When an eGPU is unplugged the KFD topology should also be destroyed
for that GPU. This never happens because the fini_sw callbacks never
get to run. Run them manually before calling amdgpu_device_ip_fini_early()
when a device has already been disconnected.
This location is intentionally chosen to make sure that the kfd locking
refcount doesn't get incremented unintentionally.
Cc: kent.russell@amd.com
Closes: https://community.frame.work/t/amd-egpu-on-linux/8691/33
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Kent Russell <kent.russell@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 6a23e7b433)
Cc: stable@vger.kernel.org
Skipping power ungate exposed some scenarios that will fail
like below:
```
amdgpu: Register(0) [regVPEC_QUEUE_RESET_REQ] failed to reach value 0x00000000 != 0x00000001n
amdgpu 0000:c1:00.0: amdgpu: VPE queue reset failed
...
amdgpu: [drm] *ERROR* wait_for_completion_timeout timeout!
```
The underlying s2idle issue that prompted this commit is going to
be fixed in BIOS.
This reverts commit 2a6c826cfe.
This was lost in the 6.19 merge so reapply it.
Fixes: 2a6c826cfe ("drm/amd: Skip power ungate during suspend for VPE")
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reported-by: Konstantin <answer2019@yandex.ru>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220812
Reported-by: Matthew Schwartz <matthew.schwartz@linux.dev>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 3925683515)
During Mode 1 reset, the ASIC undergoes a reset cycle and becomes
temporarily inaccessible via PCIe. Any attempt to access MMIO registers
during this window (e.g., from interrupt handlers or other driver threads)
can result in uncompleted PCIe transactions, leading to NMI panics or
system hangs.
To prevent this, set the `no_hw_access` flag to true immediately after
triggering the reset. This signals other driver components to skip
register accesses while the device is offline.
A memory barrier `smp_mb()` is added to ensure the flag update is
globally visible to all cores before the driver enters the sleep/wait
state.
Signed-off-by: Perry Yuan <perry.yuan@amd.com>
Reviewed-by: Yifan Zhang <yifan1.zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 7edb503fe4)
Pull PCI updates from Bjorn Helgaas:
"Enumeration:
- Enable host bridge emulation for PCI_DOMAINS_GENERIC platforms (Dan
Williams)
- Switch vmd from custom domain number allocator to the common
allocator to prevent a potential race with new non-VMD buses (Dan
Williams)
- Enable Precision Time Measurement (PTM) only if device advertises
support for a relevant role, to prevent invalid PTM Requests that
cause ACS violations that are reported as AER Uncorrectable
Non-Fatal errors (Mika Westerberg)
Resource management:
- Prevent resource tree corruption when BAR resize fails (Ilpo
Järvinen)
- Restore BARs to the original size if a BAR resize fails (Ilpo
Järvinen)
- Remove BAR release from BAR resize attempts by the xe, i915, and
amdgpu drivers so the PCI core can restore BARs if the resize fails
(Ilpo Järvinen)
- Move Resizable BAR code to rebar.c (Ilpo Järvinen)
- Add pci_rebar_size_supported() and use it in i915 and xe (Ilpo
Järvinen)
- Add pci_rebar_get_max_size() and use it in xe and amdgpu (Ilpo
Järvinen)
Power management and error handling:
- For drivers using PCI legacy suspend, save config state at suspend
so that state (not any earlier state from enumeration, probe, or
error recovery) will be restored when resuming (Lukas Wunner)
- For devices with no driver or a driver that lacks power management,
save config state at hibernate so that state (not any earlier state
from enumeration, probe, or error recovery) will be restored when
resuming (Lukas Wunner)
- Save device config space on device addition, before driver binding,
so error recovery works more reliably (Lukas Wunner)
- Drop pci_save_state() from several drivers that no longer need it
since the PCI core always does it and pci_restore_state() no longer
invalidates the saved state (Lukas Wunner)
- Document use of pci_save_state() by drivers to capture the state
they want restored during error recovery (Lukas Wunner)
Power control:
- Add a struct pci_ops.assert_perst() function pointer to
assert/deassert PCIe PERST# and implement it for the qcom driver
(Krishna Chaitanya Chundru)
- Add DT binding and pwrctrl driver for the Toshiba TC9563 PCIe
switch, which must be held in reset after poweron so the pwrctrl
driver can configure the switch via I2C before bringing up the
links (Krishna Chaitanya Chundru)
Endpoint framework:
- Convert the endpoint doorbell test to use a threaded IRQ to fix a
'sleeping while atomic' issue (Bhanu Seshu Kumar Valluri)
- Add endpoint VNTB MSI doorbell support to reduce latency between
host and endpoint (Frank Li)
New native PCIe controller drivers:
- Add CIX Sky1 host controller DT binding and driver (Hans Zhang)
- Add NXP S32G host controller DT binding and driver (Vincent
Guittot)
- Add Renesas RZ/G3S host controller DT binding and driver (Claudiu
Beznea)
- Add SpacemiT K1 host controller DT binding and driver (Alex Elder)
Amlogic Meson PCIe controller driver:
- Update DT binding to name DBI region 'dbi', not 'elbi', and update
driver to support both (Manivannan Sadhasivam)
Apple PCIe controller driver:
- Move struct pci_host_bridge allocation from pci_host_common_init()
to callers, which significantly simplifies pcie-apple (Marc
Zyngier)
Broadcom STB PCIe controller driver:
- Disable advertising ASPM L0s support correctly (Jim Quinlan)
- Add a panic/die handler to print diagnostic info in case PCIe
caused an unrecoverable abort (Jim Quinlan)
Cadence PCIe controller driver:
- Add module support for Cadence platform host and endpoint
controller driver (Manikandan K Pillai)
- Split headers into 'legacy' (LGA) and 'high perf' (HPA) to prepare
for new CIX Sky1 driver (Manikandan K Pillai)
MediaTek PCIe controller driver:
- Convert DT binding to YAML schema (Christian Marangi)
- Add Airoha AN7583 DT compatible and driver support (Christian
Marangi)
Qualcomm PCIe controller driver:
- Add Qualcomm Kaanapali to SM8550 DT binding (Qiang Yu)
- Add required 'power-domains' and 'resets' to qcom sa8775p, sc7280,
sc8280xp, sm8150, sm8250, sm8350, sm8450, sm8550, x1e80100 DT
schemas (Krzysztof Kozlowski)
- Look up OPP using both frequency and data rate (not just frequency)
so RPMh votes can account for both (Krishna Chaitanya Chundru)
Rockchip DesignWare PCIe controller driver:
- Add Rockchip RK3528 compatible strings in DT binding (Yao Zi)
STMicroelectronics STM32MP25 PCIe controller driver:
- Fix a race between link training and endpoint register
initialization (Christian Bruel)
- Align endpoint allocations to match the ATU requirements (Christian
Bruel)
Synopsys DesignWare PCIe controller driver:
- Clear L1 PM Substate Capability 'Supported' bits unless glue driver
says it's supported, which prevents users from enabling non-working
L1SS. Currently only qcom and tegra194 support L1SS (Bjorn Helgaas)
- Remove now-superfluous L1SS disable code from tegra194 (Bjorn
Helgaas)
- Configure L1SS support in dw-rockchip when DT says
'supports-clkreq' (Shawn Lin)
TI Keystone PCIe controller driver:
- Fail the probe instead of silently succeeding if ks_pcie_of_data
didn't specify Root Complex or Endpoint mode (Siddharth Vadapalli)
- Make keystone buildable as a loadable module, except on ARM32 where
hook_fault_code() is __init (Siddharth Vadapalli)"
* tag 'pci-v6.19-changes' of git://git.kernel.org/pub/scm/linux/kernel/git/pci/pci: (100 commits)
MAINTAINERS: Add Manivannan Sadhasivam as PCI/pwrctrl maintainer
MAINTAINERS: Add CIX Sky1 PCIe controller driver maintainer
PCI: sky1: Add PCIe host support for CIX Sky1
dt-bindings: PCI: Add CIX Sky1 PCIe Root Complex bindings
PCI: cadence: Add support for High Perf Architecture (HPA) controller
MAINTAINERS: Add NXP S32G PCIe controller driver maintainer
PCI: s32g: Add NXP S32G PCIe controller driver (RC)
PCI: dwc: Add register and bitfield definitions
dt-bindings: PCI: s32g: Add NXP S32G PCIe controller
PCI: Add Renesas RZ/G3S host controller driver
PCI: host-generic: Move bridge allocation outside of pci_host_common_init()
dt-bindings: PCI: Add Renesas RZ/G3S PCIe controller binding
PCI: Validate pci_rebar_size_supported() input
Documentation: PCI: Amend error recovery doc with pci_save_state() rules
treewide: Drop pci_save_state() after pci_restore_state()
PCI/ERR: Ensure error recoverability at all times
PCI/PM: Stop needlessly clearing state_saved on enumeration and thaw
PCI/PM: Reinstate clearing state_saved in legacy and !PM codepaths
PCI: dw-rockchip: Configure L1SS support
PCI: tegra194: Remove unnecessary L1SS disable code
...
Pull drm updates from Dave Airlie:
"There was a rather late merge of a new color pipeline feature, that
some userspace projects are blocked on, and has seen a lot of work in
amdgpu. This should have seen some time in -next. There is additional
support for this for Intel, that if it arrives in the next day or two
I'll pass it on in another pull request and you can decide if you want
to take it.
Highlights:
- Arm Ethos NPU accelerator driver
- new DRM color pipeline support
- amdgpu will now run discrete SI/CIK cards instead of radeon, which
enables vulkan support in userspace
- msm gets gen8 gpu support
- initial Xe3P support in xe
Full detail summary:
New driver:
- Arm Ethos-U65/U85 accel driver
Core:
- support the drm color pipeline in vkms/amdgfx
- add support for drm colorop pipeline
- add COLOR PIPELINE plane property
- add DRM_CLIENT_CAP_PLANE_COLOR_PIPELINE
- throttle dirty worker with vblank
- use drm_for_each_bridge_in_chain_scoped in drm's bridge code
- Ensure drm_client_modeset tests are enabled in UML
- add simulated vblank interrupt - use in drivers
- dumb buffer sizing helper
- move freeing of drm client memory to driver
- crtc sharpness strength property
- stop using system_wq in scheduler/drivers
- support emergency restore in drm-client
Rust:
- make slice::as_flattened usable on all supported rustc
- add FromBytes::from_bytes_prefix() method
- remove redundant device ptr from Rust GEM object
- Change how AlwaysRefCounted is implemented for GEM objects
gpuvm:
- Add deferred vm_bo cleanup to GPUVM (for rust)
atomic:
- cleanup and improve state handling interfaces
buddy:
- optimize block management
dma-buf:
- heaps: Create heap per CMA reserved location
- improve userspace documentation
dp:
- add POST_LT_ADJ_REQ training sequence
- DPCD dSC quirk for synaptics panamera devices
- helpers to query branch DSC max throughput
ttm:
- Rename ttm_bo_put to ttm_bo_fini
- allow page protection flags on risc-v
- rework pipelined eviction fence handling
amdgpu:
- enable amdgpu by default for SI/CI dGPUs
- enable DC by default on SI
- refactor CIK/SI enablement
- add ABM KMS property
- Re-enable DM idle optimizations
- DC Analog encoders support
- Powerplay fixes for fiji/iceland
- Enable DC on bonaire by default
- HMM cleanup
- Add new RAS framework
- DML2.1 updates
- YCbCr420 fixes
- DC FP fixes
- DMUB fixes
- LTTPR fixes
- DTBCLK fixes
- DMU cursor offload handling
- Userq validation improvements
- Unify shutdown callback handling
- Suspend improvements
- Power limit code cleanup
- SR-IOV fixes
- AUX backlight fixes
- DCN 3.5 fixes
- HDMI compliance fixes
- DCN 4.0.1 cursor updates
- DCN interrupt fix
- DC KMS full update improvements
- Add additional HDCP traces
- DCN 3.2 fixes
- DP MST fixes
- Add support for new SR-IOV mailbox interface
- UQ reset support
- HDP flush rework
- VCE1 support
amdkfd:
- HMM cleanups
- Relax checks on save area overallocations
- Fix GPU mappings after prefetch
radeon:
- refactor CIK/SI enablement
xe:
- Initial Xe3P support
- panic support on VRAM for display
- fix stolen size check
- Loosen used tracking restriction
- New SR-IOV debugfs structure and debugfs updates
- Hide the GPU madvise flag behind a VM_BIND flag
- Always expose VRAM provisioning data on discrete GPUs
- Allow VRAM mappings for userptr when used with SVM
- Allow pinning of p2p dma-buf
- Use per-tile debugfs where appropriate
- Add documentation for Execution Queues
- PF improvements
- VF migration recovery redesign work
- User / Kernel VRAM partitioning
- Update Tile-based messages
- Allow configfs to disable specific GT types
- VF provisioning and migration improvements
- use SVM range helpers in PT layer
- Initial CRI support
- access VF registers using dedicated MMIO view
- limit number of jobs per exec queue
- add sriov_admin sysfs tree
- more crescent island specific support
- debugfs residency counter
- SRIOV migration work
- runtime registers for GFX 35
i915:
- add initial Xe3p_LPD display version 35 support
- Enable LNL+ content adaptive sharpness filter
- Use optimized VRR guardband
- Enable Xe3p LT PHY
- enable FBC support for Xe3p_LPD display
- add display 30.02 firmware support
- refactor SKL+ watermark latency setup
- refactor fbdev handling
- call i915/xe runtime PM via function pointers
- refactor i915/xe stolen memory/display interfaces
- use display version instead of gfx version in display code
- extend i915_display_info with Type-C port details
- lots of display cleanups/refactorings
- set O_LARGEFILE in __create_shmem
- skuip guc communication warning on reset
- fix time conversions
- defeature DRRS on LNL+
- refactor intel_frontbuffer split between i915/xe/display
- convert inteL_rom interfaces to struct drm_device
- unify display register polling interfaces
- aovid lock inversion when pinning to GGTT on CHV/BXT+VTD
panel:
- Add KD116N3730A08/A12, chromebook mt8189
- JT101TM023, LQ079L1SX01,
- GLD070WX3-SL01 MIPI DSI
- Samsung LTL106AL0, Samsung LTL106AL01
- Raystar RFF500F-AWH-DNN
- Winstar WF70A8SYJHLNGA
- Wanchanglong w552946aaa
- Samsung SOFEF00
- Lenovo X13s panel
- ilitek-ili9881c - add rpi 5" support
- visionx-rm69299 - add backlight support
- edp - support AUI B116XAN02.0
bridge:
- improve ref counting
- ti-sn65dsi86 - add support for DP mode with HPD
- synopsis: support CEC, init timer with correct freq
- ASL CS5263 DP-to-HDMI bridge support
nova-core:
- introduce bitfield! macro
- introduce safe integer converters
- GSP inits to fully booted state on Ampere
- Use more future-proof register for GPU identification
nova-drm:
- select NOVA_CORE
- 64-bit only
nouveau:
- improve reclocking on tegra 186+
- add large page and compression support
msm:
- GPU:
- Gen8 support: A840 (Kaanapali) and X2-85 (Glymur)
- A612 support
- MDSS:
- Added support for Glymur and QCS8300 platforms
- DPU:
- Enabled Quad-Pipe support, unlocking higher resolutions support
- Added support for Glymur platform
- Documented DPU on QCS8300 platform as supported
- DisplayPort:
- Added support for Glymur platform
- Added support lame remapping inside DP block
- Documented DisplayPort controller on QCS8300 and SM6150/QCS615
as supported
tegra:
- NVJPG driver
panfrost:
- display JM contexts over debugfs
- export JM contexts to userspace
- improve error and job handling
panthor:
- support custom ASN_HASH for mt8196
- support mali-G1 GPU
- flush shmem write before mapping buffers uncached
- make timeout per-queue instead of per-job
mediatek:
- MT8195/88 HDMIv2/DDCv2 support
rockchip:
- dsi: add support for RK3368
amdxdna:
- enhance runtime PM
- last hardware error reading uapi
- support firmware debug output
- add resource and telemetry data uapi
- preemption support
imx:
- add driver for HDMI TX Parallel audio interface
ivpu:
- add support for user-managed preemption buffer
- add userptr support
- update JSM firware API to 3.33.0
- add better alloc/free warnings
- fix page fault in unbind all bos
- rework bind/unbind of imported buffers
- enable MCA ECC signalling
- split fw runtime and global memory buffers
- add fdinfo memory statistics
tidss:
- convert to drm logging
- logging cleanup
ast:
- refactor generation init paths
- add per chip generation detect_tx_chip
- set quirks for each chip model
atmel-hlcdc:
- set LCDC_ATTRE register in plane disable
- set correct values for plane scaler
solomon:
- use drm helper for get_modes and move_valid
sitronix:
- fix output position when clearing screens
qaic:
- support dma-buf exports
- support new firmware's READ_DATA implementation
- sahara AIC200 image table update
- add sysfs support
- add coredump support
- add uevents support
- PM support
sun4i:
- layer refactors to decouple plane from output
- improve DE33 support
vc4:
- switch to generic CEC helpers
komeda:
- use drm_ logging functions
vkms:
- configfs support for display configuration
vgem:
- fix fence timer deadlock
etnaviv:
- add HWDB entry for GC8000 Nano Ultra VIP r6205"
* tag 'drm-next-2025-12-03' of https://gitlab.freedesktop.org/drm/kernel: (1869 commits)
Revert "drm/amd: Skip power ungate during suspend for VPE"
drm/amdgpu: use common defines for HUB faults
drm/amdgpu/gmc12: add amdgpu_vm_handle_fault() handling
drm/amdgpu/gmc11: add amdgpu_vm_handle_fault() handling
drm/amdgpu: use static ids for ACP platform devs
drm/amdgpu/sdma6: Update SDMA 6.0.3 FW version to include UMQ protected-fence fix
drm/amdgpu: Forward VMID reservation errors
drm/amdgpu/gmc8: Delegate VM faults to soft IRQ handler ring
drm/amdgpu/gmc7: Delegate VM faults to soft IRQ handler ring
drm/amdgpu/gmc6: Delegate VM faults to soft IRQ handler ring
drm/amdgpu/gmc6: Cache VM fault info
drm/amdgpu/gmc6: Don't print MC client as it's unknown
drm/amdgpu/cz_ih: Enable soft IRQ handler ring
drm/amdgpu/tonga_ih: Enable soft IRQ handler ring
drm/amdgpu/iceland_ih: Enable soft IRQ handler ring
drm/amdgpu/cik_ih: Enable soft IRQ handler ring
drm/amdgpu/si_ih: Enable soft IRQ handler ring
drm/amd/display: fix typo in display_mode_core_structs.h
drm/amd/display: fix Smart Power OLED not working after S4
drm/amd/display: Move RGB-type check for audio sync to DCE HW sequence
...
Skipping power ungate exposed some scenarios that will fail
like below:
```
amdgpu: Register(0) [regVPEC_QUEUE_RESET_REQ] failed to reach value 0x00000000 != 0x00000001n
amdgpu 0000:c1:00.0: amdgpu: VPE queue reset failed
...
amdgpu: [drm] *ERROR* wait_for_completion_timeout timeout!
```
The underlying s2idle issue that prompted this commit is going to
be fixed in BIOS.
This reverts commit 2a6c826cfe.
Fixes: 2a6c826cfe ("drm/amd: Skip power ungate during suspend for VPE")
Cc: stable@vger.kernel.org
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Reported-by: Konstantin <answer2019@yandex.ru>
Closes: https://bugzilla.kernel.org/show_bug.cgi?id=220812
Reported-by: Matthew Schwartz <matthew.schwartz@linux.dev>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When trying to unload amdgpu in the SteamDeck (TTY mode), the following
set of errors happens and the system gets unstable:
[..]
[drm] Initialized amdgpu 3.64.0 for 0000:04:00.0 on minor 0
amdgpu 0000:04:00.0: [drm:amdgpu_ib_ring_tests [amdgpu]] *ERROR* IB test failed on gfx_0.0.0 (-110).
amdgpu 0000:04:00.0: amdgpu: ib ring test failed (-110).
[..]
amdgpu 0000:04:00.0: amdgpu: SMU: I'm not done with your previous command: SMN_C2PMSG_66:0x0000001E SMN_C2PMSG_82:0x00000000
amdgpu 0000:04:00.0: amdgpu: Failed to disable gfxoff!
amdgpu 0000:04:00.0: amdgpu: SMU: I'm not done with your previous command: SMN_C2PMSG_66:0x0000001E SMN_C2PMSG_82:0x00000000
amdgpu 0000:04:00.0: amdgpu: Failed to disable gfxoff!
[..]
When the driver initializes the GPU, the PSP validates all the firmware
loaded, and after that, it is not possible to load any other firmware
unless the device is reset. What is happening in the load/unload
situation is that PSP halts the GC engine because it suspects that
something is amiss. To address this issue, this commit ensures that the
GPU is reset (mode 2 reset) in the unload sequence.
Acked-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Suggested-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Rodrigo Siqueira <siqueira@igalia.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
During the suspend sequence VPE is already going to be power gated
as part of vpe_suspend(). It's unnecessary to call during calls to
amdgpu_device_set_pg_state().
It actually can expose a race condition with the firmware if s0i3
sequence starts as well. Drop these calls.
Cc: Peyton.Lee@amd.com
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 2a6c826cfe)
Cc: stable@vger.kernel.org
During the suspend sequence VPE is already going to be power gated
as part of vpe_suspend(). It's unnecessary to call during calls to
amdgpu_device_set_pg_state().
It actually can expose a race condition with the firmware if s0i3
sequence starts as well. Drop these calls.
Cc: Peyton.Lee@amd.com
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
PCI core handles releasing device's resources and their rollback in case of
failure of a BAR resizing operation. Releasing resource prior to calling
pci_resize_resource() prevents PCI core from restoring the BARs as they
were.
Remove driver-side release of BARs from the amdgpu driver.
Also remove the driver initiated assignment as pci_resize_resource() should
try to assign as much as possible. If the driver side call manages to get
more required resources assigned in some scenario, such a problem should be
fixed inside pci_resize_resource() instead.
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Tested-by: Alex Bennée <alex.bennee@linaro.org> # AVA, AMD GPU
Link: https://patch.msgid.link/20251113162628.5946-11-ilpo.jarvinen@linux.intel.com
BAR resize operation is implemented in the pci_resize_resource() and
pbus_reassign_bridge_resources() functions. pci_resize_resource() can be
called either from __resource_resize_store() from sysfs or directly by the
driver for the Endpoint Device.
The pci_resize_resource() requires that caller has released the device
resources that share the bridge window with the BAR to be resized as
otherwise the bridge window is pinned in place and cannot be changed.
pbus_reassign_bridge_resources() rolls back resources if the resize
operation fails, but rollback is performed only for the bridge windows.
Because releasing the device resources are done by the caller of the BAR
resize interface, these functions performing the BAR resize do not have
access to the device resources as they were before the resize.
pbus_reassign_bridge_resources() could try __pci_bridge_assign_resources()
after rolling back the bridge windows as they were, however, it will not
guarantee the resource are assigned due to differences in how FW and the
kernel assign the resources (alignment of the start address and tail).
To perform rollback robustly, the BAR resize interface has to be altered to
also release the device resources that share the bridge window with the BAR
to be resized.
Also, remove restoring from the entries failed list as saved list should
now contain both the bridge windows and device resources so the extra
restore is duplicated work.
Some drivers (currently only amdgpu) want to prevent releasing some
resources. Add exclude_bars param to pci_resize_resource() and make amdgpu
pass its register BAR (BAR 2 or 5), which should never be released during
resize operation. Normally 64-bit prefetchable resources do not share a
bridge window with the 32-bit only register BAR, but there are various
fallbacks in the resource assignment logic which may make the resources
share the bridge window in rare cases.
This change (together with the driver side changes) is to counter the
resource releases that had to be done to prevent resource tree corruption
in the ("PCI: Release assigned resource before restoring them") change. As
such, it likely restores functionality in cases where device resources were
released to avoid resource tree conflicts which appeared to be "working"
when such conflicts were not correctly detected by the kernel.
Reported-by: Simon Richter <Simon.Richter@hogyros.de>
Link: https://lore.kernel.org/linux-pci/f9a8c975-f5d3-4dd2-988e-4371a1433a60@hogyros.de/
Reported-by: Alex Bennée <alex.bennee@linaro.org>
Link: https://lore.kernel.org/linux-pci/874irqop6b.fsf@draig.linaro.org/
Signed-off-by: Ilpo Järvinen <ilpo.jarvinen@linux.intel.com>
[bhelgaas: squash amdgpu BAR selection from
https://lore.kernel.org/r/20251114103053.13778-1-ilpo.jarvinen@linux.intel.com]
Signed-off-by: Bjorn Helgaas <bhelgaas@google.com>
Tested-by: Alex Bennée <alex.bennee@linaro.org> # AVA, AMD GPU
Reviewed-by: Christian König <christian.koenig@amd.com>
Link: https://patch.msgid.link/20251113162628.5946-7-ilpo.jarvinen@linux.intel.com
Now that DC supports analog connectors, it has reached feature
parity with the legacy non-DC display driver on SI dGPUs.
Use the DC display driver by default on SI dGPUs, unless it is
explicitly disabled using the amdgpu.dc=0 module parameter.
DC brings proper support for DP/HDMI audio, DP MST,
10-bit colors, some HDR features, atomic modesetting, etc.
Also clarify the comment about what is missing to have full
DC support for CIK APUs.
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When amdgpu_device_health_check fails, amdgpu_ras_pre_reset
will not be called and therefore amdgpu_ras_post_reset
cannot be called either.
Signed-off-by: YiPeng Chai <YiPeng.Chai@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For S3 on vangogh, PMFW needs to be notified before the
driver powers down RLC. This already happens in smu_disable_dpms()
so drop the superfluous call in amdgpu_device_suspend().
Co-developed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
(cherry picked from commit 960e30a61e)
During gpu reset, all GPU-related resources are
inaccessible. To avoid affecting ras functionality,
suspend ras module before gpu reset and resume
it after gpu reset is complete.
V2:
Rename functions to avoid misunderstanding.
V3:
Move flush_delayed_work to amdgpu_ras_process_pause,
Move schedule_delayed_work to amdgpu_ras_process_unpause.
V4:
Rename functions.
V5:
Move the function to amdgpu_ras.c.
Signed-off-by: YiPeng Chai <YiPeng.Chai@amd.com>
Reviewed-by: Tao Zhou <tao.zhou1@amd.com>
Reviewed-by: Hawking Zhang <Hawking.Zhang@amd.com>
Acked-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This patch adds robust reset handling for user queues (userq) to improve
recovery from queue failures. The key components include:
1. Queue detection and reset logic:
- amdgpu_userq_detect_and_reset_queues() identifies failed queues
- Per-IP detect_and_reset callbacks for targeted recovery
- Falls back to full GPU reset when needed
2. Reset infrastructure:
- Adds userq_reset_work workqueue for async reset handling
- Implements pre/post reset handlers for queue state management
- Integrates with existing GPU reset framework
3. Error handling improvements:
- Enhanced state tracking with HUNG state
- Automatic reset triggering on critical failures
- VRAM loss handling during recovery
4. Integration points:
- Added to device init/reset paths
- Called during queue destroy, suspend, and isolation events
- Handles both individual queue and full GPU resets
The reset functionality works with both gfx/compute and sdma queues,
providing better resilience against queue failures while minimizing
disruption to unaffected queues.
v2: add detection and reset calls when preemption/unmaped fails.
add a per device userq counter for each user queue type.(Alex)
v3: make sure we hold the adev->userq_mutex when we call amdgpu_userq_detect_and_reset_queues. (Alex)
warn if the adev->userq_mutex is not held.
v4: make sure we have all of the uqm->userq_mutex held.
warn if the uqm->userq_mutex is not held.
v5: Use array for user queue type counters.(Alex)
all of the uqm->userq_mutex need to be held when calling detect and reset. (Alex)
v6: fix lock dep warning in amdgpu_userq_fence_dence_driver_process
v7: add the queue types in an array and use a loop in amdgpu_userq_detect_and_reset_queues (Lijo)
v8: remove atomic_set(&userq_mgr->userq_count[i], 0).
it should already be 0 since we kzalloc the structure (Alex)
v9: For consistency with kernel queues, We may want something like:
amdgpu_userq_is_reset_type_supported (Alex)
Signed-off-by: Jesse Zhang <Jesse.Zhang@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If any hardware IPs involved with the second phase of suspend fail, unwind
all steps to restore back to original state.
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If any hardware IPs involved with the first phase of suspend fail, unwind
all steps to restore back to original state.
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
For S3 on vangogh, PMFW needs to be notified before the
driver powers down RLC. This already happens in smu_disable_dpms()
so drop the superfluous call in amdgpu_device_suspend().
Co-developed-by: Mario Limonciello (AMD) <superm1@kernel.org>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello (AMD) <superm1@kernel.org>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
This commit refactors the AMDGPU userqueue management subsystem to replace
IDR (ID Allocation) with XArray for improved performance, scalability, and
maintainability. The changes address several issues with the previous IDR
implementation and provide better locking semantics.
Key changes:
1. **Global XArray Introduction**:
- Added `userq_doorbell_xa` to `struct amdgpu_device` for global queue tracking
- Uses doorbell_index as key for efficient global lookup
- Replaces the previous `userq_mgr_list` linked list approach
2. **Per-process XArray Conversion**:
- Replaced `userq_idr` with `userq_mgr_xa` in `struct amdgpu_userq_mgr`
- Maintains per-process queue tracking with queue_id as key
- Uses XA_FLAGS_ALLOC for automatic ID allocation
3. **Locking Improvements**:
- Removed global `userq_mutex` from `struct amdgpu_device`
- Replaced with fine-grained XArray locking using XArray's internal spinlocks
4. **Runtime Idle Check Optimization**:
- Updated `amdgpu_runtime_idle_check_userq()` to use xa_empty
5. **Queue Management Functions**:
- Converted all IDR operations to equivalent XArray functions:
- `idr_alloc()` → `xa_alloc()`
- `idr_find()` → `xa_load()`
- `idr_remove()` → `xa_erase()`
- `idr_for_each()` → `xa_for_each()`
Benefits:
- **Performance**: XArray provides better scalability for large numbers of queues
- **Memory Efficiency**: Reduced memory overhead compared to IDR
- **Thread Safety**: Improved locking semantics with XArray's internal spinlocks
v2: rename userq_global_xa/userq_xa to userq_doorbell_xa/userq_mgr_xa
Remove xa_lock and use its own lock.
v3: Set queue->userq_mgr = uq_mgr in amdgpu_userq_create()
v4: use xa_store_irq (Christian)
hold the read side of the reset lock while creating/destroying queues and the manager data structure. (Chritian)
Acked-by: Alex Deucher <alexander.deucher@amd.com>
Suggested-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
1. Introduced amdgpu_virt_init_critical_region during VF init.
- VFs use init_data_header_offset and init_data_header_size_kb
transmitted via PF2VF mailbox to fetch the offset of
critical regions' offsets/sizes in VRAM and save to
adev->virt.crit_region_offsets and adev->virt.crit_region_sizes_kb.
Signed-off-by: Ellen Pan <yunru.pan@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Currently SRIOV runtime will use kiq to write HDP_MEM_FLUSH_CNTL for
hdp flush. This register need to be write from CPU for nbif to aware,
otherwise it will not work.
Implement amdgpu_kiq_hdp_flush and use kiq to do gpu hdp flush during
sriov runtime.
v2:
- fallback to amdgpu_asic_flush_hdp when amdgpu_kiq_hdp_flush failed
- add function amdgpu_mes_hdp_flush
v3:
- changed returned error
Reviewed-by: Lijo Lazar <lijo.lazar@amd.com>
Signed-off-by: Victor Zhao <Victor.Zhao@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There is already a helper for telling if a block is valid, but if
IP handling wants to check if it's HW is enabled no such helper
exists.
Reviewed-by: Harry Wentland <harry.wentland@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Complete reset-on-init sequence before sysfs interfaces are created.
Devices get properly initiaized only after reset, and then only sysfs
interfaces should be made available.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Decouple the amdgpu fence from the amdgpu_job structure.
This lets us clean up the separate fence ops for the embedded
fence and other fences. This also allows us to allocate the
vm fence up front when we allocate the job.
v2: Additional cleanup suggested by Christian
v3: Additional cleanups suggested by Christian
v4: Additional cleanups suggested by David and
vm fence fix
v5: cast seqno (David)
Cc: David.Wu3@amd.com
Cc: christian.koenig@amd.com
Tested-by: David (Ming Qiang) Wu <David.Wu3@amd.com>
Reviewed-by: David (Ming Qiang) Wu <David.Wu3@amd.com>
Reviewed-by: Christian König <christian.koenig@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If a userq failed to suspend the rest of the suspend sequence may
have problems. Pass the error code up to the caller for a decision
on what to do.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
amdgpu_device_ip_suspend_phase2() calls amdgpu_ip_block_suspend()
which already sets HW block status to false when succeeding with
IP suspend. Remove the explicit call in
amdgpu_device_ip_suspend_phase2() so that the status is accurate.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Error handling was introduced in commit e095026f00 ("drm/amdgpu:
validate suspend before function call") so the comment about TODO is no
longer needed.
Fixes: e095026f00 ("drm/amdgpu: validate suspend before function call")
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Mario Limonciello <mario.limonciello@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
There has been multiple complains that 10 seconds are usually to long.
The original requirement for longer timeout came from compute tests on
AMDVLK, since that is no longer a topic reduce the timeout back to 2
seconds for all queues.
While at it also remove any special handling for compute queues under
SRIOV or pass through.
v2: fix checkpatch warning.
Signed-off-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Enabling ASPM causes randoms hangs on Tahiti and Oland on Zen4.
It's unclear if this is a platform-specific or GPU-specific issue.
Disable ASPM on SI for the time being.
Reviewed-by: Alex Deucher <alexander.deucher@amd.com>
Signed-off-by: Timur Kristóf <timur.kristof@gmail.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
If reinitialization of one of the GPUs fails after reset, it logs
failure on all subsequent GPUs eventhough they have resumed
successfully.
A sample log where only device at 0000:95:00.0 had a failure -
amdgpu 0000:15:00.0: amdgpu: GPU reset(19) succeeded!
amdgpu 0000:65:00.0: amdgpu: GPU reset(19) succeeded!
amdgpu 0000:75:00.0: amdgpu: GPU reset(19) succeeded!
amdgpu 0000:85:00.0: amdgpu: GPU reset(19) succeeded!
amdgpu 0000:95:00.0: amdgpu: GPU reset(19) failed
amdgpu 0000:e5:00.0: amdgpu: GPU reset(19) failed
amdgpu 0000:f5:00.0: amdgpu: GPU reset(19) failed
amdgpu 0000:05:00.0: amdgpu: GPU reset(19) failed
amdgpu 0000:15:00.0: amdgpu: GPU reset end with ret = -5
To avoid confusion, report the error for each device
separately and return the first error as the overall result.
Signed-off-by: Lijo Lazar <lijo.lazar@amd.com>
Reviewed-by: Asad Kamal <asad.kamal@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
