diff options
| author | Tangudu Tilak Tirumalesh <tilak.tirumalesh.tangudu@intel.com> | 2026-08-27 13:50:11 +0530 |
|---|---|---|
| committer | Shuicheng Lin <shuicheng.lin@intel.com> | 2026-08-28 13:41:47 -0700 |
| commit | d3f1b058a77d1754d469ea4eaf43e754d43cdc4c (patch) | |
| tree | 47e29dccb50433bcb044f177a16d02952cac6cfd /drivers | |
| parent | 0f0a5aba875992c4d0aaf8c6178550557fa38ce2 (diff) | |
| download | linux-next-d3f1b058a77d1754d469ea4eaf43e754d43cdc4c.tar.gz linux-next-d3f1b058a77d1754d469ea4eaf43e754d43cdc4c.zip | |
drm/xe/ggtt: invalidate engine GGTT TLBs for multi-queue GTs
Engines cache GGTT translations in their own TLBs, which the
xe_tlb_inval_ggtt() does not touch.
After a GGTT unmap the engines can still hit the stale translation and
read the old page.
Issue an INVAL_FULL (intra vf) invalidation on GGTT unmap, gated on the
GT having any multi-queue engine class.
v2:
- Use xe_gt_has_multi_queue() helper and address nits (Niranjana)
v3:
- Invalidate both the primary and media GTs, similar to
xe_ggtt_invalidate() (Matt Brost)
v5:
- Guard NULL gt and explain the multi-queue-only engine flush
(Niranjana)
Assisted-by: Claude:claude-opus-4-8
Reviewed-by: Matthew Brost <matthew.brost@intel.com>
Reviewed-by: Niranjana Vishwanathapura <niranjana.vishwanathapura@intel.com>
Signed-off-by: Tangudu Tilak Tirumalesh <tilak.tirumalesh.tangudu@intel.com>
Link: https://patch.msgid.link/20260827082011.3370672-4-tilak.tirumalesh.tangudu@intel.com
Signed-off-by: Shuicheng Lin <shuicheng.lin@intel.com>
Diffstat (limited to 'drivers')
| -rw-r--r-- | drivers/gpu/drm/xe/xe_ggtt.c | 34 |
1 files changed, 33 insertions, 1 deletions
diff --git a/drivers/gpu/drm/xe/xe_ggtt.c b/drivers/gpu/drm/xe/xe_ggtt.c index 9b58f47986f6..63e8bf193605 100644 --- a/drivers/gpu/drm/xe/xe_ggtt.c +++ b/drivers/gpu/drm/xe/xe_ggtt.c @@ -451,6 +451,7 @@ int xe_ggtt_init_early(struct xe_ggtt *ggtt) ALLOW_ERROR_INJECTION(xe_ggtt_init_early, ERRNO); /* See xe_pci_probe() */ static void xe_ggtt_invalidate(struct xe_ggtt *ggtt); +static void xe_ggtt_invalidate_engine(struct xe_ggtt *ggtt); static void xe_ggtt_initial_clear(struct xe_ggtt *ggtt) { @@ -482,8 +483,11 @@ static void ggtt_node_remove(struct xe_ggtt_node *node) xe_ggtt_clear(ggtt, xe_ggtt_node_addr(node), xe_ggtt_node_size(node)); drm_mm_remove_node(&node->base); node->base.size = 0; - if (bound && node->invalidate_on_remove) + if (bound && node->invalidate_on_remove) { xe_ggtt_invalidate(ggtt); + /* Drain engine TLBs so a recycled range can't hit a stale entry. */ + xe_ggtt_invalidate_engine(ggtt); + } mutex_unlock(&ggtt->lock); ggtt_node_fini(node); @@ -601,6 +605,34 @@ static void xe_ggtt_invalidate(struct xe_ggtt *ggtt) ggtt_invalidate_gt_tlb(ggtt->tile->media_gt); } +static void ggtt_invalidate_gt_engine_tlb(struct xe_gt *gt) +{ + int err; + + /* + * Multi-queue GTs miss engine GGTT TLB invalidations (GuC + * lite-restore skips the full context restore), so the engine + * flush is needed there. + */ + if (!gt || !xe_gt_has_multi_queue(gt)) + return; + + err = xe_tlb_inval_ggtt_full(>->tlb_inval); + xe_gt_WARN(gt, err, "Failed to invalidate engine GGTT TLBs (%pe)", + ERR_PTR(err)); +} + +/* + * Drain engine-side GGTT TLBs on teardown so a recycled range's next + * occupant can't hit a predecessor's cached translation. + */ +static void xe_ggtt_invalidate_engine(struct xe_ggtt *ggtt) +{ + /* Each GT in a tile has its own engine TLBs to drain */ + ggtt_invalidate_gt_engine_tlb(ggtt->tile->primary_gt); + ggtt_invalidate_gt_engine_tlb(ggtt->tile->media_gt); +} + /** * xe_ggtt_shift_nodes() - Shift GGTT nodes to adjust for a change in usable address range. * @ggtt: the &xe_ggtt struct instance |
