perf: Fix alloc_callchain_buffers()

Commit 927c7a9 ("perf: Fix race in callchains") introduced
a mismatch in the sizing of struct callchain_cpus_entries.

nr_cpu_ids must be used instead of num_possible_cpus(), or we
might get out of bound memory accesses on some machines.

Signed-off-by: Eric Dumazet <eric.dumazet@gmail.com>
Cc: Frederic Weisbecker <fweisbec@gmail.com>
Cc: Peter Zijlstra <a.p.zijlstra@chello.nl>
Cc: Arnaldo Carvalho de Melo <acme@redhat.com>
Cc: David Miller <davem@davemloft.net>
Cc: Stephane Eranian <eranian@google.com>
CC: stable@kernel.org
LKML-Reference: <1295980851.3588.351.camel@edumazet-laptop>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kernel/perf_event.c

@@ -1999,8 +1999,7 @@ static int alloc_callchain_buffers(void)
 	 * accessed from NMI. Use a temporary manual per cpu allocation
 	 * until that gets sorted out.
 	 */
-	size = sizeof(*entries) + sizeof(struct perf_callchain_entry *) *
-		num_possible_cpus();
+	size = offsetof(struct callchain_cpus_entries, cpu_entries[nr_cpu_ids]);
 
 	entries = kzalloc(size, GFP_KERNEL);
 	if (!entries)