[PATCH 2/2] perf tools: Fix Intel PT getting stuck in a loop

[Adrian Hunter]

Check for being stuck in a loop.  That can happen if a
decoder error results in the decoder erroneously setting
the ip to an address that is itself in an infinite loop
that consumes no packets.  The only way to be in a loop
that consumes no packets is if it consists of unconditional
branches.  So the check for being stuck is if we see
a repeating cycle of consecutive unconditional branches.

Signed-off-by: Adrian Hunter <adrian.hunter@xxxxxxxxx>
---
 .../perf/util/intel-pt-decoder/intel-pt-decoder.c  | 42 ++++++++++++++++++++++
 .../perf/util/intel-pt-decoder/intel-pt-decoder.h  |  1 +
 2 files changed, 43 insertions(+)

diff --git a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c
index 748a7a078313..e8ff6573ecec 100644
--- a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c
+++ b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.c
@@ -37,6 +37,9 @@
 
 #define INTEL_PT_RETURN 1
 
+/* Maximum number of loops with no packets consumed i.e. stuck in a loop */
+#define INTEL_PT_MAX_LOOPS 10000
+
 struct intel_pt_blk {
 	struct intel_pt_blk *prev;
 	uint64_t ip[INTEL_PT_BLK_SIZE];
@@ -114,6 +117,10 @@ struct intel_pt_decoder {
 	unsigned int fup_tx_flags;
 	unsigned int tx_flags;
 	uint64_t timestamp_insn_cnt;
+	uint64_t stuck_ip;
+	int no_progress;
+	int stuck_ip_prd;
+	int stuck_ip_cnt;
 	const unsigned char *next_buf;
 	size_t next_len;
 	unsigned char temp_buf[INTEL_PT_PKT_MAX_SZ];
@@ -263,6 +270,8 @@ static int intel_pt_ext_err(int code)
 		return INTEL_PT_ERR_OVR;
 	case -ENOSPC:
 		return INTEL_PT_ERR_LOST;
+	case -ELOOP:
+		return INTEL_PT_ERR_NELOOP;
 	default:
 		return INTEL_PT_ERR_UNK;
 	}
@@ -278,6 +287,7 @@ static const char *intel_pt_err_msgs[] = {
 	[INTEL_PT_ERR_OVR]    = "Overflow packet",
 	[INTEL_PT_ERR_LOST]   = "Lost trace data",
 	[INTEL_PT_ERR_UNK]    = "Unknown error!",
+	[INTEL_PT_ERR_NELOOP] = "Never-ending loop",
 };
 
 int intel_pt__strerror(int code, char *buf, size_t buflen)
@@ -550,6 +560,7 @@ static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
 	decoder->period_insn_cnt += insn_cnt;
 
 	if (err) {
+		decoder->no_progress = 0;
 		decoder->pkt_state = INTEL_PT_STATE_ERR2;
 		intel_pt_log_at("ERROR: Failed to get instruction",
 				decoder->ip);
@@ -589,13 +600,44 @@ static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
 	}
 
 	if (intel_pt_insn->branch == INTEL_PT_BR_UNCONDITIONAL) {
+		int cnt = decoder->no_progress++;
+
 		decoder->state.from_ip = decoder->ip;
 		decoder->ip += intel_pt_insn->length +
 				intel_pt_insn->rel;
 		decoder->state.to_ip = decoder->ip;
 		err = INTEL_PT_RETURN;
+
+		/*
+		 * Check for being stuck in a loop.  This can happen if a
+		 * decoder error results in the decoder erroneously setting the
+		 * ip to an address that is itself in an infinite loop that
+		 * consumes no packets.  When that happens, there must be an
+		 * unconditional branch.
+		 */
+		if (cnt) {
+			if (cnt == 1) {
+				decoder->stuck_ip = decoder->state.to_ip;
+				decoder->stuck_ip_prd = 1;
+				decoder->stuck_ip_cnt = 1;
+			} else if (cnt > INTEL_PT_MAX_LOOPS ||
+				   decoder->state.to_ip == decoder->stuck_ip) {
+				intel_pt_log_at("ERROR: Never-ending loop",
+						decoder->state.to_ip);
+				decoder->pkt_state = INTEL_PT_STATE_ERR_RESYNC;
+				err = -ELOOP;
+				goto out;
+			} else if (!--decoder->stuck_ip_cnt) {
+				decoder->stuck_ip_prd += 1;
+				decoder->stuck_ip_cnt = decoder->stuck_ip_prd;
+				decoder->stuck_ip = decoder->state.to_ip;
+			}
+		}
+		goto out_no_progress;
 	}
 out:
+	decoder->no_progress = 0;
+out_no_progress:
 	decoder->state.insn_op = intel_pt_insn->op;
 	decoder->state.insn_len = intel_pt_insn->length;
 
diff --git a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h
index 955263adfd8d..706c6bccc57e 100644
--- a/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h
+++ b/tools/perf/util/intel-pt-decoder/intel-pt-decoder.h
@@ -48,6 +48,7 @@ enum {
 	INTEL_PT_ERR_OVR,
 	INTEL_PT_ERR_LOST,
 	INTEL_PT_ERR_UNK,
+	INTEL_PT_ERR_NELOOP,
 	INTEL_PT_ERR_MAX,
 };
 
-- 
1.9.1

--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/