The current cleanup order during file descriptor close can lead to
a race condition where the eviction fence worker attempts to access
a destroyed mutex from the user queue manager:
[ 517.294055] DEBUG_LOCKS_WARN_ON(lock->magic != lock)
[ 517.294060] WARNING: CPU: 8 PID: 2030 at kernel/locking/mutex.c:564
[ 517.294094] Workqueue: events amdgpu_eviction_fence_suspend_worker [amdgpu]
The issue occurs because:
1. We destroy the user queue manager (including its mutex) first
2. Then try to destroy eviction fences which may have pending work
3. The eviction fence worker may try to access the already-destroyed mutex
Fix this by reordering the cleanup to:
1. First mark the fd as closing and destroy eviction fences,
which flushes any pending work
2. Then safely destroy the user queue manager after we're certain
no more fence work will be executed
The copy in amdgpu_driver_postclose_kms() needs to be removed (Christian)
Reviewed-by: Christian König <christian.koenig@amd.com>
Reviewed-by: Prike Liang <Prike.Liang@amd.com>
Reviewed-by: Arvind Yadav <Arvind.Yadav@amd.com>
Signed-off-by: Jesse Zhang <Jesse.Zhang@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
Static analysis shows that pointer "svms" cannot be NULL because it points
to the object "struct svm_range_list". Remove the extra NULL check. It is
meaningless and harms the readability of the code.
In the function svm_range_get_info() there is no possibility of failure.
Therefore, the caller of the function svm_range_get_info() does not need
a return value. Change the function svm_range_get_info() return type from
"int" to "void".
Since the function svm_range_get_info() has a return type of "void". The
caller of the function svm_range_get_info() does not need a return value.
Delete extra code.
Found by Linux Verification Center (linuxtesting.org) with SVACE.
Signed-off-by: Andrey Vatoropin <a.vatoropin@crpt.ru>
Signed-off-by: Felix Kuehling <felix.kuehling@amd.com>
Reviewed-by: Felix Kuehling <felix.kuehling@amd.com>
Signed-off-by: Alex Deucher <alexander.deucher@amd.com>
When an entity from application B is killed, drm_sched_entity_kill()
removes all jobs belonging to that entity through
drm_sched_entity_kill_jobs_work(). If application A's job depends on a
scheduled fence from application B's job, and that fence is not properly
signaled during the killing process, application A's dependency cannot be
cleared.
This leads to application A hanging indefinitely while waiting for a
dependency that will never be resolved. Fix this issue by ensuring that
scheduled fences are properly signaled when an entity is killed, allowing
dependent applications to continue execution.
Signed-off-by: Lin.Cao <lincao12@amd.com>
Reviewed-by: Philipp Stanner <phasta@kernel.org>
Signed-off-by: Christian König <christian.koenig@amd.com>
Link: https://lore.kernel.org/r/20250515020713.1110476-1-lincao12@amd.com
Updates for v6.16
CI:
- uprev mesa
GPU:
- ACD (Adaptive Clock Distribution) support for X1-85. This is required
enable the higher frequencies.
- Drop fictional `address_space_size`. For some older devices, the address
space size is limited to 4GB to avoid potential 64b rollover math problems
in the fw. For these, an `ADRENO_QUIRK_4GB_VA` quirk is added. For
everyone else we get the address space size from the SMMU `ias` (input
address sizes), which is usually 48b.
- Improve robustness when GMU HFI responses time out
- Fix crash when throttling GPU immediately during boot
- Fix for rgb565_predicator on Adreno 7c3
- Remove `MODULE_FIRMWARE()`s for GPU, the GPU can load the firmware after
probe and having partial set of fw (ie. sqe+gmu but not zap) causes problems
MDSS:
- Added SAR2130P support to MDSS driver
DPU:
- Changed to use single CTL path for flushing on DPU 5.x+
- Improved SSPP allocation code to allow sharing of SSPP between planes
- Enabled SmartDMA on SM8150, SC8180X, SC8280XP, SM8550
- Added SAR2130P support
- Disabled DSC support on MSM8937, MSM8917, MSM8953, SDM660
- Misc fixes
DP:
- Switch to use new helpers for DP Audio / HDMI codec handling
- Fixed LTTPR handling
DSI:
- Added support for SA8775P
- Added SAR2130P support
MDP4:
- Fixed LCDC / LVDS controller on
HDMI:
- Switched to use new helpers for ACR data
- Fixed old standing issue of HPD not working in some cases
Signed-off-by: Dave Airlie <airlied@redhat.com>
From: Rob Clark <robdclark@gmail.com>
Link: https://lore.kernel.org/r/CAF6AEGv2Go+nseaEwRgeZbecet-h+Pf2oBKw1CobCF01xu2XVg@mail.gmail.com
Adds basic support for the new display classes available on GB20x GPUs.
Most of the changes here deal with HW method moves, with the only other
change of note being tweaks to skip allocation of CTXDMA objects, which
aren't required on Blackwell display.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Some older NVIDIA and some newer NVIDIA hardware/firmware seems to
have issues with address only transactions (firmware rejects them).
Add an option to the core drm dp to avoid address only transactions,
This just puts the MOT flag removal on the last message of the transfer
and avoids the start of transfer transaction.
This with the flag set in nouveau, allows eDP probing on GB203 device.
Signed-off-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This commit adds support for the GB20x GPUs found on GeForce RTX 50xx
series boards.
Beyond a few miscellaneous register moves and HW class ID plumbing,
this reuses most of the code added to support GH100/GB10x.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This commit enables basic support for the GB100/GB102 Blackwell GPUs.
Beyond HW class ID plumbing there's very little change here vs GH100.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
From VOLTA_CHANNEL_GPFIFO_A onwards, HW no longer updates the GET/GP_GET
pointers in USERD following channel progress, but instead updates on a
timer for compatibility, and SW is expected to implement its own method
of tracking channel progress (typically via non-WFI semaphore release).
Nouveau has been making use of the compatibility mode up until now,
however, from BLACKWELL_CHANNEL_GPFIFO_A HW no longer supports USERD
writeback at all.
Allocate a per-channel buffer in system memory, and append a non-WFI
semaphore release to the end of each push buffer segment to simulate
the pointers previously read from USERD.
This change is implemented for Fermi (which is the first to support non-
WFI semaphore release) onwards, as readback from system memory is likely
faster than BAR1 reads.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Primarily a cleanup to allow for changes in newer CHANNEL_GPFIFO classes
to be more easily implemented.
Compared to the prior implementation, this submits userspace push buffer
segments as subroutines and uses the NV_RAMUSERD_TOP_LEVEL_GET registers
to track the main (kernel) push buffer progress.
Fixes a number of sporadic failures seen during piglit runs.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
This commit enables basic support for Hopper GPUs, and is intended
primarily as a base supporting Blackwell GPUs, which reuse most of
the code added here.
Advanced features such as Confidential Compute are not supported.
Beyond a few miscellaneous register moves and HW class ID plumbing,
the bulk of the changes implemented here are to support the GSP-RM
boot sequence used on Hopper/Blackwell GPUs, as well as a new page
table layout.
There should be no changes here that impact prior GPUs.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Co-developed-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Timur Tabi <ttabi@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
GPUs exist now with a 64-bit BAR0, which mean that BAR1 and BAR2's
indices (as passed to pci_resource_len() etc) are bumped up by one.
Modify nvkm_device.resource_addr/size() to take an enum instead of
an integer bar index, and take IORESOURCE_MEM_64 into account when
translating to the "raw" bar id.
[airlied: fixup ERR_PTR]
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
HOPPER_CHANNEL_GPFIFO_A removes the SEMAPHORE[A-D] methods that are
currently used by nouveau to implement fences on GF100 and newer.
Switch to the newer SEM methods available from VOLTA_CHANNEL_GPFIFO,
which are also available on the Hopper/Blackwell host classes.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Use data from 'struct nvkm_vmm_page/desc' to determine which PDEs need
to be mirrored to RM instead of hardcoded values for pre-Hopper page
tables.
Needed to support Hopper/Blackwell.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
The current code using NV90F1_CTRL_CMD_VASPACE_COPY_SERVER_RESERVED_PDES
not only requires changes to support the new page table layout used on
Hopper/Blackwell GPUs, but is also broken in that it always mirrors the
PDEs used for virtual address 0, rather than the area reserved for RM.
This works fine for the non-NVK case where the kernel has full control
of the VMM layout and things end up in the right place, but NVK puts
its kernel reserved area much higher in the address space.
Fixing the code to work at any VA is not enough as some parts of RM want
the reserved area in a specific location, and NVK would then hit other
assertions in RM instead.
Fortunately, it appears that RM never needs to allocate anything within
its reserved area for DRM clients, and the COPY_SERVER_RESERVED_PDES
control call primarily serves to allow RM to locate the root page table
when initialising a channel's instance block.
Flag VMMs allocated by the DRM driver as externally owned, and use
NV0080_CTRL_CMD_DMA_SET_PAGE_DIRECTORY to inform RM of the root page
table in a similar way to NVIDIA's UVM driver.
The COPY_SERVER_RESERVED_PDES paths are kept for the golden context
image and gr scrubber channel, where RM needs the reserved area.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
When mirroring BAR2 page tables to RM, we need to know the level shift
for the root page table (which is currently hardcoded), as well as the
raw PDE value (which is currently hardcoded in GP1xx-AD1xx format).
In order to support GH100/GBxxx, modify the code to determine the page
shift from per-GPU info in nvkm_vmm_page, as well as read the relevant
PDE back from the root page table rather than recalculating it.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
GH100/GBxxx have moved the register that controls where in VRAM the
the BAR0 NV_PRAMIN window points.
Add a HAL for this, as the BAR0 window is needed for BAR2 bootstrap.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
These registers have moved on GH100/GBxxx, and the GSP-RM init code uses
hardcoded values from earlier GPUs to fill GspSystemInfo.
Replace the per-GPU accessors in nvkm_pci_func with region info, and use
it when initialising GspSystemInfo.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
Add r570-specific HAL routines, and support loading of GSP-RM version
570.144 if firmware is available.
There should be no impact on r535, or non-GSP paths.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
570.144 has incompatible changes to NV0000_ALLOC_PARAMETERS.
Factor out the common code so it can be shared.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
570.86.15 uses a slightly different calculation for the size of the
sysmem buffer needed to store GSP-RM's vidmem data across suspend.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
570.86.16 has incompatible changes to NV_CHANNEL_ALLOC_PARAMS.
At the same time, remove the duplicated channel allocation code from
golden context init.
Signed-off-by: Ben Skeggs <bskeggs@nvidia.com>
Reviewed-by: Dave Airlie <airlied@redhat.com>
Reviewed-by: Timur Tabi <ttabi@nvidia.com>
Tested-by: Timur Tabi <ttabi@nvidia.com>
Signed-off-by: Dave Airlie <airlied@redhat.com>
