Lucas De Marchi <lucas.demarchi@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419071026.32370-1-imre.deak@intel.com

drm/i915: Start writeback from the shrinker

When we are called to relieve mempressue via the shrinker, the only way
we can make progress is either by discarding unwanted pages (those
objects that userspace has marked MADV_DONTNEED) or by reclaiming the
dirty objects via swap. As we know that is the only way to make further
progress, we can initiate the writeback as we invalidate the objects.
This means the objects we put onto the inactive anon lru list are
already marked for reclaim+writeback and so will trigger a wait upon the
writeback inside direct reclaim, greatly improving the success rate of
direct reclaim on i915 objects.

The corollary is that we may start a slow swap on opportunistic
mempressure from the likes of the compaction + migration kthreads. This
is limited by those threads only being allowed to shrink idle pages, but
also that if we reactivate the page before it is swapped out by gpu
activity, we only page the cost of repinning the page. The cost is most
felt when an object is reused after mempressure, which hopefully
excludes the latency sensitive tasks (as we are just extending the
impact of swap thrashing to them).

Apparently this is not the first time we've had this idea. Back in
commit 5537252b6b6d ("drm/i915: Invalidate our pages under memory
pressure") we wanted to start writeback but settled on invalidate after
Hugh Dickins warned us about a possibility of a deadlock within shmemfs
if we started writeback from shrink_slab. Looking at the callchain,
using writeback from i915_gem_shrink should be equivalent to the pageout
also employed by shrink_slab, i.e. it should not be any riskier afaict.

v2: Leave mmapings intact. At this point, the only mmapings of our
objects will be via CPU mmaps on the shmemfs filp, which are
out-of-scope for our LRU tracking. Instead leave those pages to the
inactive anon LRU page list for aging and pageout as normal.

v3: Be selective on which paths trigger writeback, in particular
excluding paths shrinking just to reclaim vm space (e.g. mmap, vmap
reapers) and avoid starting writeback on the entire process space from
within the pm freezer.

References: https://bugs.freedesktop.org/show_bug.cgi?id=108686
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Matthew Auld <matthew.auld@intel.com>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Michal Hocko <mhocko@suse.com>
Reviewed-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com> #v1
Link: https://patchwork.freedesktop.org/patch/msgid/20190420115539.29081-1-chris@chris-wilson.co.uk

drm/i915/selftests: Check that gpu reset is usable from atomic context

GPU reset is now available with GuC enabled, so re-enable our check that
this reset is usable from atomic context.

Signed-off-by: Fernando Pacheco <fernando.pacheco@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419230015.18121-6-fernando.pacheco@intel.com

Revert "drm/i915/guc: Disable global reset"

We have now prepared the guc reset paths to avoid taking struct_mutex, or
any other lock, and so it is now safe to re-enable.

References: fe62365f9f80 ("drm/i915/guc: Disable global reset")
Signed-off-by: Fernando Pacheco <fernando.pacheco@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419230015.18121-5-fernando.pacheco@intel.com

drm/i915/uc: Place uC firmware in upper range of GGTT

Currently we pin the GuC or HuC firmware image just before uploading.
Perma-pin during uC initialization instead and use the range reserved at
the top of the address space.

Moving the firmware resulted in needing to:
- use an additional pinning for the rsa signature which will be used
  during HuC auth as addresses above GUC_GGTT_TOP do not map through GTT.

v2: Remove call to set to gtt domain
    Do not restore fw gtt mapping unconditionally
    Separate out pin/unpin functions and drop usage of pin/unpin
    Use uc_fw init/fini functions to bind/unbind fw object

v3: Bind is only needed during xfer (Chris)
    Remove attempts to bind outside of xfer (Chris)
    Mark fw bind/unbind static

Signed-off-by: Fernando Pacheco <fernando.pacheco@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419230015.18121-4-fernando.pacheco@intel.com

drm/i915/uc: Reserve upper range of GGTT

GuC and HuC depend on struct_mutex for device reinitialization. Moving
away from this dependency requires perma-pinning the firmware images in
GGTT.  The upper portion of the GuC address space has a sizeable hole
(several MB) that is inaccessible by GuC. Reserve this range within GGTT
as it can comfortably hold GuC/HuC firmware images.

v2: Reserve node rather than insert (Chris)
    Simpler determination of node start/size (Daniele)
    Move reserve/release out to intel_guc.* files

v3: Reserve starting at GUC_GGTT_TOP only and bail if this
    fails (Chris)

Signed-off-by: Fernando Pacheco <fernando.pacheco@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419230015.18121-3-fernando.pacheco@intel.com

drm/i915/uc: Rename uC firmware init/fini functions

he uC firmware init function is called during GuC/HuC init early phases.
 Rename to include "_early" and properly reflect which phase we are at.

The uC firmware fini function is cleaning up the state set/created on
firmware fetch. Replace "_fini" with "_cleanup_fetch".

v2: also rename uC fw fini function

Signed-off-by: Fernando Pacheco <fernando.pacheco@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419230015.18121-2-fernando.pacheco@intel.com

drm/i915/gtt: Skip clearing the GGTT under gen6+ full-ppgtt

If we know that the user cannot access the GGTT, by virtue of having a
segregated memory area, we can skip clearing the unused entries as they
cannot be accessed.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Matthew Auld <matthew.william.auld@gmail.com>
Reviewed-by: Matthew Auld <matthew.william.auld@gmail.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419201207.5477-1-chris@chris-wilson.co.uk

drm/i915: remove DRM_AUTH from IOCTLs which also have DRM_RENDER_ALLOW

This is to work around problems with libva and vainfo.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190417112525.16848-1-christian.koenig@amd.com

drm/i915: Expose the busyspin durations for i915_wait_request

An interesting discussion regarding "hybrid interrupt polling" for NVMe
came to the conclusion that the ideal busyspin before sleeping was half
of the expected request latency (and better if it was already halfway
through that request). This suggested that we too should look again at
our tradeoff between spinning and waiting. Currently, our spin simply
tries to hide the cost of enabling the interrupt, which is good to avoid
penalising nop requests (i.e. test throughput) and not much else.
Studying real world workloads suggests that a spin of upto 500us can
dramatically boost performance, but the suggestion is that this is not
from avoiding interrupt latency per-se, but from secondary effects of
sleeping such as allowing the CPU reduce cstate and context switch away.

In a truly hybrid interrupt polling scheme, we would aim to sleep until
just before the request completed and then wake up in advance of the
interrupt and do a quick poll to handle completion. This is tricky for
ourselves at the moment as we are not recording request times, and since
we allow preemption, our requests are not on as a nicely ordered
timeline as IO. However, the idea is interesting, for it will certainly
help us decide when busyspinning is worthwhile.

v2: Expose the spin setting via Kconfig options for easier adjustment
and testing.
v3: Don't get caught sneaking in a change to the busyspin parameters.
v4: Explain more about the "hybrid interrupt polling" scheme that we
want to migrate towards.

Suggested-by: Sagar Kamble <sagar.a.kamble@intel.com>
References: http://events.linuxfoundation.org/sites/events/files/slides/lemoal-nvme-polling-vault-2017-final_0.pdf
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Sagar Kamble <sagar.a.kamble@intel.com>
Cc: Eero Tamminen <eero.t.tamminen@intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Ben Widawsky <ben@bwidawsk.net>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Sagar Kamble <sagar.a.kamble@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419182625.11186-1-chris@chris-wilson.co.uk

drm/i915: Track HAS_RPS alongside HAS_RC6 in the device info

For consistency (and elegance!), add intel_device_info.has_rps.
The immediate boon is that RPS support is now emitted along the other
capabilities in the debug log and after errors.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190419134836.5626-1-chris@chris-wilson.co.uk

drm/i915: Use drm_dev_unplug()

The driver does not currently support unbinding from a device which is
in use.  Since open file descriptors may still be pointing into kernel
memory where the device structures used to be, entirely correct kernel
panics protect the driver from being unbound as we should not be
unbinding it before those dangling pointers have been made safe.

According to the documentation found inside drivers/gpu/drm/drm_drv.c,
drm_dev_unplug() should be used instead of drm_dev_unregister() in
order to make a device inaccessible to users as soon as it is unpluged.
Follow that advice to make those possibly dangling pointers safe,
protected by DRM layer from a user who is otherwise left pointing into
possibly reused kernel memory after the driver has been unbound from
the device.  Once done, also cancel inflight operations immediately by
calling i915_gem_set_wedged().

Signed-off-by: Janusz Krzysztofik <janusz.krzysztofik@intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190405130235.7707-2-janusz.krzysztofik@linux.intel.com

drm/i915: Remove unwarranted clamping for hsw/bdw

We always start off at an "efficient frequency" and can let the system
autotune from there, eliminating the need to clamp the available range.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Sagar Arun Kamble <sagar.a.kamble@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190418205358.11450-1-chris@chris-wilson.co.uk

drm/i915: Setup the RCS ring prior to execution

We need to set the various ring registers prior to restarting the
engine, or else we may restart it after reset/resume in an ill-defined
state.

Reported-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190418132720.3716-2-chris@chris-wilson.co.uk

drm/i915: Stop overwriting RING_IMR in rcs resume

We store the engine->imr mask and set up the RING_IMR register on
restarting the engine. We do not then want to overwrite it with
an incomplete mask later as we may then lose interrupts!

Reported-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190418132720.3716-1-chris@chris-wilson.co.uk

drm/i915: Avoid use-after-free in reporting create.size

We have to avoid chasing after a userspace race!

<3>[  473.114328] BUG: KASAN: use-after-free in i915_gem_create+0x1d2/0x1f0 [i915]
<3>[  473.114389] Read of size 8 at addr ffff88815bf1d840 by task gem_flink_race/1541

<4>[  473.114464] CPU: 1 PID: 1541 Comm: gem_flink_race Tainted: G     U            5.1.0-rc4-g7d07e025e786-kasan_88+ #1
<4>[  473.114469] Hardware name: To Be Filled By O.E.M. To Be Filled By O.E.M./J4205-ITX, BIOS P1.10 09/29/2016
<4>[  473.114474] Call Trace:
<4>[  473.114488]  dump_stack+0x7c/0xbb
<4>[  473.114612]  ? i915_gem_create+0x1d2/0x1f0 [i915]
<4>[  473.114621]  print_address_description+0x65/0x270
<4>[  473.114728]  ? i915_gem_create+0x1d2/0x1f0 [i915]
<4>[  473.114839]  ? i915_gem_create+0x1d2/0x1f0 [i915]
<4>[  473.114848]  kasan_report+0x149/0x18d
<4>[  473.114962]  ? i915_gem_create+0x1d2/0x1f0 [i915]
<4>[  473.115069]  i915_gem_create+0x1d2/0x1f0 [i915]
<4>[  473.115176]  ? i915_gem_object_create.part.28+0x4b0/0x4b0 [i915]
<4>[  473.115289]  ? i915_gem_dumb_create+0x1a0/0x1a0 [i915]
<4>[  473.115297]  drm_ioctl_kernel+0x192/0x260
<4>[  473.115306]  ? drm_ioctl_permit+0x280/0x280
<4>[  473.115326]  drm_ioctl+0x67c/0x960
<4>[  473.115438]  ? i915_gem_dumb_create+0x1a0/0x1a0 [i915]
<4>[  473.115448]  ? drm_getstats+0x20/0x20
<4>[  473.115459]  ? __lock_acquire+0xa66/0x3fe0
<4>[  473.115474]  ? _raw_spin_unlock_irqrestore+0x39/0x60
<4>[  473.115485]  ? debug_object_active_state+0x2ea/0x4e0
<4>[  473.115496]  ? debug_show_all_locks+0x2d0/0x2d0
<4>[  473.115513]  do_vfs_ioctl+0x18d/0xfa0
<4>[  473.115522]  ? check_flags.part.27+0x440/0x440
<4>[  473.115532]  ? ioctl_preallocate+0x1a0/0x1a0
<4>[  473.115547]  ? __fget+0x2ac/0x410
<4>[  473.115561]  ? __ia32_sys_dup3+0xb0/0xb0
<4>[  473.115569]  ? rwlock_bug.part.0+0x90/0x90
<4>[  473.115590]  ksys_ioctl+0x35/0x70
<4>[  473.115597]  ? lockdep_hardirqs_off+0x1cb/0x2b0
<4>[  473.115608]  __x64_sys_ioctl+0x6a/0xb0
<4>[  473.115614]  ? lockdep_hardirqs_on+0x342/0x590
<4>[  473.115623]  do_syscall_64+0x97/0x400
<4>[  473.115633]  entry_SYSCALL_64_after_hwframe+0x49/0xbe
<4>[  473.115641] RIP: 0033:0x7fce590d55d7
<4>[  473.115649] Code: b3 66 90 48 8b 05 b1 48 2d 00 64 c7 00 26 00 00 00 48 c7 c0 ff ff ff ff c3 66 2e 0f 1f 84 00 00 00 00 00 b8 10 00 00 00 0f 05 <48> 3d 01 f0 ff ff 73 01 c3 48 8b 0d 81 48 2d 00 f7 d8 64 89 01 48
<4>[  473.115655] RSP: 002b:00007fce4d525ba8 EFLAGS: 00000246 ORIG_RAX: 0000000000000010
<4>[  473.115662] RAX: ffffffffffffffda RBX: 0000000000000000 RCX: 00007fce590d55d7
<4>[  473.115667] RDX: 00007fce4d525c10 RSI: 00000000c010645b RDI: 0000000000000007
<4>[  473.115672] RBP: 00007fce4d525c10 R08: 00007fce4d526700 R09: 00007fce4d526700
<4>[  473.115677] R10: 0000000000000054 R11: 0000000000000246 R12: 00000000c010645b
<4>[  473.115682] R13: 0000000000000007 R14: 0000000000000000 R15: 00007ffe0e4a7450

<3>[  473.115731] Allocated by task 1541:
<4>[  473.115766]  kmem_cache_alloc+0xce/0x290
<4>[  473.115895]  i915_gem_object_create.part.28+0x1c/0x4b0 [i915]
<4>[  473.116000]  i915_gem_create+0xe3/0x1f0 [i915]
<4>[  473.116008]  drm_ioctl_kernel+0x192/0x260
<4>[  473.116013]  drm_ioctl+0x67c/0x960
<4>[  473.116020]  do_vfs_ioctl+0x18d/0xfa0
<4>[  473.116026]  ksys_ioctl+0x35/0x70
<4>[  473.116032]  __x64_sys_ioctl+0x6a/0xb0
<4>[  473.116038]  do_syscall_64+0x97/0x400
<4>[  473.116044]  entry_SYSCALL_64_after_hwframe+0x49/0xbe

<3>[  473.116071] Freed by task 1542:
<4>[  473.116101]  kmem_cache_free+0xb7/0x2f0
<4>[  473.116205]  __i915_gem_free_objects+0x7d4/0xe10 [i915]
<4>[  473.116311]  i915_gem_create_ioctl+0xaa/0xd0 [i915]
<4>[  473.116318]  drm_ioctl_kernel+0x192/0x260
<4>[  473.116323]  drm_ioctl+0x67c/0x960
<4>[  473.116330]  do_vfs_ioctl+0x18d/0xfa0
<4>[  473.116335]  ksys_ioctl+0x35/0x70
<4>[  473.116341]  __x64_sys_ioctl+0x6a/0xb0
<4>[  473.116347]  do_syscall_64+0x97/0x400
<4>[  473.116354]  entry_SYSCALL_64_after_hwframe+0x49/0xbe

Testcase: igt/gem_flink_race/flink_close
Fixes: e163484afa8d ("drm/i915: Update size upon return from GEM_CREATE")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Michał Winiarski <michal.winiarski@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190417132507.27133-1-chris@chris-wilson.co.uk

drm/i915: Make workaround verification *optional*

Sometimes the HW doesn't even play fair, and completely forgets about
register writes. Skip verifying known troublemakers.

References: https://bugs.freedesktop.org/show_bug.cgi?id=108954
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190417075657.19456-4-chris@chris-wilson.co.uk

drm/i915: Verify the engine workarounds stick on application

Read the engine workarounds back using the GPU after loading the initial
context state to verify that we are setting them correctly, and bail if
it fails.

v2: Break out the verification into its own loop

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190417075657.19456-3-chris@chris-wilson.co.uk

drm/i915: Verify workarounds immediately after application

Immediately after writing the workaround, verify that it stuck in the
register.

References: https://bugs.freedesktop.org/show_bug.cgi?id=108954
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190417075657.19456-2-chris@chris-wilson.co.uk

drm/i915: Update DRIVER_DATE to 20190417

Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>

drm/i915/ehl: inherit icl cdclk init/uninit

The cdclk init/uninit code was changed by commit 93a643f29bcb
("drm/i915/cdclk: have only one init/uninit function") between the
versions of commit 39564ae86d51 ("drm/i915/ehl: Inherit Ice Lake
conditional code"). What got merged fails to do cdclk init/uninit on
ehl.

Fixes: 39564ae86d51 ("drm/i915/ehl: Inherit Ice Lake conditional code")
Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Cc: Bob Paauwe <bob.j.paauwe@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Bob Paauwe <bob.j.paauwe@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190416082852.18141-1-jani.nikula@intel.com

drm/i915: Introduce struct class_instance for engines across the uAPI

SSEU reprogramming of the context introduced the notion of engine class
and instance for a forwards compatible method of describing any engine
beyond the old execbuf interface. We wish to adopt this class:instance
description for more interfaces, so pull it out into a separate type for
userspace convenience.

Fixes: e46c2e99f600 ("drm/i915: Expose RPCS (SSEU) configuration to userspace (Gen11 only)")
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>
Cc: Lionel Landwerlin <lionel.g.landwerlin@intel.com>
Cc: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Cc: Tony Ye <tony.ye@intel.com>
Cc: Andi Shyti <andi@etezian.org>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Acked-by: Tony Ye <tony.ye@intel.com>
Reviewed-by: Andi Shyti <andi@etezian.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20190412071416.30097-1-chris@chris-wilson.co.uk

drm/i915: fully convert the IRQ initialization macros to intel_uncore

Make them take the uncore argument from the caller instead of passing
the implicit &dev_priv->uncore directly. This will allow us to finally
pass something that's not dev_priv->uncore in the future, and gets rid
of the implicit variables in register macros.

v2: Rebase on top of the newer patches.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1)
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410235344.31199-6-paulo.r.zanoni@intel.com

drm/i915: convert the IRQ initialization functions to intel_uncore

The IRQ initialization helpers are simple and self-contained. Continue
the transition started in the recent uncore rework to get us rid of
I915_READ/WRITE and the implicit dev_priv variables.

While the implicit dev_priv is removed from the IRQ initialization
helpers, we didn't get rid of them in the macro callers. Doing that
should be very simple now.

v2: Rebase on top of the new patches.

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1)
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410235344.31199-5-paulo.r.zanoni@intel.com

drm/i915: add GEN2_ prefix to the I{E, I, M, S}R registers

This discussion started because we use token pasting in the
GEN{2,3}_IRQ_INIT and GEN{2,3}_IRQ_RESET macros, so gen2-4 passes an
empty argument to those macros, making the code a little weird. The
original proposal was to just add a comment as the empty argument, but
Ville suggested we just add a prefix to the registers, and that indeed
sounds like a more elegant solution.

Now doing this is kinda against our rules for register naming since we
only add gens or platform names as register prefixes when the given
gen/platform changes a register that already existed before. On the
other hand, we have so many instances of IIR/IMR in comments that
adding a prefix would make the users of these register more easily
findable, in addition to make our token pasting macros actually
readable. So IMHO opening an exception here is worth it.

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410235344.31199-4-paulo.r.zanoni@intel.com

drm/i915: don't specify the IRQ register in the gen2 macros

Like the gen3+ macros, the gen2 versions of the IRQ initialization
macros take the register name in the 'type' argument. But gen2 only
has one set of registers, so there's really no need to specify the
type. This commit removes the type argument and uses the registers
directly instead of passing them through variables.

Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410235344.31199-3-paulo.r.zanoni@intel.com

drm/i915: refactor the IRQ init/reset macros

The whole point of having macros here is for the token pasting
necessary to automatically have IMR, IIR and IER selected. We don't
really need or want all the inlining that happens as a consequence.
The good thing about the current code is that it works regardless of
the relative offsets between these registers (they change after gen4,
with the usual VLV/CHV exceptions).

One thing which we can do is to split the logic of what we do with
imr/ier/iir to functions separate from the macros that pick them.
That's what we do in this commit. This allows us to get rid of the
gen8 duplicates and also all the inlining:

add/remove: 2/0 grow/shrink: 0/21 up/down: 384/-5949 (-5565)
Function                                     old     new   delta
gen3_irq_reset                                 -     233    +233
gen3_irq_init                                  -     151    +151
i8xx_irq_postinstall                         459     442     -17
gen11_irq_postinstall                        804     744     -60
ironlake_irq_postinstall                     450     353     -97
vlv_display_irq_postinstall                  348     245    -103
i965_irq_postinstall                         378     272    -106
i915_irq_postinstall                         333     227    -106
gen8_irq_power_well_post_enable              374     240    -134
ironlake_irq_reset                           397     218    -179
vlv_display_irq_reset                        616     433    -183
i965_irq_reset                               374     180    -194
cherryview_irq_reset                         379     185    -194
i915_irq_reset                               407     209    -198
ibx_irq_reset                                332     133    -199
gen5_gt_irq_postinstall                      533     332    -201
gen8_irq_power_well_pre_disable              434     204    -230
gen8_gt_irq_postinstall                      469     196    -273
gen8_de_irq_postinstall                     1200     836    -364
gen5_gt_irq_reset                            471      76    -395
gen8_gt_irq_reset                            775      99    -676
gen8_irq_reset                              1100     333    -767
gen11_irq_reset                             1959     686   -1273
Total: Before=2259222, After=2253657, chg -0.25%

v2:
 - Make checkpatch happy with a temporary which_ (Checkpatch).
 - Reorder the arguments for the INIT macros (Ville).
 - Correctly explain when the register offsets change in the commit
   message (Ville).
 - Use more line breaks in the macro calls to make the arguments look
   a little more organized/readable.
 - Update the bloat-o-meter output (minor change only).

Reviewed-by: Ville Syrjälä <ville.syrjala@linux.intel.com> (v1)
Signed-off-by: Paulo Zanoni <paulo.r.zanoni@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410235344.31199-2-paulo.r.zanoni@intel.com

Merge tag 'gvt-next-2019-04-16' of https://github.com/intel/gvt-linux into drm-intel-next-queued

gvt-next-2019-04-16

- Refine range of MCHBAR snapshot (Yakui)
- Refine out-of-sync page struct (Yakui)
- Remove unused vGPU sreg (Yan)
- Refind MMIO reg names (Xiaolin)
- Proper handling of sync/async flip (Colin)
- Proper handling of PIPE_CONTROL/MI_FLUSH_DW index mode (Xiaolin)
- EXCC reg mask fix (Colin)

Signed-off-by: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
From: Zhenyu Wang <zhenyuw@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190416084814.GH17995@zhen-hp.sh.intel.com

drm/i915: Mark up ips for RCU protection

drivers/gpu/drm/i915/intel_pm.c:8352:9: error: incompatible types in comparison expression (different address spaces)
drivers/gpu/drm/i915/intel_pm.c:8359:9: error: incompatible types in comparison expression (different address spaces)

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190412085410.10392-3-chris@chris-wilson.co.uk

drm/i915: Drop bool return from breadcrumbs signaler

Since removal of the "missed interrupt detection" nobody used the result
of whether or not we signaled anybody during that invocation, so now
remove the return value.

References: 789659f4307a ("drm/i915: Drop fake breadcrumb irq")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190416085218.431-1-chris@chris-wilson.co.uk

drm/i915/gvt: Fix incorrect mask of mmio 0x22028 in gen8/9 mmio list

According to GFX PRM on 01.org, bit 31:16 of mmio 0x22028 should be masks.

Fixes: 178657139307 ("drm/i915/gvt: vGPU context switch")
Reviewed-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Colin Xu <colin.xu@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>

drm/i915/gvt: addressed guest GPU hang with HWS index mode

with the introduce of "switch to use HWS indices rather than address",
guest GPU hang observed when running workloads which will update the
seqno to the real HW HWSP, not vitural GPU HWSP and then cause GPU hang.

this patch is to revoke index mode in PIPE_CTRL and MI_FLUSH_DW and
patch guest GPU HWSP address value to these commands.

Fixes: 54939ea0bd85 ("drm/i915: Switch to use HWS indices rather than addresses")
Reviewed-by: Zhenyu Wang <zhenyuw@linux.intel.com>
Signed-off-by: Xiaolin Zhang <xiaolin.zhang@intel.com>
Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>

Merge tag 'drm-intel-next-2019-04-04' into gvt-next

Merge back drm-intel-next for engine name definition refinement
and 54939ea0bd85 ("drm/i915: Switch to use HWS indices rather than addresses")
that would need gvt fixes to depend on.

Signed-off-by: Zhenyu Wang <zhenyuw@linux.intel.com>

drm/i915: Nuke drm_crtc_state and use intel_atomic_state instead

This is one of the patches to start replacing drm pointers
and use the intel_atomic_state and intel_crtc to derive
the necessary intel state variables required for the intel
modeset functions.

v3:
* Remove the unwanted newline (Ville)
v2:
* Flip the function arguments (Ville)
* Remove some remaining instances of drm pointers (Ville)
* Use old_crtc_state and new_crtc_state (Ville)

Suggested-by: Ville Syrjala <ville.syrjala@linux.intel.com>
Cc: Ville Syrjala <ville.syrjala@linux.intel.com>
Signed-off-by: Manasi Navare <manasi.d.navare@intel.com>
Reviewed-by: Ville Syrjala <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190415182210.13347-1-manasi.d.navare@intel.com

drm/i915/selftests: Skip live timeline/suspend tests if wedged

If the driver is wedged, we can not issue the requests to exercise the
timelines or the system across suspend, so skip the tests. live_hangcheck
is there to fail if we cannot recover.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190413125820.14112-4-chris@chris-wilson.co.uk

drm/i915: Teach intel_workarounds to use uncore mmio access

Start weaning ourselves off the implicit I915_WRITE macro madness and
start using the explicit intel_uncore mmio access.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190412202458.10653-1-chris@chris-wilson.co.uk

drm/i915/ehl: Inherit Ice Lake conditional code

Most of the conditional code for ICELAKE also applies to ELKHARTLAKE
so use IS_GEN(dev_priv, 11) even for PM and Workarounds for now.

v2: - Rename commit (Jose)
    - Include a wm workaround (Jose and Lucas)
    - Include display core init (Jose and Lucas)
v3: Add a missing case of gen greater-than 11 (Jose)

Cc: José Roberto de Souza <jose.souza@intel.com>
Cc: Lucas De Marchi <lucas.demarchi@intel.com>
Signed-off-by: Bob Paauwe <bob.j.paauwe@intel.com>
Signed-off-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190412180920.22347-1-rodrigo.vivi@intel.com

drm/i915: Handle catastrophic error on engine reset

If cat error is set, we need to clear it by acking it. Further,
if it is set, we must not do a normal request for reset.

v2: avoid goto (Chris)
v3: comment, error format, direct assign (Chris)
Bspec: 12567
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190412165353.16432-1-mika.kuoppala@linux.intel.com

drm/i915: Shortcut readiness to reset check

If the engine says it is ready for reset, it is ready
so avoid further dancing and proceed.

v2: reg (Chris)
v3: request, ack, mask from following patch (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190412165335.16347-1-mika.kuoppala@linux.intel.com

drm/i915: Suppress spurious combo PHY B warning

On ICL the DMC doesn't reinit combo PHY B so we should not warn
about its state being bogus during the display core uninit.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190411143349.17934-1-ville.syrjala@linux.intel.com
Reviewed-by: Imre Deak <imre.deak@intel.com>

drm/i915: Restore correct bxt_ddi_phy_calc_lane_lat_optim_mask() calculation

We are no longer calling bxt_ddi_phy_calc_lane_lat_optim_mask() when
intel{hdmi,dp}_compute_config() succeeds, and instead only call it
when those fail. This is fallout from the bool->int
.compute_config() conversion which failed to invert the return
value check before calling bxt_ddi_phy_calc_lane_lat_optim_mask().
Let's just replace it with an early bailout so that it's harder
to miss.

This restores the correct latency optim setting calculation
(which could fix some real failures), and avoids the
MISSING_CASE() from bxt_ddi_phy_calc_lane_lat_optim_mask()
after intel{hdmi,dp}_compute_config() has failed.

Cc: Lyude Paul <lyude@redhat.com>
Fixes: 204474a6b859 ("drm/i915: Pass down rc in intel_encoder->compute_config()")
Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109373
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190411164925.28491-1-ville.syrjala@linux.intel.com
Reviewed-by: Lyude Paul <lyude@redhat.com>

drm/i915: Flush the CSB pointer reset

The HW resets it CSB tail pointer on resetting the engine. Most of the
time. In case it doesn't (and for system resume) we write the expected
value anyway. For extra paranoia, flush the write before we invalidate
the cacheline.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190412110159.10495-1-chris@chris-wilson.co.uk

drm/i915: Fix the inconsistent RMW in WA 827

RMW is used only in the disable path. Using it in enable path
for consistency.

Suggested-by: Ville Syrjala <ville.syrjala@linux.intel.com>
Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190330011921.10397-2-radhakrishna.sripada@intel.com

drm/i915: Rename skl_wa_clkgating to the actual WA

No functional change. Renaming the function to reflect the specific WA.

Suggested-by: Ville Syrjala <ville.syrjala@linux.intel.com>
Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Signed-off-by: Radhakrishna Sripada <radhakrishna.sripada@intel.com>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190330011921.10397-1-radhakrishna.sripada@intel.com

drm/i915: Do not enable FEC without DSC

Currently we enable FEC even when DSC is no used. While that is
theoretically valid supposedly there isn't much of a benefit from
this. But more importantly we do not account for the FEC link
bandwidth overhead (2.4%) in the non-DSC link bandwidth computations.
So the code may think we have enough bandwidth when we in fact
do not.

Cc: stable@vger.kernel.org
Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Cc: Manasi Navare <manasi.d.navare@intel.com>
Fixes: 240999cf339f ("i915/dp/fec: Add fec_enable to the crtc state.")
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190326144903.6617-1-ville.syrjala@linux.intel.com
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>

drm/i915: Avoid reclaim taints from runtime-pm debug

As intel_runtime_pm_get/_put may be called from any blockable context,
we need to avoid allowing reclaim from our mallocs, as we need to
avoid tainting any mutexes held by the callers (as they may themselves
not allow for allocations as they are taken in the shrinker).

<4> [435.339331] WARNING: possible circular locking dependency detected
<4> [435.339364] 5.1.0-rc4-CI-Trybot_4116+ #1 Tainted: G     U
<4> [435.339395] ------------------------------------------------------
<4> [435.339426] gem_caching/1334 is trying to acquire lock:
<4> [435.339456] 000000004505c39b (wakeref#3){+.+.}, at: intel_engine_pm_put+0x1b/0x40 [i915]
<4> [435.339788]
but task is already holding lock:
<4> [435.339819] 00000000ee77b4ed (fs_reclaim){+.+.}, at: fs_reclaim_acquire.part.24+0x0/0x30
<4> [435.339879]
which lock already depends on the new lock.

<4> [435.339918]
the existing dependency chain (in reverse order) is:
<4> [435.339952]
-> #1 (fs_reclaim){+.+.}:
<4> [435.339998]        fs_reclaim_acquire.part.24+0x24/0x30
<4> [435.340035]        kmem_cache_alloc_trace+0x2a/0x290
<4> [435.340311]        __print_intel_runtime_pm_wakeref+0x24/0x160 [i915]
<4> [435.340590]        untrack_intel_runtime_pm_wakeref+0x16e/0x1d0 [i915]
<4> [435.340869]        intel_runtime_pm_put_unchecked+0xd/0x30 [i915]
<4> [435.341147]        __intel_wakeref_put_once+0x22/0x40 [i915]
<4> [435.341508]        i915_request_retire+0x477/0xaf0 [i915]
<4> [435.341871]        ring_retire_requests+0x86/0x160 [i915]
<4> [435.342226]        i915_retire_requests+0x58/0xc0 [i915]
<4> [435.342576]        retire_work_handler+0x5b/0x70 [i915]
<4> [435.342615]        process_one_work+0x245/0x610
<4> [435.342646]        worker_thread+0x37/0x380
<4> [435.342679]        kthread+0x119/0x130
<4> [435.342714]        ret_from_fork+0x3a/0x50
<4> [435.342739]
-> #0 (wakeref#3){+.+.}:
<4> [435.342788]        lock_acquire+0xa6/0x1c0
<4> [435.342822]        __mutex_lock+0x8c/0x960
<4> [435.342853]        atomic_dec_and_mutex_lock+0x33/0x50
<4> [435.343151]        intel_engine_pm_put+0x1b/0x40 [i915]
<4> [435.343501]        i915_request_retire+0x477/0xaf0 [i915]
<4> [435.343851]        ring_retire_requests+0x86/0x160 [i915]
<4> [435.344202]        i915_retire_requests+0x58/0xc0 [i915]
<4> [435.344543]        i915_gem_shrink+0xd8/0x5b0 [i915]
<4> [435.344835]        i915_drop_caches_set+0x17b/0x250 [i915]
<4> [435.344877]        simple_attr_write+0xb0/0xd0
<4> [435.344911]        full_proxy_write+0x51/0x80
<4> [435.344943]        vfs_write+0xbd/0x1b0
<4> [435.344972]        ksys_write+0x55/0xe0
<4> [435.345002]        do_syscall_64+0x55/0x190
<4> [435.345040]        entry_SYSCALL_64_after_hwframe+0x49/0xbe

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190409174108.19396-1-chris@chris-wilson.co.uk

drm/i915/execlists: Always reset the context's RING registers

During reset, we try and stop the active ring. This has the consequence
that we often clobber the RING registers within the context image. When
we find an active request, we update the context image to rerun that
request (if it was guilty, we replace the hanging user payload with
NOPs). However, we were ignoring an active context if the request had
completed, with the consequence that the next submission on that request
would start with RING_HEAD==0 and not the tail of the previous request,
causing all requests still in the ring to be rerun. Rare, but
occasionally seen within CI where we would spot that the context seqno
would reverse and complain that we were retiring an incomplete request.

    <0> [412.390350]   <idle>-0       3d.s2 408373352us : __i915_request_submit: rcs0 fence 1e95b:3640 -> current 3638
    <0> [412.390350]   <idle>-0       3d.s2 408373353us : __i915_request_submit: rcs0 fence 1e95b:3642 -> current 3638
    <0> [412.390350]   <idle>-0       3d.s2 408373354us : __i915_request_submit: rcs0 fence 1e95b:3644 -> current 3638
    <0> [412.390350]   <idle>-0       3d.s2 408373354us : __i915_request_submit: rcs0 fence 1e95b:3646 -> current 3638
    <0> [412.390350]   <idle>-0       3d.s2 408373356us : __execlists_submission_tasklet: rcs0 in[0]:  ctx=2.1, fence 1e95b:3646 (current 3638), prio=4
    <0> [412.390350] i915_sel-4613    0.... 408373374us : __i915_request_commit: rcs0 fence 1e95b:3648
    <0> [412.390350] i915_sel-4613    0d..1 408373377us : process_csb: rcs0 cs-irq head=2, tail=3
    <0> [412.390350] i915_sel-4613    0d..1 408373377us : process_csb: rcs0 csb[3]: status=0x00000001:0x00000000, active=0x1
    <0> [412.390350] i915_sel-4613    0d..1 408373378us : __i915_request_submit: rcs0 fence 1e95b:3648 -> current 3638
    <0> [412.390350]   <idle>-0       3..s1 408373378us : execlists_submission_tasklet: rcs0 awake?=1, active=5
    <0> [412.390350] i915_sel-4613    0d..1 408373379us : __execlists_submission_tasklet: rcs0 in[0]:  ctx=2.2, fence 1e95b:3648 (current 3638), prio=4
    <0> [412.390350] i915_sel-4613    0.... 408373381us : i915_reset_engine: rcs0 flags=4
    <0> [412.390350] i915_sel-4613    0.... 408373382us : execlists_reset_prepare: rcs0: depth<-0
    <0> [412.390350]   <idle>-0       3d.s2 408373390us : process_csb: rcs0 cs-irq head=3, tail=4
    <0> [412.390350]   <idle>-0       3d.s2 408373390us : process_csb: rcs0 csb[4]: status=0x00008002:0x00000002, active=0x1
    <0> [412.390350]   <idle>-0       3d.s2 408373390us : process_csb: rcs0 out[0]: ctx=2.2, fence 1e95b:3648 (current 3640), prio=4
    <0> [412.390350] i915_sel-4613    0.... 408373401us : intel_engine_stop_cs: rcs0
    <0> [412.390350] i915_sel-4613    0d..1 408373402us : process_csb: rcs0 cs-irq head=4, tail=4
    <0> [412.390350] i915_sel-4613    0.... 408373403us : intel_gpu_reset: engine_mask=1
    <0> [412.390350] i915_sel-4613    0d..1 408373408us : execlists_cancel_port_requests: rcs0:port0 fence 1e95b:3648, (current 3648)
    <0> [412.390350] i915_sel-4613    0.... 408373442us : intel_engine_cancel_stop_cs: rcs0
    <0> [412.390350] i915_sel-4613    0.... 408373442us : execlists_reset_finish: rcs0: depth->0
    <0> [412.390350] ksoftirq-26      3..s. 408373442us : execlists_submission_tasklet: rcs0 awake?=1, active=0
    <0> [412.390350] ksoftirq-26      3d.s1 408373443us : process_csb: rcs0 cs-irq head=5, tail=5
    <0> [412.390350] i915_sel-4613    0.... 408373475us : i915_request_retire: rcs0 fence 1e95b:3640, current 3648
    <0> [412.390350] i915_sel-4613    0.... 408373476us : i915_request_retire: __retire_engine_request(rcs0) fence 1e95b:3640, current 3648
    <0> [412.390350] i915_sel-4613    0.... 408373494us : __i915_request_commit: rcs0 fence 1e95b:3650
    <0> [412.390350] i915_sel-4613    0d..1 408373496us : process_csb: rcs0 cs-irq head=5, tail=5
    <0> [412.390350] i915_sel-4613    0d..1 408373496us : __i915_request_submit: rcs0 fence 1e95b:3650 -> current 3648
    <0> [412.390350] i915_sel-4613    0d..1 408373498us : __execlists_submission_tasklet: rcs0 in[0]:  ctx=2.1, fence 1e95b:3650 (current 3648), prio=6
    <0> [412.390350] i915_sel-4613    0.... 408373500us : i915_request_retire_upto: rcs0 fence 1e95b:3648, current 3648
    <0> [412.390350] i915_sel-4613    0.... 408373500us : i915_request_retire: rcs0 fence 1e95b:3642, current 3648
    <0> [412.390350] i915_sel-4613    0.... 408373501us : i915_request_retire: __retire_engine_request(rcs0) fence 1e95b:3642, current 3648
    <0> [412.390350] i915_sel-4613    0.... 408373514us : i915_request_retire: rcs0 fence 1e95b:3644, current 3648
    <0> [412.390350] i915_sel-4613    0.... 408373515us : i915_request_retire: __retire_engine_request(rcs0) fence 1e95b:3644, current 3648
    <0> [412.390350] i915_sel-4613    0.... 408373527us : i915_request_retire: rcs0 fence 1e95b:3646, current 3640
    <0> [412.390350]   <idle>-0       3..s1 408373569us : execlists_submission_tasklet: rcs0 awake?=1, active=1
    <0> [412.390350]   <idle>-0       3d.s2 408373569us : process_csb: rcs0 cs-irq head=5, tail=1
    <0> [412.390350]   <idle>-0       3d.s2 408373570us : process_csb: rcs0 csb[0]: status=0x00000001:0x00000000, active=0x1
    <0> [412.390350]   <idle>-0       3d.s2 408373570us : process_csb: rcs0 csb[1]: status=0x00000018:0x00000002, active=0x5
    <0> [412.390350]   <idle>-0       3d.s2 408373570us : process_csb: rcs0 out[0]: ctx=2.1, fence 1e95b:3650 (current 3650), prio=6
    <0> [412.390350]   <idle>-0       3d.s2 408373571us : process_csb: rcs0 completed ctx=2
    <0> [412.390350] i915_sel-4613    0.... 408373621us : i915_request_retire: i915_request_retire:253 GEM_BUG_ON(!i915_request_completed(request))

v2: Fixup the cancellation path to drain the CSB and reset the pointers.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190411130515.20716-2-chris@chris-wilson.co.uk

drm/i915/guc: Implement reset locally

Before causing guc and execlists to diverge further (breaking guc in the
process), take a copy of the current reset procedure and make it local to
the guc submission backend

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190411130515.20716-1-chris@chris-wilson.co.uk

drm/i915: Disable read only ppgtt support for gen11

On gen11 writing to read only ppgtt page causes a gpu hang.
This behaviour is different than with previous gen where
read only ppgtt access is supported. On those, the write
is just dropped without visible side effects.

Disable ro ppgtt support on gen11 until a solution can
be found to bring it into line with its predecessors.

References: HSDES#1807136187
References: https://bugzilla.freedesktop.org/show_bug.cgi?id=108569
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190411083034.28311-1-mika.kuoppala@linux.intel.com

drm/i915: Call i915_sw_fence_fini on request cleanup

As i915_requests are put into an RCU-freelist, they may get reused
before debugobjects notice them as being freed. On cleanup, explicitly
call i915_sw_fence_fini() so that the debugobject is properly tracked.

Reported-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Fixes: b7404c7ecb38 ("drm/i915: Bump ready tasks ahead of busywaits")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Reviewed-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190411122445.20060-1-chris@chris-wilson.co.uk

drm/i915: Clean up DSC vs. not bpp handling

No point in duplicating all this code when we can just
use a variable to hold the output bpp (the only thing
that differs between the two branches).

Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Cc: Manasi Navare <manasi.d.navare@intel.comk>
Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190326144903.6617-2-ville.syrjala@linux.intel.com
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>

drm/i915: Set DP min_bpp to 8*3 for non-RGB output formats

6bpc is only legal for RGB and RAW pixel encodings. For the rest
the minimum is 8bpc. Set our lower limit accordingly.

Signed-off-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190326142556.21176-6-ville.syrjala@linux.intel.com
Reviewed-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>

drm/i915/icl: Switch to using 12 deep CSB status FIFO

Now when we can support variable csb fifo sizes, disable legacy mode.
By disabling legacy we hope to get better hw testing coverage by
assuming everyone else have switched over.

v2: rebase

References: https://bugs.freedesktop.org/show_bug.cgi?id=110338
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Kelvin Gardiner <kelvin.gardiner@intel.com>
Cc: Paulo Zanoni <paulo.r.zanoni@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190405204657.12887-2-chris@chris-wilson.co.uk

drm/i915: Prepare for larger CSB status FIFO size

Make csb entry count variable in preparation for larger
CSB status FIFO size found on gen11+ hardware.

v2: adapt to hwsp access only (Chris)
    non continuous mmio (Daniele)
v3: entries (Chris), fix macro for checkpatch
v4: num_entries (Chris)
v5: consistency on num_entries

Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Joonas Lahtinen <joonas.lahtinen@linux.intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190405204657.12887-1-chris@chris-wilson.co.uk

drm/i915: Use Engine1 instance for gen11 pm interrupts

With gen11 the interrupt registers are shared between 2 engines,
with Engine1 instance being upper word and Engine0 instance being
lower. Annoyingly gen11 selected the pm interrupts to be in the
Engine1 instance.

Rectify the situation by shifting the access accordingly,
based on gen.

v2: comments, warn on overzealous rps_events

Bugzilla: https://bugzilla.freedesktop.org/show_bug.cgi?id=108059
Testcase: igt/i915_pm_rps@min-max-config-loaded
Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410105923.18546-6-mika.kuoppala@linux.intel.com

drm/i915/icl: Disable video turbo mode for rp control

There is no video turbo mode for gen11, so don't set it.

v2: inline (Chris)
v3: brackets (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410132436.23679-1-mika.kuoppala@linux.intel.com

drm/i915/icl: Enable media sampler powergate

Enable media sampler powergate as recommended.

v2: use REG_BIT (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410105923.18546-3-mika.kuoppala@linux.intel.com

drm/i915/icl: Apply a recommended rc6 threshold

On gen11 the recommended rc6 threshold differs from previous
gens, apply it. Move the write to a correct spot in sequence.

v2: do write in 2b, fix bspec ref (Michal)

Bspec: 33149
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Acked-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410105923.18546-2-mika.kuoppala@linux.intel.com

drm/i915: Use dedicated rc6 enabling sequence for gen11

In order not to inflate gen9 rc6 enabling sequence with
gen11 specifics, use a separate function for it.

Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410105923.18546-1-mika.kuoppala@linux.intel.com

drm/i915: Bump ready tasks ahead of busywaits

Consider two tasks that are running in parallel on a pair of engines
(vcs0, vcs1), but then must complete on a shared engine (rcs0). To
maximise throughput, we want to run the first ready task on rcs0 (i.e.
the first task that completes on either of vcs0 or vcs1). When using
semaphores, however, we will instead queue onto rcs in submission order.

To resolve this incorrect ordering, we want to re-evaluate the priority
queue when each of the request is ready. Normally this happens because
we only insert into the priority queue requests that are ready, but with
semaphores we are inserting ahead of their readiness and to compensate
we penalize those tasks with reduced priority (so that tasks that do not
need to busywait should naturally be run first). However, given a series
of tasks that each use semaphores, the queue degrades into submission
fifo rather than readiness fifo, and so to counter this we give a small
boost to semaphore users as their dependent tasks are completed (and so
we no longer require any busywait prior to running the user task as they
are then ready themselves).

v2: Fixup irqsave for schedule_lock (Tvrtko)

Testcase: igt/gem_exec_schedule/semaphore-codependency
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Cc: Dmitry Rogozhkin <dmitry.v.rogozhkin@intel.com>
Cc: Dmitry Ermilov <dmitry.ermilov@intel.com>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190409152922.23894-1-chris@chris-wilson.co.uk

drm/i915: Only reset the pinned kernel contexts on resume

On resume, we know that the only pinned contexts in danger of seeing
corruption are the kernel context, and so we do not need to walk the
list of all GEM contexts as we tracked them on each engine.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410190120.830-1-chris@chris-wilson.co.uk

drm/i915/dp: Expose force_dsc_enable through debugfs

Currently we use force_dsc_enable to force DSC from IGT, but
we dont expose this value to userspace through debugfs.
This patch exposes this through the same dsc_fec_support
debugfs node per connector so that we can restore its value
back after the tests are completed.

Suggested-by: Imre Deak <imre.deak@intel.com>
Cc: Imre Deak <imre.deak@intel.com>
Cc: Lyude Paul <lyude@redhat.com>
Cc: Anusha Srivatsa <anusha.srivatsa@intel.com>
Signed-off-by: Manasi Navare <manasi.d.navare@intel.com>
Reviewed-by: Imre Deak <imre.deak@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190405224821.32435-1-manasi.d.navare@intel.com

drm/i915/icl: Don't warn on spurious interrupts

There is a chance we can see spurious interrupts in live
now. We have more engines enabled and that with more elaborate
access patterns with pm and display, increases the chances
hardware just makes a social call, without anything to work on.

Remove the error as we have tests to actually probe if
we really miss interrupt, instead of getting spurious ones.

Note that now we do write to intr_dw even with a zero
value. This is considered advantegous as the write
is an ack that sw is done.

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410132124.21795-2-mika.kuoppala@linux.intel.com

drm/i915/icl: Handle rps interrupts without irq lock

Unlike previous gens, we already hold the irq_lock on
entering the rps handler so we can't use it as it is.

Make a gen11 specific rps interrupt handler without
locking.

v2: return early (Chris)

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/20190410132124.21795-1-mika.kuoppala@linux.intel.com

ALSA: hda: Fix racy display power access

snd_hdac_display_power() doesn't handle the concurrent calls carefully
enough, and it may lead to the doubly get_power or put_power calls,
when a runtime PM and an async work get called in racy way.

This patch addresses it by reusing the bus->lock mutex that has been
used for protecting the link state change in ext bus code, so that it
can protect against racy display state changes.  The initialization of
bus->lock was moved from snd_hdac_ext_bus_init() to
snd_hdac_bus_init() as well accordingly.

Testcase: igt/i915_pm_rpm/module-reload #glk-dsi
Reported-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Imre Deak <imre.deak@intel.com>
Signed-off-by: Takashi Iwai <tiwai@suse.de>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Link: https://patchwork.freedesktop.org/patch/msgid/s5h8swiunph.wl-tiwai@suse.de

drm/i915/dp: revert back to max link rate and lane count on eDP

Commit 7769db588384 ("drm/i915/dp: optimize eDP 1.4+ link config fast
and narrow") started to optize the eDP 1.4+ link config, both per spec
and as preparation for display stream compression support.

Sadly, we again face panels that flat out fail with parameters they
claim to support. Revert, and go back to the drawing board.

v2: Actually revert to max params instead of just wide-and-slow.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=109959
Fixes: 7769db588384 ("drm/i915/dp: optimize eDP 1.4+ link config fast and narrow")
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Manasi Navare <manasi.d.navare@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Cc: Matt Atwood <matthew.s.atwood@intel.com>
Cc: "Lee, Shawn C" <shawn.c.lee@intel.com>
Cc: Dave Airlie <airlied@gmail.com>
Cc: intel-gfx@lists.freedesktop.org
Cc: <stable@vger.kernel.org> # v5.0+
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Manasi Navare <manasi.d.navare@intel.com>
Tested-by: Albert Astals Cid <aacid@kde.org> # v5.0 backport
Tested-by: Emanuele Panigati <ilpanich@gmail.com> # v5.0 backport
Tested-by: Matteo Iervasi <matteoiervasi@gmail.com> # v5.0 backport
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190405075220.9815-1-jani.nikula@intel.com

drm/i915/icl: Fix port disable sequence for mipi-dsi

Re-enable clock gating of DDI clocks.

v2: Fix the default ddi clk state for mipi-dsi (Imre)

Fixes: 1026bea00381 ("drm/i915/icl: Ungate DSI clocks")
Signed-off-by: Vandita Kulkarni <vandita.kulkarni@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1553513202-13863-2-git-send-email-vandita.kulkarni@intel.com

drm/i915/icl: Ungate ddi clocks before IO enable

IO enable sequencing needs ddi clocks enabled.
These clocks will be gated at a later point in
the enable sequence.

v2: Fix the commit header (Uma)
v3: Remove the redundant read (Ville)

Fixes: 949fc52af19e ("drm/i915/icl: add pll mapping for DSI")
Signed-off-by: Vandita Kulkarni <vandita.kulkarni@intel.com>
Reviewed-by: Uma Shankar <uma.shankar@intel.com>
Signed-off-by: Jani Nikula <jani.nikula@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/1553513202-13863-1-git-send-email-vandita.kulkarni@intel.com

drm/i915/guc: Replace preempt_client lookup with engine->preempt_context

Circumvent the dance we currently perform to find the preempt_client and
lookup its HW context for this engine, as we know we have already pinned
the preempt_context on the engine.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190408091728.20207-15-chris@chris-wilson.co.uk

drm/i915/selftests: Mark live_forcewake_ops as unreliable

A couple of machines in the farm show quite frequent errors in the
powerwells not being released. Either there is an external agent
interferring with the powerwells, or the powerwell doesn't quite behave
as we anticipate -- either way, the test is not reliable enough to be
enabled by default in CI. It has served its immediate purpose in
providing coverage as we made tweaks to forcewake, so keep it available
for future testing.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=110210
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Mika Kuoppala <mika.kuoppala@linux.intel.com>
Reviewed-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190407192649.14750-1-chris@chris-wilson.co.uk

drm/i915/psr: Do not enable PSR in interlaced mode for all GENs

This interlaced mode restriction applies to all gens, not only to
Haswell.

Also while at it updating the debug message to.

Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190406005112.27205-4-jose.souza@intel.com

drm/i915/psr: Initialize PSR mutex even when sink is not reliable

Even when driver is reloaded and hits this scenario the PSR mutex
should be initialized, otherwise reading PSR debugfs status will
execute mutex_lock() over a mutex that was not initialized.

Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190406005112.27205-3-jose.souza@intel.com

drm/i915: Remove unused VLV/CHV PSR registers

PSR support for VLV and CHV was dropped in commit ce3508fd2a77
("drm/i915/psr: Nuke PSR support for VLV and CHV") so no need to keep
this registers around.

Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190406005112.27205-2-jose.souza@intel.com

drm/i915/psr: Update PSR2 SU corruption workaround comment

Turn out it is not a DMC bug it is actually a HW one, so this
workaround will be needed for current gens, lets update the comment
and remove the FIXME.

BSpec: 7723
Cc: Dhinakaran Pandiyan <dhinakaran.pandiyan@intel.com>
Cc: Rodrigo Vivi <rodrigo.vivi@intel.com>
Reviewed-by: Rodrigo Vivi <rodrigo.vivi@intel.com>
Signed-off-by: José Roberto de Souza <jose.souza@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190406005112.27205-1-jose.souza@intel.com

drm/i915: Consolidate the timeline->barrier

The timeline is strictly ordered, so by inserting the timeline->barrier
request into the timeline->last_request it naturally provides the same
barrier. Consolidate the pair of barriers into one as they serve the
same purpose.

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190408091728.20207-4-chris@chris-wilson.co.uk

drm/i915: Use static allocation for i915_globals_park()

In order to avoid the malloc inside i915_globals_park() occurring
underneath a lock connected to the shrinker (thus causing circular
lockdeps warnings), move the rcu_worker to a global.

<4> [39.085073] ======================================================
<4> [39.085273] WARNING: possible circular locking dependency detected
<4> [39.085552] 5.1.0-rc3-CI-Trybot_4088+ #1 Tainted: G     U
<4> [39.085752] ------------------------------------------------------
<4> [39.085949] kswapd0/32 is trying to acquire lock:
<4> [39.086121] 00000000004b5f91 (wakeref#3){+.+.}, at: intel_engine_pm_put+0x1b/0x40 [i915]
<4> [39.086493]
but task is already holding lock:
<4> [39.086682] 00000000dd009a9a (fs_reclaim){+.+.}, at: __fs_reclaim_acquire+0x0/0x30
<4> [39.086910]
which lock already depends on the new lock.

<4> [39.087139]
the existing dependency chain (in reverse order) is:
<4> [39.087356]
-> #2 (fs_reclaim){+.+.}:
<4> [39.087604]        fs_reclaim_acquire.part.24+0x24/0x30
<4> [39.087785]        kmem_cache_alloc_trace+0x2a/0x290
<4> [39.087998]        i915_globals_park+0x22/0xa0 [i915]
<4> [39.088478]        idle_work_handler+0x1df/0x220 [i915]
<4> [39.089016]        process_one_work+0x245/0x610
<4> [39.089447]        worker_thread+0x37/0x380
<4> [39.089956]        kthread+0x119/0x130
<4> [39.090374]        ret_from_fork+0x3a/0x50
<4> [39.090868]
-> #1 (wakeref#4){+.+.}:
<4> [39.091569]        __mutex_lock+0x8c/0x960
<4> [39.092054]        atomic_dec_and_mutex_lock+0x33/0x50
<4> [39.092521]        intel_gt_pm_put+0x1b/0x40 [i915]
<4> [39.093047]        intel_engine_park+0xeb/0x1d0 [i915]
<4> [39.093514]        __intel_wakeref_put_once+0x10/0x30 [i915]
<4> [39.094062]        i915_request_retire+0x477/0xaf0 [i915]
<4> [39.094547]        ring_retire_requests+0x86/0x160 [i915]
<4> [39.095110]        i915_retire_requests+0x58/0xc0 [i915]
<4> [39.095587]        i915_gem_wait_for_idle.part.22+0xb2/0xf0 [i915]
<4> [39.096142]        switch_to_kernel_context_sync+0x2a/0x70 [i915]
<4> [39.096633]        i915_gem_init+0x59c/0x9c0 [i915]
<4> [39.097174]        i915_driver_load+0xd96/0x1880 [i915]
<4> [39.097640]        i915_pci_probe+0x29/0xa0 [i915]
<4> [39.098145]        pci_device_probe+0xa1/0x120
<4> [39.098607]        really_probe+0xf3/0x3e0
<4> [39.099031]        driver_probe_device+0x10a/0x120
<4> [39.099599]        device_driver_attach+0x4b/0x50
<4> [39.100033]        __driver_attach+0x97/0x130
<4> [39.100525]        bus_for_each_dev+0x74/0xc0
<4> [39.100954]        bus_add_driver+0x13f/0x210
<4> [39.101441]        driver_register+0x56/0xe0
<4> [39.101891]        do_one_initcall+0x58/0x2e0
<4> [39.102319]        do_init_module+0x56/0x1ea
<4> [39.102805]        load_module+0x2701/0x29e0
<4> [39.103231]        __se_sys_finit_module+0xd3/0xf0
<4> [39.103727]        do_syscall_64+0x55/0x190
<4> [39.104153]        entry_SYSCALL_64_after_hwframe+0x49/0xbe
<4> [39.104736]
-> #0 (wakeref#3){+.+.}:
<4> [39.105437]        lock_acquire+0xa6/0x1c0
<4> [39.105923]        __mutex_lock+0x8c/0x960
<4> [39.106345]        atomic_dec_and_mutex_lock+0x33/0x50
<4> [39.106897]        intel_engine_pm_put+0x1b/0x40 [i915]
<4> [39.107375]        i915_request_retire+0x477/0xaf0 [i915]
<4> [39.107930]        ring_retire_requests+0x86/0x160 [i915]
<4> [39.108412]        i915_retire_requests+0x58/0xc0 [i915]
<4> [39.108934]        i915_gem_shrink+0xd8/0x5b0 [i915]
<4> [39.109431]        i915_gem_shrinker_scan+0x59/0x130 [i915]
<4> [39.109884]        do_shrink_slab+0x131/0x3e0
<4> [39.110380]        shrink_slab+0x228/0x2c0
<4> [39.110810]        shrink_node+0x177/0x460
<4> [39.111317]        balance_pgdat+0x239/0x580
<4> [39.111743]        kswapd+0x186/0x570
<4> [39.112221]        kthread+0x119/0x130
<4> [39.112641]        ret_from_fork+0x3a/0x50

Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Tvrtko Ursulin <tvrtko.ursulin@intel.com>
Link: https://patchwork.freedesktop.org/patch/msgid/20190408091728.20207-3-chris@chris-wilson.co.uk