drm/amdgpu/atom: add support for new DIGxEncoderControl cmd table

New digital encoder setup table for elm/baf.

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

drm/amdgpu/atom: add support for new SetPixelClock table

New version of the SetPixelClock table for elm/baf.  The
new table calculates the pll dividers and handles spread
spectrum calculations and setup.

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

drm/amdgpu/atom: add SetDCEClock helper

New cmd table for ELM/BAF for setting the dispclock or
dprefclock.

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

drm/amdgpu: update atombios.h (v2)

update to internal version 893

v2: Pull in gfx_info changes from 898

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

drm/amdgpu: bump the afmt limit for CZ, ST, Polaris

Fixes array overflow on these chips.

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

drm/amdgpu: use defines for CRTCs and AMFT blocks

Prerequiste for the next patch which ups the limits.

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

drm/amdgpu: add ELM/BAF DCE11 configs (v2)

Add support for the display configuration on elm/baf.

v2: add missing Stoney case

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

drm/amdgpu: add ELM/BAF asic types

New asic types for ellesmere and baffin.

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

drm/amd: add DCE 11.2 register headers

Add register headers for DCE (Display and Composition Engine)
11.2.

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

drm/radeon: Allow setting shader registers using DMA/COPY packet3 on SI.

Mesa uses a COPY_DATA packet to copy the grid size for indirect dispatches
into COMPUTE_USER_DATA_*.

Setting those registers with a SET_SH_REG packet is allowed, not allowing
them with other packets seems like an oversight.

v2: Clarify commit message.

Signed-off-by: Bas Nieuwenhuizen <bas@basnieuwenhuizen.nl>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove sorting of CS BOs

Not needed any more.

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: group BOs by log2 of the size on the LRU v2

This allows us to have small BOs on the LRU before big ones.

v2: fix of by one and list corruption bug

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 apply quirks for now.

This isn't being used so drop it.

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

drm/amdgpu: fix error checking when reuse vmid on same ring

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: only update last_flush when vmid doesn't have other new owner

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: double fence slot

we introduced vmid fence, so one hw submission could produce two fences.

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: enable sdma clockgating on ST

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

drm/amdgpu: enable sdma clockgating on CZ

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

drm/amdgpu/sdma: rename fiji cg functions

They care common for all sdma 3.0 parts

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

drm/amdgpu: enable gmc clockgating for ST

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

drm/amdgpu: enable gmc clockgating for CZ

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

drm/amdgpu/vi: rename fiji cg functions

They can be used for other VI parts.

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

drm/amdgpu: enable gfx clockgating for ST (v2)

v2: just enable MGCG for now since CGCG causes hangs

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

drm/amdgpu: enable gfx clockgating for CZ

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

drm/amdgpu/gfx: rework fiji cg functions so they can be shared

They can be shared with other asics with minor modifications.

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

drm/amdgpu: add a new set of rlc function pointers

Different asics tend to have different ways to interact
with the RLC.  This just covers enter/exit of safe mode
for updating CG and PG state, but could be extended to
cover other RLC operations in the future if necessary.

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

drm/amdgpu/gfx: adjust gfx_v8_0_send_serdes_cmd for ST

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

drm/amdgpu/gfx8: rename send_serdes_cmd

So it can be shared with CZ/ST.

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

drm/amdgpu/gmc: add proper CG flags for fiji

We were already enabling these CG features, this uses
the standard interface for doing so.

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

drm/amdgpu/common: add proper CG flags for fiji

We were already enabling these CG features, this uses
the standard interface for doing so.

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

drm/amdgpu/sdma: add proper CG flags for fiji

We were already enabling these CG features, this uses
the standard interface for doing so.

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

drm/amdgpu/gfx: add proper CG flags for fiji

We were already enabling these CG features, this uses
the standard interface for doing so.

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

drm/amdgpu: add new CG flag for ROM clockgating

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

drm/ttm: implement LRU add callbacks v2

This allows fine grained control for the driver where to add a BO into the LRU.

v2: fix typo in comment

Reviewed-by: Sinclair Yeh <syeh@vmware.com>
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/ttm: add optional LRU removal callback v2

Useful for driver specific LRU handling.

v2: fix typo in comment

Reviewed-by: Sinclair Yeh <syeh@vmware.com>
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/ttm: remove unused validation sequence

Not used any more.

Reviewed-by: Sinclair Yeh <syeh@vmware.com>
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/ttm: remove lazy parameter from ttm_bo_wait

Not used any more.

Reviewed-by: Sinclair Yeh <syeh@vmware.com>
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/ttm: remove use_ticket parameter from ttm_bo_reserve

Not used any more.

Reviewed-by: Sinclair Yeh <syeh@vmware.com>
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/ttm: don't wait for BO on initial allocation

When we use an extern reservation object that otherwise waits for every
fence registered with it.

Reviewed-by: Sinclair Yeh <syeh@vmware.com>
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 the coding style in amdgpu_ring.c

No functional change.

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 the ring name for debugfs (v2)

Instead of hard coding just another name in the ring code.

v2: squash in Tom's rebase 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: reduce the ring size for SDMA

Those are way too large.

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: reduce the ring size for GFX

Those are way too large.

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 max_dw in ring_init

Instead of specifying the total ring size calculate that from the maximum
number of dw a submission can have and the number of concurrent submissions.

This fixes UVD with 8 concurrent submissions or more.

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/amd/powerplay: fix fan speed percent setting error on Fiji

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

drm/amd/powerplay: fix fan speed percent setting error on Tonga

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

drm/ttm: Fix TTM BO accounting

TTM BO accounting is out of sync with how memory is really allocated
in ttm[_dma]_tt_alloc_page_directory. This resulted in excessive
estimated overhead with many small allocations.

ttm_dma_tt_alloc_page_directory makes a single allocation for three
arrays: pages, DMA and CPU addresses. It uses drm_calloc_large, which
uses kmalloc internally for allocations smaller than PAGE_SIZE.
ttm_round_pot should be a good approximation of its memory usage both
above and below PAGE_SIZE.

Reviewed-by: Thomas Hellstrom <thellstrom@vmware.com>
Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
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/amd: make a type-safe cgs_device struct. (v2)

This is just a type-safety things to avoid everyone taking void *,
it doesn't change anything.

v2: agd5f: split out the dal changes into a separate patch.

Signed-off-by: Dave Airlie <airlied@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: handle more than 10 UVD sessions (v2)

Change History
--------------

v2:
- Make firmware version check correctly. Firmware
  versions >= 1.80 should all support 40 UVD
  instances.
- Replace AMDGPU_MAX_UVD_HANDLES with max_handles
  variable.

v1:
- The firmware can handle upto 40 UVD sessions.

Signed-off-by: Arindam Nath <arindam.nath@amd.com>
Signed-off-by: Ayyappa Chandolu <ayyappa.chandolu@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd: make some function-local tables static const

These tables were initialized on stack on each call, avoid that
and save a little bit of text size.

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

drm/amd/powerplay: mark phm_master_table_* structs as const

Also adjust phm_construct_table to take a const pointer

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

drm/amd/powerplay: Mark pem_event_action chains as const

As these arrays were of pointer to pointer type, they were
pointer to pointer to const. Make them pointer to const
pointer to const.

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

drm/amdgpu: Mark all instances of struct drm_info_list as const

All these are compile time constand and the
drm_debugfs_create/remove_files functions take a const
pointer argument.

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

drm/amd/scheduler: Mark amdgpu_sched_ops const

This marks the struct amdgpu_sched_ops const and
adjusts amd_sched_init to take a const pointer
for the ops param. The ops member of
struct amd_gpu_scheduler is also changed to const.

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

drm/amd: Mark some tables as const

This patch marks some compile-time constant tables 'const'.
The tables marked in this patch are the low hanging fruit
where little other changes were necesary to avoid casting
away constness etc. Also mark some tables that are private
to a file as static.

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

drm/amd/powerplay: fix stutter setup in mclk level init

Stale ifdef.

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

drm/radeon: add support for SET_APPEND_CNT packet3 (v2)

This adds support to the command parser for the set append counter
packet3, this is required to support atomic counters on
evergreen/cayman GPUs.

v2: fixup some of the hardcoded numbers with real register names
(Christian)

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

drm/radeon: handle more than 10 UVD sessions

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Arindam Nath <arindam.nath@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: add support for loading new UVD fw

Signed-off-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Arindam Nath <arindam.nath@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: reuse VMIDs already assigned to a process

If we don't need to flush we can easily use another VMID
already assigned to the process.

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: add a fence after the VM flush

This way we can track when the flush is done.

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: use a sync object for VMID fences v2

v2: rebase & cleanup

This way we can store more than one fence as user for each VMID.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com> (v1)
Reviewed-by: Chunming Zhou <david1.zhou@amd.com> (v1)
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: merge VM manager and VM context ID structure

No need to have two of them any more.

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: Set PFLIP_SUBMITTED for crtc after address update

Also add some pflip debug prints.

This change allows us to wait on pflip status until the new surface address
is actually submitted to the register.

This reverts ed3020e923240829dcdfd3343f6e91dc02c63775
drm/amdgpu: Move MMIO flip out of spinlocked region
The original change assumed DAL will aquire locks inside DAL
implemetion of page_flip callback which eventaully didn't happen.

This moves the flip before status update which makes sense for the
non-DAL code pathes as well.

Signed-off-by: Andrey Grodzovsky <Andrey.Grodzovsky@amd.com>
Reviewed-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/radeon: Support DRM_MODE_PAGE_FLIP_ASYNC

When this flag is set, we program the hardware to execute the flip
during horizontal blank (i.e. for the next scanline) instead of during
vertical blank (i.e. for the next frame).

Currently this is only supported on ASICs which have a page flip
completion interrupt (>= R600), and only if the use_pflipirq parameter
has value 2 (the default).

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

drm/amd/powerplay: add deep sleep divider id into DPM table on Tonga

Add a proper implementation for setting the deep sleep divider.

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

drm/amdgpu: optionally enable GART debugfs file

Keeping the pages array around can use a lot of system memory
when you want a large GART.

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 GART page addr array

Not needed any more.

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 BO pages instead of GART array

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: change parameter passing in the VM code

Make it more flexible by passing src and page addresses
directly instead of the structures they contain.

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 the GTT power of two limit

As far as I can see that isn't neccessary any more.

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/amd/powerplay: use min_clock_in_sr for deep sleep feature.

This comes from the display handling code.

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/amdgpu: Enable clockgating in UVD6 for Stoney

This patch enables clockgating for the UVD6 block in Stoney.

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

drm/amd/amdgpu: Enable clockgating for UVD5 on Tonga

This patch enables clock gating for the UVD5 block with
Tonga.

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

drm/amd/amdgpu: Add SW clock gating support to UVD 5 and 6

This patch adds support for software clock gating to UVD 5
and UVD 6 blocks with a preliminary commented out hardware
gating routine.

Currently hardware gating does not work so it's not activated.

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

drm/radeon: don't include RADEON_HPD_NONE in HPD IRQ enable bitsets

The values of all but the RADEON_HPD_NONE members of the radeon_hpd_id
enum transform 1:1 into bit positions within the 'enabled' bitset as
assembled by evergreen_hpd_init():

  enabled |= 1 << radeon_connector->hpd.hpd;

However, if ->hpd.hpd happens to equal RADEON_HPD_NONE == 0xff, UBSAN
reports

  UBSAN: Undefined behaviour in drivers/gpu/drm/radeon/evergreen.c:1867:16
  shift exponent 255 is too large for 32-bit type 'int'
  [...]
  Call Trace:
   [<ffffffff818c4d35>] dump_stack+0xbc/0x117
   [<ffffffff818c4c79>] ? _atomic_dec_and_lock+0x169/0x169
   [<ffffffff819411bb>] ubsan_epilogue+0xd/0x4e
   [<ffffffff81941cbc>] __ubsan_handle_shift_out_of_bounds+0x1fb/0x254
   [<ffffffffa0ba7f2e>] ? atom_execute_table+0x3e/0x50 [radeon]
   [<ffffffff81941ac1>] ? __ubsan_handle_load_invalid_value+0x158/0x158
   [<ffffffffa0b87700>] ? radeon_get_pll_use_mask+0x130/0x130 [radeon]
   [<ffffffff81219930>] ? wake_up_klogd_work_func+0x60/0x60
   [<ffffffff8121a35e>] ? vprintk_default+0x3e/0x60
   [<ffffffffa0c603c4>] evergreen_hpd_init+0x274/0x2d0 [radeon]
   [<ffffffffa0c603c4>] ? evergreen_hpd_init+0x274/0x2d0 [radeon]
   [<ffffffffa0bd196e>] radeon_modeset_init+0x8ce/0x18d0 [radeon]
   [<ffffffffa0b71d86>] radeon_driver_load_kms+0x186/0x350 [radeon]
   [<ffffffffa03b6b16>] drm_dev_register+0xc6/0x100 [drm]
   [<ffffffffa03bc8c4>] drm_get_pci_dev+0xe4/0x490 [drm]
   [<ffffffff814b83f0>] ? kfree+0x220/0x370
   [<ffffffffa0b687c2>] radeon_pci_probe+0x112/0x140 [radeon]
   [...]
  =====================================================================
  radeon 0000:01:00.0: No connectors reported connected with modes

At least on x86, there should be no user-visible impact as there

  1 << 0xff == 1 << (0xff & 31) == 1 << 31

holds and 31 > RADEON_MAX_HPD_PINS. Thus, this patch is a cosmetic one.

All of the above applies analogously to evergreen_hpd_fini(),
r100_hpd_init(), r100_hpd_fini(), r600_hpd_init(), r600_hpd_fini(),
rs600_hpd_init() and rs600_hpd_fini()

Silence UBSAN by checking ->hpd.hpd for RADEON_HPD_NONE before oring it
into the 'enabled' bitset in the *_init()- or the 'disabled' bitset in
the *_fini()-functions respectively.

Signed-off-by: Nicolai Stange <nicstange@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: refine code for code style.

White space fix.

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

drm/amdgpu: No need to stop hw init although vce's state was not true.

This is not a fatal error.

v2: add comment why ignore the error here.

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

drm/amdgpu: fix issue that can't set vce clock gate.

Need to soft reset VCE as part of the clockgating
sequence.

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

drm/amdgpu: use ref to keep job alive

this is to fix fatal page fault error that occured if:
job is signaled/released after its timeout work is already
put to the global queue (in this case the cancel_delayed_work
will return false), which will lead to NX-protection error
page fault during job_timeout_func.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: rework TDR in scheduler (v2)

Add two callbacks to scheduler to maintain jobs, and invoked for
job timeout calculations. Now TDR measures time gap from
job is processed by hw.

v2:
fix typo

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: get rid of incorrect TDR

original time out detect routine is incorrect, cuz it measures
the gap from job scheduled, but we should only measure the
gap from processed by hw.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: put job to list before done

the mirror_list will be used for later time out detect
feature.  This is needed to properly detect a GPU
timeout with the scheduler.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: delay job free to when it's finished (v2)

for those jobs submitted through scheduler, do not
free it immediately after scheduled, instead free it
in global workqueue by its sched fence signaling
callback function.

v2:
call uf's bo_undef after job_run()
call job's sync free after job_run()
no static inline __amdgpu_job_free() anymore, just use
kfree(job) to replace it.

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Reviewed-by: Chunming Zhou <david1.zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/dce11: fix vertical bars appear on monitor

Fixed mc stop and resume hardware programming sequence.

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

drm/amdgpu: use sched_job_init to initialize sched_job

Consolidate job initialization in one place rather than
duplicating it in multiple places.

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

drm/amdgpu: patch cond exec for SDMA

More ground work for conditional execution on SDMA
necessary for preemption.

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/amdgpu: support cond exec

This adds the groundwork for conditional execution on
SDMA which is necessary for preemption.

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/amdgpu: improve vmid assigment V2

V2: the signaled items on the LRU maintain their order

Signed-off-by: Chunming Zhou <David1.Zhou@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: mark amdgpu_allowed_register_entry tables as 'const'

Signed-off-by: Nils Wallménius <nils.wallmenius@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: do not store bios_header_start in amdgpu_device

It is only used locally in amdgpu_get_bios

Signed-off-by: Nils Wallménius <nils.wallmenius@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: delete unused struct member suspend from radeon_device

Signed-off-by: Nils Wallménius <nils.wallmenius@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: delete unused struct member suspend from amdgpu_device

Signed-off-by: Nils Wallménius <nils.wallmenius@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: hard reset r600 and newer GPU when hibernating.

Some GPU block like UVD and VCE require hard reset to be properly
resume if there is no real powerdown of the asic like during various
hibernation step. This patch perform such hard reset.

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

drm/radeon: allow to force hard GPU reset.

In some cases, like when freezing for hibernation, we need to be
able to force hard reset even if no engine are stuck. This patch
add a bool option to current asic reset callback to allow to force
hard reset on asic that supports it.

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

drm/radeon: add driver option to disable vce block.

Quite few suspend/hibernation bugs are related to this block. Add
an option to disable those as a work around.

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

drm/radeon: add driver option to disable uvd block.

Quite few suspend/hibernation bugs are related to this block. Add
an option to disable those as a work around.

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

drm/radeon: consolidate cik vce initialization and startup code.

This match the exact same control flow as existing code. It just
use goto instead of multiple levels of if/else. It also clarify
early initialization failures by clearing rdev->has_vce doing so
does not change end result from hardware point of view, it only
avoids printing more error messages down the line and thus only
the original error is reported.

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

drm/radeon: consolidate si vce initialization and startup code.

This match the exact same control flow as existing code. It just
use goto instead of multiple levels of if/else. It also clarify
early initialization failures by clearing rdev->has_vce doing so
does not change end result from hardware point of view, it only
avoids printing more error messages down the line and thus only
the original error is reported.

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

drm/radeon: consolidate ni vce initialization and startup code.

This match the exact same control flow as existing code. It just
use goto instead of multiple levels of if/else. It also clarify
early initialization failures by clearing rdev->has_vce doing so
does not change end result from hardware point of view, it only
avoids printing more error messages down the line and thus only
the original error is reported.

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