Re: [patch 3/3] clockevents: Fix resume logic - updated version

[Andrew Morton <akpm@linux-foundation.org>]

On Sat, 12 May 2007 11:18:09 +0200 Thomas Gleixner <tglx@linutronix.de> wrote:

> > It's peculiar that the hang happens when acpi_evaluate_object() hits its
> > return statement.  Any theories there?
> 
> Only stack or memory corruption come into mind, but I have no clue how
> this is related to the resume logic changes.

So I had the brilliant idea of turning on some kernel debugging.  It's
a shame that CONFIG_SOFTWARE_SUSPEND disables CONFIG_DEBUG_PAGEALLOC.

[   73.533454] swsusp: Basic memory bitmaps created
[   73.550429] Stopping tasks ... BUG: at kernel/lockdep.c:2414 check_flags()
[   73.550988]  [<c0104c14>] show_trace_log_lvl+0x1a/0x30
[   73.551143]  [<c0105769>] show_trace+0x12/0x14
[   73.551279]  [<c01057c1>] dump_stack+0x15/0x17
[   73.551412]  [<c0132732>] check_flags+0x93/0x13d
[   73.551554]  [<c0135558>] lock_acquire+0x28/0x7f
[   73.551691]  [<c0310e35>] _spin_lock+0x2b/0x38
[   73.551827]  [<c013dc43>] refrigerator+0x16/0xc7
[   73.551965]  [<c0125d2e>] get_signal_to_deliver+0x32/0x387
[   73.552124]  [<c010336d>] do_notify_resume+0x91/0x6a9
[   73.552271]  [<c0103df1>] work_notifysig+0x13/0x1a
[   73.552413]  =======================
[   73.552507] irq event stamp: 3075
[   73.552595] hardirqs last  enabled at (3075): [<c0103e51>] syscall_exit_work+0x11/0x26
[   73.552821] hardirqs last disabled at (3074): [<c0103d35>] syscall_exit+0x9/0x1a
[   73.553046] softirqs last  enabled at (2778): [<c01209f2>] __do_softirq+0x92/0x9a
[   73.553255] softirqs last disabled at (2693): [<c0120a27>] do_softirq+0x2d/0x46
[   73.559504] done.
[   73.559569] Shrinking memory...  \b-\bdone (0 pages freed)
[   73.646511] Freed 0 kbytes in 0.08 seconds (0.00 MB/s)
[   73.649595] platform sonypi: freeze
[   73.649707] platform bluetooth: freeze
[   73.649817] usb_endpoint usbdev5.1_ep81: PM: suspend 0->1, parent 5-0:1.0 already 2
[   73.650023] hub 5-0:1.0: PM: suspend 2-->1

<snippage>

[   73.739499] ipw2200 0000:06:0b.0: freeze
[   73.743860] eth1: Going into suspend...
[   73.748444] e100 0000:06:08.0: freeze

at this point I lost netconsole (earlier testing was without netconsole
btw)


The lockdep spew is coming out of here:

static void check_flags(unsigned long flags)
{
#if defined(CONFIG_DEBUG_LOCKDEP) && defined(CONFIG_TRACE_IRQFLAGS)
	if (!debug_locks)
		return;

	if (irqs_disabled_flags(flags))
-->		DEBUG_LOCKS_WARN_ON(current->hardirqs_enabled);
	else
		DEBUG_LOCKS_WARN_ON(!current->hardirqs_enabled);

and the callsite is:

void refrigerator(void)
{
	/* Hmm, should we be allowed to suspend when there are realtime
	   processes around? */
	long save;

-->	task_lock(current);
	if (freezing(current)) {
		frozen_process();
		task_unlock(current);
	} else {


I don't really know what lockdep is complaining about there.  I assume I'm
not supposed to, given that whoever wrote that couldn't be bothered
documenting any of it.

I _think_ it means that lockdep believes that local irqs are enabled
(according to its state tracking), only it turns out that they're not.