DEBUG_LOCKS_WARN_ON() within lockdep trace_hardirqs_on_caller()

Hello,

Current 3.1.0-git3 (03873-g2c9c838), x86_64, has triggered an DEBUG_LOCKS_WARN_ON() within trace_hardirqs_on_caller()

[ 4364.656167] ------------[ cut here ]------------
[ 4364.656186] WARNING: at kernel/lockdep.c:2532 trace_hardirqs_on_caller+0x6d/0x166()
[ 4364.656191] Hardware name: Aspire 5741G    
[ 4364.656264] Pid: 26796, comm: conftest Not tainted 3.1.0-dbg-03873-g2c9c838-dirty #652
[ 4364.656269] Call Trace:
[ 4364.656279]  [<ffffffff8103e970>] warn_slowpath_common+0x7e/0x96
[ 4364.656289]  [<ffffffff81477056>] ? __bad_area_nosemaphore+0x35/0x1b9
[ 4364.656296]  [<ffffffff8103e99d>] warn_slowpath_null+0x15/0x17
[ 4364.656303]  [<ffffffff81072770>] trace_hardirqs_on_caller+0x6d/0x166
[ 4364.656310]  [<ffffffff81072876>] trace_hardirqs_on+0xd/0xf
[ 4364.656317]  [<ffffffff81477056>] __bad_area_nosemaphore+0x35/0x1b9
[ 4364.656324]  [<ffffffff8147722f>] bad_area+0x45/0x4c
[ 4364.656333]  [<ffffffff814850f9>] do_page_fault+0x2da/0x48a
[ 4364.656343]  [<ffffffff810ec2d0>] ? do_brk+0x24e/0x2a2
[ 4364.656352]  [<ffffffff8124642d>] ? trace_hardirqs_off_thunk+0x3a/0x3c
[ 4364.656360]  [<ffffffff81482975>] page_fault+0x25/0x30
[ 4364.656366] ---[ end trace 9870b4ffc9ea6750 ]---

I guess, local_irq_enable() gets called 2 times in do_page_fault(). First,

1051 if (user_mode_vm(regs)) {
1052 local_irq_enable();
1053 error_code |= PF_USER;
1054 } else {
1055 if (regs->flags & X86_EFLAGS_IF)
1056 local_irq_enable();
1057 }

and later within one of bad_area-cases, e.g.,
1106 vma = find_vma(mm, address);
1107 if (unlikely(!vma)) {
1108 bad_area(regs, error_code, address);
1109 return;
1110 }

hitting assertion `!irqs_disabled()' failure in lockdep trace_hardirqs_on_caller().

Sergey

I hit this a few days ago too. http://lkml.org/lkml/2011/7/23/167
It started happening as soon as the x86 merge hit Linus tree (3.0 was fine).

Dave