External clients which import our bo's wait only
for exclusive dmabuf-fences, not on shared ones,
ditto for bo's which we import from external
providers and write to.

Therefore attach exclusive fences on prime shared buffers
if our exported buffer gets imported by an external
client, or if we import a buffer from an external
exporter.

See discussion in thread:
https://lists.freedesktop.org/archives/dri-devel/2016-October/122370.html

Prime export tested on Intel iGPU + AMD Tonga dGPU as
DRI3/Present Prime render offload, and with the Tonga
standalone as primary gpu.

v2: Add a wait for all shared fences before prime export,
    as suggested by Christian Koenig.

v3: - Mark buffer prime_exported in amdgpu_gem_prime_pin,
    so we only use the exclusive fence when exporting a
    bo to external clients like a separate iGPU, but not
    when exporting/importing from/to ourselves as part of
    regular DRI3 fd passing.

    - Propagate failure of reservation_object_wait_rcu back
    to caller.

v4: - Switch to a prime_shared_count counter instead of a
      flag, which gets in/decremented on prime_pin/unpin, so
      we can switch back to shared fences if all clients
      detach from our exported bo.

    - Also switch to exclusive fence for prime imported bo's.

v5: - Drop lret, instead use int ret -> long ret, as proposed
      by Christian.

Bugzilla: https://bugs.freedesktop.org/show_bug.cgi?id=95472
Tested-by: Mike Lothian <mike@fireburn.co.uk> (v1)
Signed-off-by: NMario Kleiner <mario.kleiner.de@gmail.com>
Reviewed-by: Christian König <christian.koenig@amd.com>.
Cc: Christian König <christian.koenig@amd.com>
Cc: Michel Dänzer <michel.daenzer@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

On CZ/ST systems with AZ rather than ACP audio, we need to bail
early in hw_fini since there is nothing to do.

bug: https://bugs.freedesktop.org/show_bug.cgi?id=98276Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Just to be clean should we ever run into -ENOMEM during module init.

v2: fix typo in commit message
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Some fences might be alive even after we have stopped the scheduler leading
to warnings about leaked objects from the SLUB allocator.

Fix this by allocating/freeing the SLUB allocator from the module
init/fini functions just like we do it for hw fences.

v2: make variable static, add link to bug

Fixes: https://bugs.freedesktop.org/show_bug.cgi?id=97500Reported-by: NGrazvydas Ignotas <notasas@gmail.com>
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

This should only happen on boards TV connectors which do not
have a ddc bus for those connectors.  None of the asics supported
by amdgpu support tv, so we shouldn't hit this, but check
to be on the safe side (e.g., bios bug for example).
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

If the platform does not support hybrid graphics or ATPX dGPU
power control.
Acked-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Newer iceland parts require new smc firmware.
Reviewed-by: NHuang Rui <ray.huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Newer tonga parts require new smc firmware.
Reviewed-by: NHuang Rui <ray.huang@amd.com>
Reviewed-by: NEric Huang <JinHuiEric.Huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Consolidate existing quirks. Fixes stability issues
on some kickers.
Acked-by: NHuang Rui <ray.huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

the value of last_mclk_dpm_enable_mask will be changed if
other clients(vce,dal) trigger set power state between enable
and disable uvd dpm.
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This fixes a regression in all these drivers since the cache
mode tracking was fixed for mixed mappings. It uses the new
arch API to add the VRAM range to the PAT mapping tracking
tables.

Fixes: 87744ab3 (mm: fix cache mode tracking in vm_insert_mixed())
Reviewed-by: Christian König <christian.koenig@amd.com>.
Signed-off-by: NDave Airlie <airlied@redhat.com>

Older firmware versions don't support 3 rings.

fixes:
https://bugs.freedesktop.org/show_bug.cgi?id=98016

v2: use define for fw version
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

To free fences, call_rcu() is used, which calls amdgpu_fence_free()
after a grace period. During teardown, there is no guarantee all
callbacks have finished, so amdgpu_fence_slab may be destroyed before
all fences have been freed. If we are lucky, this results in some slab
warnings, if not, we get a crash in one of rcu threads because callback
is called after amdgpu has already been unloaded.

Fix it with a rcu_barrier().

Fixes: b4413535 ("drm/amdgpu: RCU protected amdgpu_fence_release")
Acked-by: NChunming Zhou <david1.zhou@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NGrazvydas Ignotas <notasas@gmail.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Looks like .last_flush reference is left at teardown.
Leak reported by CONFIG_SLUB_DEBUG.

Fixes: 41d9eb2c ("drm/amdgpu: add a fence after the VM flush")
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NGrazvydas Ignotas <notasas@gmail.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Flush any outstanding thermal work before tearing down
the dpm driver.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Cancel any pending reset work when we tear down the driver.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NJimQu <Jim.Qu@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

No need to retain previous settings as this is the first time
we set pg_flags.  Probably a copy/paste typo from the CZ code.
Avoids confusion.

No change in behavior as adev is kzallocated.
Reviewed-by: NTom St Denis <tom.stdenis@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

It's technically an MC register so make sure we initialize it
in the MC module rather than the gfx module.  Since other bits
in the same register are used to enable ATC CG features make
sure we apply the golden setting first.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NEvan Quan <evan.quan@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We need this for more than just DCE.  Move it out of the DCE modules
and into the device code.  This way we can be sure the scratch registers
are initialized properly before we run asic_init which happens before
DCE IPs are restored.

Fixes atombios hangs in asic_init.
Reviewed-by: NJimQu <Jim.Qu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This is already handled by the dce IP modules in their
suspend and resume code.  No need to do it again.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This removes the 'write' and 'force' from get_user_pages() and replaces
them with 'gup_flags' to make the use of FOLL_FORCE explicit in callers
as use of this flag can result in surprising behaviour (and hence bugs)
within the mm subsystem.
Signed-off-by: NLorenzo Stoakes <lstoakes@gmail.com>
Acked-by: NChristian König <christian.koenig@amd.com>
Acked-by: NJesper Nilsson <jesper.nilsson@axis.com>
Acked-by: NMichal Hocko <mhocko@suse.com>
Reviewed-by: NJan Kara <jack@suse.cz>
Signed-off-by: NLinus Torvalds <torvalds@linux-foundation.org>

Previous code was just a copy/paste from KV.
Reviewed-by: NTom St Denis <tom.stdenis@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Otherwise, you can't disable dpm.

Tested-by and Reviewed-by: Rex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

IP types are not an index.  Each asic may have number and
type of IPs.  Properly check the the type rather than
using the type id as an index.

v2: fix all the IPs to not use IP type as an idx as well.
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Limit clocks on a specific HD86xx part to avoid
crashes (while awaiting an appropriate PP fix).
Signed-off-by: NTom St Denis <tom.stdenis@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

debugfs_create_file() returns NULL on error, it only returns error
pointers if debugfs isn't enabled in the config and we checked for that
earlier so it can't happen.

Fixes: 4f4824b5 ('drm/amd/amdgpu: Convert ring debugfs entries to binary')
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NDan Carpenter <dan.carpenter@oracle.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

UVD and VCE CG are handled specially, however the previous
fix for this skipped late init for those blocks rather than
just CG.  Just protect the CG function call.  No functional
change since UVD and VCE don't currently utilize a late_init
function.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Sometimes during multiple reboots, the system hangs
during bootup. The issue is very random and happens
once in around 50 reboots or so.

It seems if clockgating is enabled before late init,
the GFX engine sometimes does not respond.

This patch changes the ordering a little so that
both powergating and clockgating are enabled only
after late init calls.
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NArindam Nath <arindam.nath@amd.com>
Tested-by: NSunil Uttarwar <Sunil.Uttarwar1@amd.com>
Reviewed-by: NTom St Denis <tom.stdenis@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Ensure that we really only report a GPU reset if one has happened since the
creation of the context.
Signed-off-by: NNicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

When setting up the RLC, only disable the CGCG and
CGLS bits rather than clearing the entire register
to avoid losing the golden settings.
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

When DisplayPort AUX channel i2c adapter is registered, drm_connector's
kdev member is used as a parent, so we get sysfs structure like:
  /drm/card1/card1-DP-2/i2c-12
Because of that, there is a problem when drm core (and not the driver)
calls drm_connector_unregister(), it removes parent sysfs entries
('card1-DP-2' in our example) while the i2c adapter is still registered.
Later we get a WARN when we try to unregister the i2c adapter:

  WARNING: CPU: 3 PID: 1374 at fs/sysfs/group.c:243 sysfs_remove_group+0x14c/0x150
  sysfs group ffffffff82911e40 not found for kobject 'i2c-12'

To fix it, we can use the .early_unregister hook to unregister the i2c
adapter before drm_connector's sysfs is torn down.
Signed-off-by: NGrazvydas Ignotas <notasas@gmail.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This reverts commit 1dcd32fb9c54334ec948a0f18174a748d6b14364.

The block size is indeed an equal match, so this can cause performance regressions.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

If this happens (and it recently did), we free a structure while part of
it is still in use, which results in non-obvious crashes. The way it's
detached is not trivial (DRM core has to call the connector .destroy
callback and things must be torn down in the right order), so better
detect it and warn early.
Signed-off-by: NGrazvydas Ignotas <notasas@gmail.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

All other amdgpu/dce_v* files have this call, it's only mysteriously
missing from dce_v11_0.c since the file was added and causes leaks.

Fixes: aaa36a97 ("drm/amdgpu: Add initial VI support")
Signed-off-by: NGrazvydas Ignotas <notasas@gmail.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Successful sw_init() and hw_init() states are tracked, but not
late_init(). Various error paths may result in amdgpu_fini() being
called before .late init is done, so late_init needs to be tracked
to avoid unexpected or multiple .late_fini() calls.
Signed-off-by: NGrazvydas Ignotas <notasas@gmail.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Include the CIK asics in the ifdef.
Reviewed-By: NEmily Deng <Emily.Deng@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This is needed for proper VCE DPM on some APUs.

v2: fix the asic list
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>