Lijo Lazar [Mon, 30 Jan 2023 04:18:39 +0000 (09:48 +0530)]
drm/amdgpu: Add fallback path for discovery info
If SOC doesn't expose dedicated vram, discovery region may be
available through system memory. Rename the existing interface to
generic read_binary_from_mem and add a fallback path to read from system
memory.
Lijo Lazar [Mon, 30 Jan 2023 04:08:09 +0000 (09:38 +0530)]
drm/amdgpu: Read discovery info from system memory
On certain ASICs, discovery info is available at reserved region in system
memory. The location is available through ACPI interface. Add API to read
discovery info from there.
Mukul Joshi [Tue, 31 Jan 2023 16:23:50 +0000 (11:23 -0500)]
drm/amdkfd: Enable SVM on Native mode
This patch enables SVM capability on GFX9.4.3 when
run in Native mode. It also sets best_prefetch and
best_restore locations to CPU as there is no VRAM.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com> Acked-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lijo Lazar [Mon, 16 Jan 2023 05:25:38 +0000 (10:55 +0530)]
drm/amdgpu: Add flags for partition mode query
It's not required to take lock on all cases while querying partition
mode. Querying partition mode during KFD init process doesn't need to
take a lock. Init process after a switch will already be happening under
lock. Control the behaviour by adding flags to xcp_query_partition_mode.
Lijo Lazar [Thu, 9 Mar 2023 07:34:56 +0000 (13:04 +0530)]
drm/amdgpu: Add mode-2 reset in SMU v13.0.6
Modifications to mode-2 reset flow for SMU v13.0.6 ASICs.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com> Reviewed-by: Le Ma <le.ma@amd.com> Reviewed-by: Asad Kamal <asad.kamal@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lijo Lazar [Fri, 10 Mar 2023 06:03:37 +0000 (11:33 +0530)]
drm/amd/pm: Notify PMFW about driver unload cases
On SMU v13.0.6 APUs, FW will need to take some actions if driver is going
to halt RLC. Notify PMFW that driver is not going to manage device so
that FW takes care of the required actions.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com> Reviewed-by: Le Ma <le.ma@amd.com> Reviewed-by: Asad Kamal <asad.kamal@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lijo Lazar [Fri, 10 Mar 2023 12:41:25 +0000 (18:11 +0530)]
drm/amd/pm: Update PMFW headers for version 85.54
It adds message support for FW notification on driver unload.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com> Reviewed-by: Le Ma <le.ma@amd.com> Reviewed-by: Asad Kamal <asad.kamal@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Le Ma [Thu, 16 Mar 2023 09:42:49 +0000 (17:42 +0800)]
drm/amdgpu: change the print level to warn for ip block disabled
Avoid to mislead users as it's not a real error.
Signed-off-by: Le Ma <le.ma@amd.com> Reviewed-by: Asad Kamal <asad.kamal@amd.com> Reviewed-by: Amber Lin <Amber.Lin@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
On GFXIP9.4.3 APP APU where there is no dedicated VRAM domain handle
VRAM BO allocation requests on CPU domain and validate them on GTT.
Support for handling multi-socket and multi-numa partitions within a
socket will be added by future patches, this enables 1P NPS1 asic
bringup configuration.
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com> Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
drm/amdgpu: Handle VRAM dependencies on GFXIP9.4.3
[For 1P NPS1 mode driver bringup]
Changes required to initialize the amdgpu driver with frontdoor firmware
loading and discovery=2 with the native mode SBIOS that enables CPU GPU
unified interleaved memory.
sudo modprobe amdgpu discovery=2
Once PSP TMR region is reported via the ACPI interface, the dependency
on the ip_discovery.bin will be removed.
Choice of where to allocate driver table is given to each IP version. In
general, both GTT and VRAM domains will be considered. If one of the
tables has a strict restriction for VRAM domain, then only VRAM domain
is considered.
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
(lijo: Modified the handling for SMU Tables) Signed-off-by: Lijo Lazar <lijo.lazar@amd.com> Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Hawking Zhang [Wed, 28 Dec 2022 10:18:38 +0000 (18:18 +0800)]
drm/amdgpu: Add mmhub v1_8_0 ras err status registers
add new ras error status registers introduced in
mmhub v1_8_0 to log mmea and mm_cane ras err, including
MMEAx_UE|CE_ERR_STATUS_LO|HI
MM_CANE_UE|CE_ERR_STATUS_LO|HI
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com> Reviewed-by: Tao Zhou <tao.zhou1@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Hawking Zhang [Fri, 3 Feb 2023 08:10:37 +0000 (16:10 +0800)]
drm/amdgpu: Add common helper to reset ras error
Add common helper to reset ras error status. It
applies to IP blocks that follow the new ras error
logging register design, and need to write 0 to
reset the error status. For IP blocks that don't
support the new design, please still implement ip
specific helper.
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com> Reviewed-by: Tao Zhou <tao.zhou1@amd.com> Reviewed-by: Lijo Lazar <lijo.lazar@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Hawking Zhang [Thu, 2 Feb 2023 12:54:08 +0000 (20:54 +0800)]
drm/amdgpu: Add common helper to query ras error (v2)
Add common helper to query ras error status and
log error information, including memory block id
and erorr count. The helpers are applicable to IP
blocks that follow the new ras error logging design.
For IP blocks that don't support the new design,
please still implement ip specific helper to query
ras error.
v2: optimize struct amdgpu_ras_err_status_reg_entry
and the implementaion in helper (Lijo/Tao)
Signed-off-by: Hawking Zhang <Hawking.Zhang@amd.com> Reviewed-by: Tao Zhou <tao.zhou1@amd.com> Reviewed-by: Lijo Lazar <lijo.lazar@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Jonathan Kim [Thu, 2 Feb 2023 16:10:08 +0000 (11:10 -0500)]
drm/amdkfd: bind cpu and hiveless gpu to a hive if xgmi connected
If a CPU and GPU are xGMI connected but the GPU is hiveless with
respect to other GPUs, create a new CPU-GPU hive using the GPU's PCI
device location ID as the new hive ID to maintain fine grain memory
access usage.
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>
Philip Yang [Tue, 24 Jan 2023 15:10:14 +0000 (10:10 -0500)]
drm/amdkfd: Cleanup KFD nodes creation
kfd node allocation outside kfd->num_nodes loop is not needed and causes
memory leak because kfd->num_nodes is at least equal to 1.
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>
This allows backing ttm_tt structure with pages from different NUMA
pools.
Tested-by: Graham Sider <graham.sider@amd.com> Reviewed-by: Felix Kuehling <felix.kuehling@amd.com> Signed-off-by: Christian König <christian.koenig@amd.com> Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lijo Lazar [Fri, 20 Jan 2023 07:12:00 +0000 (12:42 +0530)]
drm/amdgpu: Fix mqd init on GFX v9.4.3
For MQD init, an XCC's queue is selected with GRBM select. However, for
initialization of MQD, values read from logical XCC0 registers are used.
This results in garbage values being read from XCC0 whose queue is not
selected. Change to read from the right XCC for MQD initialization.
Lijo Lazar [Thu, 19 Jan 2023 09:17:22 +0000 (14:47 +0530)]
drm/amdgpu: Use unique doorbell range per xcc
Program different ranges in each XCC with MEC_DOORBELL_RANGE_LOWER/HIGHER.
Keeping the same range causes CPF in other XCCs also to be busy when an IB
packet is submitted to KCQ. Only the XCC which processes the packet
comes back to idle afterwards and this causes other CPs not be idle.
This in turn affects clockgating behavior as RLC doesn't get idle
interrupt.
LOWER/HIGHER covers only KIQ/KCQs which are per XCC queues. Assigning
different ranges doesn't seem to have any side effect as user queue ranges
are outside of this range. User queue tests - PM4 through KFD and AQL
through rocr - have the same results after this change.
Lijo Lazar [Tue, 17 Jan 2023 11:24:49 +0000 (16:54 +0530)]
drm/amdgpu: Keep SDMAv4.4.2 active during reset
During ASIC wide reset, SDMA shouldn't be clockgated and be ready to
accept freeze requests from PMFW. For that, don't stop SDMA engine
during reset and keep the clocks active.
GFXIP 9.4.3 could be in APU or carveout mode but we cannot use the
xgmi.connected_to_cpu flag to identify the iolinks type. Use appropriate
APU or Carveout mode based condition to report xgmi connection in kfd
topology.
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
James Zhu [Tue, 10 Jan 2023 14:01:33 +0000 (09:01 -0500)]
drm/amdgpu: increase AMDGPU_MAX_HWIP_RINGS
[WA] Increase AMDGPU_MAX_HWIP_RINGS to 64 to support more compute
ring resource. Later need redesign with queue/prirority/scheduler
factors to reduce AMDGPU_MAX_HWIP_RINGS.
Signed-off-by: James Zhu <James.Zhu@amd.com> Acked-by: Christian König <christian.koenig@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Graham Sider [Thu, 5 Jan 2023 15:58:07 +0000 (10:58 -0500)]
drm/amdgpu: Correct dGPU MTYPE settings for gfx943
Revert temporary dGPU VRAM MTYPE setting and align with expected
coherency protocol.
Signed-off-by: Graham Sider <Graham.Sider@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Felix Kuehling [Tue, 29 Nov 2022 17:45:26 +0000 (12:45 -0500)]
drm/amdgpu: Allocate GART table in RAM for AMD APU
Some AMD APUs may not have a dedicated VRAM. On such platforms the GART
table should be allocated on the system memory. When real vram size is
zero, place the GART table in system memory and create an SG BO to make
it GPU accessible.
v2: fix includes
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com>
(rajneesh: removed set_memory_wc workaround) Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Harish Kasiviswanathan <Harish.Kasiviswanathan@amd.com> Signed-off-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lijo Lazar [Tue, 20 Dec 2022 08:51:57 +0000 (14:21 +0530)]
drm/amdgpu: Add FGCG logic for GFX v9.4.3
Add logic for fine grain clock gating logic for GFX v9.4.3. The feature
will be controlled using CG flags. Also, make a change so that RLC safe
mode entry/exit is done only once during CG update sequence.
On AMD APP APUs, to make UTCL2 snoop CPU caches, its not sufficient to
rely on xgmi connected flag so add the logic to use is_app_apu to
program the PDE_REQUEST_PHYSICAL bit correctly for gfxhub and mmhub
both.
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Amber Lin [Mon, 28 Nov 2022 16:26:02 +0000 (11:26 -0500)]
amd/amdgpu: Set MTYPE_UC for access over PCIe
For GFX v9_4_3, set MTYPE_UC for memory access over PCIe.
v4 - add missing indentation pointed out by Felix and add his
reviewed-by tag.
v3 - add missing logic for the svm path.
v2 - add amdgpu_xgmi_same_hive to separate access over xgmi from pcie
Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Amber Lin <Amber.Lin@amd.com> Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lijo Lazar [Thu, 15 Dec 2022 07:43:29 +0000 (13:13 +0530)]
drm/amdgpu: Fix GFX 9.4.3 dma address capability
ASICs with GFX 9.4.3 support 48-bit addressing.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com> Acked-by: Christian König <christian.koenig@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Mukul Joshi [Fri, 9 Dec 2022 14:03:01 +0000 (09:03 -0500)]
drm/amdkfd: Setup current_logical_xcc_id in MQD
Setup rolling current_logical_xcc_id in MQD for GFX9.4.3
to ensure each queue starts at a different place and prevent
hotspotting issues. Also, remove updating current_logical_xcc_id
during queue update.
Suggested-by: Joseph Greathouse <Joseph.Greathouse@amd.com> Signed-off-by: Mukul Joshi <mukul.joshi@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Le Ma [Fri, 9 Dec 2022 11:44:05 +0000 (19:44 +0800)]
drm/amdgpu: correct the vmhub index when page fault occurs
The AMDGPU_GFXHUB was bind to each xcc in the logical order.
Thus convert the node_id to logical xcc_id to index the
correct AMDGPU_GFXHUB. And "node_id / 4" can get the correct
AMDGPU_MMHUB0 index.
Signed-off-by: Le Ma <le.ma@amd.com> Tested-by: Asad kamal <asad.kamal@amd.com> Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Mukul Joshi [Thu, 8 Dec 2022 17:08:17 +0000 (12:08 -0500)]
drm/amdkfd: Update packet manager for GFX9.4.3
In GFX 9.4.3, there can be more than 8 SDMA engines.
As a result, extended_engine_sel and engine_sel fields
in MAP_QUEUES packet need to be updated to allow correct
mapping of SDMA queues to these SDMA engines.
Signed-off-by: Mukul Joshi <mukul.joshi@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Lijo Lazar [Thu, 24 Nov 2022 08:53:58 +0000 (14:23 +0530)]
drm/amdgpu: Skip TMR allocation if not required
On ASICs with PSPv13.0.6, TMR is reserved at boot time. There is no need
to allocate TMR region by driver. However, it's still required to send
SETUP_TMR command to PSP.
Lijo Lazar [Wed, 16 Nov 2022 11:45:47 +0000 (17:15 +0530)]
drm/amdgpu: Switch to SOC partition funcs
For GFXv9.4.3, use SOC level partition switch implementation rather than
keeping them at GFX IP level. Change the exisiting implementation in
GFX IP for keeping partition mode and restrict it to only GFX related
switch.
Add function to initialize soc configuration information for GC 9.4.3
ASICs. Use it to map IPs and other SOC related information once IP
configuration information is available through discovery.
For GC9.4.3 compute partition related callbacks are initialized as part
of configuration init.
Switching the partition mode configuration of ASIC is SOC
level function rather than something at GFX core level. Add
partition mode switch functions as SOC specific callbacks.
Implement the XCP manager callbacks needed for partition
switch for GC 9.4.3 based ASICs.
Within a device, an accelerator core partition can be constituted with
different IP instances. These partitions are spatial in nature. Number
of partitions which can exist at the same time depends on the 'partition
mode'. Add a manager entity which is responsible for switching between
different partition modes and maintaining partitions. It is also
responsible for suspend/resume of different partitions.
On GPXIP 9.4.3 APU, in no carveout mode there is no real vram heap and
could be emulated by the driver over the interleaved NUMA system memory
and the APU could also be in the carveout mode during early development
stage or otherwise for debugging purpose so introduce a new member in
amdgpu_gmc to figure out whether the APU is in the native mode as per
the production configuration. AMD_IS_APU cannot be used for Accelerated
Processing Platform APUs as it might be used in a different context on
previous generations or on small APUs.
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com> Reviewed-by: Felix Kuehling <Felix.Kuehling@amd.com> Tested-by: Graham Sider <graham.sider@amd.com> Signed-off-by: Rajneesh Bhardwaj <rajneesh.bhardwaj@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Philip Yang [Mon, 14 Nov 2022 22:35:43 +0000 (17:35 -0500)]
drm/amdgpu: more GPU page fault info for GC v9.4.3
Output IH cookie node_id and translate it to the corresponding AID id
and XCC id, to help debug the GPU page fault.
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>
Shiwu Zhang [Fri, 11 Nov 2022 07:54:52 +0000 (15:54 +0800)]
drm/amdgpu: fix kcq mqd_backup buffer double free for multi-XCD
For gfx_v9_4_3 and beyond, struct kiq has its own mqd_backup pointer
rather than using the last pointer from mec struct. Then the kfree
operation on the pointer from the mec struct should be removed otherwise
it will cause double free on the first kcq's mqd_backup buffer on XCD1.
Signed-off-by: Shiwu Zhang <shiwu.zhang@amd.com> Reviewed-by: Le Ma <le.ma@amd.com> Signed-off-by: Alex Deucher <alexander.deucher@amd.com>