perf kvm: Option to print events that exceed a duration
authorDavid Ahern <dsahern@gmail.com>
Thu, 8 Aug 2013 01:56:38 +0000 (21:56 -0400)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Mon, 12 Aug 2013 13:31:04 +0000 (10:31 -0300)
This is useful to spot high latency blips. It is normal for HLT reasons
to have long exit times, so strip those from the duration check.

v2: changed threshold to duration per acme's request

Signed-off-by: David Ahern <dsahern@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Ingo Molnar <mingo@kernel.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Namhyung Kim <namhyung@kernel.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Runzhen Wang <runzhen@linux.vnet.ibm.com>
Cc: Xiao Guangrong <xiaoguangrong@linux.vnet.ibm.com>
Link: http://lkml.kernel.org/r/1375926999-75129-1-git-send-email-dsahern@gmail.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
tools/perf/builtin-kvm.c
tools/perf/perf.h

index 2ceec817a2fba181cb0c9e06876d52c28c648d5f..fa2f3d79886aa9d623ac8e6bd9568b9db89a8713 100644 (file)
@@ -106,6 +106,7 @@ struct perf_kvm_stat {
        u64 total_time;
        u64 total_count;
        u64 lost_events;
+       u64 duration;
 
        const char *pid_str;
        struct intlist *pid_list;
@@ -473,7 +474,7 @@ static bool update_kvm_event(struct kvm_event *event, int vcpu_id,
 static bool handle_end_event(struct perf_kvm_stat *kvm,
                             struct vcpu_event_record *vcpu_record,
                             struct event_key *key,
-                            u64 timestamp)
+                            struct perf_sample *sample)
 {
        struct kvm_event *event;
        u64 time_begin, time_diff;
@@ -510,12 +511,24 @@ static bool handle_end_event(struct perf_kvm_stat *kvm,
        vcpu_record->start_time = 0;
 
        /* seems to happen once in a while during live mode */
-       if (timestamp < time_begin) {
+       if (sample->time < time_begin) {
                pr_debug("End time before begin time; skipping event.\n");
                return true;
        }
 
-       time_diff = timestamp - time_begin;
+       time_diff = sample->time - time_begin;
+
+       if (kvm->duration && time_diff > kvm->duration) {
+               char decode[32];
+
+               kvm->events_ops->decode_key(kvm, &event->key, decode);
+               if (strcmp(decode, "HLT")) {
+                       pr_info("%" PRIu64 " VM %d, vcpu %d: %s event took %" PRIu64 "usec\n",
+                                sample->time, sample->pid, vcpu_record->vcpu_id,
+                                decode, time_diff/1000);
+               }
+       }
+
        return update_kvm_event(event, vcpu, time_diff);
 }
 
@@ -562,7 +575,7 @@ static bool handle_kvm_event(struct perf_kvm_stat *kvm,
                return handle_begin_event(kvm, vcpu_record, &key, sample->time);
 
        if (kvm->events_ops->is_end_event(evsel, sample, &key))
-               return handle_end_event(kvm, vcpu_record, &key, sample->time);
+               return handle_end_event(kvm, vcpu_record, &key, sample);
 
        return true;
 }
@@ -1429,6 +1442,8 @@ static int kvm_events_live(struct perf_kvm_stat *kvm,
                OPT_STRING('k', "key", &kvm->sort_key, "sort-key",
                        "key for sorting: sample(sort by samples number)"
                        " time (sort by avg time)"),
+               OPT_U64(0, "duration", &kvm->duration,
+                   "show events other than HALT that take longer than duration usecs"),
                OPT_END()
        };
        const char * const live_usage[] = {
@@ -1467,6 +1482,8 @@ static int kvm_events_live(struct perf_kvm_stat *kvm,
                        usage_with_options(live_usage, live_options);
        }
 
+       kvm->duration *= NSEC_PER_USEC;   /* convert usec to nsec */
+
        /*
         * target related setups
         */
index 32bd102c32b6205b55ba4b78628963d39852575f..cf20187eee0a7345373dfc053c580bc1e77c4bb8 100644 (file)
 #ifndef NSEC_PER_SEC
 # define NSEC_PER_SEC                  1000000000ULL
 #endif
+#ifndef NSEC_PER_USEC
+# define NSEC_PER_USEC                 1000ULL
+#endif
 
 static inline unsigned long long rdclock(void)
 {