vc4: Align following shader recs to 16 bytes.

Otherwise, the low address bits will end up being interpreted as attribute
counts.

2 files changed, 10 insertions, 2 deletions

diff --git a/src/gallium/drivers/vc4/vc4_simulator_validate.c b/src/gallium/drivers/vc4/vc4_simulator_validate.c
index e16b3728eac..a660f94c58a 100644
--- a/src/gallium/drivers/vc4/vc4_simulator_validate.c
+++ b/src/gallium/drivers/vc4/vc4_simulator_validate.c
@@ -143,7 +143,8 @@ validate_gl_shader_state(VALIDATE_ARGS)
 	*(uint32_t *)validated = (exec->shader_rec_p +
 				  exec->shader_state[i].addr);
 
-	exec->shader_rec_p += gl_shader_rec_size(exec->shader_state[i].addr);
+	exec->shader_rec_p +=
+		roundup(gl_shader_rec_size(exec->shader_state[i].addr), 16);
 
 	return 0;
 }
@@ -461,7 +462,13 @@ validate_shader_rec(struct drm_device *dev,
 	pkt_v = exec->shader_rec_v;
 	memcpy(pkt_v, pkt_u, packet_size);
 	exec->shader_rec_u += packet_size;
-	exec->shader_rec_v += packet_size;
+	/* Shader recs have to be aligned to 16 bytes (due to the attribute
+	 * flags being in the low bytes), so round the next validated shader
+	 * rec address up.  This should be safe, since we've got so many
+	 * relocations in a shader rec packet.
+	 */
+	BUG_ON(roundup(packet_size, 16) - packet_size > nr_relocs * 4);
+	exec->shader_rec_v += roundup(packet_size, 16);
 	exec->shader_rec_size -= packet_size;
 
 	for (i = 0; i < nr_relocs; i++) {
diff --git a/src/gallium/drivers/vc4/vc4_simulator_validate.h b/src/gallium/drivers/vc4/vc4_simulator_validate.h
index 28dd2f65e70..52f664adbc1 100644
--- a/src/gallium/drivers/vc4/vc4_simulator_validate.h
+++ b/src/gallium/drivers/vc4/vc4_simulator_validate.h
@@ -41,6 +41,7 @@
 #define kfree(ptr) free(ptr)
 #define krealloc(ptr, size, args) realloc(ptr, size)
 #define roundup(x, y) align(x, y)
+#define BUG_ON(condition) assert(!(condition))
 
 static inline int
 copy_from_user(void *dst, void *src, size_t size)