OSDN Git Service

perf: Copy parent's address filter offsets on clone
authorAlexander Shishkin <alexander.shishkin@linux.intel.com>
Fri, 15 Feb 2019 11:56:54 +0000 (13:56 +0200)
committerArnaldo Carvalho de Melo <acme@redhat.com>
Fri, 22 Feb 2019 19:52:07 +0000 (16:52 -0300)
When a child event is allocated in the inherit_event() path, the VMA
based filter offsets are not copied from the parent, even though the
address space mapping of the new task remains the same, which leads to
no trace for the new task until exec.

Reported-by: Mansour Alharthi <malharthi9@gatech.edu>
Signed-off-by: Alexander Shishkin <alexander.shishkin@linux.intel.com>
Tested-by: Mathieu Poirier <mathieu.poirier@linaro.org>
Acked-by: Peter Zijlstra <peterz@infradead.org>
Cc: Jiri Olsa <jolsa@redhat.com>
Fixes: 375637bc5249 ("perf/core: Introduce address range filtering")
Link: http://lkml.kernel.org/r/20190215115655.63469-2-alexander.shishkin@linux.intel.com
Signed-off-by: Arnaldo Carvalho de Melo <acme@redhat.com>
kernel/events/core.c

index 5aeb4c7..2d89efc 100644 (file)
@@ -1255,6 +1255,7 @@ static void put_ctx(struct perf_event_context *ctx)
  *           perf_event_context::lock
  *         perf_event::mmap_mutex
  *         mmap_sem
+ *           perf_addr_filters_head::lock
  *
  *    cpu_hotplug_lock
  *      pmus_lock
@@ -10312,6 +10313,20 @@ perf_event_alloc(struct perf_event_attr *attr, int cpu,
                        goto err_per_task;
                }
 
+               /*
+                * Clone the parent's vma offsets: they are valid until exec()
+                * even if the mm is not shared with the parent.
+                */
+               if (event->parent) {
+                       struct perf_addr_filters_head *ifh = perf_event_addr_filters(event);
+
+                       raw_spin_lock_irq(&ifh->lock);
+                       memcpy(event->addr_filters_offs,
+                              event->parent->addr_filters_offs,
+                              pmu->nr_addr_filters * sizeof(unsigned long));
+                       raw_spin_unlock_irq(&ifh->lock);
+               }
+
                /* force hw sync on the address filters */
                event->addr_filters_gen = 1;
        }