drm/amdgpu: replace readq/writeq with atomic64 operations

what we really want is a read or write that is guaranteed to be 64 bits
at a time, atomic64 operations are supported on all architectures

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

gpu: drm: amd: powerplay: Remove logically dead code

Result of pointer airthmentic is never null

fix coverity defect:1451876

Signed-off-by: Hariprasad Kelam <hariprasad.kelam@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: Zero initialize some variables

Clang warns (only Navi warning shown but Arcturus warns as well):

drivers/gpu/drm/amd/amdgpu/../powerplay/navi10_ppt.c:1534:4: warning:
variable 'asic_default_power_limit' is used uninitialized whenever '?:'
condition is false [-Wsometimes-uninitialized]
                        smu_read_smc_arg(smu, &asic_default_power_limit);
                        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/amd/amdgpu/../powerplay/inc/amdgpu_smu.h:588:3: note:
expanded from macro 'smu_read_smc_arg'
        ((smu)->funcs->read_smc_arg? (smu)->funcs->read_smc_arg((smu), (arg)) : 0)
         ^~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/amd/amdgpu/../powerplay/navi10_ppt.c:1550:30: note:
uninitialized use occurs here
                smu->default_power_limit = asic_default_power_limit;
                                           ^~~~~~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/amd/amdgpu/../powerplay/navi10_ppt.c:1534:4: note:
remove the '?:' if its condition is always true
                        smu_read_smc_arg(smu, &asic_default_power_limit);
                        ^
drivers/gpu/drm/amd/amdgpu/../powerplay/inc/amdgpu_smu.h:588:3: note:
expanded from macro 'smu_read_smc_arg'
        ((smu)->funcs->read_smc_arg? (smu)->funcs->read_smc_arg((smu), (arg)) : 0)
         ^
drivers/gpu/drm/amd/amdgpu/../powerplay/navi10_ppt.c:1517:35: note:
initialize the variable 'asic_default_power_limit' to silence this
warning
        uint32_t asic_default_power_limit;
                                         ^
                                          = 0
1 warning generated.

As the code is currently written, if read_smc_arg were ever NULL, arg
would fail to be initialized but the code would continue executing as
normal because the return value would just be zero.

There are a few different possible solutions to resolve this class
of warnings which have appeared in these drivers before:

1. Assume the function pointer will never be NULL and eliminate the
   wrapper macros.

2. Have the wrapper macros initialize arg when the function pointer is
   NULL.

3. Have the wrapper macros return an error code instead of 0 when the
   function pointer is NULL so that the callsites can properly bail out
   before arg can be used.

4. Initialize arg at the top of its function.

Number four is the path of least resistance right now as every other
change will be driver wide so do that here. I only make the comment
now as food for thought.

Fixes: b4af964e75c4 ("drm/amd/powerplay: make power limit retrieval as asic specific")
Link: https://github.com/ClangBuiltLinux/linux/issues/627
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

MAINTAINERS: update amdkfd maintainer (v3)

I'm leaving the role of amdkfd maintainer. Therefore, update the relevant
entry in the MAINTAINERS file with the name of the new maintainer.

Good Luck!

v3: update mailing list, file list (Alex)

Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> (v2)
Signed-off-by: Oded Gabbay <oded.gabbay@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Fix GPU reset crash regression.

amdgpu_ip_block.status.hw for GMC wasn't set to
false on suspend during GPU reset and so on resume gmc_v9_0_resume
wasn't called.
Caused by 'drm/amdgpu: fix double ucode load by PSP(v3)'

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: check before issuing messages for max sustainable clocks

Those messages are not supported on Arcturus and should not be
issued.

Affected ASIC: Arcturus

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

drm/amdgpu/discovery: move common discovery code out of navi1*_reg_base_init()

move amdgpu_discovery_reg_base_init() from navi1*_reg_base_init() to a
common function nv_reg_base_init().

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/soc15: fix external_rev_id for navi14

fix the hard code external_rev_id.

Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: tiancyin <tianci.yin@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: update ras sysfs feature info

remove confused ras error type info

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: skip pcie params override on Arcturus V2

This is not supported on Arcturus.

Affected ASIC: Arcturus

V2: minor cosmetic fix

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

drm/amdgpu: Fix panic during gpu reset

Clear the flag after hw suspend, otherwise it skips the corresponding hw
resume.

Signed-off-by: xinhui pan <xinhui.pan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: pin the csb buffer on hw init for gfx v8

Without this pin, the csb buffer will be filled with inconsistent
data after S3 resume. And that will causes gfx hang on gfxoff
exit since this csb will be executed then.

Signed-off-by: Likun Gao <Likun.Gao@amd.com>
Tested-by: Paul Gover <pmw.gover@yahoo.co.uk>
Reviewed-by: Feifei Xu <Feifei.Xu@amd.com>
Reviewed-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Block immediate flips for non-fast updates

[Why]
Underflow can occur in the case where we change buffer pitch, DCC state,
rotation or mirroring for a plane while also performing an immediate
flip. It can also generate a p-state warning stack trace on DCN1 which
is typically observed during the cursor handler pipe locking because of
how frequent cursor updates can occur.

[How]
Store the update type on each CRTC - every plane will have access to
the CRTC state if it's flipping. If the update type is not
UPDATE_TYPE_FAST then the immediate flip should be disallowed.

No changes to the target vblank sequencing need to be done, we just
need to ensure that the surface registers do a double buffered update.

Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Reviewed-by: David Francis <david.francis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Validate dc_plane_info and dc_plane_size in atomic check

[Why]
Pitch, DCC, rotation and mirroring can result in updates that are not
UPDATE_TYPE_FAST but UPDATE_TYPE_MED instead. DC needs dc_plane_info
and dc_plane_size to make this determination and we aren't currently
passing this into DC during atomic check.

Underflow (visible or non-visible) can occur if we don't validate this
correctly. This also will generally trigger p-state warnings, typically
via the cursor handler when locking.

[How]
Get the framebuffer tiling flags and generate the required structures
for DC in dm_determine_update_type_for_commit.

Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Reviewed-by: David Francis <david.francis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: enable Navi12 kfd support for amdgpu

Navi12 has the same interface as Navi10

Signed-off-by: shaoyunl <shaoyun.liu@amd.com>
Reviewed-by: Jack Xiao <Jack.Xiao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Add missing NV12 asic IDs

Add missing navi12 asic ids.

Signed-off-by: Roman Li <Roman.Li@amd.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Add nv12 DC ip block

Load DC and amdgpu display manager

Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Add ASICREV_IS_NAVI macros

They are used by DC to determine ASIC revs.

Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Roman Li <Roman.Li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: enable DPG mode for Navi12

Enable Dynamic Power Gating VCN for Navi12.

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Reviewed-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add VCN ip block for Navi12

Add VCN2 ip block for Navi12

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add Navi12 VCN firmware support

Add Navi12 to VCN family

Signed-off-by: Boyuan Zhang <boyuan.zhang@amd.com>
Reviewed-by: Leo Liu <leo.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add psp ip block for navi12

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Jack Xiao <Jack.Xiao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add smu ip block for navi12

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Jack Xiao <Jack.Xiao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: start autoload till RLCG fw for navi12

rlc save restore list is not ready yet for navi12

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Jack Xiao <Jack.Xiao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/psp11: add psp support for navi12

Same as other navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/smu11: add smu support for navi12

Same as other Navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: correct smu rlc handshake enablement bit

Correct the enablement bit of SMU RLC handshake.

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

drm/amdgpu: add CGTT_GS_NGG_CLK_CTRL register to gc header

gc 10.1.2 introduced this new register

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/sdma5: add golden settings for navi12 (v2)

common golden settings are put in golden_settings_sdma_5 array

v2: update settings (Alex)

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Jack Xiao <Jack.Xiao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: add golden settings for navi12 (v2)

Add initial golden settings for navi12 gfx.

v2: update settings

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Jack Xiao <Jack.Xiao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: enable virtual display for navi12

Virtual display is a sw display interface for
bring up and virtualization or for cards without
display hardware.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: set tcp harvest for navi12

Same as navi10.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add ip blocks for navi12

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gmc10: set gart size and vm size for navi12

Same as other navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/sdma5: add placeholder for navi12 golden settings

None yet.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/sdma5: declare sdma firmwares for navi12

Declare the firmwares and load the proper ones for navi12.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: set rlc funcs for navi12

Same as other navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: set number of me(c)/pipe/queue for navi12

Same as other navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: add placeholder for navi12 golden settings

Not used yet.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: declare cp/rlc firmwares for navi12

Set the name properly to load the right ucode.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: add gfx config for navi12

got from mmCP_MAX_CONTEXT and mmPA_SC_FIFO_SIZE

v2: squash all navi asics together because the
settings are the same.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu/gfx10: set gfx cg for navi12

Same as other navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: set nbio/hdp cg for navi12

Same as navi10.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: initialize cg/pg flags and external rev id for navi12

don't enable any cg/pg features yet.

v2: calculate external revision id from revision id so that we can
    differentiate navi12 A0 from A1 directly.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use front door firmware loading for navi12

Same as other navi asics.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: set asic family and ip blocks for navi12

same with navi10

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add gpu_info firmware for navi12

gpu_info firmare store asic configuration details.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add navi12 asic type

Add asic type.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: initialize reg base for navi12

Set up the register offset map for navi12.

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add ip offset header for navi12 (v2)

This adds the absolute offsets of each IP regiser block.

v2: Squash in MP1 update

Signed-off-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: update SDMA V4 microcode init

Removed loading duplicate instances of SDMA FW for Arcturus.
We use a single image for all instances.

Signed-off-by: John Clements <john.clements@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: extend PSP FW loading support to 8 SDMA instances

Arcturus has 8 instances of SDMA.  Update host to PSP interface
to handle it.

Signed-off-by: John Clements <john.clements@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: disable MEC2 JT context init for Arcturus

We don't need to handle it like other asics.

Signed-off-by: John Clements <john.clements@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: update PSP CMD fail response status print

Print the response in hex with the apprpriate mask.

Signed-off-by: John Clements <john.clements@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add PSP KDB loading support for Arcturus

Add support for the arcturus specific psp metadata to the
amdgpu firmware and properly parse it when loading it.

Signed-off-by: John Clements <john.clements@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add PSP SW init support for Arcturus

Add arcturus cases to psp init sewquence.

Signed-off-by: John Clements <john.clements@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: removed duplicate line

Remove duplicate break.

Signed-off-by: John Clements <john.clements@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: correct navi10 vcn powergate

vcn dpm on is a prerequisite for vcn power gate control.

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: enable SW SMU power profile switch support in KFD

Hook up the SW SMU power profile switch in KFD routine.

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Kevin Wang <kevin1.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: support power profile retrieval and setting on arcturus

Enable arcturus power profile retrieval and setting.

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Kevin Wang <kevin1.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: guard consistency between CPU copy and local VRAM

This can prevent CPU to use the out-dated copy.

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: replace AMDGPU_RAS_UE with AMDGPU_RAS_SUCCESS

ce can also trigger interrupt, and even both ce and ue error can be
found in one ras query, distinguishing between ce and ue in interrupt
handler is uncessary.

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Suggested-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: only uncorrectable error needs gpu reset

we only read error information for correctable error in interrupt
handler, gpu reset is unnecessary since there is no data lost
in correctable error

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: update the calc algorithm of umc ecc error count

the initial value of ecc error count can be adjusted

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: implement umc ras init function

enable umc ce interrupt and initialize ecc error count

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: support ce interrupt in ras module

correctable error can also trigger interrupt in some ras blocks

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add error address query for umc ras

umc error address query can get ce/ue error address and clear error
status

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: apply umc_for_each_channel macro to umc_6_1

use umc_for_each_channel to make code simpler

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add macro of umc for each channel

common function for all umc versions, loop for each umc channel is
a frequent used operation in umc block, define it as a macro to
simplify code

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: initialize new parameters and functions for amdgpu_umc structure

add initialization for new members of amdgpu_umc structure

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add more parameters and functions to amdgpu_umc structure

expose more parameters and functions of specific umc version to common
umc layer, so amdgpu_umc layer and other blocks could access them

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove the clear of MCA_ADDR

clearing MCA_STATUS is enough to reset the whole MCA, writing zero to
MCA_ADDR is unnecessary

Signed-off-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: honor hw limit on fetching metrics data for navi10

too frequently to update mertrics table will cause smu internal error.

Signed-off-by: Kevin Wang <kevin1.wang@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Don't replace the dc_state for fast updates

[Why]
DRM private objects have no hw_done/flip_done fencing mechanism on their
own and cannot be used to sequence commits accordingly.

When issuing commits that don't touch the same set of hardware resources
like page-flips on different CRTCs we can run into the issue below
because of this:

1. Client requests non-blocking Commit #1, has a new dc_state #1,
state is swapped, commit tail is deferred to work queue

2. Client requests non-blocking Commit #2, has a new dc_state #2,
state is swapped, commit tail is deferred to work queue

3. Commit #2 work starts, commit tail finishes,
atomic state is cleared, dc_state #1 is freed

4. Commit #1 work starts,
commit tail encounters null pointer deref on dc_state #1

In order to change the DC state as in the private object we need to
ensure that we wait for all outstanding commits to finish and that
any other pending commits must wait for the current one to finish as
well.

We do this for MEDIUM and FULL updates. But not for FAST updates, nor
would we want to since it would cause stuttering from the delays.

FAST updates that go through dm_determine_update_type_for_commit always
create a new dc_state and lock the DRM private object if there are
any changed planes.

We need the old state to validate, but we don't actually need the new
state here.

[How]
If the commit isn't a full update then the use after free can be
resolved by simply discarding the new state entirely and retaining
the existing one instead.

With this change the sequence above can be reexamined. Commit #2 will
still free Commit #1's reference, but before this happens we actually
added an additional reference as part of Commit #2.

If an update comes in during this that needs to change the dc_state
it will need to wait on Commit #1 and Commit #2 to finish. Then it'll
swap the state, finish the work in commit tail and drop the last
reference on Commit #2's dc_state.

Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=204181
Fixes: 004b3938e637 ("drm/amd/display: Check scaling info when determing update type")
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: David Francis <david.francis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Skip determining update type for async updates

[Why]
By passing through the dm_determine_update_type_for_commit for atomic
commits that can be done asynchronously we are incurring a
performance penalty by locking access to the global private object
and holding that access until the end of the programming sequence.

This is also allocating a new large dc_state on every access in addition
to retaining all the references on each stream and plane until the end
of the programming sequence.

[How]
Shift the determination for async update before validation. Return early
if it's going to be an async update.

Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: David Francis <david.francis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Allow cursor async updates for framebuffer swaps

[Why]
We previously allowed framebuffer swaps as async updates for cursor
planes but had to disable them due to a bug in DRM with async update
handling and incorrect ref counting. The check to block framebuffer
swaps has been added to DRM for a while now, so this check is redundant.

The real fix that allows this to properly in DRM has also finally been
merged and is getting backported into stable branches, so dropping
this now seems to be the right time to do so.

[How]
Drop the redundant check for old_fb != new_fb.

With the proper fix in DRM, this should also fix some cursor stuttering
issues with xf86-video-amdgpu since it double buffers the cursor.

IGT tests that swap framebuffers (-varying-size for example) should
also pass again.

Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: David Francis <david.francis@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix unsigned variable instance compared to less than zero

Currenly the error check on variable instance is always false because
it is a uint32_t type and this is never less than zero. Fix this by
making it an int type.

Addresses-Coverity: ("Unsigned compared against 0")
Fixes: 7d0e6329dfdc ("drm/amdgpu: update more sdma instances irq support")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: Allow changing of fan_control in smu_v11_0

[Why]
Before this change, the fan control state on smu_v11 was not able to be
changed because the capability check for checking if the fan control
capability existed was inverted.

[How]
The capability check for fan control in smu_v11_0_auto_fan_control was
inverted, to correctly check for the absence, instead of presence of fan
control capabilities.

Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Matt Coffin <mcoffin13@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: fix a few spelling mistakes

There are a few spelling mistakes "unknow" -> "unknown" and
"enabeld" -> "enabled". Fix these.

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

gpu: drm: radeon: Fix a possible null-pointer dereference in radeon_connector_set_property()

In radeon_connector_set_property(), there is an if statement on line 743
to check whether connector->encoder is NULL:
    if (connector->encoder)

When connector->encoder is NULL, it is used on line 755:
    if (connector->encoder->crtc)

Thus, a possible null-pointer dereference may occur.

To fix this bug, connector->encoder is checked before being used.

This bug is found by a static analysis tool STCheck written by us.

Signed-off-by: Jia-Ju Bai <baijiaju1990@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/powerplay: fix off-by-one upper bounds limit checks

There are two occurrances of off-by-one upper bound checking of indexes
causing potential out-of-bounds array reads. Fix these.

Addresses-Coverity: ("Out-of-bounds read")
Fixes: cb33363d0e85 ("drm/amd/powerplay: add smu feature name support")
Fixes: 6b294793e384 ("drm/amd/powerplay: add smu message name support")
Signed-off-by: Colin Ian King <colin.king@canonical.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: Fix EEH during kexec

During kexec some adapters hit an EEH since they are not properly
shut down in the radeon_pci_shutdown() function. Adding
radeon_suspend_kms() fixes this issue.

Signed-off-by: KyleMahlkuch <kmahlkuc@linux.vnet.ibm.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: Extend CU mask to 8 SEs (v3)

Following bitmap layout logic introduced by:
"drm/amdgpu: support get_cu_info for Arcturus".

v2: squash in fixup for gfx_v9_0.c (Alex)
v3: squash in debug print output fix

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

drm/amdgpu: support get_cu_info for Arcturus

This change is because SE/SH layout on Arcturus is 8*1, different from
4*2(or 4*1) on Vega ASICs.

Currently the cu bitmap array is 4x4 size, and besides the bitmap is used widely
across SW stack. To mostly reduce the scale of impact, we make the cu bitmap
array compatible with SE/SH layout on Arcturus. Then the store of cu bits of
each shader array for Arcturus will be like below:
    SE0,SH0 --> bitmap[0][0]
    SE1,SH0 --> bitmap[1][0]
    SE2,SH0 --> bitmap[2][0]
    SE3,SH0 --> bitmap[3][0]
    SE4,SH0 --> bitmap[0][1]
    SE5,SH0 --> bitmap[1][1]
    SE6,SH0 --> bitmap[2][1]
    SE7,SH0 --> bitmap[3][1]

Signed-off-by: Le Ma <le.ma@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Fix pcie_bw on Vega20

The registers used for VG20 are different in that certain performance
counters were split off to TXCLK3/4. Vega10/12 doesn't have this, so add
a new vg20_get_pcie_usage to reflect this change.

Signed-off-by: Kent Russell <kent.russell@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Update NBIO headers to add TXCLK3/4

These are added for VG20, and are needed for PCIe bandwidth.

Signed-off-by: Kent Russell <kent.russell@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Add amdgpu_asic_funcs.reset_method for Vega20

Fixes GPU reset crash.

Signed-off-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Mark KFD VRAM allocations for wipe on release

Memory used by KFD applications can contain sensitive information that
should not be leaked to other processes. The current approach to prevent
leaks is to clear VRAM at allocation time. This is not effective because
memory can be reused in other ways without being cleared. Synchronously
clearing memory on the allocation path also carries a significant
performance penalty.

Stop clearing memory at allocation time. Instead mark the memory for
wipe on release.

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

drm/amdgpu: Implement VRAM wipe on release

Wipe VRAM memory containing sensitive data when moving or releasing
BOs. Clearing the memory is pipelined to minimize any impact on
subsequent memory allocation latency. Use of a poison value should
help debug future use-after-free bugs.

When moving BOs, the existing ttm_bo_pipelined_move ensures that the
memory won't be reused before being wiped.

When releasing BOs, the BO is fenced with the memory fill operation,
which results in queuing the BO for a delayed delete.

v2: Move amdgpu_amdkfd_unreserve_memory_limit into
amdgpu_bo_release_notify so that KFD can use memory that's still
being cleared in the background

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

drm/amdgpu: Add flag to wipe VRAM on release

This memory allocation flag will be used to indicate BOs containing
sensitive data that should not be leaked to other processes.

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

drm/ttm: Add release_notify callback to ttm_bo_driver

This notifies the driver that a BO is about to be released.

Releasing a BO also invokes the move_notify callback from
ttm_bo_cleanup_memtype_use, but that happens too late for anything
that would add fences to the BO and require a delayed delete.

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

drm/amd/display: Use switch table for dc_to_smu_clock_type

Using a static int array will cause errors if the given dm_pp_clk_type
is out-of-bounds. For robustness, use a switch table, with a default
case to handle all invalid values.

v2: 0 is a valid clock type for smu_clk_type. Return SMU_CLK_COUNT
    instead on invalid mapping.

Signed-off-by: Leo Li <sunpeng.li@amd.com>
Reviewed-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Use proper enum conversion functions

clang warns:

drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm_pp_smu.c:336:8:
warning: implicit conversion from enumeration type 'enum smu_clk_type'
to different enumeration type 'enum amd_pp_clock_type'
[-Wenum-conversion]
                                        dc_to_smu_clock_type(clk_type),
                                        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
drivers/gpu/drm/amd/amdgpu/../display/amdgpu_dm/amdgpu_dm_pp_smu.c:421:14:
warning: implicit conversion from enumeration type 'enum
amd_pp_clock_type' to different enumeration type 'enum smu_clk_type'
[-Wenum-conversion]
                                        dc_to_pp_clock_type(clk_type),
                                        ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

There are functions to properly convert between all of these types, use
them so there are no longer any warnings.

Fixes: a43913ea50a5 ("drm/amd/powerplay: add function get_clock_by_type_with_latency for navi10")
Fixes: e5e4e22391c2 ("drm/amd/powerplay: add interface to get clock by type with latency for display (v2)")
Link: https://github.com/ClangBuiltLinux/linux/issues/586
Signed-off-by: Nathan Chancellor <natechancellor@gmail.com>
Reviewed-by: Leo Li <sunpeng.li@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix double ucode load by PSP(v3)

previously the ucode loading of PSP was repreated, one executed in
phase_1 init/re-init/resume and the other in fw_loading routine

Avoid this double loading by clearing ip_blocks.status.hw in suspend or reset
prior to the FW loading and any block's hw_init/resume

v2:
still do the smu fw loading since it is needed by bare-metal

v3:
drop the change in reinit_early_sriov, just clear all block's status.hw
in the head place and set the status.hw after hw_init done is enough

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

drm/amdgpu: fix incorrect judge on sos fw version

for SRIOV the SOS fw of PSP is loaded in hypervisor thus
guest won't tell the version of it, and judging feature by
reading the sos fw version in guest side is completely wrong

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

drm/amdgpu: cleanup vega10 SRIOV code path

we can simplify all those unnecessary function under
SRIOV for vega10 since:
1) PSP L1 policy is by force enabled in SRIOV
2) original logic always set all flags which make itself
   a dummy step

besides,
1) the ih_doorbell_range set should also be skipped
for VEGA10 SRIOV.
2) the gfx_common registers should also be skipped
for VEGA10 SRIOV.

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

drm/amd/powerplay: sort feature status index by asic feature id for smu

before this change, the pp_feature sysfs show feature enable state by
logic feature id, it is not easy to read.
this change will sort pp_features show index by asic feature id.

before:
features high: 0x00000623 low: 0xb3cdaffb
00. DPM_PREFETCHER       ( 0) : enabeld
01. DPM_GFXCLK           ( 1) : enabeld
02. DPM_UCLK             ( 3) : enabeld
03. DPM_SOCCLK           ( 4) : enabeld
04. DPM_MP0CLK           ( 5) : enabeld
05. DPM_LINK             ( 6) : enabeld
06. DPM_DCEFCLK          ( 7) : enabeld
07. DS_GFXCLK            (10) : enabeld
08. DS_SOCCLK            (11) : enabeld
09. DS_LCLK              (12) : disabled
10. PPT                  (23) : enabeld
11. TDC                  (24) : enabeld
12. THERMAL              (33) : enabeld
13. RM                   (35) : disabled
......

after:
features high: 0x00000623 low: 0xb3cdaffb
00. DPM_PREFETCHER       ( 0) : enabeld
01. DPM_GFXCLK           ( 1) : enabeld
02. DPM_GFX_PACE         ( 2) : disabled
03. DPM_UCLK             ( 3) : enabeld
04. DPM_SOCCLK           ( 4) : enabeld
05. DPM_MP0CLK           ( 5) : enabeld
06. DPM_LINK             ( 6) : enabeld
07. DPM_DCEFCLK          ( 7) : enabeld
08. MEM_VDDCI_SCALING    ( 8) : enabeld
09. MEM_MVDD_SCALING     ( 9) : enabeld
10. DS_GFXCLK            (10) : enabeld
11. DS_SOCCLK            (11) : enabeld
12. DS_LCLK              (12) : disabled
13. DS_DCEFCLK           (13) : enabeld
......

Signed-off-by: Kevin Wang <kevin1.wang@amd.com>
Reviewed-by: Kenneth Feng <kenneth.feng@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: enable KFD support for navi14

Same as navi10.

Reviewed-by: Xiaojie Yuan <xiaojie.yuan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: disable inject for failed subblocks of gfx

some subblocks of gfx fail in inject test, disable them

Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: support gfx ras error injection and err_cnt query

check gfx error count in both ras querry function and
ras interrupt handler.

gfx ras is still disabled by default due to known stability
issue found in gpu reset.

Signed-off-by: Dennis Li <Dennis.Li@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>