diff --git a/libipt/src/pt_block_decoder.c b/libipt/src/pt_block_decoder.c
index 2d8cbf58..6709863a 100644
--- a/libipt/src/pt_block_decoder.c
+++ b/libipt/src/pt_block_decoder.c
@@ -45,6 +45,10 @@
 static int pt_blk_proceed_trailing_event(struct pt_block_decoder *,
 					 struct pt_block *);
 
+static int pt_blk_proceed_no_event_cached(struct pt_block_decoder *decoder,
+					  struct pt_block *block,
+					  struct pt_block_cache *bcache,
+					  const struct pt_mapped_section *msec);
 static int pt_blk_set_trig_anchor(struct pt_block_decoder *decoder)
 {
 	if (!decoder)
@@ -2431,12 +2435,14 @@ pt_blk_proceed_no_event_fill_cache(struct pt_block_decoder *decoder,
 	if (status < 0)
 		return status;
 
+	
+	int valid_cache = pt_bce_is_valid(bce);
 	/* If we don't have a valid cache entry, yet, fill the cache some more.
 	 *
 	 * On our way back, we add a cache entry for this instruction based on
 	 * the cache entry of the succeeding instruction.
 	 */
-	if (!pt_bce_is_valid(bce)) {
+	if (!valid_cache) {
 		/* If we exceeded the maximum number of allowed steps, we insert
 		 * a trampoline to the next instruction.
 		 *
@@ -2515,7 +2521,14 @@ pt_blk_proceed_no_event_fill_cache(struct pt_block_decoder *decoder,
 	 * Cache updates are atomic so even if the two versions were not
 	 * identical, we wouldn't care because they are both correct.
 	 */
-	return pt_bcache_add(bcache, ioff, bce);
+	status = pt_bcache_add(bcache, ioff, bce);
+	if (status < 0)
+		return status;
+
+	if (valid_cache)
+		return pt_blk_proceed_no_event_cached(decoder, block,
+						bcache, msec);
+	return status;
 }
 
 /* Proceed at a potentially truncated instruction.