Don't immediately write the seqno into the breadcrumb slot, but wait
until we've attempted to flush the writes; that is we need to ensure the
memory is coherent prior to updating the breadcrumb so that any
observers who see the new seqno can proceed.
Signed-off-by: Chris Wilson <chris@xxxxxxxxxxxxxxxxxx>
---
.../gpu/drm/i915/gt/intel_ring_submission.c | 23 +++++++++++++------
1 file changed, 16 insertions(+), 7 deletions(-)
diff --git a/drivers/gpu/drm/i915/gt/intel_ring_submission.c b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
index f70b903a98bc..c99017e3c3c9 100644
--- a/drivers/gpu/drm/i915/gt/intel_ring_submission.c
+++ b/drivers/gpu/drm/i915/gt/intel_ring_submission.c
@@ -446,31 +446,40 @@ static u32 *gen6_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
return cs;
}
-#define GEN7_XCS_WA 32
+#define GEN7_XCS_WA 8
static u32 *gen7_xcs_emit_breadcrumb(struct i915_request *rq, u32 *cs)
{
+ const u32 mi_flush_seqno =
+ MI_FLUSH_DW | MI_INVALIDATE_TLB |
+ MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX;
int i;
GEM_BUG_ON(i915_request_active_timeline(rq)->hwsp_ggtt != rq->engine->status_page.vma);
GEM_BUG_ON(offset_in_page(i915_request_active_timeline(rq)->hwsp_offset) != I915_GEM_HWS_SEQNO_ADDR);
- *cs++ = MI_FLUSH_DW | MI_INVALIDATE_TLB |
- MI_FLUSH_DW_OP_STOREDW | MI_FLUSH_DW_STORE_INDEX;
+ /* Stalling flush before we update the actual breadcrumb */
+ *cs++ = mi_flush_seqno;
+ *cs++ = (I915_GEM_HWS_SEQNO_ADDR + 4) | MI_FLUSH_DW_USE_GTT;
+ *cs++ = rq->fence.seqno;
+
+ /* Write the seqno into the breadcrumb */
+ *cs++ = mi_flush_seqno;
*cs++ = I915_GEM_HWS_SEQNO_ADDR | MI_FLUSH_DW_USE_GTT;
*cs++ = rq->fence.seqno;
+ /* Delay for the seqno to land! */
for (i = 0; i < GEN7_XCS_WA; i++) {
*cs++ = MI_STORE_DWORD_INDEX;
*cs++ = I915_GEM_HWS_SEQNO_ADDR;
*cs++ = rq->fence.seqno;
}
- *cs++ = MI_FLUSH_DW;
- *cs++ = 0;
- *cs++ = 0;
+ /* One final stall! */
+ *cs++ = mi_flush_seqno;
+ *cs++ = I915_GEM_HWS_SEQNO_ADDR | MI_FLUSH_DW_USE_GTT;
+ *cs++ = rq->fence.seqno;
*cs++ = MI_USER_INTERRUPT;
- *cs++ = MI_NOOP;
rq->tail = intel_ring_offset(rq, cs);
assert_ring_tail_valid(rq->ring, rq->tail);
--
2.25.0
_______________________________________________
Intel-gfx mailing list
Intel-gfx@xxxxxxxxxxxxxxxxxxxxx
https://lists.freedesktop.org/mailman/listinfo/intel-gfx
