[Intel-gfx] [PATCH v4 1/6] drm/i915/vlv: Added a rendering specific Hw WA 'WaTlbInvalidateStoreDataBefore'

sourab.gupta at intel.com sourab.gupta at intel.com
Mon Mar 24 18:30:02 CET 2014


From: Akash Goel <akash.goel at intel.com>

Added a new rendering specific Workaround 'WaTlbInvalidateStoreDataBefore'.
In this WA, before pipecontrol with TLB invalidate set, need to add 2 MI
Store data commands.

v2: Modified the WA comment (Ville)

v3: Added the vlv identifier with WA name (Damien)

v4: Reworked based on Chris' comments (WA moved to gen7 ring flush func,
sending 6 dwords instead of 8)

Signed-off-by: Sourab Gupta <sourab.gupta at intel.com>
Signed-off-by: Akash Goel <akash.goel at intel.com>
---
 drivers/gpu/drm/i915/intel_ringbuffer.c | 23 +++++++++++++++++++++++
 1 file changed, 23 insertions(+)

diff --git a/drivers/gpu/drm/i915/intel_ringbuffer.c b/drivers/gpu/drm/i915/intel_ringbuffer.c
index 87d1a2d..75cac4e 100644
--- a/drivers/gpu/drm/i915/intel_ringbuffer.c
+++ b/drivers/gpu/drm/i915/intel_ringbuffer.c
@@ -309,6 +309,29 @@ gen7_render_ring_flush(struct intel_ring_buffer *ring,
 	u32 scratch_addr = ring->scratch.gtt_offset + 128;
 	int ret;
 
+	if (invalidate_domains && IS_VALLEYVIEW(ring->dev)) {
+		/*
+		 * WaTlbInvalidateStoreDataBefore:vlv
+		 * This workaround is applicable in case the flush call has
+		 * arrived in context of invalidate_all_caches function.
+		 * Before pipecontrol with TLB invalidate set, need 2 store
+		 * data commands (such as MI_STORE_DATA_IMM or MI_STORE_DATA_INDEX)
+		 * Without this, hardware cannot guarantee the command after the
+		 * PIPE_CONTROL with TLB inv will not use the old TLB values.
+		 * FIXME: should apply to snb, ivb
+		 */
+		int i;
+		ret = intel_ring_begin(ring, 3 * 2);
+		if (ret)
+			return ret;
+		for (i = 0; i < 2; i++) {
+			intel_ring_emit(ring, MI_STORE_DWORD_INDEX);
+			intel_ring_emit(ring, I915_GEM_HWS_SCRATCH_ADDR);
+			intel_ring_emit(ring, 0);
+		}
+		intel_ring_advance(ring);
+	}
+
 	/*
 	 * Ensure that any following seqno writes only happen when the render
 	 * cache is indeed flushed.
-- 
1.8.5.1




More information about the Intel-gfx mailing list