perf top: Fall back to cpu-clock-tick hrtimer sampling if no cycle counter available
authorIngo Molnar <mingo@elte.hu>
Sun, 7 Jun 2009 15:31:52 +0000 (17:31 +0200)
committerIngo Molnar <mingo@elte.hu>
Sun, 7 Jun 2009 15:31:52 +0000 (17:31 +0200)
On architectures/CPUs without PMU support but with perfcounters
enabled 'perf top' currently fails because it cannot create a
cycle based hw-perfcounter.

Fall back to the cpu-clock-tick sw-perfcounter in this case, which
is hrtimer based and will always work (as long as perfcounters
is enabled).

Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
LKML-Reference: <new-submission>
Signed-off-by: Ingo Molnar <mingo@elte.hu>
tools/perf/builtin-top.c
tools/perf/util/usage.c

index fdc1d5863b012ae3da2aa41b339693f55a77f3b8..6da30a140e863fe2796a06307c8f2e464c3c9fe2 100644 (file)
@@ -527,58 +527,81 @@ static void mmap_read(void)
        }
 }
 
-static int __cmd_top(void)
+int nr_poll;
+int group_fd;
+
+static void start_counter(int i, int counter)
 {
        struct perf_counter_attr *attr;
-       pthread_t thread;
-       int i, counter, group_fd, nr_poll = 0;
        unsigned int cpu;
+
+       cpu = profile_cpu;
+       if (target_pid == -1 && profile_cpu == -1)
+               cpu = i;
+
+       attr = attrs + counter;
+
+       attr->sample_type       = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
+       attr->freq              = freq;
+
+try_again:
+       fd[i][counter] = sys_perf_counter_open(attr, target_pid, cpu, group_fd, 0);
+
+       if (fd[i][counter] < 0) {
+               int err = errno;
+
+               error("sys_perf_counter_open() syscall returned with %d (%s)\n",
+                       fd[i][counter], strerror(err));
+
+               if (err == EPERM)
+                       die(" No permission - are you root?\n");
+               /*
+                * If it's cycles then fall back to hrtimer
+                * based cpu-clock-tick sw counter, which
+                * is always available even if no PMU support:
+                */
+               if (attr->type == PERF_TYPE_HARDWARE
+                       && attr->config == PERF_COUNT_CPU_CYCLES) {
+
+                       warning(" ... trying to fall back to cpu-clock-ticks\n");
+                       attr->type = PERF_TYPE_SOFTWARE;
+                       attr->config = PERF_COUNT_CPU_CLOCK;
+                       goto try_again;
+               }
+               exit(-1);
+       }
+       assert(fd[i][counter] >= 0);
+       fcntl(fd[i][counter], F_SETFL, O_NONBLOCK);
+
+       /*
+        * First counter acts as the group leader:
+        */
+       if (group && group_fd == -1)
+               group_fd = fd[i][counter];
+
+       event_array[nr_poll].fd = fd[i][counter];
+       event_array[nr_poll].events = POLLIN;
+       nr_poll++;
+
+       mmap_array[i][counter].counter = counter;
+       mmap_array[i][counter].prev = 0;
+       mmap_array[i][counter].mask = mmap_pages*page_size - 1;
+       mmap_array[i][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
+                       PROT_READ, MAP_SHARED, fd[i][counter], 0);
+       if (mmap_array[i][counter].base == MAP_FAILED)
+               die("failed to mmap with %d (%s)\n", errno, strerror(errno));
+}
+
+static int __cmd_top(void)
+{
+       pthread_t thread;
+       int i, counter;
        int ret;
 
        for (i = 0; i < nr_cpus; i++) {
                group_fd = -1;
-               for (counter = 0; counter < nr_counters; counter++) {
-
-                       cpu     = profile_cpu;
-                       if (target_pid == -1 && profile_cpu == -1)
-                               cpu = i;
-
-                       attr = attrs + counter;
-
-                       attr->sample_type       = PERF_SAMPLE_IP | PERF_SAMPLE_TID;
-                       attr->freq              = freq;
-
-                       fd[i][counter] = sys_perf_counter_open(attr, target_pid, cpu, group_fd, 0);
-                       if (fd[i][counter] < 0) {
-                               int err = errno;
-
-                               error("syscall returned with %d (%s)\n",
-                                       fd[i][counter], strerror(err));
-                               if (err == EPERM)
-                                       printf("Are you root?\n");
-                               exit(-1);
-                       }
-                       assert(fd[i][counter] >= 0);
-                       fcntl(fd[i][counter], F_SETFL, O_NONBLOCK);
-
-                       /*
-                        * First counter acts as the group leader:
-                        */
-                       if (group && group_fd == -1)
-                               group_fd = fd[i][counter];
-
-                       event_array[nr_poll].fd = fd[i][counter];
-                       event_array[nr_poll].events = POLLIN;
-                       nr_poll++;
-
-                       mmap_array[i][counter].counter = counter;
-                       mmap_array[i][counter].prev = 0;
-                       mmap_array[i][counter].mask = mmap_pages*page_size - 1;
-                       mmap_array[i][counter].base = mmap(NULL, (mmap_pages+1)*page_size,
-                                       PROT_READ, MAP_SHARED, fd[i][counter], 0);
-                       if (mmap_array[i][counter].base == MAP_FAILED)
-                               die("failed to mmap with %d (%s)\n", errno, strerror(errno));
-               }
+               for (counter = 0; counter < nr_counters; counter++)
+                       start_counter(i, counter);
        }
 
        /* Wait for a minimal set of events before starting the snapshot */
index 2cad286e437169442f53270c3b92ba8c1447636d..e16bf9a707e89ff249e2785bf4e4996cf076e143 100644 (file)
@@ -9,29 +9,29 @@ static void report(const char *prefix, const char *err, va_list params)
 {
        char msg[1024];
        vsnprintf(msg, sizeof(msg), err, params);
-       fprintf(stderr, "%s%s\n", prefix, msg);
+       fprintf(stderr, " %s%s\n", prefix, msg);
 }
 
 static NORETURN void usage_builtin(const char *err)
 {
-       fprintf(stderr, "\n usage: %s\n", err);
+       fprintf(stderr, "\n Usage: %s\n", err);
        exit(129);
 }
 
 static NORETURN void die_builtin(const char *err, va_list params)
 {
-       report("fatal: ", err, params);
+       report(" Fatal: ", err, params);
        exit(128);
 }
 
 static void error_builtin(const char *err, va_list params)
 {
-       report("error: ", err, params);
+       report(" Error: ", err, params);
 }
 
 static void warn_builtin(const char *warn, va_list params)
 {
-       report("warning: ", warn, params);
+       report(" Warning: ", warn, params);
 }
 
 /* If we are in a dlopen()ed .so write to a global variable would segfault