perf_counter: Add unique counter id

Stephan raised the issue that we currently cannot distinguish between
similar counters within a group (PERF_RECORD_GROUP uses the config
value as identifier).

Therefore, generate a new ID for each counter using a global u64
sequence counter.

Reported-by: Stephane Eranian <eranian@googlemail.com>
Signed-off-by: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Corey Ashford <cjashfor@linux.vnet.ibm.com>
Cc: Marcelo Tosatti <mtosatti@redhat.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: John Kacur <jkacur@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

2 files changed, 12 insertions, 5 deletions

diff --git a/include/linux/perf_counter.h b/include/linux/perf_counter.h
index 9ec20fc6bd36..4845a214b9e7 100644
--- a/include/linux/perf_counter.h
+++ b/include/linux/perf_counter.h
@@ -114,8 +114,9 @@ enum perf_counter_record_format {
  * in increasing order of bit value, after the counter value.
  */
 enum perf_counter_read_format {
-        PERF_FORMAT_TOTAL_TIME_ENABLED  =  1,
-        PERF_FORMAT_TOTAL_TIME_RUNNING  =  2,
+        PERF_FORMAT_TOTAL_TIME_ENABLED  =  1U << 0,
+        PERF_FORMAT_TOTAL_TIME_RUNNING  =  1U << 1,
+        PERF_FORMAT_ID                  =  1U << 2,
 };
 
 /*
@@ -290,7 +291,7 @@ enum perf_event_type {
          *      { u32                   cpu, res; } && PERF_RECORD_CPU
          *
          *      { u64                   nr;
-         *        { u64 event, val; }   cnt[nr];  } && PERF_RECORD_GROUP
+         *        { u64 id, val; }      cnt[nr];  } && PERF_RECORD_GROUP
          *
          *      { u16                   nr,
          *                              hv,
@@ -503,6 +504,7 @@ struct perf_counter {
         struct rcu_head                 rcu_head;
 
         struct pid_namespace            *ns;
+        u64                             id;
 #endif
 };
 
diff --git a/kernel/perf_counter.c b/kernel/perf_counter.c
index caa012cfe49a..978ecfcc7aaf 100644
--- a/kernel/perf_counter.c
+++ b/kernel/perf_counter.c
@@ -1510,6 +1510,8 @@ perf_read_hw(struct perf_counter *counter, char __user *buf, size_t count)
         if (counter->hw_event.read_format & PERF_FORMAT_TOTAL_TIME_RUNNING)
                 values[n++] = counter->total_time_running +
                         atomic64_read(&counter->child_total_time_running);
+        if (counter->hw_event.read_format & PERF_FORMAT_ID)
+                values[n++] = counter->id;
         mutex_unlock(&counter->child_mutex);
 
         if (count < n * sizeof(u64))
@@ -2303,7 +2305,7 @@ static void perf_counter_output(struct perf_counter *counter,
                 u32 pid, tid;
         } tid_entry;
         struct {
-                u64 event;
+                u64 id;
                 u64 counter;
         } group_entry;
         struct perf_callchain_entry *callchain = NULL;
@@ -2416,7 +2418,7 @@ static void perf_counter_output(struct perf_counter *counter,
                         if (sub != counter)
                                 sub->pmu->read(sub);
 
-                        group_entry.event = sub->hw_event.config;
+                        group_entry.id = sub->id;
                         group_entry.counter = atomic64_read(&sub->count);
 
                         perf_output_put(&handle, group_entry);
@@ -3375,6 +3377,8 @@ done:
         return counter;
 }
 
+static atomic64_t perf_counter_id;
+
 /**
  * sys_perf_counter_open - open a performance counter, associate it to a task/cpu
  *
@@ -3470,6 +3474,7 @@ SYSCALL_DEFINE5(perf_counter_open,
         mutex_unlock(&current->perf_counter_mutex);
 
         counter->ns = get_pid_ns(current->nsproxy->pid_ns);
+        counter->id = atomic64_inc_return(&perf_counter_id);
 
         fput_light(counter_file, fput_needed2);