freedreno: emit markers to scratch registers

Emit markers by writing to scratch registers in order to "triangulate"
gpu lockup position from post-mortem register dump.  By comparing
register values in post-mortem dump to command-stream, it is possible to
narrow down which DRAW_INDX caused the lockup.

Signed-off-by: Rob Clark <robclark@freedesktop.org>

4 files changed, 33 insertions, 0 deletions

diff --git a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
index 6b8ea02f69..d4e07af72d 100644
--- a/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
+++ b/src/gallium/drivers/freedreno/a3xx/fd3_emit.c
@@ -342,6 +342,8 @@ fd3_emit_state(struct fd_context *ctx, uint32_t dirty)
 {
 	struct fd_ringbuffer *ring = ctx->ring;
 
+	emit_marker(ring, 5);
+
 	if (dirty & FD_DIRTY_SAMPLE_MASK) {
 		OUT_PKT0(ring, REG_A3XX_RB_MSAA_CONTROL, 1);
 		OUT_RING(ring, A3XX_RB_MSAA_CONTROL_DISABLE |
diff --git a/src/gallium/drivers/freedreno/freedreno_draw.h b/src/gallium/drivers/freedreno/freedreno_draw.h
index 7fb0abe211..cf83a36032 100644
--- a/src/gallium/drivers/freedreno/freedreno_draw.h
+++ b/src/gallium/drivers/freedreno/freedreno_draw.h
@@ -51,6 +51,14 @@ fd_draw(struct fd_context *ctx, enum pc_di_primtype primtype,
 {
 	struct fd_ringbuffer *ring = ctx->ring;
 
+	/* for debug after a lock up, write a unique counter value
+	 * to scratch7 for each draw, to make it easier to match up
+	 * register dumps to cmdstream.  The combination of IB
+	 * (scratch6) and DRAW is enough to "triangulate" the
+	 * particular draw that caused lockup.
+	 */
+	emit_marker(ring, 7);
+
 	OUT_PKT3(ring, CP_DRAW_INDX, idx_bo ? 5 : 3);
 	OUT_RING(ring, 0x00000000);        /* viz query info. */
 	OUT_RING(ring, DRAW(primtype, src_sel,
@@ -60,6 +68,8 @@ fd_draw(struct fd_context *ctx, enum pc_di_primtype primtype,
 		OUT_RELOC(ring, idx_bo, idx_offset, 0, 0);
 		OUT_RING (ring, idx_size);
 	}
+
+	emit_marker(ring, 7);
 }
 
 #endif /* FREEDRENO_DRAW_H_ */
diff --git a/src/gallium/drivers/freedreno/freedreno_util.c b/src/gallium/drivers/freedreno/freedreno_util.c
index 0462e5fb51..7056edd2c1 100644
--- a/src/gallium/drivers/freedreno/freedreno_util.c
+++ b/src/gallium/drivers/freedreno/freedreno_util.c
@@ -31,6 +31,8 @@
 
 #include "freedreno_util.h"
 
+unsigned marker_cnt;
+
 enum adreno_rb_depth_format
 fd_pipe2depth(enum pipe_format format)
 {
diff --git a/src/gallium/drivers/freedreno/freedreno_util.h b/src/gallium/drivers/freedreno/freedreno_util.h
index 047c62f156..f8672339cf 100644
--- a/src/gallium/drivers/freedreno/freedreno_util.h
+++ b/src/gallium/drivers/freedreno/freedreno_util.h
@@ -94,6 +94,7 @@ pipe_surface_format(struct pipe_surface *psurf)
 
 #define LOG_DWORDS 0
 
+static inline void emit_marker(struct fd_ringbuffer *ring, int scratch_idx);
 
 static inline void
 OUT_RING(struct fd_ringbuffer *ring, uint32_t data)
@@ -175,9 +176,27 @@ static inline void
 OUT_IB(struct fd_ringbuffer *ring, struct fd_ringmarker *start,
 		struct fd_ringmarker *end)
 {
+	/* for debug after a lock up, write a unique counter value
+	 * to scratch6 for each IB, to make it easier to match up
+	 * register dumps to cmdstream.  The combination of IB and
+	 * DRAW (scratch7) is enough to "triangulate" the particular
+	 * draw that caused lockup.
+	 */
+	emit_marker(ring, 6);
+
 	OUT_PKT3(ring, CP_INDIRECT_BUFFER_PFD, 2);
 	fd_ringbuffer_emit_reloc_ring(ring, start, end);
 	OUT_RING(ring, fd_ringmarker_dwords(start, end));
+
+	emit_marker(ring, 6);
+}
+
+static inline void
+emit_marker(struct fd_ringbuffer *ring, int scratch_idx)
+{
+	extern unsigned marker_cnt;
+	OUT_PKT0(ring, REG_AXXX_CP_SCRATCH_REG0 + scratch_idx, 1);
+	OUT_RING(ring, ++marker_cnt);
 }
 
 #endif /* FREEDRENO_UTIL_H_ */