drm/amdgpu: Fix race condition in amdgpu_mn_unregister

Exchange locking order of adev->mn_lock and mm_sem, so that
rmn->mm->mmap_sem can be taken safely, protected by adev->mn_lock,
when amdgpu_mn_destroy runs concurrently.

Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

drm/amdgpu: cleanup gem init/finit

Remove the double housekeeping and use something sane to
forcefuly delete BOs on unload.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: rework GEM info printing

Print BOs grouped per client.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: print the GPU offset as well in gem_info

To easily find which memory is used.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: optionally print the pin count in gem_info as well

Usefull when debugging page flipping.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: print the BO size only once in amdgpu_gem_info

Splitting it into KB/MB is just confusing.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: print pid as integer

Not sure why somebody thought that this is a long.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove page flip work queue v3

Just use the system queue now that we don't block any more.

v2: handle DAL as well.
v3: agd: split DAL changes out

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Mykola Lysenko <mykola.lysenko@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com> (v1)

drm/amdgpu: stop blocking for page filp fences

Just register an callback and reschedule the work item if necessary.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: stop calling amdgpu_gpu_reset from the flip code

We don't return -EDEADLK any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove fence reset detection leftovers

wait_event() never returns before the fence was signaled.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Fix race condition in MMU notifier release

The release notifier can get called a second time from
mmu_notifier_unregister depending on a race between
__mmu_notifier_release and amdgpu_mn_destroy. Use
mmu_notifier_unregister_no_release to avoid this.

Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>

drm/radeon: Fix WARN_ON if DRM_DP_AUX_CHARDEV is enabled

Rafael Antognolli's new DRM_DP_AUX_CHARDEV feature causes a WARN_ON
if drm_dp_aux->dev == drm_connector->kdev and drm_dp_aux_unregister()
is called after drm_connector_unregister(). radeon is the only driver
affected by this besides i915. (amdgpu calls drm_dp_aux_unregister()
before drm_connector_unregister().)

Cc: Rafael Antognolli <rafael.antognolli@intel.com>
Cc: Ville Syrjälä <ville.syrjala@linux.intel.com>
Cc: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Lukas Wunner <lukas@wunner.de>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/vi: move uvd tiling config setup into uvd code

Split uvd and gfx programming.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/vi: move sdma tiling config setup into sdma code

Split sdma and gfx programming.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/cik: move uvd tiling config setup into uvd code

Split uvd and gfx programming.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/cik: move sdma tiling config setup into sdma code

Split sdma and gfx programming.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx7: rework gpu_init()

Split the sw and hw parts into separate functions.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx: clean up harvest configuration (v2)

Read back harvest configuration from registers and simplify
calculations.  No need to program the raster config registers.
These are programmed as golden registers and the user mode
drivers program them as well.

v2: rebase on Tom's patches

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix coding style in amdgpu_ctx.c

Don't use pointer arithmetic and fix the indentation.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: nuke the kernel context

Not used any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use separate scheduler entity for VCE submissions

This allows us to remove the kernel context and use a better
priority for the submissions.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use separate scheduler entity for UVD submissions

This allows us to remove the kernel context and use a better
priority for the submissions.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use separate scheduler entitiy for buffer moves

This allows us to remove the global kernel context.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use SDMA round robin for VM updates v3

Distribute the load on both rings.

v2: use a loop for the initialization
v3: agd: rebase on upstream

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove is_pte_ring

Not used for anything.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use per VM entity for page table updates (v2)

Updates from different VMs can be processed independently.

v2: agd: rebase on upstream

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove the userptr rmn->lock

Avoid a lock inversion problem by just using the mmap_sem to
protect the entries of the intervall tree.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>

drm/amd/include: Update dce 8 headers for dal

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx7: Fix whitespace

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx7: Simplify wptr/rptr functions

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx7: LOC reduction in gfx_v7_0_setup_rb

Reduce for loop with bitmask to simple complement and mask

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx7: Simplify bitmask creation

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx7: Reduce linecount in table init

Replaces switch statements with direct assignments to
reduce line count significantly.

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add powerplay valid check to avoid null point.

In case CONFIG_DRM_AMD_POWERPLAY is defined and amdgpu.powerplay=0.
some functions in powrplay can also be called by DAL. and the input parameter is *adev.
if just check point not NULL was not enough and will lead to NULL point error.

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

drm/amd/powerplay: Enable low mem pstate when cancel_high

Signed-off-by: Eric Yang <eric.yang2@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: Use correct clock in cz_apply_state_adjust_rules

Signed-off-by: Eric Yang <eric.yang2@amd.com>
Reviewed-by: Eagle Yeh <eagle.yeh@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: get real display device num by cgs interface

Signed-off-by: Rex Zhu <Rex.Zhu@amd.com>
Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: Use engine clock limit calculated by dal

Use min required system clock calculated by dal

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: Make declarations of functions exposed to DAL type-safe.

Signed-off-by: David Rokhvarg <David.Rokhvarg@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: implement functions in carrizo for DAL.

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

drm/amd/powerplay: export interface to DAL.

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

drm/amd/powerplay: change struct name.

amd_pp_dal_clock_info to amd_pp_simple_clock_info.

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

drm/amd: Adding IVSRC register headers

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Move MMIO flip out of spinlocked region

Prior actual  MMIO flip we need to acquire DAL mutex to guard
our target state which get modified on reset mode.
Assign page flip status before actual flip to handle
the possible race condition with interrupt.

Signed-off-by: Vitaly Prosyak <vitaly.prosyak@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Don't crash system if we can't get crtc

Signed-off-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move sync into job object

No need to keep that for every IB.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: send VCE IB tests directly to the ring again

We need the IB test for GPU resets as well and
the scheduler should be stoped then.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: send UVD IB tests directly to the ring again

We need the IB test for GPU resets as well and
the scheduler should be stoped then.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: rename amdgpu_sched.c to amdgpu_job.c

That's probably a better matching name.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: cleanup in kernel job submission

Add a job_alloc_with_ib helper and proper job submission.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: send SDMA/GFX IB tests directly to the ring again

There is no point in sending them through the scheduler.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: directly return fence from ib_schedule

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move ring from IBs into job

We can't submit to multiple rings at the same time anyway.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: make pad_ib a ring function v3

The padding depends on the firmware version and we need that for BO moves as
well, not only for VM updates.

v2: new approach of making pad_ib a ring function
v3: fix typo in macro name

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: cleanup user fence handling in the CS

Don't keep that around twice.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add proper job alloc/free functions

And use them in the CS instead of allocating IBs and jobs separately.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix num_ibs check

Specifying no IBs on command submission is invalid, stop crashing
badly when somebody tries it.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Cc: stable@vger.kernel.org
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove AMDGPU_NUM_SYNCS

Just a leftover from semaphores.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove adev and fence from amdgpu_sync_free

Just leftovers from the semaphores.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucer@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: check userptrs mm earlier

Instead of when we try to bind it check the usermm when
we try to use it in the IOCTLs.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: Avoid double gpu reset by adding a timeout on IB ring tests.

When the radeon driver resets a gpu, it attempts to test whether all the
rings can successfully handle an IB.  If these rings fail to respond, the
process will wait forever.  Another gpu reset can't happen at this point,
as the current reset holds a lock required to do so.  Instead, make all
the IB tests run with a timeout, so the system can attempt to recover
in this case.

While this doesn't fix the underlying issue with card resets failing, it
gives the system a higher chance of recovering.  These timeouts have been
confirmed to help both a Tathi and Hawaii card recover after a gpu reset.

This also adds a new function, radeon_fence_wait_timeout, that behaves like
fence_wait_timeout.  It is used instead of fence_wait_timeout as it continues
to work during a reset.  radeon_fence_wait is changed to be implemented
using this function.

V2:
 - Changed the timeout to 1s, as the default 10s from radeon_wait_timeout was
too long.  A timeout of 100ms was tested and found to be too short.
 - Changed radeon_fence_wait_timeout to behave more like fence_wait_timeout.

Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Matthew Dawson <matthew@mjdsystems.ca>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx: minor code cleanup

Drop needless function wrapper.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: separate pushing CS to scheduler

Move that out of the main IOCTL function.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add amdgpu_set_ib_value helper (v2)

And use it in UVD/VCE command patching.

v2: squash in Christian's fix

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix size estimation for clear IB

We only need a few dw here.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/smu: skip SMC ucode loading on SR-IOV capable boards (v2)

VBIOS does this for us in asic_init.

v2: update iceland as well

Reviewed-by: Monk Liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gmc8: skip MC ucode loading on SR-IOV capable boards

VBIOS does this for us in asic_init.

Reviewed-by: Monk Liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: always repost cards that support SR-IOV

Generally a good idea between VM sessions.  We need a way to
detect VM pass-through in general and always run asic_init in
that case.

Reviewed-by: Monk Liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: track whether the asic supports SR-IOV

Required to make desicions about certain code pathes.

Reviewed-by: Monk Liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add check for atombios GPU virtualization table

This table is found on boards that support SR-IOV.  This will
be used to determine if the board supports SR-IOV and allow
the driver to take specific action in certain cases.

Reviewed-by: Monk Liu <monk.liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove unused function

amdgpu_boot_test_post_card() is not used anywhere.  Probably
a leftover from the original port from radeon.

Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Tom St Denis <tom.stdenis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd: add pm domain for ACP IP sub blocks

ACP IP have internal DMA controller, DW I2S controller and DSPs
as separate power tiles. DMA and I2S devices are added to generic
pm domain, so that entire IP can be powered off/on at appropriate
times. Unused DSPs are made to be powered off though they are powered
on during ACP pm domain power on sequence.

Signed-off-by: Maruthi Bayyavarapu <maruthi.bayyavarapu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd: add ACP driver support

This adds the ACP (Audio CoProcessor) IP driver and wires
it up to the amdgpu driver.  The ACP block provides the DMA
engine for i2s based ALSA driver. This is required for audio
on APUs that utilize an i2s codec.

Reviewed-by: Jammy Zhou <Jammy.Zhou@amd.com>
Reviewed-by: Maruthi Bayyavarapu <maruthi.bayyavarapu@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Murali Krishna Vemuri <murali-krishna.vemuri@amd.com>
Signed-off-by: Maruthi Bayyavarapu <maruthi.bayyavarapu@amd.com>
Signed-off-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove power of two limit for vramlimit

That works with other values as well.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: optimize amdgpu_vm_update_ptes a bit

Don't calculate the end address multiple times.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: cleanup comments in VM code

Neither the global nor the local mutex exists any more and
amdgpu doesn't support cayman.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: optimize VM fencing

No need to fence every page table, just the page directory is enough.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: split VM mappings into smaller operations (v3)

If we can't copy entries from the GTT or fill them with one command split
up the mapping operation into multiple ones.

v2: agd: rebase on upstream
v3: squash in Christian's fix

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use BOs GART instance for mapping addresses v4

That allows the VM code to use GART BOs from other driver instances.

v2: don't use copy optimization for foreign GARTs, that won't work.
v3: some more comment cleanups
v4: agd: rebase on upstream

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: move more logic into amdgpu_vm_map_gart v3

No need to duplicate that code over and over again. Also stop using the
flags to determine if we need to map the addresses.

v2: constify the pages_addr
v3: rebased, fix typo in commit message

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove nonsense IB size checks

Those are just leftovers from the time we wrote the VM
updates directly to the ring.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use WARN_ON_ONCE instead of BUG_ON in the SA

Crashing the system doesn't helps at all. Also properly return
-EINVAL if size or alignment are outside valid ranges.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: drop a dummy wakeup scheduler

since the dependency job is also scheduled by the same
scheduler with the job depended on it, no need to
call wake up scheduler when the dep is scheduled.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd: add dce8 enum register header

This adds the DCE8 enum header.

Acked-by: Harry Wentland <harry.wentland@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add some hwmgr functions for sysfs interface on Tonga

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add some hwmgr functions for sysfs interface on Carrizo

These add the interfaces for manual clock control.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add hwmgr's functions for Fiji sysfs interfaces.

These add the interfaces for manual clock control.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: add some sysfs interfaces for powerplay.

The new sysfs interfaces:
pp_num_states: Read-only, return the number of all pp states, 0 if powerplay is not available.
pp_cur_state: Read-only, return the index number of current pp state.
pp_force_state: Read-write, to write a power state index will switch to selected state forcedly and
enable forced state mode, disable forced state mode. such as "echo >...".
pp_table: Read-write, binary output, to be used to read or write the dpm table, the maximum
file size is 4KB of page size.
pp_dpm_sclk: Read-write, reading will return a dpm levels list, to write an index number will force
powerplay to set the corresponding dpm level.
pp_dpm_mclk: same as sclk.
pp_dpm_pcie: same as sclk.

And add new setting "manual" to the existing interface power_dpm_force_performance_level.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Eric Huang <JinHuiEric.Huang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove rptr checking

With the scheduler enabled we don't need that any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove the ring lock v2

It's not needed any more because all access goes through the scheduler now.

v2: Update commit message.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use a global LRU list for VMIDs

With the scheduler enabled managing per ring LRUs don't
make much sense any more.

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: merge vm_grab_id and vm_fence v2

No need for an extra function any more.

v2: comment cleanups

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: grab VMID before submitting job v5

This allows the scheduler to handle the dependencies on ID contention as well.

v2: grab id only once
v3: use a separate lock for the VMIDs
v4: cleanup after semaphore removal
v5: minor coding style change

Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add VM pointer to id trace

Because of the scheduler all traces come from the same thread now and
can't be distincted otherwise.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: drop hard_reset module parameter

It doesn't currently do anything and there's no need for it
going forward since pci config reset will be required as a
fallback even when we have fine grained reset implemented.

Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add a debugfs property to trigger a GPU reset

Ported from similar code in radeon.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Ken Wang <Qingqing.Wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: post card after hard reset

Posting is required after a pci config reset.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Ken Wang <Qingqing.Wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: clean up asic level reset for VI

Drop soft reset, always use pci config reset.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Ken Wang <Qingqing.Wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: clean up asic level reset for CI

Drop soft reset, always use pci config reset.

Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Ken Wang <Qingqing.Wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>