tracing/ring-buffer: fix non cpu hotplug case

Impact: fix warning with irqsoff tracer

The ring buffer allocates its buffers on pre-smp time (early_initcall).
It means that, at first, only the boot cpu buffer is allocated and
the ring-buffer cpumask only has the boot cpu set (cpu_online_mask).

Later, the secondary cpu will show up and the ring-buffer will be notified
about this event: the appropriate buffer will be allocated and the cpumask
will be updated.

Unfortunately, if !CONFIG_CPU_HOTPLUG, the ring-buffer will not be
notified about the secondary cpus, meaning that the cpumask will have
only the cpu boot set, and only one cpu buffer allocated.

We fix that by using cpu_possible_mask if !CONFIG_CPU_HOTPLUG.

This patch fixes the following warning with irqsoff tracer running:

[  169.317794] WARNING: at kernel/trace/trace.c:466 update_max_tr_single+0xcc/0xf3()
[  169.318002] Hardware name: AMILO Li 2727
[  169.318002] Modules linked in:
[  169.318002] Pid: 5624, comm: bash Not tainted 2.6.29-rc8-tip-02636-g6aafa6c #11
[  169.318002] Call Trace:
[  169.318002]  [<ffffffff81036182>] warn_slowpath+0xea/0x13d
[  169.318002]  [<ffffffff8100b9d6>] ? ftrace_call+0x5/0x2b
[  169.318002]  [<ffffffff8100b9d6>] ? ftrace_call+0x5/0x2b
[  169.318002]  [<ffffffff8100b9d1>] ? ftrace_call+0x0/0x2b
[  169.318002]  [<ffffffff8101ef10>] ? ftrace_modify_code+0xa9/0x108
[  169.318002]  [<ffffffff8106e27f>] ? trace_hardirqs_off+0x25/0x27
[  169.318002]  [<ffffffff8149afe7>] ? _spin_unlock_irqrestore+0x1f/0x2d
[  169.318002]  [<ffffffff81064f52>] ? ring_buffer_reset_cpu+0xf6/0xfb
[  169.318002]  [<ffffffff8106637c>] ? ring_buffer_reset+0x36/0x48
[  169.318002]  [<ffffffff8106aeda>] update_max_tr_single+0xcc/0xf3
[  169.318002]  [<ffffffff8100bc17>] ? sysret_check+0x22/0x5d
[  169.318002]  [<ffffffff8106e3ea>] stop_critical_timing+0x142/0x204
[  169.318002]  [<ffffffff8106e4cf>] trace_hardirqs_on_caller+0x23/0x25
[  169.318002]  [<ffffffff8149ac28>] trace_hardirqs_on_thunk+0x3a/0x3c
[  169.318002]  [<ffffffff8100bc17>] ? sysret_check+0x22/0x5d
[  169.318002] ---[ end trace db76cbf775a750cf ]---

Because this tracer may try to swap two cpu ring buffers for an
unregistered cpu on the ring buffer.

This patch might also fix a fair loss of traces due to unallocated buffers
for secondary cpus.

Reported-by: Ingo Molnar <mingo@elte.hu>
Signed-off-by: Frederic Weisbecker <fweisbec@gmail.com>
Acked-b: Steven Rostedt <rostedt@goodmis.org>
LKML-Reference: <1237470453-5427-1-git-send-email-fweisbec@gmail.com>
Signed-off-by: Ingo Molnar <mingo@elte.hu>

kernel/trace/ring_buffer.c

@@ -577,8 +577,17 @@ struct ring_buffer *ring_buffer_alloc(unsigned long size, unsigned flags)
 	if (buffer->pages == 1)
 		buffer->pages++;
 
+	/*
+	 * In case of non-hotplug cpu, if the ring-buffer is allocated
+	 * in early initcall, it will not be notified of secondary cpus.
+	 * In that off case, we need to allocate for all possible cpus.
+	 */
+#ifdef CONFIG_HOTPLUG_CPU
 	get_online_cpus();
 	cpumask_copy(buffer->cpumask, cpu_online_mask);
+#else
+	cpumask_copy(buffer->cpumask, cpu_possible_mask);
+#endif
 	buffer->cpus = nr_cpu_ids;
 
 	bsize = sizeof(void *) * nr_cpu_ids;