In GuC submit fini, forcefully tear down any exec queues by disabling
CTs, stopping the scheduler (which cleans up lost G2H), killing all
remaining queues, and resuming scheduling to allow any remaining cleanup
actions to complete and signal any remaining fences.
Split guc_submit_fini into device related and software only part. Using
device-managed and drm-managed action guarantees the correct ordering of
cleanup.
Fixes: dd08ebf6c3 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: stable@vger.kernel.org
Reviewed-by: Zhanjun Dong <zhanjun.dong@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20260310225039.1320161-3-zhanjun.dong@intel.com
(cherry picked from commit a6ab444a111a59924bd9d0c1e0613a75a0a40b89)
Signed-off-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
This was done entirely with mindless brute force, using
git grep -l '\<k[vmz]*alloc_objs*(.*, GFP_KERNEL)' |
xargs sed -i 's/\(alloc_objs*(.*\), GFP_KERNEL)/\1)/'
to convert the new alloc_obj() users that had a simple GFP_KERNEL
argument to just drop that argument.
Note that due to the extreme simplicity of the scripting, any slightly
more complex cases spread over multiple lines would not be triggered:
they definitely exist, but this covers the vast bulk of the cases, and
the resulting diff is also then easier to check automatically.
For the same reason the 'flex' versions will be done as a separate
conversion.
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
This is the result of running the Coccinelle script from
scripts/coccinelle/api/kmalloc_objs.cocci. The script is designed to
avoid scalar types (which need careful case-by-case checking), and
instead replace kmalloc-family calls that allocate struct or union
object instances:
Single allocations: kmalloc(sizeof(TYPE), ...)
are replaced with: kmalloc_obj(TYPE, ...)
Array allocations: kmalloc_array(COUNT, sizeof(TYPE), ...)
are replaced with: kmalloc_objs(TYPE, COUNT, ...)
Flex array allocations: kmalloc(struct_size(PTR, FAM, COUNT), ...)
are replaced with: kmalloc_flex(*PTR, FAM, COUNT, ...)
(where TYPE may also be *VAR)
The resulting allocations no longer return "void *", instead returning
"TYPE *".
Signed-off-by: Kees Cook <kees@kernel.org>
clangd reports many "unused header" warnings throughout the Xe driver.
Start working to clean this up by removing unnecessary includes in our
.c files and/or replacing them with explicit includes of other headers
that were previously being included indirectly.
By far the most common offender here was unnecessary inclusion of
xe_gt.h. That likely originates from the early days of xe.ko when
xe_mmio did not exist and all register accesses, including those
unrelated to GTs, were done with GT functions.
There's still a lot of additional #include cleanup that can be done in
the headers themselves; that will come as a followup series.
v2:
- Squash the 79-patch series down to a single patch. (MattB)
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20260115032803.4067824-2-matthew.d.roper@intel.com
Signed-off-by: Matt Roper <matthew.d.roper@intel.com>
We now have proper infrastructure to accurately check the LRC timestamp
without toggling the scheduling state for non-VFs. For VFs, it is still
possible to get an inaccurate view if the context is on hardware. We
guard against free-running contexts on VFs by banning jobs whose
timestamps are not moving. In addition, VFs have a timeslice quantum
that naturally triggers context switches when more than one VF is
running, thus updating the LRC timestamp.
For multi-queue, it is desirable to avoid scheduling toggling in the TDR
because this scheduling state is shared among many queues. Furthermore,
this change simplifies the GuC state machine. The trade-off for VF cases
seems worthwhile.
v5:
- Add xe_lrc_timestamp helper (Umesh)
v6:
- Reduce number of tries on stuck timestamp (VF testing)
- Convert job timestamp save to a memory copy (VF testing)
v7:
- Save ctx timestamp to LRC when start VF job (VF testing)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Umesh Nerlige Ramappa <umesh.nerlige.ramappa@intel.com>
Link: https://patch.msgid.link/20260110012739.2888434-8-matthew.brost@intel.com
Deregistering queues in the TDR introduces unnecessary complexity,
requiring reference-counting techniques to function correctly,
particularly to prevent use-after-free (UAF) issues while a
deregistration initiated from the TDR is in progress.
All that's needed in the TDR is to kick the queue off the hardware,
which is achieved by disabling scheduling. Queue deregistration should
be handled in a single, well-defined point in the cleanup path, tied to
the queue's reference count.
v4:
- Explain why extra ref were needed prior to this patch (Niranjana)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Link: https://patch.msgid.link/20260110012739.2888434-5-matthew.brost@intel.com
Use new pending job list iterator and new helper functions in Xe to
avoid reaching into DRM scheduler internals.
Part of this change involves removing pending jobs debug information
from debugfs and devcoredump. As agreed, the pending job list should
only be accessed when the scheduler is stopped. However, it's not
straightforward to determine whether the scheduler is stopped from the
shared debugfs/devcoredump code path. Additionally, the pending job list
provides little useful information, as pending jobs can be inferred from
seqnos and ring head/tail positions. Therefore, this debug information
is being removed.
v4:
- Add comment around DRM_GPU_SCHED_STAT_NO_HANG (Niranjana)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Link: https://patch.msgid.link/20260110012739.2888434-3-matthew.brost@intel.com
Since engines in the same class can be divided across multiple groups,
the GuC does not allow scheduler groups to be active if there are
multi-lrc contexts. This means that:
1) if a MLRC context is registered when we enable scheduler groups, the
GuC will silently ignore the configuration
2) if a MLRC context is registered after scheduler groups are enabled,
the GuC will disable the groups and generate an adverse event.
The expectation is that the admin will ensure that all apps that use
MLRC on PF have been terminated before scheduler groups are created. A
check is added anyway to make sure we don't still have contexts waiting
to be cleaned up laying around. A check is also added at queue creation
time to block MLRC queue creation if scheduler groups have been enabled.
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251218223846.1146344-19-daniele.ceraolospurio@intel.com
Some upcoming KLVs are sized based on the engine counts, so we need
those defines to be moved to a separate file to include them from
guc_klv_abi.h (which is already included by guc_fwif.h).
Instead of moving just the engine-related defines, it is cleaner to
move all scheduler-related defines (i.e., everything engine or context
related). Note that the legacy GuC defines have not been moved and have
instead been dropped because Xe doesn't support any GuC old enough to
still use them.
While at it, struct guc_ctxt_registration_info has been moved to
guc_submit.c since it doesn't come from the GuC specs (we added it to
make things simpler in our code).
Signed-off-by: Daniele Ceraolo Spurio <daniele.ceraolospurio@intel.com>
Cc: Michal Wajdeczko <michal.wajdeczko@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251218223846.1146344-16-daniele.ceraolospurio@intel.com
If an exec queue is idle, there is no need to issue a schedule disable
to the GuC when suspending the queue’s execution. Opportunistically skip
this step if the queue is idle and not a parallel queue. Parallel queues
must have their scheduling state flipped in the GuC due to limitations
in how submission is implemented in run_job().
Also if all pagefault queues can skip the schedule disable during a
switch to dma-fence mode, do not schedule a resume for the pagefault
queues after the next submission.
v2:
- Don't touch the LRC tail is queue is suspended but enabled in run_job
(CI)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Thomas Hellström <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/20251212182847.1683222-5-matthew.brost@intel.com
As all queues of a multi queue group use the primary queue of the group
to interface with GuC. Hence there is a dependency between the queues of
the group. So, when primary queue of a multi queue group is cleaned up,
also trigger a cleanup of the secondary queues also. During cleanup, stop
and re-start submission for all queues of a multi queue group to avoid
any submission happening in parallel when a queue is being cleaned up.
v2: Initialize group->list_lock, add fs_reclaim dependency, remove
unwanted secondary queues cleanup (Matt Brost)
v3: Properly handle cleanup of multi-queue group (Matt Brost)
v4: Fix IS_ENABLED(CONFIG_LOCKDEP) check (Matt Brost)
Revert stopping/restarting of submissions on queues of the
group in TDR as it is not needed.
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20251211010249.1647839-28-niranjana.vishwanathapura@intel.com
Add support for queues of a multi queue group to set
their priority within the queue group by adding property
DRM_XE_EXEC_QUEUE_SET_PROPERTY_MULTI_QUEUE_PRIORITY.
This is the only other property supported by secondary
queues of a multi queue group, other than
DRM_XE_EXEC_QUEUE_SET_PROPERTY_MULTI_QUEUE.
v2: Add kernel doc for enum xe_multi_queue_priority,
Add assert for priority values, fix includes and
declarations (Matt Brost)
v3: update uapi kernel-doc (Matt Brost)
v4: uapi change due to rebase
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20251211010249.1647839-23-niranjana.vishwanathapura@intel.com
Implement GuC commands and response along with the Context
Group Page (CGP) interface for multi queue support.
Ensure that only primary queue (q0) of a multi queue group
communicate with GuC. The secondary queues of the group only
need to maintain LRCA and interface with drm scheduler.
Use primary queue's submit_wq for all secondary queues of a multi
queue group. This serialization avoids any locking around CGP
synchronization with GuC.
v2: Fix G2H_LEN_DW_MULTI_QUEUE_CONTEXT value, add more comments
(Matt Brost)
v3: Minor code refactro, use xe_gt_assert
v4: Use xe_guc_ct_wake_waiters(), remove vf recovery support
(Matt Brost)
Signed-off-by: Stuart Summers <stuart.summers@intel.com>
Signed-off-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Link: https://patch.msgid.link/20251211010249.1647839-22-niranjana.vishwanathapura@intel.com
If wait for ring space started just before migration, it can delay
the recovery process, by waiting without bailout path for up to 2
seconds.
Two second wait for recovery is not acceptable, and if the ring was
completely filled even without the migration temporarily stopping
execution, then such a wait will result in up to a thousand new jobs
(assuming constant flow) being added while the wait is happening.
While this will not cause data corruption, it will lead to warning
messages getting logged due to reset being scheduled on a GT under
recovery. Also several seconds of unresponsiveness, as the backlog
of jobs gets progressively executed.
Add a bailout condition, to make sure the recovery starts without
much delay. The recovery is expected to finish in about 100 ms when
under moderate stress, so the condition verification period needs to be
below that - settling at 64 ms.
The theoretical max time which the recovery can take depends on how
many requests can be emitted to engine rings and be pending execution.
While stress testing, it was possible to reach 10k pending requests
on rings when a platform with two GTs was used. This resulted in max
recovery time of 5 seconds. But in real life situations, it is very
unlikely that the amount of pending requests will ever exceed 100,
and for that the recovery time will be around 50 ms - well within our
claimed limit of 100ms.
Fixes: a4dae94aad ("drm/xe/vf: Wakeup in GuC backend on VF post migration recovery")
Signed-off-by: Tomasz Lis <tomasz.lis@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://patch.msgid.link/20251204200820.2206168-1-tomasz.lis@intel.com
Introduce pause/unpause() helpers which stop/start further runs of
submission tasks on given GuC and can be called from PF context. This
is in preparation of usecases where we simply need to stop/start the
scheduler without losing GuC state and don't require dealing with VF
migration.
v7: Reword commit message (Matthew Brost)
Signed-off-by: Raag Jadav <raag.jadav@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Ashutosh Dixit <ashutosh.dixit@intel.com>
Link: https://patch.msgid.link/20251030122357.128825-3-raag.jadav@intel.com
The LRC software ring tail is reset to the first unsignaled pending
job's head.
Fix the re-emission logic to begin submitting from the first unsignaled
job detected, rather than scanning all pending jobs, which can cause
imbalance.
v2:
- Include missing local changes
v3:
- s/skip_replay/restore_replay (Tomasz)
Fixes: c25c1010df ("drm/xe/vf: Replay GuC submission state on pause / unpause")
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://patch.msgid.link/20251121152750.240557-1-matthew.brost@intel.com
In normal operation, a registered exec queue is disabled and
deregistered through the GuC, and freed only after the GuC confirms
completion. However, if the driver is forced to unbind while the exec
queue is still running, the user may call exec_destroy() after the GuC
has already been stopped and CT communication disabled.
In this case, the driver cannot receive a response from the GuC,
preventing proper cleanup of exec queue resources. Fix this by directly
releasing the resources when GuC is not running.
Here is the failure dmesg log:
"
[ 468.089581] ---[ end trace 0000000000000000 ]---
[ 468.089608] pci 0000:03:00.0: [drm] *ERROR* GT0: GUC ID manager unclean (1/65535)
[ 468.090558] pci 0000:03:00.0: [drm] GT0: total 65535
[ 468.090562] pci 0000:03:00.0: [drm] GT0: used 1
[ 468.090564] pci 0000:03:00.0: [drm] GT0: range 1..1 (1)
[ 468.092716] ------------[ cut here ]------------
[ 468.092719] WARNING: CPU: 14 PID: 4775 at drivers/gpu/drm/xe/xe_ttm_vram_mgr.c:298 ttm_vram_mgr_fini+0xf8/0x130 [xe]
"
v2: use xe_uc_fw_is_running() instead of xe_guc_ct_enabled().
As CT may go down and come back during VF migration.
Fixes: dd08ebf6c3 ("drm/xe: Introduce a new DRM driver for Intel GPUs")
Cc: stable@vger.kernel.org
Cc: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Shuicheng Lin <shuicheng.lin@intel.com>
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Link: https://lore.kernel.org/r/20251010172529.2967639-2-shuicheng.lin@intel.com
A queue must be in the submission backend's tracking state before the
LRC is created to avoid a race condition where the LRC's GGTT addresses
are not properly fixed up during VF post-migration recovery.
Move the queue initialization—which adds the queue to the submission
backend's tracking state—before LRC creation.
Also wait on pending GGTT fixups before allocating LRCs to avoid racing
with fixups.
v2:
- Wait on VF GGTT fixes before creating LRC (testing)
v5:
- Adjust comment in code (Tomasz)
- Reduce race window
v7:
- Only wakeup waiters in recovery path (CI)
- Wakeup waiters on abort
- Use GT warn on (Michal)
- Fix kernel doc for LRC ring size function (Tomasz)
v8:
- Guard against migration not supported or no memirq (CI)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://lore.kernel.org/r/20251008214532.3442967-28-matthew.brost@intel.com
Fixup GuC submission pause / unpause functions to properly replay any
possible state lost during VF post migration recovery.
v3:
- Add helpers for revert / replay (Tomasz)
- Add comment around WQ NOPs (Tomasz)
v7:
- Only fixup / replay parallel queues once (Testing)
- Skip unpause step on queues created after resfix done (Testing)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://lore.kernel.org/r/20251008214532.3442967-27-matthew.brost@intel.com
If VF post-migration recovery is in progress, the recovery flow will
rebuild all GuC submission state. In this case, exit all waiters to
ensure that submission queue scheduling can also be paused. Avoid taking
any adverse actions after aborting the wait.
As part of waking up the GuC backend, suspend_wait can now return
-EAGAIN indicating the waiter should be retried. If the caller is
running on work item, that work item need to be requeued to avoid a
deadlock for the work item blocking the VF migration recovery work item.
v3:
- Don't block in preempt fence work queue as this can interfere with VF
post-migration work queue scheduling leading to deadlock (Testing)
- Use xe_gt_recovery_inprogress (Michal)
v5:
- Use static function for vf_recovery (Michal)
- Add helper to wake CT waiters (Michal)
- Move some code to following patch (Michal)
- Adjust commit message to explain suspend_wait returning -EAGAIN (Michal)
- Add kernel doc to suspend_wait around returning -EAGAIN
v7:
- Add comment on why a shared wait queue is need on VFs (Michal)
- Guard again suspend_wait signaling early on resfix donw (Tomasz)
v8:
- Fix kernel doc (CI)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Michal Wajdeczko <michal.wajdeczko@intel.com>
Link: https://lore.kernel.org/r/20251008214532.3442967-18-matthew.brost@intel.com
With well-behaved software, a GT reset should never occur, nor should it
happen during VF post-migration recovery. If it does, trigger a warning
but suppress the GT reset, as VF post-migration recovery is expected to
bring the VF back to a working state.
v3:
- Better commit message (Tomasz)
v5:
- Use xe_gt_WARN_ON (Michal)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://lore.kernel.org/r/20251008214532.3442967-17-matthew.brost@intel.com
Now that we save the job's head during submission, it's no longer
necessary to adjust the LRC ring head during resubmission. Instead, a
software-based adjustment of the tail will overwrite the old jobs in
place. For some odd reason, adjusting the LRC ring head didn't work on
parallel queues, which was causing issues in our CI.
v5:
- Add comment in guc_exec_queue_start explaning why the function works
(Auld)
v7:
- Only adjust first state on first unsignaled job (Auld)
v8:
- Break unsignaled job handling to separate patch (Auld)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://lore.kernel.org/r/20251008214532.3442967-7-matthew.brost@intel.com
VF migration requires jobs to remain pending so they can be replayed
after the VF comes back. Previously, LR job fences were intentionally
signaled immediately after submission to avoid the risk of exporting
them, as these fences do not naturally signal in a timely manner and
could break dma-fence contracts. A side effect of this approach was that
LR jobs were never added to the DRM scheduler’s pending list, preventing
them from being tracked for later resubmission.
We now avoid signaling LR job fences and ensure they are never exported;
Xe already guards against exporting these internal fences. With that
guarantee in place, we can safely track LR jobs in the scheduler’s
pending list so they are eligible for resubmission during VF
post-migration recovery (and similar recovery paths).
An added benefit is that LR queues now gain the DRM scheduler’s built-in
flow control over ring usage rather than rejecting new jobs in the exec
IOCTL if the ring is full.
v2:
- Ensure DRM scheduler TDR doesn't run for LR jobs
- Stack variable for killed_or_banned_or_wedged
v4:
- Clarify commit message (Tomasz)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://lore.kernel.org/r/20251008214532.3442967-5-matthew.brost@intel.com
Add explicit tracking in the GuC submission state to record the source
of a pending enable (TDR vs. queue resume path vs. submission).
Disambiguating the origin lets the GuC submission state machine apply
the correct recovery/replay behavior.
This helps VF restore: when the device comes back, the state machine knows
whether the pending enable stems from timeout recovery, from a queue resume
sequence, or submission and can gate sequencing and fixups accordingly.
v4:
- Clarify commit message (Tomasz)
Signed-off-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Tomasz Lis <tomasz.lis@intel.com>
Link: https://lore.kernel.org/r/20251008214532.3442967-4-matthew.brost@intel.com
