oprofile, perf: Use per-cpu framework
authorRobert Richter <robert.richter@amd.com>
Thu, 23 Feb 2012 16:07:06 +0000 (17:07 +0100)
committerRobert Richter <robert.richter@amd.com>
Fri, 22 Jun 2012 14:31:20 +0000 (16:31 +0200)
This changes oprofile_perf.c to use the per-cpu framework.

Using the per-cpu framework should avoid error like the following:

 arch/arm/oprofile/../../../drivers/oprofile/oprofile_perf.c:28:28: error: variably modified 'perf_events' at file scope

Reported-by: William Cohen <wcohen@redhat.com>
Cc: Will Deacon <will.deacon@arm.com>
Signed-off-by: Robert Richter <robert.richter@amd.com>
drivers/oprofile/oprofile_perf.c

index efc4b7f..f3cfa0b 100644 (file)
@@ -1,5 +1,6 @@
 /*
  * Copyright 2010 ARM Ltd.
+ * Copyright 2012 Advanced Micro Devices, Inc., Robert Richter
  *
  * Perf-events backend for OProfile.
  */
@@ -25,7 +26,7 @@ static int oprofile_perf_enabled;
 static DEFINE_MUTEX(oprofile_perf_mutex);
 
 static struct op_counter_config *counter_config;
-static struct perf_event **perf_events[NR_CPUS];
+static DEFINE_PER_CPU(struct perf_event **, perf_events);
 static int num_counters;
 
 /*
@@ -38,7 +39,7 @@ static void op_overflow_handler(struct perf_event *event,
        u32 cpu = smp_processor_id();
 
        for (id = 0; id < num_counters; ++id)
-               if (perf_events[cpu][id] == event)
+               if (per_cpu(perf_events, cpu)[id] == event)
                        break;
 
        if (id != num_counters)
@@ -74,7 +75,7 @@ static int op_create_counter(int cpu, int event)
 {
        struct perf_event *pevent;
 
-       if (!counter_config[event].enabled || perf_events[cpu][event])
+       if (!counter_config[event].enabled || per_cpu(perf_events, cpu)[event])
                return 0;
 
        pevent = perf_event_create_kernel_counter(&counter_config[event].attr,
@@ -91,18 +92,18 @@ static int op_create_counter(int cpu, int event)
                return -EBUSY;
        }
 
-       perf_events[cpu][event] = pevent;
+       per_cpu(perf_events, cpu)[event] = pevent;
 
        return 0;
 }
 
 static void op_destroy_counter(int cpu, int event)
 {
-       struct perf_event *pevent = perf_events[cpu][event];
+       struct perf_event *pevent = per_cpu(perf_events, cpu)[event];
 
        if (pevent) {
                perf_event_release_kernel(pevent);
-               perf_events[cpu][event] = NULL;
+               per_cpu(perf_events, cpu)[event] = NULL;
        }
 }
 
@@ -257,12 +258,12 @@ void oprofile_perf_exit(void)
 
        for_each_possible_cpu(cpu) {
                for (id = 0; id < num_counters; ++id) {
-                       event = perf_events[cpu][id];
+                       event = per_cpu(perf_events, cpu)[id];
                        if (event)
                                perf_event_release_kernel(event);
                }
 
-               kfree(perf_events[cpu]);
+               kfree(per_cpu(perf_events, cpu));
        }
 
        kfree(counter_config);
@@ -277,8 +278,6 @@ int __init oprofile_perf_init(struct oprofile_operations *ops)
        if (ret)
                return ret;
 
-       memset(&perf_events, 0, sizeof(perf_events));
-
        num_counters = perf_num_counters();
        if (num_counters <= 0) {
                pr_info("oprofile: no performance counters\n");
@@ -298,9 +297,9 @@ int __init oprofile_perf_init(struct oprofile_operations *ops)
        }
 
        for_each_possible_cpu(cpu) {
-               perf_events[cpu] = kcalloc(num_counters,
+               per_cpu(perf_events, cpu) = kcalloc(num_counters,
                                sizeof(struct perf_event *), GFP_KERNEL);
-               if (!perf_events[cpu]) {
+               if (!per_cpu(perf_events, cpu)) {
                        pr_info("oprofile: failed to allocate %d perf events "
                                        "for cpu %d\n", num_counters, cpu);
                        ret = -ENOMEM;