LKML Archive on lore.kernel.org
help / color / mirror / Atom feed
* [PATCH 0/2] Timer sync lock checking
@ 2008-10-23 19:35 Johannes Berg
  2008-10-23 19:44 ` [PATCH 1/2] lockdep: implement full check without irq checking Johannes Berg
                   ` (2 more replies)
  0 siblings, 3 replies; 8+ messages in thread
From: Johannes Berg @ 2008-10-23 19:35 UTC (permalink / raw)
  To: Thomas Gleixner; +Cc: Linux Kernel list, Peter Zijlstra, Ingo Molnar

[-- Attachment #1: Type: text/plain, Size: 4095 bytes --]

Hi,

Here are two patches that implement checking with lockdep that nobody
tries to implement something like this:

void timerfn(unsigned long data)
{
	spin_lock(&lock);
	...
	spin_unlock(&lock);
}

...

setup
	spin_lock_init(&lock);
	setup_timer(&timer, timerfn, 0);

...

tear_down
	spin_lock_bh(&lock);
	del_timer_sync(&timer);
	spin_unlock_bh(&lock);

Because of the usage of the fake lock in the timer code, I first needed
to teach lockdep about a new "check" value which means that it doesn't
check irq-safety. This is required because we patch 2 takes a fake lock
around the timerfn, and del_timer_sync() also takes this fake lock, but
quite obviously the former runs in softirq context and the latter
doesn't necessarily, but clearly it doesn't need to disable softirqs
either.

Unlike with the workqueue debugging where I did the same thing, I don't
actually know about any such bug. My test code resulted in the warning
below, as expected.

johannes

[   75.083150] =======================================================
[   75.083163] [ INFO: possible circular locking dependency detected ]
[   75.083170] 2.6.27-wl-03435-g9b45bb6-dirty #92
[   75.083177] -------------------------------------------------------
[   75.083184] rmmod/4365 is trying to acquire lock:
[   75.083191]  (&test_timer){-...}, at: [<c00000000005c834>] .del_timer_sync+0x0/0x94
[   75.083219] 
[   75.083220] but task is already holding lock:
[   75.083228]  (&lock){-+..}, at: [<d0000000001090b4>] .modexit+0x38/0x70 [timer_test]
[   75.083253] 
[   75.083254] which lock already depends on the new lock.
[   75.083257] 
[   75.083263] 
[   75.083265] the existing dependency chain (in reverse order) is:
[   75.083272] 
[   75.083273] -> #1 (&lock){-+..}:
[   75.083290]        [<c00000000007fc4c>] .lock_acquire+0xa4/0xec
[   75.083305]        [<c0000000003ee0ec>] ._spin_lock+0x50/0xac
[   75.083320]        [<d00000000010902c>] .timerfn+0x2c/0x7c [timer_test]
[   75.083331]        [<c00000000005bdcc>] .run_timer_softirq+0x214/0x2e4
[   75.083343]        [<c0000000000560cc>] .__do_softirq+0xd8/0x1c4
[   75.083354]        [<c00000000000c298>] .do_softirq+0x5c/0xb8
[   75.083366]        [<c000000000055b08>] .irq_exit+0x74/0xe0
[   75.083376]        [<c00000000001e0e8>] .timer_interrupt+0xe4/0x12c
[   75.083389]        [<c000000000003614>] decrementer_common+0x114/0x180
[   75.083400]        [<c000000000011ea8>] .cpu_idle+0xd0/0x200
[   75.083411]        [<c0000000003eee04>] .rest_init+0x8c/0xa4
[   75.083423]        [<c000000000588bc0>] .start_kernel+0x4a0/0x4c8
[   75.083437]        [<c000000000007568>] .start_here_common+0x3c/0x54
[   75.083449] 
[   75.083450] -> #0 (&test_timer){-...}:
[   75.083464]        [<c00000000007fc4c>] .lock_acquire+0xa4/0xec
[   75.083474]        [<c00000000005c878>] .del_timer_sync+0x44/0x94
[   75.083486]        [<d0000000001090c0>] .modexit+0x44/0x70 [timer_test]
[   75.083497]        [<c00000000008c320>] .sys_delete_module+0x278/0x314
[   75.083509]        [<c0000000000076d4>] syscall_exit+0x0/0x40
[   75.083520] 
[   75.083521] other info that might help us debug this:
[   75.083523] 
[   75.083530] 1 lock held by rmmod/4365:
[   75.083536]  #0:  (&lock){-+..}, at: [<d0000000001090b4>] .modexit+0x38/0x70 [timer_test]
[   75.083557] 
[   75.083558] stack backtrace:
[   75.083563] Call Trace:
[   75.083571] [c00000020ed8b8d0] [c00000000000f860] .show_stack+0x6c/0x174 (unreliable)
[   75.083586] [c00000020ed8b980] [c00000000007e008] .print_circular_bug_tail+0xc8/0xec
[   75.083598] [c00000020ed8ba50] [c00000000007f4d0] .__lock_acquire+0x10ac/0x1784
[   75.083609] [c00000020ed8bb50] [c00000000007fc4c] .lock_acquire+0xa4/0xec
[   75.083621] [c00000020ed8bc10] [c00000000005c878] .del_timer_sync+0x44/0x94
[   75.083633] [c00000020ed8bca0] [d0000000001090c0] .modexit+0x44/0x70 [timer_test]
[   75.083645] [c00000020ed8bd30] [c00000000008c320] .sys_delete_module+0x278/0x314
[   75.083657] [c00000020ed8be30] [c0000000000076d4] syscall_exit+0x0/0x40


[-- Attachment #2: This is a digitally signed message part --]
[-- Type: application/pgp-signature, Size: 836 bytes --]

^ permalink raw reply	[flat|nested] 8+ messages in thread

end of thread, other threads:[~2008-10-30 11:25 UTC | newest]

Thread overview: 8+ messages (download: mbox.gz / follow: Atom feed)
-- links below jump to the message on this page --
2008-10-23 19:35 [PATCH 0/2] Timer sync lock checking Johannes Berg
2008-10-23 19:44 ` [PATCH 1/2] lockdep: implement full check without irq checking Johannes Berg
2008-10-29 15:23   ` Peter Zijlstra
2008-10-30 10:36     ` Johannes Berg
2008-10-30 11:15       ` Peter Zijlstra
2008-10-30 11:25         ` Johannes Berg
2008-10-23 19:46 ` [PATCH 2/2] timer: implement lockdep deadlock detection Johannes Berg
2008-10-23 20:37 ` [PATCH 3/2] tasklet: " Johannes Berg

This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).