Skip to contentSkip to conversation

perf intel-pt: Add a config for max loops without consuming a packet

PATCH Applied Adrian Hunter 3 participants 3 messages Jul 1, 2021 – Jul 7, 2021 perf events

lore
Share & tools
lore

Adrian Hunter

author reviewer
#0 lore raw

PATCH 1/1 perf intel-pt: Add a config for max loops without consuming a packet

The Intel PT decoder limits the number of unconditional branches (e.g. jmps) decoded without consuming any trace packets. Generally, a loop needs a conditional branch which generates a TNT packet, whereas a "ret" instruction will generate a TIP or TNT packet. So exceeding the limit is assumed to be a never-ending loop, which can happen if there has been a decoding error putting the decoder at the wrong place in the code.

Up until now, the limit of 10000 has been enough but some analytic purposes have been reported to exceed that.

Increase the limit to 100000, and make it configurable via perf config intel-pt.max-loops. Also amend the "Never-ending loop" message to mention the configuration entry.

---
 tools/perf/Documentation/perf-config.txt            |  6 ++++++
 tools/perf/util/intel-pt-decoder/intel-pt-decoder.c | 13 +++++++++----
 tools/perf/util/intel-pt-decoder/intel-pt-decoder.h |  1 +
 tools/perf/util/intel-pt.c                          |  5 +++++
 4 files changed, 21 insertions(+), 4 deletions(-)
tools/perf/Documentation/perf-config.txt +6 -0
index b0872c801866..3bb75c1f25e8 100644
@@ -706,6 +706,12 @@ intel-pt.*::
 		If set, Intel PT decoder will set the mispred flag on all
 		branches.
 
+	intel-pt.max-loops::
+		If set and non-zero, the maximum number of unconditional
+		branches decoded without consuming any trace packets. If
+		the maximum is exceeded there will be a "Never-ending loop"
+		error. The default is 100000.
+
 auxtrace.*::
 
 	auxtrace.dumpdir::
tools/perf/util/intel-pt-decoder/intel-pt-decoder.c +9 -4
index cb2520abf261..5ab631702769 100644
@@ -41,8 +41,11 @@
 
 #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
+/*
+ * Default maximum number of loops with no packets consumed i.e. stuck in a
+ * loop.
+ */
+#define INTEL_PT_MAX_LOOPS 100000
 
 struct intel_pt_blk {
 	struct intel_pt_blk *prev;
@@ -220,6 +223,7 @@ struct intel_pt_decoder {
 	uint64_t timestamp_insn_cnt;
 	uint64_t sample_insn_cnt;
 	uint64_t stuck_ip;
+	int max_loops;
 	int no_progress;
 	int stuck_ip_prd;
 	int stuck_ip_cnt;
@@ -315,6 +319,7 @@ struct intel_pt_decoder *intel_pt_decoder_new(struct intel_pt_params *params)
 	decoder->vm_tm_corr_dry_run = params->vm_tm_corr_dry_run;
 	decoder->first_timestamp    = params->first_timestamp;
 	decoder->last_reliable_timestamp = params->first_timestamp;
+	decoder->max_loops          = params->max_loops ? params->max_loops : INTEL_PT_MAX_LOOPS;
 
 	decoder->flags              = params->flags;
 
@@ -483,7 +488,7 @@ 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",
+	[INTEL_PT_ERR_NELOOP] = "Never-ending loop (refer perf config intel-pt.max-loops)",
 };
 
 int intel_pt__strerror(int code, char *buf, size_t buflen)
@@ -1168,7 +1173,7 @@ static int intel_pt_walk_insn(struct intel_pt_decoder *decoder,
 				decoder->stuck_ip = decoder->state.to_ip;
 				decoder->stuck_ip_prd = 1;
 				decoder->stuck_ip_cnt = 1;
-			} else if (cnt > INTEL_PT_MAX_LOOPS ||
+			} else if (cnt > decoder->max_loops ||
 				   decoder->state.to_ip == decoder->stuck_ip) {
 				intel_pt_log_at("ERROR: Never-ending loop",
 						decoder->state.to_ip);
tools/perf/util/intel-pt-decoder/intel-pt-decoder.h +1 -0
index 714c475808c0..4b5e79fcf557 100644
@@ -270,6 +270,7 @@ struct intel_pt_params {
 	uint32_t tsc_ctc_ratio_d;
 	enum intel_pt_param_flags flags;
 	unsigned int quick;
+	int max_loops;
 };
 
 struct intel_pt_decoder;
tools/perf/util/intel-pt.c +5 -0
index 9e0539d3a79c..1073c56a512c 100644
@@ -124,6 +124,7 @@ struct intel_pt {
 	u64 noretcomp_bit;
 	unsigned max_non_turbo_ratio;
 	unsigned cbr2khz;
+	int max_loops;
 
 	unsigned long num_events;
 
@@ -1210,6 +1211,7 @@ static struct intel_pt_queue *intel_pt_alloc_queue(struct intel_pt *pt,
 	params.vm_time_correlation = pt->synth_opts.vm_time_correlation;
 	params.vm_tm_corr_dry_run = pt->synth_opts.vm_tm_corr_dry_run;
 	params.first_timestamp = pt->first_timestamp;
+	params.max_loops = pt->max_loops;
 
 	if (pt->filts.cnt > 0)
 		params.pgd_ip = intel_pt_pgd_ip;
@@ -3508,6 +3510,9 @@ static int intel_pt_perf_config(const char *var, const char *value, void *data)
 	if (!strcmp(var, "intel-pt.mispred-all"))
 		pt->mispred_all = perf_config_bool(var, value);
 
+	if (!strcmp(var, "intel-pt.max-loops"))
+		perf_config_int(&pt->max_loops, var, value);
+
 	return 0;
 }
 
Signature · 1 line
2.17.1
Reviewed-by on v1 1/1
#1 lore raw
Quoted 8 lines · Adrian Hunter

The Intel PT decoder limits the number of unconditional branches (e.g. jmps) decoded without consuming any trace packets. Generally, a loop needs a conditional branch which generates a TNT packet, whereas a "ret" instruction will generate a TIP or TNT packet. So exceeding the limit is assumed to be a never-ending loop, which can happen if there has been a decoding error putting the decoder at the wrong place in the code.

Quoted 8 lines

Up until now, the limit of 10000 has been enough but some analytic purposes have been reported to exceed that.

Increase the limit to 100000, and make it configurable via perf config intel-pt.max-loops. Also amend the "Never-ending loop" message to mention the configuration entry.

Thanks. That is useful.

-Andi

maintainer Applied
#2 lore raw

Em Thu, Jul 01, 2021 at 10:54:53AM -0700, Andi Kleen escreveu:

Quoted 17 lines · Adrian Hunter
Quoted 15 lines

The Intel PT decoder limits the number of unconditional branches (e.g. jmps) decoded without consuming any trace packets. Generally, a loop needs a conditional branch which generates a TNT packet, whereas a "ret" instruction will generate a TIP or TNT packet. So exceeding the limit is assumed to be a never-ending loop, which can happen if there has been a decoding error putting the decoder at the wrong place in the code.

Up until now, the limit of 10000 has been enough but some analytic purposes have been reported to exceed that.

Increase the limit to 100000, and make it configurable via perf config intel-pt.max-loops. Also amend the "Never-ending loop" message to mention the configuration entry.

Quoted 7 lines
Quoted 1 line

Thanks. That is useful.

Thanks, applied.

- Arnaldo