Although the memory of pid_to_task can be allocated via calloc according
to the value of /proc/sys/kernel/pid_max, it cannot handle the case when
pid_max is changed after 'perf sched record' has created its perf.data.
If the new pid_max configured in 'perf sched replay' is smaller than the
old pid_max configured in 'perf sched record', then it will cause the
assertion failure problem.
To solve this problem, we realloc the memory of pid_to_task stepwise
once the passed-in pid parameter in register_pid is larger than the
current pid_max.
Example:
Test environment: x86_64 with 160 cores
 $ cat /proc/sys/kernel/pid_max
 163840
 $ perf sched record ls
 $ echo 5000 > /proc/sys/kernel/pid_max
 $ cat /proc/sys/kernel/pid_max
 5000
Before this patch:
 $ perf sched replay
 run measurement overhead: 221 nsecs
 sleep measurement overhead: 55356 nsecs
 the run test took 
1000011 nsecs
 the sleep test took 
1060940 nsecs
 perf: builtin-sched.c:337: register_pid: Assertion `!(pid >= (unsigned
 long)pid_max)' failed.
 Aborted
After this patch:
 $ perf sched replay
 run measurement overhead: 221 nsecs
 sleep measurement overhead: 55611 nsecs
 the run test took 
1000026 nsecs
 the sleep test took 
1060486 nsecs
 nr_run_events:        10
 nr_sleep_events:      1562
 nr_wakeup_events:     5
 task      0 (                  :1:         1), nr_events: 1
 task      1 (                  :2:         2), nr_events: 1
 task      2 (                  :3:         3), nr_events: 1
 task      3 (                  :5:         5), nr_events: 1
 ...
Signed-off-by: Yunlong Song <yunlong.song@huawei.com>
Cc: Paul Mackerras <paulus@samba.org>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Wang Nan <wangnan0@huawei.com>
Link: http://lkml.kernel.org/r/1427809596-29559-5-git-send-email-yunlong.song@huawei.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
                        pid_max = MAX_PID;
                BUG_ON((sched->pid_to_task = calloc(pid_max, sizeof(struct task_desc *))) == NULL);
        }
-       BUG_ON(pid >= (unsigned long)pid_max);
+       if (pid >= (unsigned long)pid_max) {
+               BUG_ON((sched->pid_to_task = realloc(sched->pid_to_task, (pid + 1) *
+                       sizeof(struct task_desc *))) == NULL);
+               while (pid >= (unsigned long)pid_max)
+                       sched->pid_to_task[pid_max++] = NULL;
+       }
 
        task = sched->pid_to_task[pid];