tracing/mm: don't trace mm_page_pcpu_drain on offline cpus
authorShreyas B. Prabhu <shreyas@linux.vnet.ibm.com>
Thu, 28 May 2015 22:44:22 +0000 (15:44 -0700)
committerLinus Torvalds <torvalds@linux-foundation.org>
Fri, 29 May 2015 01:25:18 +0000 (18:25 -0700)
Since tracepoints use RCU for protection, they must not be called on
offline cpus.  trace_mm_page_pcpu_drain 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:265 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
     1 lock held by swapper/5/0:
      #0:  (&(&zone->lock)->rlock){..-...}, at: [<c0000000002073b0>] .free_pcppages_bulk+0x70/0x920

    stack backtrace:
     CPU: 5 PID: 0 Comm: swapper/5 Not tainted 4.1.0-rc1+ #9
     Call Trace:
       .dump_stack+0x98/0xd4 (unreliable)
       .lockdep_rcu_suspicious+0x108/0x170
       .free_pcppages_bulk+0x60c/0x920
       .free_hot_cold_page+0x208/0x280
       .destroy_context+0x90/0xd0
       .__mmdrop+0x58/0x160
       .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 mm_page_pcpu_drain trace point into
TRACE_EVENT_CONDITION where condition is cpu_online(smp_processor_id())

Signed-off-by: Shreyas B. Prabhu <shreyas@linux.vnet.ibm.com>
Reviewed-by: Preeti U Murthy <preeti@linux.vnet.ibm.com>
Acked-by: Steven Rostedt <rostedt@goodmis.org>
Signed-off-by: Andrew Morton <akpm@linux-foundation.org>
Signed-off-by: Linus Torvalds <torvalds@linux-foundation.org>
include/trace/events/kmem.h

index aa86354..f7554fd 100644 (file)
@@ -276,12 +276,35 @@ DEFINE_EVENT(mm_page, mm_page_alloc_zone_locked,
        TP_ARGS(page, order, migratetype)
 );
 
-DEFINE_EVENT_PRINT(mm_page, mm_page_pcpu_drain,
+TRACE_EVENT_CONDITION(mm_page_pcpu_drain,
 
        TP_PROTO(struct page *page, unsigned int order, int migratetype),
 
        TP_ARGS(page, order, migratetype),
 
+       /*
+        * 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())),
+
+       TP_STRUCT__entry(
+               __field(        unsigned long,  pfn             )
+               __field(        unsigned int,   order           )
+               __field(        int,            migratetype     )
+       ),
+
+       TP_fast_assign(
+               __entry->pfn            = page ? page_to_pfn(page) : -1UL;
+               __entry->order          = order;
+               __entry->migratetype    = migratetype;
+       ),
+
        TP_printk("page=%p pfn=%lu order=%d migratetype=%d",
                pfn_to_page(__entry->pfn), __entry->pfn,
                __entry->order, __entry->migratetype)