github-mirror/linux
1
0
Fork 0
mirror of https://github.com/torvalds/linux.git synced 2026-05-26 08:02:27 +02:00
Code Issues Actions Packages Projects Releases Wiki Activity Graph

Driver Changes:

Browse Source 
- Fix a memory leak (Mika)
 - Fix a 64-bit division (Michal Wajdeczko)
 - vf migration fix (Matt Brost)
 - LRC pause Fix (Tomasz lis)
 -----BEGIN PGP SIGNATURE-----
 
 iHUEABYKAB0WIQRskUM7w1oG5rx2IZO4FpNVCsYGvwUCaTIFzAAKCRC4FpNVCsYG
 vzbpAP9l6oGkwKlAHNVMJRbzBXa8b42KmtbFk4rQkg6TvVrnsgD+M/rVo84Ldg/D
 rb07JTqzHgxlziIpzwri6UPRr8+9KAc=
 =Ax3c
 -----END PGP SIGNATURE-----

Merge tag 'drm-xe-next-fixes-2025-12-04' of https://gitlab.freedesktop.org/drm/xe/kernel into drm-next

Driver Changes:
- Fix a memory leak (Mika)
- Fix a 64-bit division (Michal Wajdeczko)
- vf migration fix (Matt Brost)
- LRC pause Fix (Tomasz lis)

Signed-off-by: Dave Airlie <airlied@redhat.com>

From: Thomas Hellstrom <thomas.hellstrom@linux.intel.com>
Link: https://patch.msgid.link/aTIGiHJnnMtqbDOO@fedora
This commit is contained in:
Dave Airlie 2025-12-05 10:21:11 +10:00
commit 86fafc584c
5 changed files with 36 additions and 23 deletions
Show Stats Download Patch File Download Diff File Expand all files Collapse all files

5
drivers/gpu/drm/xe/xe_gpu_scheduler.h
View File

@ -54,13 +54,14 @@ static inline void xe_sched_tdr_queue_imm(struct xe_gpu_scheduler *sched)
static inline void xe_sched_resubmit_jobs(struct xe_gpu_scheduler *sched)
{
struct drm_sched_job *s_job;
bool restore_replay = false;
list_for_each_entry(s_job, &sched->base.pending_list, list) {
struct drm_sched_fence *s_fence = s_job->s_fence;
struct dma_fence *hw_fence = s_fence->parent;
if (to_xe_sched_job(s_job)->skip_emit ||
(hw_fence && !dma_fence_is_signaled(hw_fence)))
restore_replay |= to_xe_sched_job(s_job)->restore_replay;
if (restore_replay || (hw_fence && !dma_fence_is_signaled(hw_fence)))
sched->base.ops->run_job(s_job);
}
}

2
drivers/gpu/drm/xe/xe_gt_sriov_pf_config.c
View File

@ -711,7 +711,7 @@ static u64 pf_profile_fair_ggtt(struct xe_gt *gt, unsigned int num_vfs)
if (num_vfs > 56)
return SZ_64M - SZ_8M;
return rounddown_pow_of_two(shareable / num_vfs);
return rounddown_pow_of_two(div_u64(shareable, num_vfs));
}
/**

47
drivers/gpu/drm/xe/xe_guc_submit.c
View File

@ -822,7 +822,7 @@ static void submit_exec_queue(struct xe_exec_queue *q, struct xe_sched_job *job)
xe_gt_assert(guc_to_gt(guc), exec_queue_registered(q));
if (!job->skip_emit || job->last_replay) {
if (!job->restore_replay || job->last_replay) {
if (xe_exec_queue_is_parallel(q))
wq_item_append(q);
else
@ -881,10 +881,10 @@ guc_exec_queue_run_job(struct drm_sched_job *drm_job)
if (!killed_or_banned_or_wedged && !xe_sched_job_is_error(job)) {
if (!exec_queue_registered(q))
register_exec_queue(q, GUC_CONTEXT_NORMAL);
if (!job->skip_emit)
if (!job->restore_replay)
q->ring_ops->emit_job(job);
submit_exec_queue(q, job);
job->skip_emit = false;
job->restore_replay = false;
}
/*
@ -2112,6 +2112,18 @@ static void guc_exec_queue_revert_pending_state_change(struct xe_guc *guc,
q->guc->resume_time = 0;
}
static void lrc_parallel_clear(struct xe_lrc *lrc)
{
struct xe_device *xe = gt_to_xe(lrc->gt);
struct iosys_map map = xe_lrc_parallel_map(lrc);
int i;
for (i = 0; i < WQ_SIZE / sizeof(u32); ++i)
parallel_write(xe, map, wq[i],
FIELD_PREP(WQ_TYPE_MASK, WQ_TYPE_NOOP) |
FIELD_PREP(WQ_LEN_MASK, 0));
}
/*
* This function is quite complex but only real way to ensure no state is lost
* during VF resume flows. The function scans the queue state, make adjustments
@ -2135,8 +2147,8 @@ static void guc_exec_queue_pause(struct xe_guc *guc, struct xe_exec_queue *q)
guc_exec_queue_revert_pending_state_change(guc, q);
if (xe_exec_queue_is_parallel(q)) {
struct xe_device *xe = guc_to_xe(guc);
struct iosys_map map = xe_lrc_parallel_map(q->lrc[0]);
/* Pairs with WRITE_ONCE in __xe_exec_queue_init */
struct xe_lrc *lrc = READ_ONCE(q->lrc[0]);
/*
* NOP existing WQ commands that may contain stale GGTT
@ -2144,14 +2156,14 @@ static void guc_exec_queue_pause(struct xe_guc *guc, struct xe_exec_queue *q)
* seems to get confused if the WQ head/tail pointers are
* adjusted.
*/
for (i = 0; i < WQ_SIZE / sizeof(u32); ++i)
parallel_write(xe, map, wq[i],
FIELD_PREP(WQ_TYPE_MASK, WQ_TYPE_NOOP) |
FIELD_PREP(WQ_LEN_MASK, 0));
if (lrc)
lrc_parallel_clear(lrc);
}
job = xe_sched_first_pending_job(sched);
if (job) {
job->restore_replay = true;
/*
* Adjust software tail so jobs submitted overwrite previous
* position in ring buffer with new GGTT addresses.
@ -2241,17 +2253,18 @@ static void guc_exec_queue_unpause_prepare(struct xe_guc *guc,
struct xe_exec_queue *q)
{
struct xe_gpu_scheduler *sched = &q->guc->sched;
struct drm_sched_job *s_job;
struct xe_sched_job *job = NULL;
bool restore_replay = false;
list_for_each_entry(s_job, &sched->base.pending_list, list) {
job = to_xe_sched_job(s_job);
list_for_each_entry(job, &sched->base.pending_list, drm.list) {
restore_replay |= job->restore_replay;
if (restore_replay) {
xe_gt_dbg(guc_to_gt(guc), "Replay JOB - guc_id=%d, seqno=%d",
q->guc->id, xe_sched_job_seqno(job));
xe_gt_dbg(guc_to_gt(guc), "Replay JOB - guc_id=%d, seqno=%d",
q->guc->id, xe_sched_job_seqno(job));
q->ring_ops->emit_job(job);
job->skip_emit = true;
q->ring_ops->emit_job(job);
job->restore_replay = true;
}
}
if (job)

1
drivers/gpu/drm/xe/xe_pagefault.c
View File

@ -102,7 +102,6 @@ static int xe_pagefault_handle_vma(struct xe_gt *gt, struct xe_vma *vma,
/* Lock VM and BOs dma-resv */
xe_validation_ctx_init(&ctx, &vm->xe->val, &exec, (struct xe_val_flags) {});
drm_exec_init(&exec, 0, 0);
drm_exec_until_all_locked(&exec) {
err = xe_pagefault_begin(&exec, vma, tile->mem.vram,
needs_vram == 1);

4
drivers/gpu/drm/xe/xe_sched_job_types.h
View File

@ -63,8 +63,8 @@ struct xe_sched_job {
bool ring_ops_flush_tlb;
/** @ggtt: mapped in ggtt. */
bool ggtt;
/** @skip_emit: skip emitting the job */
bool skip_emit;
/** @restore_replay: job being replayed for restore */
bool restore_replay;
/** @last_replay: last job being replayed */
bool last_replay;
/** @ptrs: per instance pointers. */