HDCP compliant Repeaters can support max of 127 devices and max
depth of 7 for downstream topology.

If these max limits are exceeded, repeater will set the
topology error flags MAX_CASCADE_EXCEEDED and/or MAX_DEVS_EXCEEDED
in Bstatus followed by asserting READY/CP_IRQ for HDCP transmitter.

This patch check for these error flags as soon as READY bit is asserted.
Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
[seanpaul fixed checkpatch alignment issue]
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1516254488-4971-5-git-send-email-ramalingam.c@intel.com

Second stage of HDCP authentication starts at CP_IRQ or at the
assertion of READY bit from Repeater.

Till then repeater will be authenticating with its downstream devices.
So authenticated device count, depth and ksv_list readable from
repeaters are valid only after assertion of READY/CP_IRQ.

This patch makes sure that READY is polled before reading any
topology information.
Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1516254488-4971-4-git-send-email-ramalingam.c@intel.com

Second stage of HDCP authentication required only if the HDCP
sink is a repeater.

This patch imposes above condition on second stage HDCP authentication.
Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1516254488-4971-3-git-send-email-ramalingam.c@intel.com

This patch extends the Key load process and hdcp initialization for
few more capable intel platforms i.e. HSW, BDW and BXT+.
Signed-off-by: Ramalingam C <ramalingam.c@intel.com>
[seanpaul fixed checkpatch issues]
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/1516254488-4971-2-git-send-email-ramalingam.c@intel.com

drivers/gpu/drm/i915/intel_dp.c: In function ‘intel_dp_hdcp_check_link’:
drivers/gpu/drm/i915/intel_dp.c:5191:26: error: ?: using integer constants in boolean context [-Werror=int-in-bool-context]
   return ret >= 0 ? -EIO : ret;

Fixes: 20f24d77 ("drm/i915: Implement HDCP for DisplayPort")
Signed-off-by: Chris Wilson <chris@chris-wilson.co.uk>
Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Cc: Ramalingam C <ramalingam.c@intel.com>
Cc: Sean Paul <seanpaul@chromium.org>
Cc: Jani Nikula <jani.nikula@linux.intel.com>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180118161025.22700-1-chris@chris-wilson.co.uk

Instead of always trying to disable HDCP. Only run hdcp_disable when the
state is not UNDESIRED. This will catch cases where it's enabled and
also cases where enable failed and the state is left in DESIRED mode.

Note that things won't blow up if disable is attempted while already
disabled, it's just bad form.
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180109185330.16853-1-seanpaul@chromium.org

Port E doesn't have HDCP support, and Port F is disabled. Don't setup
the hdcp shim on those.
Reviewed-by: Daniel Vetter <daniel@ffwll.ch>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180109185401.16911-1-seanpaul@chromium.org

This patch adds HDCP support for DisplayPort connectors by implementing
the intel_hdcp_shim.

Most of this is straightforward read/write from/to DPCD registers. One
thing worth pointing out is the Aksv output bit. It wasn't easily
separable like it's HDMI counterpart, so it's crammed in with the rest
of it.

Changes in v2:
- Moved intel_hdcp_check_link out of intel_dp_check_link and only call
  it on short pulse. Since intel_hdcp_check_link does its own locking,
  this ensures we don't deadlock when intel_dp_check_link is called
  holding connection_mutex.
- Rebased on drm-intel-next
Changes in v3:
- Initialize new worker
Changes in v4:
- Use intel_hdcp_init (Daniel)
- Check for reauth requests in check_link (Ram)
Changes in v5:
- None
Changes in v6:
- Fix build warnings when printing ssize_t

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Ramalingam C <ramalingam.c@intel.com>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-10-seanpaul@chromium.org

This patch adds HDCP support for HDMI connectors by implementing
the intel_hdcp_shim.

Nothing too special, just a bunch of DDC reads/writes.

Changes in v2:
- Rebased on drm-intel-next
Changes in v3:
- Initialize new worker
Changes in v4:
- Remove SKL_ prefix from most register names (Daniel)
- Wrap sanity checks in WARN_ON (Daniel)
- Consolidate the enable/disable functions into one toggle fn
- Use intel_hdcp_init (Daniel)
Changes in v5:
- checkpatch whitespace nits
Changes in v6:
- None

Cc: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Ramalingam C <ramalingam.c@intel.com>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-9-seanpaul@chromium.org

Once the Aksv is available in the PCH, we need to get it on the wire to
the receiver via DDC. The hardware doesn't allow us to read the value
directly, so we need to tell GMBUS to source the Aksv internally and
send it to the right offset on the receiver.

The way we do this is to initiate an indexed write where the index is
the Aksv register offset. We write dummy values to GMBUS3 as if we were
sending the key, and the hardware slips in the "real" values when it
goes out.

Changes in v2:
- None
Changes in v3:
- Uses new index write feature (Ville)
Changes in v4:
- None
Changes in v5:
- checkpatch whitespace fix
Changes in v6:
- None

Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-8-seanpaul@chromium.org

This patch enables the indexed write feature of the GMBUS to concatenate
2 consecutive messages into one. The criteria for an indexed write is
that both messages are writes, the first is length == 1, and the second
is length > 0. The first message is sent out by the GMBUS as the slave
command, and the second one is sent via the GMBUS FIFO as usual.

Changes in v3:
- Added to series
Changes in v4:
- Combine indexed reads and writes (Ville)
Changes in v5:
- checkpatch whitespace nits
Changes in v6:
- None
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Suggested-by: Ville Syrjälä <ville.syrjala@linux.intel.com>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-7-seanpaul@chromium.org

This patch adds the framework required to add HDCP support to intel
connectors. It implements Aksv loading from fuse, and parts 1/2/3
of the HDCP authentication scheme.

Note that without shim implementations, this does not actually implement
HDCP. That will come in subsequent patches.

Changes in v2:
- Don't open code wait_fors (Chris)
- drm_hdcp.c under MIT license (Daniel)
- Move intel_hdcp_disable() call above ddi_disable (Ram)
- Fix // comments (I wore a cone of shame for 12 hours to atone) (Daniel)
- Justify intel_hdcp_shim with comments (Daniel)
- Fixed async locking issues by adding hdcp_mutex (Daniel)
- Don't alter connector_state in enable/disable (Daniel)
Changes in v3:
- Added hdcp_mutex/hdcp_value to make async reasonable
- Added hdcp_prop_work to separate link checking & property setting
- Added new helper for atomic_check state tracking (Daniel)
- Moved enable/disable into atomic_commit with matching helpers
- Moved intel_hdcp_check_link out of all locks when called from dp
- Bumped up ksv_fifo timeout (noticed failure on one of my dongles)
Changes in v4:
- Remove SKL_ prefix from most register names (Daniel)
- Move enable/disable back to modeset path (Daniel)
- s/get_random_long/get_random_u32/ (Daniel)
- Remove mode_config.mutex lock in prop_work (Daniel)
- Add intel_hdcp_init to handle init of conn components (Daniel)
- Actually check return value of attach_property
- Check Bksv is valid before trying to authenticate (Ram)
Changes in v5:
- checkpatch whitespace changes
- s/DRM_MODE_CONTENT_PROTECTION_OFF/DRM_MODE_CONTENT_PROTECTION_UNDESIRED/
- Fix ksv list wait timeout (actually wait 5s)
- Increase the R0 timeout to 300ms (Ram)
Changes in v6:
- SPDX license

Cc: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Ramalingam C <ramalingm.c@intel.com>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-6-seanpaul@chromium.org

In preparation for implementing HDCP in i915, add some HDCP related
register offsets and defines. The dpcd register offsets will go in
drm_dp_helper.h whereas the ddc offsets along with generic HDCP stuff
will get stuffed in drm_hdcp.h, which is new.

Changes in v2:
- drm_hdcp.h gets MIT license (Daniel)
Changes in v3:
- None
Changes in v4:
- None
Changes in v5:
- None
Changes in v6:
- SPDX license

Cc: Daniel Vetter <daniel.vetter@intel.com>
Reviewed-by: Ramalingam C <ramalingm.c@intel.com>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-5-seanpaul@chromium.org

This patch adds a new optional connector property to allow userspace to enable
protection over the content it is displaying. This will typically be implemented
by the driver using HDCP.

The property is a tri-state with the following values:
- OFF: Self explanatory, no content protection
- DESIRED: Userspace requests that the driver enable protection
- ENABLED: Once the driver has authenticated the link, it sets this value

The driver is responsible for downgrading ENABLED to DESIRED if the link becomes
unprotected. The driver should also maintain the desiredness of protection
across hotplug/dpms/suspend.

If this looks familiar, I posted [1] this 3 years ago. We have been using this
in ChromeOS across exynos, mediatek, and rockchip over that time.

Changes in v2:
 - Pimp kerneldoc for content_protection_property (Daniel)
 - Drop sysfs attribute
Changes in v3:
 - None
Changes in v4:
- Changed kerneldoc to recommend userspace polling (Daniel)
- Changed kerneldoc to briefly describe how to attach the property (Daniel)
Changes in v5:
- checkpatch whitespace noise
- Change DRM_MODE_CONTENT_PROTECTION_OFF to DRM_MODE_CONTENT_PROTECTION_UNDESIRED
Changes in v6:
- None
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Sean Paul <seanpaul@chromium.org>

[1] https://lists.freedesktop.org/archives/dri-devel/2014-December/073336.html
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-4-seanpaul@chromium.org

This patch adds a little more control to a couple wait_for routines such
that we can avoid open-coding read/wait/timeout patterns which:
 - need the value of the register after the wait_for
 - run arbitrary operation for the read portion

This patch also chooses the correct sleep function (based on
timers-howto.txt) for the polling interval the caller specifies.

Changes in v2:
- Added to the series
Changes in v3:
- Rebased on drm-intel-next-queued and the new Wmin/max _wait_for
- Removed msleep option
Changes in v4:
- Removed ; for OP in _wait_for (Chris)
- Moved reg_value definition above ret (Chris)
Changes in v4:
- checkpatch whitespace fix
Changes in v5:
- None
Changes in v6:
- None
Suggested-by: Chris Wilson <chris@chris-wilson.co.uk>
Reviewed-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Reviewed-by: Chris Wilson <chris@chris-wilson.co.uk>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-3-seanpaul@chromium.org

I'm adding some stuff below it and it's killing my editor's vibe.

Changes in v2:
- Added to the series
Changes in v3:
- None
Changes in v4:
- None
Changes in v5:
- None
Changes in v6:
- None

Cc: Manasi Navare <manasi.d.navare@intel.com>
Acked-by: Daniel Vetter <daniel.vetter@ffwll.ch>
Signed-off-by: Sean Paul <seanpaul@chromium.org>
Link: https://patchwork.freedesktop.org/patch/msgid/20180108195545.218615-2-seanpaul@chromium.org

drm/imx: format modifier support

- Add tiled prefetch support to PRE
- Add format modifier support to PRG and imx-drm-core
- Use runtime PM to control PRG clock
- Allow building ipu-v3 under COMPILE_TEST

* tag 'imx-drm-next-2018-01-02' of git://git.pengutronix.de/git/pza/linux:
  gpu: ipu-v3: allow to build with COMPILE_TEST
  drm/imx: advertise supported plane format modifiers
  drm/imx: add FB modifier support
  gpu: ipu-v3: prg: add modifier support
  gpu: ipu-v3: pre: add tiled prefetch support
  gpu: ipu-v3: prg: switch to runtime PM

This series builds upon the set of fixes previously submitted to move
Armada DRM closer to atomic modeset.  We're nowhere near yet, but this
series helps to get us closer by unifying some of the differences
between the primary and overlay planes.

New features added allows userspace to disable the primary plane if
overlay is full screen and there's nothing obscuring the colorkey -
this saves having to fetch an entire buffer containing nothing but
colorkey when displaying full screen video.

[airlied: fixup for atomic plane helper rename:
a01cb8ba
Author: Ville Syrjälä <ville.syrjala@linux.intel.com>
Date:   Wed Nov 1 22:16:19 2017 +0200

    drm: Move drm_plane_helper_check_state() into drm_atomic_helper.c
]

* 'drm-armada-devel-4.15' of git://git.armlinux.org.uk/~rmk/linux-arm: (29 commits)
  drm/armada: expand overlay trace entry
  drm/armada: implement primary plane update
  drm/armada: extract register generation from armada_drm_primary_set()
  drm/armada: wait for previous work when moving overlay window
  drm/armada: move overlay plane register update generation
  drm/armada: re-organise overlay register update generation
  drm/armada: disable planes at next blanking period
  drm/armada: avoid work allocation
  drm/armada: allow armada_drm_plane_work_queue() to silently fail
  drm/armada: use drm_plane_helper_check_state()
  drm/armada: only enable HSMOOTH if scaling horizontally
  drm/armada: move writes of LCD_SPU_SRAM_PARA1 under lock
  drm/armada: move regs into armada_plane_work
  drm/armada: move event sending into armada_plane_work
  drm/armada: move fb retirement into armada_plane_work
  drm/armada: move overlay plane work out from under spinlock
  drm/armada: clear plane enable bit when disabling
  drm/armada: clean up armada_drm_crtc_plane_disable()
  drm/armada: allow the primary plane to be disabled
  drm/armada: wait and cancel any pending frame work at disable
  ...

Highlights this time:
1. Fix for a nasty Kconfig dependency chain issue from Philipp.
2. Occlusion query buffer address added to the cmdstream validator by
Christian.
3. Fixes and cleanups to the job handling from me. This allows us to
turn on the GPU performance profiling added in the last cycle.
It is also prep work for hooking in the DRM GPU scheduler, which I hope
to land for the next cycle.

* 'etnaviv/next' of https://git.pengutronix.de/git/lst/linux: (32 commits)
  drm/etnaviv: use memset32 to init pagetable
  drm/etnaviv: move submit free out of critical section
  drm/etnaviv: re-enable perfmon support
  drm/etnaviv: couple runtime PM management to submit object lifetime
  drm/etnaviv: move GPU active handling to bo pin/unpin
  drm/etnaviv: move cmdbuf into submit object
  drm/etnaviv: use submit exec_state for perfmon sampling
  drm/etnaviv: move exec_state to submit object
  drm/etnaviv: move PMRs to submit object
  drm/etnaviv: refcount the submit object
  drm/etnaviv: move ww_acquire_ctx out of submit object
  drm/etnaviv: move object unpinning to submit cleanup
  drm/etnaviv: attach in fence to submit and move fence wait to fence_sync
  drm/etnaviv: rename submit fence to out_fence
  drm/etnaviv: move object fence attachment to gem_submit path
  drm/etnaviv: simplify submit_create
  drm/etnaviv: add lockdep annotations to buffer manipulation functions
  drm/etnaviv: hold GPU lock while inserting END command
  drm/etnaviv: move workqueue to be per GPU
  drm/etnaviv: remove switch_context member from etnaviv_gpu
  ...

Merge tag 'exynos-drm-next-for-v4.16' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos into drm-next

Remove lagacy IPP driver
- This driver isn't used anymore so remove it. Marek is preparing new one
  which includes completely rewritten API so this driver will be replaced
  with the new version[1] later.
And cleanups.

[1] https://patches.linaro.org/cover/118386/

* tag 'exynos-drm-next-for-v4.16' of git://git.kernel.org/pub/scm/linux/kernel/git/daeinki/drm-exynos:
  drm/exynos: ipp: Remove Exynos DRM IPP subsystem
  drm/exynos/decon: Add include guard to the Exynos7 header
  drm/exynos/decon: Move headers from global to local place
  drm/exynos: decon5433: Remove unnecessary platform_get_resource() error check

- Add CWSR (compute wave save restore) support for GFX8 (Carrizo)
- Fix SDMA user-mode queues support for GFX7 (Kaveri)
- Add SDMA user-mode queues support for GFX8 (Carrizo)
- Allow HWS (hardware scheduling) to schedule multiple processes concurrently
- Add debugfs support
- Simplify process locking and lock dependencies
- Refactoring topology code to prepare for dGPU support + fixes to that code
  - Add option to generate dummy/virtual CRAT table when its missing or deformed
  - Recognize CPUs other then APUs as compute entities
- Various clean ups and bug fixes

I have not yet sent the dGPU topology code because it depends on a patch
for the PCI subsystem that adds PCIe atomics support. Once that patch is
upstreamed we can continue with the rest of the dGPU code.

* tag 'drm-amdkfd-next-2017-12-24' of git://people.freedesktop.org/~gabbayo/linux: (53 commits)
  drm/amdgpu: Add support for reporting VRAM usage
  drm/amdkfd: Ignore ACPI CRAT for non-APU systems
  drm/amdkfd: Module option to disable CRAT table
  drm/amdkfd: Add AQL Queue Memory flag on topology
  drm/amdkfd: Fixup incorrect info in the CZ CRAT table
  drm/amdkfd: Add perf counters to topology
  drm/amdkfd: Add topology support for dGPUs
  drm/amdkfd: Add topology support for CPUs
  drm/amdkfd: Fix sibling_map[] size
  drm/amdkfd: Simplify counting of memory banks
  drm/amdkfd: Turn verbose topology messages into pr_debug
  drm/amdkfd: sync IOLINK defines to thunk spec
  drm/amdkfd: Support enumerating non-GPU devices
  drm/amdkfd: Decouple CRAT parsing from device list update
  drm/amdkfd: Reorganize CRAT fetching from ACPI
  drm/amdkfd: Group up CRAT related functions
  drm/amdkfd: Fix memory leaks in kfd topology
  drm/amdkfd: Topology: Fix location_id
  drm/amdkfd: Update number of compute unit from KGD
  drm/amd: Remove get_vmem_size from KGD-KFD interface
  ...

Now that memset32 is available, the open-coded pagetable initialization
loop can be replaced.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

There is no need to hold the GPU lock while freeing the submit
object. Only move the retired submits from the GPU active list to
a temporary retire list under the GPU lock.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

Now that the PMR lifetime issues are solved we can safely re-enable
performance counter profiling support.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

As long as there is an active submit, we want the GPU to stay awake. This
is slightly complicated by the fact that we really want to wake the GPU
at the last possible moment to achieve maximum power savings.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

The active count is used to check if the BO is idle, where idle is defined
as not active on the GPU and all VM mappings and reference counts dropped
to the initial state. As the idling of the mappings and references now only
happens in the submit cleanup, the active state handling must be moved to
the same location in order to keep the userspace semantics.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

Less dynamic allocations and slims down the cmdbuf object to only the
required information, as everything else is already available in the
submit object.

This also simplifies buffer and mappings lifetime management, as they
are now exlusively attached to the submit object and not additionally
to the cmdbuf.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

The GPU exec state may have changed at the time when the perfmon sampling
is done, as it reflects the state of the last submission, not the current
GPU execution state.

So for proper sampling we must use the submit exec_state.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

We'll need this in some places where only the submit is available. Also
this is a first step at slimming down the cmdbuf object.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

To make them available to the event worker even after the actual
command stream execution has finished.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

The submit object lifetime will get extended to the actual GPU
execution. As multiple users will depend on this, add a kref to
properly control destruction of the object.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

The acquire_ctx is special in that it needs to be released from the same
thread as has been used to initialize it. This collides with the intention to
extend the submit lifetime beyond the gem_submit function with potentially
other threads doing the final cleanup.

Move the ww_acquire_ctx to the function local stack as suggested in the
documentation.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

This is safe to call in all paths, as the BO_PINNED flag tells us if the BO
needs unpinning.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

Simplifies the cleanup path and moves fence waiting to a central location.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

This is the fence passed out on a sucessful GPU submit. Make the name
more clear.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>

The object fencing has nothing to do with the actual GPU buffer submit,
so move it to the gem submit path to have a cleaner split.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

Use kzalloc so other code doesn't need to worry about uninitialized members.
Drop the non-standard GFP flags, as we really don't want to fail the submit
when under slight memory pressure. Remove one level of indentation by using
an early return if the allocation failed. Also remove the unused drm device
member.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

When manipulating the kernel command buffer the GPU mutex must be held, as
otherwise different callers might try to replace the same part of the
buffer, wreacking havok in the GPU execution.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>

Inserting the END command when suspending the GPU is changing the
command buffer state, which requires the GPU to be held.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>
Reviewed-by: Christian Gmeiner <christian.gmeiner@gmail.com>

While the etnaviv workqueue needs to be ordered, as we rely on work items
being executed in queuing order, this is only true for a single GPU.
Having a shared workqueue for all GPUs in the system limits concurrency
artificially.

Getting each GPU its own ordered workqueue still meets our ordering
expectations and enables retire workers to run concurrently.
Signed-off-by: Lucas Stach <l.stach@pengutronix.de>
Reviewed-by: Philipp Zabel <p.zabel@pengutronix.de>