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>
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>
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>
[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>
[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: 986430446c ("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>
[Why]
We'll exit optimized power state to do link detection but we won't enter
back into the optimized power state.
This could potentially block s2idle entry depending on the sequencing,
but it also means we're losing some power during the transition period.
[How]
Hook up the handler like DCN21. It was also missed like the
exit_optimized_pwr_state callback.
Fixes: 64b1d0e8d5 ("drm/amd/display: Add DCN3.1 HWSEQ")
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Eric Yang <Eric.Yang2@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[Why]
Otherwise SMU won't mark Display as idle when trying to perform s2idle.
[How]
Mark the bit in the dcn31 codepath, doesn't apply to older ASIC.
It needed to be split from phy refclk off to prevent entering s2idle
when PSR was engaged but driver was not ready.
Fixes: 118a331516 ("drm/amd/display: Add DCN3.1 clock manager support")
Tested-by: Daniel Wheeler <daniel.wheeler@amd.com>
Reviewed-by: Eric Yang <Eric.Yang2@amd.com>
Acked-by: Rodrigo Siqueira <Rodrigo.Siqueira@amd.com>
Signed-off-by: Nicholas Kazlauskas <nicholas.kazlauskas@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
[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>
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>
[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: 986430446c ("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>
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>
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>
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>
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>
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>
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>
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>
Chips with no display hardware should return false for
DC support.
v2: drop Arcturus and Aldebaran
Fixes: f7f12b2582 ("drm/amdgpu: default to true in amdgpu_device_asic_has_dc_support")
Reviewed-by: Evan Quan <evan.quan@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Reported-by: Tareque Md.Hanif <tarequemd.hanif@yahoo.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
By setting mp1_state as PP_MP1_STATE_UNLOAD, MP1 will do some proper cleanups and
put itself into a state ready for PNP. That can workaround some random resuming
failure observed on BOCO capable platforms.
Signed-off-by: Evan Quan <evan.quan@amd.com>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Guchun Chen <guchun.chen@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If the platform suspend happens to fail and the power rail
is not turned off, the GPU will be in an unknown state on
resume, so reset the asic so that it will be in a known
good state on resume even if the platform suspend failed.
v2: handle s0ix
Acked-by: Luben Tuikov <luben.tuikov@amd.com>
Acked-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
It's not only supported by HG/PX laptops. It's supported
by all dGPUs which supports BOCO/BACO functionality (runtime
D3).
BOCO - Bus Off, Chip Off. The entire chip is powered off.
This is controlled by ACPI.
BACO - Bus Active, Chip Off. The chip still shows up
on the PCI bus, but the device itself is powered
down.
v2: fix missed HG/PX reference
Reviewed-by: Evan Quan <evan.quan@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
AMD systems currently lay out MCA bank types such that the type of bank
number "i" is either the same across all CPUs or is Reserved/Read-as-Zero.
For example:
Bank # | CPUx | CPUy
0 LS LS
1 RAZ UMC
2 CS CS
3 SMU RAZ
Future AMD systems will lay out MCA bank types such that the type of
bank number "i" may be different across CPUs.
For example:
Bank # | CPUx | CPUy
0 LS LS
1 RAZ UMC
2 CS NBIO
3 SMU RAZ
Change the structures that cache MCA bank types to be per-CPU and update
smca_get_bank_type() to handle this change.
Move some SMCA-specific structures to amd.c from mce.h, since they no
longer need to be global.
Break out the "count" for bank types from struct smca_hwid, since this
should provide a per-CPU count rather than a system-wide count.
Apply the "const" qualifier to the struct smca_hwid_mcatypes array. The
values in this array should not change at runtime.
Signed-off-by: Yazen Ghannam <yazen.ghannam@amd.com>
Signed-off-by: Borislav Petkov <bp@suse.de>
Link: https://lore.kernel.org/r/20211216162905.4132657-3-yazen.ghannam@amd.com
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>
Cc: stable@vger.kernel.org
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>
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>
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>
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>
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>
Backmerging for v5.16-rc5. Resolves a conflict between drm-misc-next
and drm-misc-fixes in the vc4 driver.
Signed-off-by: Thomas Zimmermann <tzimmermann@suse.de>
