Signal a fault to the scheduler on an illegal instruction or register
access violation instead of kicking of the reset handler directly.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Problem: After GPU reset on dGPUs with gfx8 compute ring
1.0.0 fails to pass the ring test. Ring registers inspection
shows that it's active and no hang is observed (rptr == wptr)
No significant diffs were observed between CP_HQD* registers
for the ring in good and bad shape.

Fix: No clear reason why but reversing the order of ring tests
fixes the problem.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Extract the function of fw loading out of powerplay.
Do fw loading between hw_init/resuem_phase1 and phase2
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

gfx and sdma can be initialized before smu.
Reviewed-by: NEvan Quan <evan.quan@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

AMDGPU_FW_LOAD_DIRECT is used for bring up.
Now it don't work any more. so remove the support.

v2: Add warning message if user select
   AMDGPU_FW_LOAD_DIRECT/AMDGPU_FW_LOAD_PSP on VI.
Reviewed-by: NEvan Quan <evan.quan@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

[Why]
1. we never submit IBs to the KIQ
2. there seems to be ~500ms delay during amdgpu resume spent in KIQ,
   hence pointing toward interrupts are not working correctly.

[How]
remove interrupt handling for KIQ.
Signed-off-by: NShirish S <shirish.s@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>i
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Move in_suspend flag to adev from gfx, so
can be used in other ip blocks, also keep
consistent with gpu_in_reset flag.
Reviewed-by: NEvan Quan <evan.quan@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Everything that isn't related to the IH ring.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NHuang Rui <ray.huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

the CG related registers have been programed in golden setting
PG register default value is 0.
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NHang Zhou <hang.zhou@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

before halt rlc/cp, need to
1. enter rlc safe mode
2. wait rlc/cp idle
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NHang Zhou <hang.zhou@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

change function name gfx_v6/7/8/9_0_gpu_init to
gfx_v6/7/8/9_0_constants_init.
this function is just for init gfx constants such
as max pipes, render backends...
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

We don't need that in the per ASIC code.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

That only worked by pure coincident. Completely remove the shifting and
always apply correct PAGE_SHIFT.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

adev->gfx.rlc has the values from rlc_hdr already processed by
le32_to_cpu.  Using the rlc_hdr values on big-endian machines causes
a kernel Oops due to writing well outside of the array (0x24000000
instead of 0x24).
Signed-off-by: NA. Wilcox <AWilcox@Wilcox-Tech.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

1. initialize kiq before initialize gfx ring.
2. set kiq ring ready immediately when kiq initialize
   successfully.
3. split function gfx_v8_0_kiq_resume into two functions.
   gfx_v8_0_kiq_resume is for kiq initialize.
   gfx_v8_0_kcq_resume is for kcq initialize.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Send all kcq unmap_queue packets and then wait for
complete.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

There are no any logical changes here.

1. if kcq can be enabled via kiq, we don't need to
   do kiq ring test.
2. amdgpu_ring_test_ring function can be used to
   sync the ring complete, remove the duplicate code.

v2: alloc 6 (not 7) dws for unmap_queues
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This is required by gfx hw and can fix the rlc hang when
do s3 stree test on Cz/St.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NHang Zhou <hang.zhou@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Try to kill waves on the SQ.

v2: only for the GFX ring for now.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NHuang Rui <ray.huang@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This is required by gfx hw and can fix the rlc hang when
do s3 stree test on Cz/St.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NHang Zhou <hang.zhou@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Unify to set power ungate state at the begin of suspend/fini.
Remove the workaround code for gfx off feature in
amdgpu_device.c.
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

v2: Rebase
v3: Use defines for CP_SQ and CP_ECC_ERROR interrupts.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

1. add amdgpu_device_ip_late_set_pg_state function for
   set pg state.
2. delete duplicate pg state setting on gfx_v8_0's late_init.
Reviewed-by: NEvan Quan <evan.quan@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

gfx ip block can call set_powergating_by_smu to set gfx pg state if
necessary.
Reviewed-by: NEvan Quan <evan.quan@amd.com>
Signed-off-by: NRex Zhu <Rex.Zhu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Access to SQ_EDC_INFO requires selecting register instance and
hence mutex lock when accessing GRBM_GFX_INDEX for which a work
is schedueled from IH. But SQ interrupt can be raised on many instances
at once which means queuing work will usually succeed for the first one
but fail for the rest since the work takes time to process. To avoid
losing info about other interrupt instances call the parsing function
directly from high IRQ when current work hasn't finished and avoid
accessing SQ_EDC_INFO in that case.

v2:
Simplify high IRQ and BH handlers synchronization using work_pending.
Remove {READ,WRITE}_ONCE notations since smp_{r,w}mb are implicit
compiler barriers.

v3:
Remove exlicit memory barriers as scedule_work has r/w barriers.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Switch to using reg fields defines istead of magic values.
Add SH_ID and PRIV fields reading for instr. and err cases.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Acked-by: NChristian König <christian.koenig@amd.com>
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

SQ can generate interrupts and installs the ISR to
handle the SQ interrupts.

Add parsing SQ data in interrupt handler.

v2:
Remove CZ only limitation.
Rebase.
Signed-off-by: NDavid Panariti <David.Panariti@amd.com>
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: NNicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

ISR will DRM_ERROR ECC error message.

v2:
Remove CZ only limitation.
Rebase.
Signed-off-by: NDavid Panariti <David.Panariti@amd.com>
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: NNicolai Hähnle <nicolai.haehnle@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This patch is extends the usage of WB in
gfx8's ib test which was originally
implemented in the below upstream patch
"ed9324af drm/amdgpu: change gfx9 ib test to use WB"

For reference below are the reasons for switching
to WB:

1)Because when doing IB test we don't want to involve KIQ health
status affect, and since SCRATCH register access is go through
KIQ that way GFX IB test would failed due to KIQ fail.

2)acccessing SCRATCH register cost much more time than WB method
because SCRATCH register access runs through KIQ which at least could
begin after GPU world switch back to current Guest VF
Signed-off-by: NShirish S <shirish.s@amd.com>
Reviewed-by: NChunming Zhou <david1.zhou@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NLeo Liu <leo.liu@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NLeo Liu <leo.liu@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Signed-off-by: NLeo Liu <leo.liu@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Otherwise, the SQ may skip some of the register writes, or shader waves may
be allocated where we don't expect them, so that as a result we don't actually
reset all of the register SRAMs. This can lead to spurious ECC errors later on
if a shader uses an uninitialized register.
Signed-off-by: NNicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Otherwise, the SQ may skip some of the register writes, or shader waves may
be allocated where we don't expect them, so that as a result we don't actually
reset all of the register SRAMs. This can lead to spurious ECC errors later on
if a shader uses an uninitialized register.
Signed-off-by: NNicolai Hähnle <nicolai.haehnle@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>
Cc: stable@vger.kernel.org

Also remove code duplication in write and read regs functions.
This also fixes potential missing unlock in amdgpu_debugfs_regs_write
in case get_user would fail.

v2: Add SRBM mutex locking.
v3: Fix TO counter and fix comment location.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Stuffing the PASID mapping into the VM flush isn't flexible enough since
the PASID mapping changes not as often as we need a VM flush.

v2: add missing use of gmc_v7_0_emit_pasid_mapping
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NFelix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

This reverts commit 75737cb4.

Fixes compute rings test failure on bare metal during full GPU reset.

RCA:
the ring buffer has to be filled with valid packets (such as NOPs) first
before submitting MAP_QUEUEs packet into KIQ. Once a compute engine is mapped,
it will immediately execute the ring buffer if the RTPR is not equal to the
WTPR from the MQD. It could lead to engine hang if the ring buffer filled
with random data.
Signed-off-by: NAndrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: NChristian König <christian.koenig@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Move the CSA area to the top of the VA space to avoid clashing with
HMM/ATC in the lower range on GFX9.

v2: wrong sign noticed by Roger, rebase on CSA_VADDR cleanup, handle VA
hole on GFX9 as well.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NAlex Deucher <alexander.deucher@amd.com>
Acked-by: NMonk Liu <monk.liu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

Instead of repeating this multiple times.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Reviewed-by: NAlex Deucher <alexander.deucher@amd.com>
Acked-by: NMonk Liu <monk.liu@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>

All HDP invalidation and most flush can now be replaced by the generic
ASIC function.
Signed-off-by: NChristian König <christian.koenig@amd.com>
Acked-by: NChunming Zhou <david1.zhou@amd.com>
Signed-off-by: NAlex Deucher <alexander.deucher@amd.com>