sched/debug: Fix deadlock when enabling sched events
authorJosh Poimboeuf <jpoimboe@redhat.com>
Mon, 13 Jun 2016 07:32:09 +0000 (02:32 -0500)
committerIngo Molnar <mingo@kernel.org>
Tue, 14 Jun 2016 10:47:21 +0000 (12:47 +0200)
I see a hang when enabling sched events:

  echo 1 > /sys/kernel/debug/tracing/events/sched/enable

The printk buffer shows:

  BUG: spinlock recursion on CPU#1, swapper/1/0
   lock: 0xffff88007d5d8c00, .magic: dead4ead, .owner: swapper/1/0, .owner_cpu: 1
  CPU: 1 PID: 0 Comm: swapper/1 Not tainted 4.7.0-rc2+ #1
  Hardware name: QEMU Standard PC (i440FX + PIIX, 1996), BIOS 1.8.1-20150318_183358- 04/01/2014
  ...
  Call Trace:
   <IRQ>  [<ffffffff8143d663>] dump_stack+0x85/0xc2
   [<ffffffff81115948>] spin_dump+0x78/0xc0
   [<ffffffff81115aea>] do_raw_spin_lock+0x11a/0x150
   [<ffffffff81891471>] _raw_spin_lock+0x61/0x80
   [<ffffffff810e5466>] ? try_to_wake_up+0x256/0x4e0
   [<ffffffff810e5466>] try_to_wake_up+0x256/0x4e0
   [<ffffffff81891a0a>] ? _raw_spin_unlock_irqrestore+0x4a/0x80
   [<ffffffff810e5705>] wake_up_process+0x15/0x20
   [<ffffffff810cebb4>] insert_work+0x84/0xc0
   [<ffffffff810ced7f>] __queue_work+0x18f/0x660
   [<ffffffff810cf9a6>] queue_work_on+0x46/0x90
   [<ffffffffa00cd95b>] drm_fb_helper_dirty.isra.11+0xcb/0xe0 [drm_kms_helper]
   [<ffffffffa00cdac0>] drm_fb_helper_sys_imageblit+0x30/0x40 [drm_kms_helper]
   [<ffffffff814babcd>] soft_cursor+0x1ad/0x230
   [<ffffffff814ba379>] bit_cursor+0x649/0x680
   [<ffffffff814b9d30>] ? update_attr.isra.2+0x90/0x90
   [<ffffffff814b5e6a>] fbcon_cursor+0x14a/0x1c0
   [<ffffffff81555ef8>] hide_cursor+0x28/0x90
   [<ffffffff81558b6f>] vt_console_print+0x3bf/0x3f0
   [<ffffffff81122c63>] call_console_drivers.constprop.24+0x183/0x200
   [<ffffffff811241f4>] console_unlock+0x3d4/0x610
   [<ffffffff811247f5>] vprintk_emit+0x3c5/0x610
   [<ffffffff81124bc9>] vprintk_default+0x29/0x40
   [<ffffffff811e965b>] printk+0x57/0x73
   [<ffffffff810f7a9e>] enqueue_entity+0xc2e/0xc70
   [<ffffffff810f7b39>] enqueue_task_fair+0x59/0xab0
   [<ffffffff8106dcd9>] ? kvm_sched_clock_read+0x9/0x20
   [<ffffffff8103fb39>] ? sched_clock+0x9/0x10
   [<ffffffff810e3fcc>] activate_task+0x5c/0xa0
   [<ffffffff810e4514>] ttwu_do_activate+0x54/0xb0
   [<ffffffff810e5cea>] sched_ttwu_pending+0x7a/0xb0
   [<ffffffff810e5e51>] scheduler_ipi+0x61/0x170
   [<ffffffff81059e7f>] smp_trace_reschedule_interrupt+0x4f/0x2a0
   [<ffffffff81893ba6>] trace_reschedule_interrupt+0x96/0xa0
   <EOI>  [<ffffffff8106e0d6>] ? native_safe_halt+0x6/0x10
   [<ffffffff8110fb1d>] ? trace_hardirqs_on+0xd/0x10
   [<ffffffff81040ac0>] default_idle+0x20/0x1a0
   [<ffffffff8104147f>] arch_cpu_idle+0xf/0x20
   [<ffffffff81102f8f>] default_idle_call+0x2f/0x50
   [<ffffffff8110332e>] cpu_startup_entry+0x37e/0x450
   [<ffffffff8105af70>] start_secondary+0x160/0x1a0

Note the hang only occurs when echoing the above from a physical serial
console, not from an ssh session.

The bug is caused by a deadlock where the task is trying to grab the rq
lock twice because printk()'s aren't safe in sched code.

Signed-off-by: Josh Poimboeuf <jpoimboe@redhat.com>
Cc: Linus Torvalds <torvalds@linux-foundation.org>
Cc: Matt Fleming <matt@codeblueprint.co.uk>
Cc: Mel Gorman <mgorman@techsingularity.net>
Cc: Mike Galbraith <efault@gmx.de>
Cc: Peter Zijlstra <peterz@infradead.org>
Cc: Srikar Dronamraju <srikar@linux.vnet.ibm.com>
Cc: Thomas Gleixner <tglx@linutronix.de>
Cc: stable@vger.kernel.org
Fixes: cb2517653fcc ("sched/debug: Make schedstats a runtime tunable that is disabled by default")
Link: http://lkml.kernel.org/r/20160613073209.gdvdybiruljbkn3p@treble
Signed-off-by: Ingo Molnar <mingo@kernel.org>
kernel/sched/fair.c

index 4e33ad1..a2348de 100644 (file)
@@ -3246,7 +3246,7 @@ static inline void check_schedstat_required(void)
                        trace_sched_stat_iowait_enabled()  ||
                        trace_sched_stat_blocked_enabled() ||
                        trace_sched_stat_runtime_enabled())  {
-               pr_warn_once("Scheduler tracepoints stat_sleep, stat_iowait, "
+               printk_deferred_once("Scheduler tracepoints stat_sleep, stat_iowait, "
                             "stat_blocked and stat_runtime require the "
                             "kernel parameter schedstats=enabled or "
                             "kernel.sched_schedstats=1\n");