tracing/mm: don't trace kmem_cache_free on offline cpus

Since tracepoints use RCU for protection, they must not be called on
offline cpus.  trace_kmem_cache_free can be called on an offline cpu in
this scenario caught by LOCKDEP:

    ===============================
    [ INFO: suspicious RCU usage. ]
    4.1.0-rc1+ #9 Not tainted
    -------------------------------
    include/trace/events/kmem.h:148 suspicious rcu_dereference_check() usage!

    other info that might help us debug this:

    RCU used illegally from offline CPU!
    rcu_scheduler_active = 1, debug_locks = 1
    no locks held by swapper/1/0.

    stack backtrace:
    CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.1.0-rc1+ #9
    Call Trace:
      .dump_stack+0x98/0xd4 (unreliable)
      .lockdep_rcu_suspicious+0x108/0x170
      .kmem_cache_free+0x344/0x4b0
      .__mmdrop+0x4c/0x160
      .idle_task_exit+0xf0/0x100
      .pnv_smp_cpu_kill_self+0x58/0x2c0
      .cpu_die+0x34/0x50
      .arch_cpu_idle_dead+0x20/0x40
      .cpu_startup_entry+0x708/0x7a0
      .start_secondary+0x36c/0x3a0
      start_secondary_prolog+0x10/0x14

Fix this by converting kmem_cache_free trace point into
TRACE_EVENT_CONDITION where condition is cpu_online(smp_processor_id())

Signed-off-by: Shreyas B. Prabhu <[email protected]>
Reported-by: Aneesh Kumar K.V <[email protected]>
Reviewed-by: Preeti U Murthy <[email protected]>
Acked-by: Steven Rostedt <[email protected]>
Signed-off-by: Andrew Morton <[email protected]>
Signed-off-by: Linus Torvalds <[email protected]>

include/trace/events/kmem.h

@@ -140,11 +140,22 @@ DEFINE_EVENT(kmem_free, kfree,
 	TP_ARGS(call_site, ptr)
 );
 
-DEFINE_EVENT(kmem_free, kmem_cache_free,
+DEFINE_EVENT_CONDITION(kmem_free, kmem_cache_free,
 
 	TP_PROTO(unsigned long call_site, const void *ptr),
 
-	TP_ARGS(call_site, ptr)
+	TP_ARGS(call_site, ptr),
+
+	/*
+	 * This trace can be potentially called from an offlined cpu.
+	 * Since trace points use RCU and RCU should not be used from
+	 * offline cpus, filter such calls out.
+	 * While this trace can be called from a preemptable section,
+	 * it has no impact on the condition since tasks can migrate
+	 * only from online cpus to other online cpus. Thus its safe
+	 * to use raw_smp_processor_id.
+	 */
+	TP_CONDITION(cpu_online(raw_smp_processor_id()))
 );
 
 TRACE_EVENT(mm_page_free,