drm/amd/powerplay: tidy up ret checks in amd_powerplay.c (v3)

v2: squash in regression fix (Rex)
v3: Squash in regression fix (Rex)

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: refine code in amd_powerplay.c (v2)

1. use flag PP_DPM_DISABLED within powerplay
   notify amdgpu dpm state by cgs interface.
2. delete redundant virtualization check in
   powerplay

v2: squash in fix for hwmgr_init (Rex)

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>

amdgpu/pp: rewrite fiji pwr virus upload code.

Along the same lines as rewriting the polaris code, this rewrites
the fiji code, and reduces the driver size by ~40k.

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

amdgpu/pp: rewrite polaris pwrvirus upload code.

This reduces the pwrvirus table size by 30k, by moving the
sequences of writes to the data register into blocks.

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

amdgpu/nbio: use constant nbio_hdp_flush_reg structs.

This removes the init path as well, since the init path
just did some constant init of some structs.

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

amdgpu/soc15: make the pcie index/data registers constant.

These don't seem to change at runtime, and the initialisers
are constant data. This could be improved by not selecting
the apu/non-apu path on each pcie read/write access.

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

amdgpu/pp: constify soft_dummy_pp_table.

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

amdgpu/pp: use array_size to size the pwrvirus tables.

This avoids fragile hardcoding of array size.

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

amdgfx/gfx: don't use static objects for ce/de meta. (v2)

This isn't safe if we have multiple GPUs plugged in, since
there is only one copy of this struct in the bss, just allocate
on stack, it's 40/108 bytes which should be safe.

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/amdgpu: Add a new flag for SR-IOV to share memory between PF & VF

Add ATOM_VRAM_BLOCK_SRIOV_MSG_SHARE_RESERVATION to identify whether
driver need to reserve VRAM for SR-IOV shared memory.

Signed-off-by: Horace Chen <horace.chen@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: export new interfaces in amd_pm_funcs

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/amdgpu: add comments in struct amd_pm_funcs define

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 dmesg log level in powerplay

Use pr_debug to prevent spamming unimportant dmesg.

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/amdgpu: create powerplay by cgs interface

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/amdgpu: add cgs interface to register pp handle

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/amdgpu: delete pp_enable in adev

amdgpu not care powerplay or dpm is enabled.
just check ip functions and pp functions

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/amdgpu: delete dead code about fw load check

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: fix memory leak in powerplay

cgs device not free.

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/amdgpu: move amdgpu_ucode_init_bo to amdgpu_device.c

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/amdgpu: move common pm sysfs code to amdgpu_device.c

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/amdgpu: Handle GPUVM fault storms

When many wavefronts cause VM faults at the same time, it can
overwhelm the interrupt handler and cause IH ring overflows before
the driver can notify or kill the faulting application.

As a workaround I'm introducing limited per-VM fault credit. After
that number of VM faults have occurred, further VM faults are
filtered out at the prescreen stage of processing.

This depends on the PASID in the interrupt packet, so it currently
only works for KFD contexts.

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

drm/amd/chash: Fix typo

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

drm/amd/powerplay: fix spelling mistake: "dividable" -> "divisible"

Trivial fix to spelling mistakes in pr_err error message and ASSERT
messages.

Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

amdgpu/pp: reduce size of vega10_fuses_default

I've no idea why this is like this, why store 64-bit fields
as a string, and then parse the strings, this is just over
engineered.

Reduce the size of the amdgpu.o by 80k.

   text    data     bss     dec     hex filename
1331332   17982    1008 1350322  149ab2 amdgpu.o
1244668   17982    1008 1263658  13482a amdgpu.o

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

drm/amdgpu: use designated initialiser for thermal_irq_src.

This fixes the 0-day build warning.

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

amdgpu/pp: slim down the pwr virus tables.

This is what I'd call slightly overengineered, we waste 40k on
storing a value that is write or end, when we could just use the
register value to denote end.

Remove the virus command parameter, and save
   text    data     bss     dec     hex filename
1412724   17982    1008 1431714  15d8a2 ../drm-next-build/drivers/gpu/drm/amd/amdgpu/amdgpu.o
1331332   17982    1008 1350322  149ab2 ../drm-next-build/drivers/gpu/drm/amd/amdgpu/amdgpu.o

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

amdgpu/pp: move amdgpu_fuses_default into static const.

There is no reason that this gets passed back into the function
from outside the file, just reference the table directly.

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

amdgpu/pp: move PhwVega10_Magic to static const.

This isn't used outside this file.

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

amdgpu/pp: remove ci_smc/smumgr split.

This split serves no purpose, and we can make a bunch of functions
static now.

There are lots of cases of this sort of split in the powerplay code,
please start cleaning them up. Ideally the function table is in the
same file as all the implementations used in it.

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

drm/amdgpu/vega10: static constify channel_number

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

drm/amdgpu/pp: constify some powerplay tables

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

amdgpu/powerplay: constify large struct

This moves this from being global data to global rodata, I'm
sure it would be easy to move it to being local data.

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

amdgpu: don't ask about CHASH just default it for now.

If we bump this up a level, we can ask about it, for now,
just default to what amdgpu does.

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

drm/amdgpu: map compute rings by least recently used pipe

This patch provides a guarantee that the first n queues allocated by
an application will be on different pipes. Where n is the number of
pipes available from the hardware.

This helps avoid ring aliasing which can result in work executing in
time-sliced mode instead of truly parallel mode.

Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add option for force enable multipipe policy for compute

Useful for testing the effects of multipipe compute without recompiling.

Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use multipipe compute policy on non PL11 asics

A performance regression for OpenCL tests on Polaris11 had this feature
disabled for all asics.

Instead, disable it selectively on the affected asics.

Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Andres Rodriguez <andresx7@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix vf error handling

The error handling for virtual functions assumed a single
vf per VM and didn't properly account for bare metal.  Make
the error arrays per device and add locking.

Reviewed-by: Gavin Wan <gavin.wan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: clarify license in amdgpu_trace_points.c

It was not clear.  The rest of the driver is MIT/X11.

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

drm/amdgpu: Add gem_prime_mmap support

v2: drop hdp invalidate/flush.
v3: honor pgoff during prime mmap. Add a barrier after cpu access.
v4: drop begin/end_cpu_access() for now, revisit later.

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

drm/amd/powerplay: delete dead code in smumgr

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: delete SMUM_FIELD_MASK

repeated defining in hwmgr.h

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: delete SMUM_WAIT_INDIRECT_FIELD

repeated defining in hwmgr.h
use PHM_WAIT_INDIRECT_FIELD instand.

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: delete SMUM_READ_FIELD

repeated defining in hwmgr.h

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: delete SMUM_SET_FIELD

repeated defining in hwmgr.h

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: delete SMUM_READ_VFPF_INDIRECT_FIELD

repeated defining in hwmgr.h

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: delete SMUM_WRITE_VFPF_INDIRECT_FIELD

repeated defining in hwmgr.h

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: delete SMUM_WRITE_FIELD

the macro is as same as PHM_WRITE_FIELD

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: delete SMU_WRITE_INDIRECT_FIELD

the macro is as same as PHM_WRITE_INDIRECT_FIELD

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: move macros to hwmgr.h

the macro is not relevant to SMU,
so rename SMU_WAIT_FIELD_UNEQUAL to
PHM_WAIT_FIELD_UNEQUAL and move to hwmgr.h

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: move PHM_WAIT_VFPF_INDIRECT_FIELD to hwmgr.h

the macro is not relevant to SMU, so move to hwmgr.h
and rename to PHM_WAIT_VFPF_INDIRECT_FIELD

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: move SMUM_WAIT_VFPF_INDIRECT_FIELD_UNEQUAL to hwmgr.h

the macro is not relevant to SMU, so move to hwmgr.h
and rename to PHM_WAIT_VFPF_INDIRECT_FIELD_UNEQUAL

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: move SMUM_WAIT_INDIRECT_FIELD_UNEQUAL to hwmgr.h

the macro is not relevent to SMU, so move to hwmgr.h
and rename to PHM_WAIT_INDIRECT_FIELD_UNEQUAL

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: add new helper functions in hwmgr.h

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: use SMU_IND_INDEX/DATA_11 pair

in VFPF macros to support virtualization

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: refine powerplay code.

delete struct smumgr, put smu backend function table
in struct hwmgr

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: delete dead code in hwmgr.h

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: refine interface in struct pp_smumgr_func

unify to use struct hwmgr as function parameter in
smumgr.

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/amdgpu: simplify pinning into visible VRAM

Just set the CPU access required flag when we pin it.

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 firmware memoryleak(v2)

this fix memory leak due to request_firmware after driver
unloaded

v2:
release gmc firmware for gmc6/7/8 as well

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

drm/amdgpu:fix uvd ring fini routine(v2)

fix missing finish uvd enc_ring.
v2:
since the adev pointer check in already in ring_fini
so drop the check outsider

Signed-off-by: Monk Liu <Monk.Liu@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/amdgpu/sriov:alloc KIQ MQD in VRAM(v2)

this way after KIQ MQD released in drv unloading, CPC
can still let KIQ access this MQD thus RLCV SAVE_VF
will not fail

v2:
always use VRAM domain for KIQ MQD no matter BM or SRIOV

Signed-off-by: Monk Liu <Monk.Liu@amd.com>
Acked-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:unmap KCQ in gfx hw_fini(v2)

v2:
move kcq_disable out of SRIOV, make it genearal

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

drm/amdgpu:halt when vm fault

only with this way we can debug the VMC page fault issue

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: Add copy_pte_num_dw member in amdgpu_vm_pte_funcs

Use it to replace the hard coded value in amdgpu_vm_bo_update_mapping().

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

drm/amdgpu: Fix a bug in amdgpu_fill_buffer()

When max_bytes is not 8 bytes aligned and bo size is larger than
max_bytes, the last 8 bytes in a ttm node may be left unchanged.
For example, on pre SDMA 4.0, max_bytes = 0x1fffff, and the bo size
is 0x200000, the problem will happen.

In order to fix the problem, we separately store the max nums of
PTEs/PDEs a single operation can set in amdgpu_vm_pte_funcs
structure, rather than inferring it from bytes limit of SDMA
constant fill, i.e. fill_max_bytes.

Together with the fix, we replace the hard code value "10" in
amdgpu_vm_bo_update_mapping() with the corresponding values from
structure amdgpu_vm_pte_funcs.

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

drm/amdgpu: Correct bytes limit for SDMA 3.0 copy and fill

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

drm/amdgpu: use 2MB fragment size for GFX6,7 and 8

Use 2MB fragment size by default for older hardware generations as well.

Signed-off-by: Christian König <christian.koenig@amd.com>
Acked-by: John Bridgman <john.bridgman@amd.com>
Reviewed-by: Roger He <Hongbo.He@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Fix driver reloading failure

SRIOV doesn't implement PMC capability of PCIe, so it can't update
power state by reading PMC register.

Currently, amdgpu driver doesn't disable pci device when removing
driver, the enable_cnt of pci device will not be decrease to 0.
When reloading driver, pci_enable_device will do nothing as
enable_cnt is not zero. And power state will not be updated as PMC
is not support.
So current_state of pci device is not D0 state and pci_enable_msi
return fail.

Add pci_disable_device when remmoving driver to fix the issue.

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

drm/amd/powerplay: refine phm_register_thermal_interrupt interface

currently, not all asics implement this callback function
so not return error to avoid powerplay initialize failed
in those asices

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/amdgpu: add vega10/raven mmhub/athub golden settings

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: change alert temperature range

Change to more meaningful range that triggers thermal
interrupts.

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: implement register thermal interrupt for Vega10

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: add register thermal interrupt in hwmgr_hw_init

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/amdgpu: add cgs query info of pci bus devfn

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/amdgpu: Partial revert of iova debugfs

We discovered that on some devices even with iommu enabled
you can access all of system memory through the iommu translation.

Therefore, we revert the read method to the translation only service
and drop the write method completely.

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

drm/amd/amgpu: update vega10 sdma golden setting

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amgpu: update raven sdma golden setting

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Junwei Zhang <Jerry.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/sriov:fix memory leak after gpu reset

GPU reset will require all hw doing hw_init thus
ucode_init_bo will be invoked again, which lead to
memory leak

skip the fw_buf allocation during sriov gpu reset to avoid
memory leak.

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:make ctx_add_fence interruptible(v2)

otherwise a gpu hang will make application couldn't be killed
under timedout=0 mode

v2:
Fix memoryleak job/job->s_fence issue
unlock mn
remove the ERROR msg after waiting being interrupted

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/sriov:init csb for gfxv9

RLC need CSB registers initiated under SRIOV during world switch
otherwise the clear state buffer behav will not be recovered to
current VF scheme after switch back

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

drm/amdgpu/sriov:increate mailbox polling timeout

increase timeout to 12 seconds,because there may have multiple
FLR waiting for done, the waiting time of events may be long,
increase to 12s to reduce timeout failure.

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

drm/amdgpu/sriov:fix page fault issue of driver unload

bo_free on csa is too late to put in amdgpu_fini because that
time ttm is already finished,
Move it earlier to avoid the page fault.

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

drm/amdgpu:use formal register to trigger hdp invalidate

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:hdp flush should be put it initialized

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

drm/amdgpu:insert TMZ_BEGIN

FRAME_CONTROL(begin) is needed for vega10 due to ucode logic change,
it can fix some CTS random fail under gfx preemption enabled mode.

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

drm/amdgpu/sriov:don't load psp fw during gpu reset

At least for SRIOV we found reload PSP fw during
gpu reset cause PSP hang.

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

drm/amdgpu/sriov:move in_reset to adev and rename

currently in_reset is only used in sriov gpu reset, and it
will be used for other non-gfx hw component later, like
PSP, so move it from gfx to adev and rename to in_sriov_reset
make more sense.

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:no kiq in IH

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/sriov:fix missing error handling

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: Add GPU reset functionality for Vega10

V2

Signed-off-by: Ken Wang <Ken.Wang@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/ttm: Remove TTM dma tracepoint since it's not required anymore

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

drm/amd/amdgpu: remove usage of ttm trace

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

drm/amd/amdgpu: add support for iova_to_phys to replace TTM trace (v5)

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(v2): Add domain to iova debugfs
(v3): Add true read/write methods to access system memory of pages
      mapped to the device
(v4): Move get_domain call out of loop and return on error
(v5): Just use kmap/kunmap

drm/amd/amdgpu: Fold TTM debugfs entries into array (v2)

Signed-off-by: Tom St Denis <tom.stdenis@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(v2): add domains and avoid strcmp

drm/amdgpu: fix checkpatch.pl warning to amdgpu_drv.c

fix checkpatch.pl WARNING:
Prefer 'unsigned int' to bare use of 'unsigned'

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

drm/amdgpu: remove the clearance of vce 4.0 interrupt mask

Requested by SRIOV, the clearance of the bit moved into firmware

Signed-off-by: Leo Liu <leo.liu@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/sdma3: set wptr shadow atomically

Port it from sdma4 for wptr polling usage.

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

drm/amdgpu/sdma3: Enable sdma wptr polling for SRIOV

When hypervisor triggering FLR for one of VFs, need to enable sdma
wptr polling to avoid missing wptr update if enabling doorbell.

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

drm/ttm: cleanup ttm_page_alloc_dma.c

Remove unused defines and variables. Also stop computing the
gfp_flags when they aren't used.

No intended 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>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Track pending retry faults in IH and VM (v2)

IH tracks pending retry faults in a hash table for fast lookup in
interrupt context. Each VM has a short FIFO of pending VM faults for
processing in a bottom half.

The IH prescreening stage adds retry faults and filters out repeated
retry interrupts to minimize the impact of interrupt storms.

It's the VM's responsibility remove pending faults once they are
handled. For now this is only done when the VM is destroyed.

v2:
- Made the hash table smaller and the FIFO longer. I never want the
  FIFO to fill up, because that would make prescreen take longer.
  128 pending page faults should be enough to keep migrations busy.

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