perf: Robustify task_function_call()
authorPeter Zijlstra <peterz@infradead.org>
Wed, 24 Feb 2016 17:45:51 +0000 (18:45 +0100)
committerIngo Molnar <mingo@kernel.org>
Thu, 25 Feb 2016 07:44:29 +0000 (08:44 +0100)
Since there is no serialization between task_function_call() doing
task_curr() and the other CPU doing context switches, we could end
up not sending an IPI even if we had to.

And I'm not sure I still buy my own argument we're OK.

Signed-off-by: Peter Zijlstra (Intel) <peterz@infradead.org>
Cc: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: Jiri Olsa <jolsa@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: dvyukov@google.com
Cc: eranian@google.com
Cc: oleg@redhat.com
Cc: panand@redhat.com
Cc: sasha.levin@oracle.com
Cc: vince@deater.net
Link: http://lkml.kernel.org/r/20160224174948.340031200@infradead.org
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/events/core.c

index 25edabd207de748f147bdcd41b41a25428725d5e..614614821f00a02928439b068903904d81233808 100644 (file)
@@ -64,8 +64,17 @@ static void remote_function(void *data)
        struct task_struct *p = tfc->p;
 
        if (p) {
-               tfc->ret = -EAGAIN;
-               if (task_cpu(p) != smp_processor_id() || !task_curr(p))
+               /* -EAGAIN */
+               if (task_cpu(p) != smp_processor_id())
+                       return;
+
+               /*
+                * Now that we're on right CPU with IRQs disabled, we can test
+                * if we hit the right task without races.
+                */
+
+               tfc->ret = -ESRCH; /* No such (running) process */
+               if (p != current)
                        return;
        }
 
@@ -92,13 +101,17 @@ task_function_call(struct task_struct *p, remote_function_f func, void *info)
                .p      = p,
                .func   = func,
                .info   = info,
-               .ret    = -ESRCH, /* No such (running) process */
+               .ret    = -EAGAIN,
        };
+       int ret;
 
-       if (task_curr(p))
-               smp_call_function_single(task_cpu(p), remote_function, &data, 1);
+       do {
+               ret = smp_call_function_single(task_cpu(p), remote_function, &data, 1);
+               if (!ret)
+                       ret = data.ret;
+       } while (ret == -EAGAIN);
 
-       return data.ret;
+       return ret;
 }
 
 /**
@@ -169,19 +182,6 @@ static bool is_kernel_event(struct perf_event *event)
  *    rely on ctx->is_active and therefore cannot use event_function_call().
  *    See perf_install_in_context().
  *
- * This is because we need a ctx->lock serialized variable (ctx->is_active)
- * to reliably determine if a particular task/context is scheduled in. The
- * task_curr() use in task_function_call() is racy in that a remote context
- * switch is not a single atomic operation.
- *
- * As is, the situation is 'safe' because we set rq->curr before we do the
- * actual context switch. This means that task_curr() will fail early, but
- * we'll continue spinning on ctx->is_active until we've passed
- * perf_event_task_sched_out().
- *
- * Without this ctx->lock serialized variable we could have race where we find
- * the task (and hence the context) would not be active while in fact they are.
- *
  * If ctx->nr_events, then ctx->is_active and cpuctx->task_ctx are set.
  */
 
@@ -212,7 +212,7 @@ static int event_function(void *info)
         */
        if (ctx->task) {
                if (ctx->task != current) {
-                       ret = -EAGAIN;
+                       ret = -ESRCH;
                        goto unlock;
                }