drm/amd/display: Refactor vendor specific link training sequence

[Why]
Current implementation is not scalable and retrofits the existing
standard link training code for purposes outside of its original design.

[How]
Refactor vendor specific link training sequence into its own separate
function to be called instead of the standard link training function.

Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Wenjing Liu <Wenjing.Liu@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: George Shen <George.Shen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Limit max link cap with LTTPR caps

[Why]
Max link rate should be limited to the maximum link rate support by any
LTTPR that are connected, including when operating in transparent mode.

[How]
Include transparent mode when factoring in LTTPR max supported link
rate.

Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Wesley Chalmers <wesley.chalmers@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: George Shen <George.Shen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: fix B0 TMDS deepcolor no dislay issue

[why]
B0 PHY C map to F, D map to G driver use logic instance, dmub does the
remap. Driver still need use the right PHY instance to access right HW.

[how]
use phyical instance when program PHY register.

[note]
could move resync_control programming to dmub next.

Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Dmytro Laktyushkin <Dmytro.Laktyushkin@amd.com>
Reviewed-by: Jun Lei <Jun.Lei@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Charlene Liu <Charlene.Liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Check the memory can be accesssed by ttm_device_clear_dma_mappings.

If the event guard is enabled and VF doesn't receive an ack from PF for full access,
the guest driver load crashes.
This is caused due to the call to ttm_device_clear_dma_mappings with non-initialized
mman during driver tear down.

This patch adds the necessary condition to check if the mman initialization passed or not
and takes the path based on the condition output.

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

drm/amdkfd: correct sdma queue number in kfd device init (v3)

This patch keeps the setting of sdma queue number to the same
after recent KFD code refactor. Additionally, improve code to
use switch case to list IP version to complete kfd device_info
structure filling for IH version assignment. This makes consistency
with the IP parse code in amdgpu_discovery.c.

v2: use dev_warn for the default switch case;
    set default sdma queue per engine(8) and IH handler to v9. (Jonathan)

v3: Fix missed IP version check of Raven.

Fixes: f0dc99a6f742bc ("drm/amdkfd: add kfd_device_info_init function")
Signed-off-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Graham Sider <Graham.Sider@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Access the FRU on Aldebaran

This is supported, although the offset is different from VG20, so fix
that with a variable and enable getting the product name and serial
number from the FRU. Do this for all SKUs since all SKUs have the FRU

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: Only overwrite serial if field is empty

On Aldebaran, the serial may be obtained from the FRU. Only overwrite
the serial with the unique_id if the serial is empty. This will support
printing serial numbers for mGPU devices where there are 2 unique_ids
for the 2 GPUs, but only one serial number for the board

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: Enable unique_id for Aldebaran

It's supported, so support the unique_id sysfs file

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: Increase potential product_name to 64 characters

Having seen at least 1 42-character product_name, bump the number up to
64, and put that definition into amdgpu.h to make future adjustments
simpler.

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: Remove the redundant code of psp bootloader functions

The psp bootloader functions code of psp_v13_0.c had been
optimized before. According the code style of psp_v13_0.c
to remove the redundant code of psp_v11_0.c.

v2: squash in drop unused variable (Alex)

Signed-off-by: yipechai <YiPeng.Chai@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Call amdgpu_device_unmap_mmio() if device is unplugged to prevent crash in GPU initialization failure

[Why]
In amdgpu_driver_load_kms, when amdgpu_device_init returns error during driver modprobe, it
will start the error handle path immediately and call into amdgpu_device_unmap_mmio as well
to release mapped VRAM. However, in the following release callback, driver stills visits the
unmapped memory like vcn.inst[i].fw_shared_cpu_addr in vcn_v3_0_sw_fini. So a kernel crash occurs.

[How]
call amdgpu_device_unmap_mmio() if device is unplugged to prevent invalid memory address in
vcn_v3_0_sw_fini() when GPU initialization failure.

Signed-off-by: Leslie Shi <Yuliang.Shi@amd.com>
Reviewed-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Acked-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fixup bad vram size on gmc v8

Some boards(like RX550) seem to have garbage in the upper
16 bits of the vram size register.  Check for
this and clamp the size properly.  Fixes
boards reporting bogus amounts of vram.

after add this patch,the maximum GPU VRAM size is 64GB,
otherwise only 64GB vram size will be used.

Signed-off-by: Zongmin Zhou<zhouzongmin@kylinos.cn>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Fix USB4 null pointer dereference in update_psp_stream_config

[Why]
A porting error on a previous patch left the block of code that
causes the crash from a NULL pointer dereference.

More specifically, we try to access link_enc before it's assigned in
the USB4 case in the following assignment:

config.dio_output_idx = link_enc->transmitter - TRANSMITTER_UNIPHY_A;

[How]
That assignment occurs later depending on the ASIC version. It's only
needed on DCN31 and only after link_enc is already assigned.

Fixes: 986430446c917b ("drm/amd/display: fix a crash on USB4 over C20 PHY")
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Send Message to SMU on aldebaran passthrough for sbr handling

For Aldebaran chip passthrough case we need to intimate SMU
about special handling for SBR.On older chips we send
LightSBR to SMU, enabling the same for Aldebaran. Slight
difference, compared to previous chips, is on Aldebaran, SMU
would do a heavy reset on SBR. Hence, the word Heavy
instead of Light SBR is used for SMU to differentiate.

Reviewed by: Shaoyun.liu <Shaoyun.liu@amd.com>
Signed-off-by: sashank saye <sashank.saye@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Don't inherit GEM object VMAs in child process

When an application having open file access to a node forks, its shared
mappings also get reflected in the address space of child process even
though it cannot access them with the object permissions applied. With the
existing permission checks on the gem objects, it might be reasonable to
also create the VMAs with VM_DONTCOPY flag so a user space application
doesn't need to explicitly call the madvise(addr, len, MADV_DONTFORK)
system call to prevent the pages in the mapped range to appear in the
address space of the child process. It also prevents the memory leaks
due to additional reference counts on the mapped BOs in the child
process that prevented freeing the memory in the parent for which we had
worked around earlier in the user space inside the thunk library.

Additionally, we faced this issue when using CRIU to checkpoint restore
an application that had such inherited mappings in the child which
confuse CRIU when it mmaps on restore. Having this flag set for the
render node VMAs helps. VMAs mapped via KFD already take care of this so
this is needed only for the render nodes.

To limit the impact of the change to user space consumers such as OpenGL
etc, limit it to KFD BOs only.

Acked-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: David Yat Sin <david.yatsin@amd.com>
Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: reset queue which consumes RAS poison (v2)

CP supports unmap queue with reset mode which only destroys specific queue without affecting others.
Replacing whole gpu reset with reset queue mode for RAS poison consumption
saves much time, and we can also fallback to gpu reset solution if reset
queue fails.

v2: Return directly if process is NULL;
    Reset queue solution is not applicable to SDMA, fallback to legacy
way;
    Call kfd_unref_process after lookup process.

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

drm/amdkfd: add reset queue function for RAS poison (v2)

The new interface unmaps queues with reset mode for the process consumes
RAS poison, it's only for compute queue.

v2: rename the function to reset_queues.

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

drm/amdkfd: add reset parameter for unmap queues

So we can set reset mode for unmap operation, no functional change.

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

drm/amdgpu: add gpu reset control for umc page retirement

Add a reset parameter for umc page retirement, let user decide whether
call gpu reset in umc page retirement.

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

drm/amdgpu: Modify indirect register access for gfx9 sriov

Expand RLCG interface for new GC read & write commands.
New interface will only be used if the PF enables the flag in pf2vf msg.

v2: Added a description for the scratch registers

Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Reviewed-by: David Nieto <david.nieto@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: get xgmi info before ip_init

Driver needs to call get_xgmi_info() before ip_init
to determine whether it needs to handle a pending hive reset.

Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Reviewed-by: David Nieto <david.nieto@amd.com>
Reviewed by: shaoyun.liu <Shaoyun.lui@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Modify indirect register access for amdkfd_gfx_v9 sriov

Modify GC register access from MMIO to RLCG if the indirect
flag is set

Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Reviewed-by: David Nieto <david.nieto@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Modify indirect register access for gmc_v9_0 sriov

Modify GC register access from MMIO to RLCG if the
indirect flag is set

v2: Replaced ternary operator with if-else for better
readability

Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Reviewed-by: David Nieto <david.nieto@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Add *_SOC15_IP_NO_KIQ() macro definitions

Add helper macros to change register access
from direct to indirect.

Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Reviewed-by: David Nieto <david.nieto@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Filter security violation registers

Recently, there is security policy update under SRIOV.
We need to filter the registers that hit the violation
and move the code to the host driver side so that
the guest driver can execute correctly.

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

drm/amdgpu: add support for IP discovery gc_info table v2

Used on gfx9 based systems. Fixes incorrect CU counts reported
in the kernel log.

Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1833
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Fix warning comparing pointer to 0

Fix the following coccicheck warning:

./drivers/gpu/drm/amd/display/dc/clk_mgr/dcn301/vg_clk_mgr.c:744:35-36:
WARNING comparing pointer to 0.

Reported-by: Abaci Robot <abaci@linux.alibaba.com>
Signed-off-by: Jiapeng Chong <jiapeng.chong@linux.alibaba.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: clean up some leftovers from bring up

Some old registers leftover from pre-silicon.  No longer
relevant on real hardware.  Remove.

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

drm/amdgpu: Separate vf2pf work item init from virt data exchange

We want to be able to call virt data exchange conditionally
after gmc sw init to reserve bad pages as early as possible.
Since this is a conditional call, we will need
to call it again unconditionally later in the init sequence.

Refactor the data exchange function so it can be
called multiple times without re-initializing the work item.

v2: Cleaned up the code. Kept the original call to init_exchange_data()
inside early init to initialize the work item, afterwards call
exchange_data() when needed.

Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Reviewed By: Shaoyun.liu <Shaoyun.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: use max() and min() to make code cleaner

Use max() and min() in order to make code cleaner.

Reported-by: Zeal Robot <zealci@zte.com.cn>
Signed-off-by: Changcheng Deng <deng.changcheng@zte.com.cn>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: When the VCN(1.0) block is suspended, powergating is explicitly enabled

Play a video on the raven (or PCO, raven2) platform, and then do the S3
test. When resume, the following error will be reported:

amdgpu 0000:02:00.0: [drm:amdgpu_ring_test_helper [amdgpu]] *ERROR* ring
vcn_dec test failed (-110)
[drm:amdgpu_device_ip_resume_phase2 [amdgpu]] *ERROR* resume of IP block
<vcn_v1_0> failed -110
amdgpu 0000:02:00.0: amdgpu: amdgpu_device_ip_resume failed (-110).
PM: dpm_run_callback(): pci_pm_resume+0x0/0x90 returns -110

[why]
When playing the video: The power state flag of the vcn block is set to
POWER_STATE_ON.

When doing suspend: There is no change to the power state flag of the
vcn block, it is still POWER_STATE_ON.

When doing resume: Need to open the power gate of the vcn block and set
the power state flag of the VCN block to POWER_STATE_ON.
But at this time, the power state flag of the vcn block is already
POWER_STATE_ON. The power status flag check in the "8f2cdef drm/amd/pm:
avoid duplicate powergate/ungate setting" patch will return the
amdgpu_dpm_set_powergating_by_smu function directly.
As a result, the gate of the power was not opened, causing the
subsequent ring test to fail.

[how]
In the suspend function of the vcn block, explicitly change the power
state flag of the vcn block to POWER_STATE_OFF.

BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1828
Signed-off-by: chen gong <curry.gong@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/pm: Fix xgmi link control on aldebaran

Fix the message argument.
0: Allow power down
1: Disallow power down

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/pm: restore SMU version print statement for dGPUs

This is still needed for thoes in case the firmware fails to load
then the message is the only way to tell what firmware was on them

Suggested-by: Lijo Lazar <lijo.Lazar@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: fix svm_bo release invalid wait context warning

Add svm_range_bo_unref_async to schedule work to wait for svm_bo
eviction work done and then free svm_bo. __do_munmap put_page
is atomic context, call svm_range_bo_unref_async to avoid warning
invalid wait context. Other non atomic context call svm_range_bo_unref.

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

drm/amdgpu: introduce new amdgpu_fence object to indicate the job embedded fence

The job embedded fence donesn't initialize the flags at
dma_fence_init(). Then we will go a wrong way in
amdgpu_fence_get_timeline_name callback and trigger a null pointer panic
once we enabled the trace event here. So introduce new amdgpu_fence
object to indicate the job embedded fence.

[  156.131790] BUG: kernel NULL pointer dereference, address: 00000000000002a0
[  156.131804] #PF: supervisor read access in kernel mode
[  156.131811] #PF: error_code(0x0000) - not-present page
[  156.131817] PGD 0 P4D 0
[  156.131824] Oops: 0000 [#1] PREEMPT SMP PTI
[  156.131832] CPU: 6 PID: 1404 Comm: sdma0 Tainted: G           OE     5.16.0-rc1-custom #1
[  156.131842] Hardware name: Gigabyte Technology Co., Ltd. Z170XP-SLI/Z170XP-SLI-CF, BIOS F20 11/04/2016
[  156.131848] RIP: 0010:strlen+0x0/0x20
[  156.131859] Code: 89 c0 c3 0f 1f 80 00 00 00 00 48 01 fe eb 0f 0f b6 07 38 d0 74 10 48 83 c7 01 84 c0 74 05 48 39 f7 75 ec 31 c0 c3 48 89 f8 c3 <80> 3f 00 74 10 48 89 f8 48 83 c0 01 80 38 00 75 f7 48 29 f8 c3 31
[  156.131872] RSP: 0018:ffff9bd0018dbcf8 EFLAGS: 00010206
[  156.131880] RAX: 00000000000002a0 RBX: ffff8d0305ef01b0 RCX: 000000000000000b
[  156.131888] RDX: ffff8d03772ab924 RSI: ffff8d0305ef01b0 RDI: 00000000000002a0
[  156.131895] RBP: ffff9bd0018dbd60 R08: ffff8d03002094d0 R09: 0000000000000000
[  156.131901] R10: 000000000000005e R11: 0000000000000065 R12: ffff8d03002094d0
[  156.131907] R13: 000000000000001f R14: 0000000000070018 R15: 0000000000000007
[  156.131914] FS:  0000000000000000(0000) GS:ffff8d062ed80000(0000) knlGS:0000000000000000
[  156.131923] CS:  0010 DS: 0000 ES: 0000 CR0: 0000000080050033
[  156.131929] CR2: 00000000000002a0 CR3: 000000001120a005 CR4: 00000000003706e0
[  156.131937] DR0: 0000000000000000 DR1: 0000000000000000 DR2: 0000000000000000
[  156.131942] DR3: 0000000000000000 DR6: 00000000fffe0ff0 DR7: 0000000000000400
[  156.131949] Call Trace:
[  156.131953]  <TASK>
[  156.131957]  ? trace_event_raw_event_dma_fence+0xcc/0x200
[  156.131973]  ? ring_buffer_unlock_commit+0x23/0x130
[  156.131982]  dma_fence_init+0x92/0xb0
[  156.131993]  amdgpu_fence_emit+0x10d/0x2b0 [amdgpu]
[  156.132302]  amdgpu_ib_schedule+0x2f9/0x580 [amdgpu]
[  156.132586]  amdgpu_job_run+0xed/0x220 [amdgpu]

v2: fix mismatch warning between the prototype and function name (Ray, kernel test robot)

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

drm/amd/pm: skip setting gfx cgpg in the s0ix suspend-resume

In the s0ix entry need retain gfx in the gfxoff state,so here need't
set gfx cgpg in the S0ix suspend-resume process. Moreover move the S0ix
check into SMU12 can simplify the code condition check.

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

drivers/amd/pm: smu13: use local variable adev

Since this variable was made available by the previous commit, use
it to make function access cleaner.

Suggested-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/radeon: Fix syntax errors in comments

Delete the redundant word 'we'.

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

amdgpu: fix some comment typos

Signed-off-by: Yann Dirson <ydirson@free.fr>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

amdgpu: fix some kernel-doc markup

Those are not today pulled by the sphinx doc, but better be ready.

Signed-off-by: Yann Dirson <ydirson@free.fr>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: include description of some of the GC microcontrollers

This is Alex' description from the "Looking for clarifications around gfx/kcq/kiq"
thread, edited to fit as ReST.

Original text: https://www.spinics.net/lists/amd-gfx/msg71383.html

Originally-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Yann Dirson <ydirson@free.fr>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: include description of AMDGPU hardware structure

This describes in broad lines the how an AMD GPU is organized, in
terms of hardware blocks.

This is Alex' description from the "gpu block diagram" thread, edited to
fit as ReST.

Original text: https://www.spinics.net/lists/amd-gfx/msg71543.html

Originally-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Yann Dirson <ydirson@free.fr>
Acked-by: Harry Wentland <harry.wentland@amd.com>
Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/pm: fix a potential gpu_metrics_table memory leak

Memory is allocated for gpu_metrics_table in renoir_init_smc_tables(),
but not freed in int smu_v12_0_fini_smc_tables(). Free it!

Fixes: 95868b85764a ("drm/amd/powerplay: add Renoir support for gpu metrics export")

Signed-off-by: Lang Yu <lang.yu@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: fix gmc bo pin count leak in SRIOV

[Why]
gmc bo will be pinned during loading amdgpu and reset in SRIOV while
only unpinned in unload amdgpu

[How]
add amdgpu_in_reset and sriov judgement to skip pin bo

v2: fix wrong judgement

Signed-off-by: Jingwen Chen <Jingwen.Chen2@amd.com>
Reviewed-by: Horace Chen <horace.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/amdgpu: fix psp tmr bo pin count leak in SRIOV

[Why]
psp tmr bo will be pinned during loading amdgpu and reset in SRIOV while
only unpinned in unload amdgpu

[How]
add amdgpu_in_reset and sriov judgement to skip pin bo

v2: fix wrong judgement

Signed-off-by: Jingwen Chen <Jingwen.Chen2@amd.com>
Reviewed-by: Horace Chen <horace.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/pm: Skip power state allocation

Power states are not valid for arcturus and aldebaran, no need to
allocate memory.

Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Yang Wang <kevinyang.wang@amd.com>
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: correct the wrong cached state for GMC on PICASSO

Pair the operations did in GMC ->hw_init and ->hw_fini. That
can help to maintain correct cached state for GMC and avoid
unintention gate operation dropping due to wrong cached state.

BugLink: https://gitlab.freedesktop.org/drm/amd/-/issues/1828
Signed-off-by: Evan Quan <evan.quan@amd.com>
Acked-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: use adev_to_drm to get drm_device pointer

Updated for consistency when accessing drm_device from amdgpu driver.

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

drm/amdgpu: move smu_debug_mask to a more proper place

As the smu_context will be invisible from outside(of power). Also,
the smu_debug_mask can be shared around all power code instead of
some specific framework(swSMU) only.

Signed-off-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: SRIOV flr_work should use down_write

Host initiated VF FLR may fail if someone else is
already holding a read_lock. Change from down_write_trylock
to down_write to guarantee the reset goes through.

Signed-off-by: Victor Skvortsov <victor.skvortsov@amd.com>
Reviewed by: Shaoyun.liu <Shaoyun.liu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: 3.2.166

Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Signed-off-by: Aric Cyr <aric.cyr@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: implement dc_mode_memclk

why:
Need interface to lower clocks when in dc (power save)
mode. Must be able to work with p_state unsupported cases
Can cause flicker when OS notifies us of dc state change

how:
added dal3 interface for KMD
added pathway to query smu for this softmax
added blank before clock change to override underflow
added logic to change clk based on pstatesupport and softmax
added logic in prepare/optimize_bw to conform while changing
clocks

Reviewed-by: Aric Cyr <Aric.Cyr@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Martin Leung <Martin.Leung@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: ODM + MPO window on only one half of ODM

[Why]
For ODM + MPO window on one half of ODM, only 3 pipes should
be allocated and scaling parameters adjusted to handle this case

[How]
Fix pipe allocation when MPO viewport is only on one side of ODM
split, and modify scaling paramters.
Added diags test cases for ODM + windows MPO, where MPO window is
on right half, left half, and both halves or ODM.

Reviewed-by: Dmytro Laktyushkin <Dmytro.Laktyushkin@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Eric Bernstein <eric.bernstein@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Reset DMCUB before HW init

[Why]
If the firmware wasn't reset by PSP or HW and is currently running
then the firmware will hang or perform underfined behavior when we
modify its firmware state underneath it.

[How]
Reset DMCUB before setting up cache windows and performing HW init.

Reviewed-by: Aurabindo Jayamohanan Pillai <Aurabindo.Pillai@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: [FW Promotion] Release 0.0.97

Reviewed-by: Aric Cyr <Aric.Cyr@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Anthony Koo <Anthony.Koo@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Force det buf size to 192KB with 3+ streams and upscaling

[WHY]
This workaround resolves underflow caused by incorrect DST_Y_PREFETCH.
Overriding to 192KB DET buf size until the DST_Y_PREFETCH calc is fixed.

Reviewed-by: Eric Yang <Eric.Yang2@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Michael Strauss <michael.strauss@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: parse and check PSR SU caps

[why]
Adding a function to read PSR capabilities
and ALPM capabilities.
Also adding a helper function to validate if
the sink and the driver support PSR SU.
[how]
- isolated all PSR and ALPM reading calls to a separate funciton
- set all required PSR caps
- added a helper function to check if PSR SU is supported by sink
and the driver

Reviewed-by: Roman Li <Roman.Li@amd.com>
Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Mikita Lipski <mikita.lipski@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Add src/ext ID info for dummy service

[Why]
Current error log of dummy irq service doesn't have
src/ext ID info in the log.

[How]
Add src/ext ID in ack/set of dummy irq service.

Reviewed-by: Wayne Lin <Wayne.Lin@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Solomon Chiu <solomon.chiu@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Add debugfs entry for ILR

[Why & How]
In order to know the intermediate link rates supported by the eDP
panel and test to select the optimized link rate to save power,
create a new debugfs entry "ilr_setting" for
setting ILR.

Reviewed-by: Aurabindo Jayamohanan Pillai <Aurabindo.Pillai@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Wayne Lin <Wayne.Lin@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Set exit_optimized_pwr_state for DCN31

[Why]
SMU now respects the PHY refclk disable request from driver.

This causes a hang during hotplug when PHY refclk was disabled
because it's not being re-enabled and the transmitter control
starts on dc_link_detect.

[How]
We normally would re-enable the clk with exit_optimized_pwr_state
but this is only set on DCN21 and DCN301. Set it for dcn31 as well.

This fixes DMCUB timeouts in the PHY.

Fixes: 64b1d0e8d500 ("drm/amd/display: Add DCN3.1 HWSEQ")

Reviewed-by: Eric Yang <Eric.Yang2@amd.com>
Acked-by: Pavle Kotarac <Pavle.Kotarac@amd.com>
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm:amdgpu:remove unneeded variable

return value form directly instead of
taking this in another redundant variable.

Reported-by: Zeal Robot <zealci@zte.com.cm>
Signed-off-by: chiminghao <chi.minghao@zte.com.cn>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: split amdgpu/index for readability

This starts to make the formated index much more manageable to the reader.

Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Yann Dirson <ydirson@free.fr>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drivers/amd/pm: drop statement to print FW version for smu_v13

Update smu_v13 to match smu_v12 and smu_v11 behavior where this is
fetched from debugfs rather than in kernel logs on every boot.

Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/pm: fix reading SMU FW version from amdgpu_firmware_info on YC

This value does not get cached into adev->pm.fw_version during
startup for smu13 like it does for other SMU like smu12.

Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: fix function scopes

This turns previously global functions into static, thus removing
compile-time warnings such as:

 warning: no previous prototype for 'get_highest_allowed_voltage_level'
 [-Wmissing-prototypes]
 742 | unsigned int get_highest_allowed_voltage_level(uint32_t chip_family, uint32_t hw_internal_rev, uint32_t pci_revision_id)
     |              ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
 warning: no previous prototype for 'rv1_vbios_smu_send_msg_with_param'
 [-Wmissing-prototypes]
 102 | int rv1_vbios_smu_send_msg_with_param(struct clk_mgr_internal *clk_mgr, unsigned int msg_id, unsigned int param)
     |     ^~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~

Changes since v1:
- As suggested by Rodrigo Siqueira:
  1. Rewrite function signatures to make them more readable.
  2. Get rid of unused functions in order to remove 'defined but not
     used' warnings.

Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Isabella Basso <isabbasso@riseup.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Reduce stack size for dml31 UseMinimumDCFCLK

Use the struct display_mode_lib pointer instead of passing lots of large
arrays as parameters by value.

Addresses this warning (resulting in failure to build a RHEL debug kernel
with Werror enabled):

../drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.c: In function ‘UseMinimumDCFCLK’:
../drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.c:7478:1: warning: the frame size of 2128 bytes is larger than 2048 bytes [-Wframe-larger-than=]

NOTE: AFAICT this function previously had no observable effect, since it
only modified parameters passed by value and doesn't return anything.
Now it may modify some values in struct display_mode_lib passed in by
reference.

Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd/display: Reduce stack size for dml31_ModeSupportAndSystemConfigurationFull

Move code using the Pipe struct to a new helper function.

Works around[0] this warning (resulting in failure to build a RHEL debug
kernel with Werror enabled):

../drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.c: In function ‘dml31_ModeSupportAndSystemConfigurationFull’:
../drivers/gpu/drm/amd/amdgpu/../display/dc/dml/dcn31/display_mode_vba_31.c:5740:1: warning: the frame size of 2144 bytes is larger than 2048 bytes [-Wframe-larger-than=]

The culprit seems to be the Pipe struct, so pull the relevant block out
into its own sub-function. (This is porting
commit a62427ef9b55 ("drm/amd/display: Reduce stack size for dml21_ModeSupportAndSystemConfigurationFull")
from dml31 to dml21)

[0] AFAICT this doesn't actually reduce the total amount of stack which
can be used, just moves some of it from
dml31_ModeSupportAndSystemConfigurationFull to the new helper function,
so the former happens to no longer exceed the limit for a single
function.

Reviewed-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Michel Dänzer <mdaenzer@redhat.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: re-format file header comments

Fix the warning below:

 warning: Cannot understand  * \file amdgpu_ioc32.c
 on line 2 - I thought it was a doc line

Changes since v1:
- As suggested by Alexander Deucher:
  1. Reduce diff to minimum as this DOC section doesn't provide much
     value.

Signed-off-by: Isabella Basso <isabbasso@riseup.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove unnecessary variables

This fixes the warnings below, and also drops the display_count
variable, as it's unused.

 In function 'svm_range_map_to_gpu':
 warning: variable 'bo_va' set but not used [-Wunused-but-set-variable]
 1172 |         struct amdgpu_bo_va bo_va;
      |                             ^~~~~
 ...
 In function 'dcn201_update_clocks':
 warning: variable 'enter_display_off' set but not used [-Wunused-but-set-variable]
 132 |         bool enter_display_off = false;
     |              ^~~~~~~~~~~~~~~~~

Changes since v1:
- As suggested by Rodrigo Siqueira:
  1. Drop display_count variable.
- As suggested by Felix Kuehling:
  1. Remove block surrounding amdgpu_xgmi_same_hive.

Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Signed-off-by: Isabella Basso <isabbasso@riseup.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: fix amdgpu_ras_mca_query_error_status scope

This commit fixes the compile-time warning below:

 warning: no previous prototype for ‘amdgpu_ras_mca_query_error_status’
 [-Wmissing-prototypes]

Changes since v1:
- As suggested by Alexander Deucher:
  1. Make function static instead of adding prototype.

Signed-off-by: Isabella Basso <isabbasso@riseup.net>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amd: move variable to local scope

`edp_stream` is only used when backend is enabled on eDP, don't
declare the variable outside that scope.

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

drm/amd: add some extra checks that is_dig_enabled is defined

There are a few places that this isn't checked that could potentially
be a NULL pointer access.

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

drm/amdgpu: Reduce SG bo memory usage for mGPUs

For userptr bo, if adev is not in IOMMU isolation mode, RAM direct map
to GPU, multiple GPUs use same system memory dma mapping address, they
can share the original mem->bo in attachment to reduce dma address array
memory usage.

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

drm/amdgpu: Detect if amdgpu in IOMMU direct map mode

If host and amdgpu IOMMU is not enabled or IOMMU is pass through mode,
set adev->ram_is_direct_mapped flag which will be used to optimize
memory usage for multi GPU mappings.

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

Documentation/gpu: Add amdgpu and dc glossary

In the DC driver, we have multiple acronyms that are not obvious most of
the time; the same idea is valid for amdgpu. This commit introduces a DC
and amdgpu glossary in order to make it easier to navigate through our
driver.

Changes since V3:
 - Yann: Add new acronyms to amdgpu glossary
 - Daniel: Add link between dc and amdgpu glossary

Changes since V2:
 - Add MMHUB

Changes since V1:
 - Yann: Divide glossary based on driver context.
 - Alex: Make terms more consistent and update CPLIB
 - Add new acronyms to the glossary

Reviewed-by: Yann Dirson <ydirson@free.fr>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: Add basic overview of DC pipeline

This commit describes how DCN works by providing high-level diagrams
with an explanation of each component. In particular, it details the
Global Sync signals.

Change since V2:
 - Add a comment about MMHUBBUB.

Reviewed-by: Yann Dirson <ydirson@free.fr>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: How to collect DTN log

Introduce how to collect DTN log from debugfs.

Reviewed-by: Yann Dirson <ydirson@free.fr>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: Document pipe split visual confirmation

Display core provides a feature that makes it easy for users to debug
Pipe Split. This commit introduces how to use such a debug option.

Reviewed-by: Yann Dirson <ydirson@free.fr>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: Document amdgpu_dm_visual_confirm debugfs entry

Display core provides a feature that makes it easy for users to debug
Multiple planes by enabling a visual notification at the bottom of each
plane. This commit introduces how to use such a feature.

Reviewed-by: Yann Dirson <ydirson@free.fr>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

Documentation/gpu: Reorganize DC documentation

Display core documentation is not well organized, and it is hard to find
information due to the lack of sections. This commit reorganizes the
documentation layout, and it is preparation work for future changes.

Changes since V1:
- Christian: Group amdgpu documentation together.
- Daniel: Drop redundant amdgpu prefix.
- Jani: Create index pages.
- Yann: Mirror display folder in the documentation.

Reviewed-by: Yann Dirson <ydirson@free.fr>
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add support for SMU debug option

SMU firmware expects the driver maintains error context
and doesn't interact with SMU any more when SMU errors
occurred. That will aid in debugging SMU firmware issues.

Add SMU debug option support for this request, it can be
enabled or disabled via amdgpu_smu_debug debugfs file.
Use a 32-bit mask to indicate corresponding debug modes.
Currently, only one mode(HALT_ON_ERROR) is supported.
When enabled, it brings hardware to a kind of halt state
so that no one can touch it any more in the envent of SMU
errors.

The dirver interacts with SMU via sending messages. And
threre are three ways to sending messages to SMU in current
implementation. Handle them respectively as following:

1, smu_cmn_send_smc_msg_with_param() for normal timeout cases

  Halt on any error.

2, smu_cmn_send_msg_without_waiting()/smu_cmn_wait_for_response()
for longer timeout cases

  Halt on errors apart from ETIME. Otherwise this way won't work.
  Let the user handle ETIME error in such a case.

3, smu_cmn_send_msg_without_waiting() for no waiting cases

  Halt on errors apart from ETIME. Otherwise second way won't work.

== Command Guide ==

1, enable HALT_ON_ERROR mode

 # echo 0x1 > /sys/kernel/debug/dri/0/amdgpu_smu_debug

2, disable HALT_ON_ERROR mode

 # echo 0x0 > /sys/kernel/debug/dri/0/amdgpu_smu_debug

v5:
 - Use bit mask to allow more debug features.(Evan)
 - Use WRAN() instead of BUG().(Evan)

v4:
 - Set to halt state instead of a simple hang.(Christian)

v3:
 - Use debugfs_create_bool().(Christian)
 - Put variable into smu_context struct.
 - Don't resend command when timeout.

v2:
 - Resend command when timeout.(Lijo)
 - Use debugfs file instead of module parameter.

Signed-off-by: Lang Yu <lang.yu@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: introduce a kind of halt state for amdgpu device

It is useful to maintain error context when debugging
SW/FW issues. Introduce amdgpu_device_halt() for this
purpose. It will bring hardware to a kind of halt state,
so that no one can touch it any more.

Compare to a simple hang, the system will keep stable
at least for SSH access. Then it should be trivial to
inspect the hardware state and see what's going on.

v2:
 - Set adev->no_hw_access earlier to avoid potential crashes.(Christian)

Suggested-by: Christian Koenig <christian.koenig@amd.com>
Suggested-by: Andrey Grodzovsky <andrey.grodzovsky@amd.com>
Signed-off-by: Lang Yu <lang.yu@amd.com>
Reviewed-by: Christian Koenig <christian.koenig@amd.co>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: check df_funcs and its callback pointers

in case they are not avaiable in early phase

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

drm/amdgpu: don't override default ECO_BITs setting

Leave this bit as hardware default setting

Signed-off-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: correct register access for RLC_JUMP_TABLE_RESTORE

should count on GC IP base address

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

drm/amdgpu: read and authenticate ip discovery binary

read and authenticate ip discovery binary getting from
vram first, if it is not valid, read and authenticate
the one getting from file

Signed-off-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 helper to verify ip discovery binary signature

To be used to check ip discovery binary signature

Signed-off-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: rename discovery_read_binary helper

add _from_vram in the funciton name to diffrentiate
the one used to read from file

Signed-off-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 helper to load ip_discovery binary from file

To be used when ip_discovery binary is not carried by vbios

Signed-off-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: fix incorrect VCN revision in SRIOV

Guest OS will setup VCN instance 1 which is disabled as an enabled instance and
execute initialization work on it, but this causes VCN ib ring test failure
on the disabled VCN instance during modprobe:

amdgpu 0000:00:08.0: amdgpu: ring vcn_enc_1.0 uses VM inv eng 5 on hub 1
amdgpu 0000:00:08.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on vcn_dec_0 (-110).
amdgpu 0000:00:08.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on vcn_enc_0.0 (-110).
[drm:amdgpu_device_delayed_init_work_handler [amdgpu]] *ERROR* ib ring test failed (-110).

v2: drop amdgpu_discovery_get_vcn_version and rename sriov_config to
vcn_config
v3: modify VCN's revision in SR-IOV and bare-metal

Fixes: baf3f8f3740625 ("drm/amdgpu: handle SRIOV VCN revision parsing")
Signed-off-by: Leslie Shi <Yuliang.Shi@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: add modifiers in amdgpu_vkms_plane_init()

Fix following warning in SRIOV during modprobe:

amdgpu 0000:00:08.0: GFX9+ requires FB check based on format modifier
WARNING: CPU: 0 PID: 1023 at drivers/gpu/drm/amd/amdgpu/amdgpu_display.c:1150 amdgpu_display_framebuffer_init+0x8e7/0xb40 [amdgpu]

Signed-off-by: Leslie Shi <Yuliang.Shi@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: disable default navi2x co-op kernel support

This patch reverts the following:
commit 48733b224fa7ba ("drm/amdkfd: add Navi2x to GWS init conditions")

Disable GWS usage in default settings for now due to FW bugs.

Signed-off-by: Jonathan Kim <jonathan.kim@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: add Navi2x to GWS init conditions

Initalize GWS on Navi2x with mec2_fw_version >= 0x42.

Signed-off-by: Graham Sider <Graham.Sider@amd.com>
Reviewed-and-tested-by: Jonathan Kim <jonathan.kim@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: only hw fini SMU fisrt for ASICs need that

We found some headaches on ASICs don't need that,
so remove that for them.

Suggested-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Lang Yu <lang.yu@amd.com>
Reviewed-by: Kevin Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: remove power on/off SDMA in SMU hw_init/fini()

Currently, we don't find some neccesities to power on/off
SDMA in SMU hw_init/fini(). It makes more sense in SDMA
hw_init/fini().

Signed-off-by: Lang Yu <lang.yu@amd.com>
Reviewed-by: Kevin Wang <kevinyang.wang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdkfd: Make KFD support on Hawaii experimental

Hawaii support is mostly untested these days. ROCm user mode also
depends on custom firmware for AQL packet processing, that was never
pushed upstream due to quality regressions in graphics driver testing.

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

drm/amdkfd: Don't split unchanged SVM ranges

If an existing SVM range overlaps an svm_range_set_attr call, we would
normally split it in order to update only the overlapping part.
However, if the attributes of the existing range would not be changed
splitting it is unnecessary.

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

drm/amdkfd: Fix svm_range_is_same_attrs

The existing function doesn't compare the access bitmaps and flags.
This can result in failure to update those attributes in existing
ranges when all other attributes remained unchanged.

Because the access and flags attributes modify only some bits in the
respective bitmaps, we cannot compare them directly. Instead we need to
check whether applying the attributes to a particular range would
change the bitmaps.

A PREFETCH_LOC attribute must always trigger a migration, even if the
attribute value remains unchanged. E.g. if some pages were migrated due
to a CPU page fault, a prefetch must still be executed to migrate pages
back to VRAM.

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

drm/amdkfd: Fix error handling in svm_range_add

Add null-pointer check after the last svm_range_new call. This was
originally reported by Zhou Qingyang <zhou1615@umn.edu> based on a
static analyzer.

To avoid duplicating the unwinding code from svm_range_handle_overlap,
I merged the two functions into one.

Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com>
Cc: Zhou Qingyang <zhou1615@umn.edu>
Reviewed-by: Philip Yang <Philip.Yang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>

drm/amdgpu: Handle fault with same timestamp

Remove not unique timestamp WARNING as same timestamp interrupt happens
on some chips,

Drain fault need to wait for the processed_timestamp to be truly greater
than the checkpoint or the ring to be empty to be sure no stale faults
are handled.

Bug: https://gitlab.freedesktop.org/drm/amd/-/issues/1818
Signed-off-by: Philip Yang <Philip.Yang@amd.com>
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>