Re: oops on resume

[Dominik Brodowski <linux@dominikbrodowski.de>]

On Sat, Aug 07, 2004 at 03:03:42PM -0400, Brian J. Murrell wrote:
> On Sat, 2004-08-07 at 15:24 +0200, Dominik Brodowski wrote:
> > 
> > Might it be that interrupts are still disabled during the RESUME
> > notification, and 
> > 
> > #define write_sequnlock_irq(lock)                                       \
> >         do { write_sequnlock(lock); local_irq_enable(); } while(0)
> > 
> > tries to enable them?
> 
> But that sounds like such a fundamental issue that surely I would not be
> the first person to find it, especially considering how long even I was
> seeing it before I found the time to work it up into a bug report.

Problem is, I don't use the resume funcitonality on my notebook as that's
only available with the "vga" video driver instead of "i830"; and I can't
even try it now as my notebook is in quite a sick state [disk corruption!]
caused either by Ingo's voluntary-preempt patches or some experimenting by
myself, which accessed free'd data as if it were non-free'd...

Anyways, I've prepared a patch to work around the "local_irq_enable()" call.
Could you test it, please?

	Dominik

diff -ruN linux-original/arch/i386/kernel/timers/timer_tsc.c linux/arch/i386/kernel/timers/timer_tsc.c
--- linux-original/arch/i386/kernel/timers/timer_tsc.c	2004-08-05 17:44:47.000000000 +0200
+++ linux/arch/i386/kernel/timers/timer_tsc.c	2004-08-08 18:23:27.791458408 +0200
@@ -265,7 +265,8 @@
 {
 	struct cpufreq_freqs *freq = data;
 
-	write_seqlock_irq(&xtime_lock);
+	if (val != CPUFREQ_RESUMECHANGE)
+		write_seqlock_irq(&xtime_lock);
 	if (!ref_freq) {
 		ref_freq = freq->old;
 		loops_per_jiffy_ref = cpu_data[freq->cpu].loops_per_jiffy;
@@ -291,7 +292,9 @@
 		}
 #endif
 	}
-	write_sequnlock_irq(&xtime_lock);
+
+	if (val != CPUFREQ_RESUMECHANGE)
+		write_sequnlock_irq(&xtime_lock);
 
 	return 0;
 }