Some results from Alexander's acpi_leave_sleep_state patch

[Huw Rogers <count0-tC47gz4GrgtWk0Htik3J/w@public.gmane.org>]

Alexander's patch had some interesting effects on my S3 resume problems
with my SiS 648FX/963 / P4 / Radeon 9600 laptop
(http://bugme.osdl.org/show_bug.cgi?id=2048). Although the system was
still locked up, the display was no longer inoperable to the point that
the battery and AC had to be removed to get it back, and I could use a
regular power cycle to recover. A mild improvement.

S4 (CONFIG_PM_DISK) resume booted (wrongly) via POST -> boot loader, and
Linux attempted to repeat the suspend during it's boot! Previously the
S4 resume did go straight to Linux resume as it should. From the console
log (unfortunately not written to disk for posterity for either S3 or S4),
S4 seems confused about whether it's suspending or resuming during both
suspension and resume (I see lines mentioning resume activity during
suspend and vice versa).

This is a SiS 648FX/963 chipset, HT P4, Radeon M-10 (9600 Mobile) laptop.

-Huw

Alexander Malysh wrote:
> what are you think about attached patch?
> 
> Note: at least on my laptop gpes must be enabled before '_WAK' method 
> execution, otherwise after '_WAK' was executed interrupts are enabled and gpe 
> register were restored with wrong values...
> 
> @Karol: can you please get this patch a try?
> 
> On Sunday 15 February 2004 22:00, Alexander Malysh wrote:
> > On Sunday 15 February 2004 21:07, you wrote:
> > > On Sun, Feb 15, 2004 at 04:47:19PM +0100, Alexander Malysh wrote:
> > > > Hi again,
> > > >
> > > > after some more testing, I have found a cause for this.
> > > > We disable non wakeup gpes (and store those values) with interrupts
> > > > disabled, but enable those with interrupts enabled which were already
> > > > overwritten while handling of wakeup interrupt[1]. So we must restore
> > > > non wakeup gpes as long interrupts are disabled. Attached patch should
> > > > fix it.
> > > >
> > > >
> > > > --- linux-2.6.2-linus/drivers/acpi/hardware/hwsleep.c~orig	2004-02-15
> > > > 16:31:50.928455040 +0100 +++
> > > > linux-2.6.2-linus/drivers/acpi/hardware/hwsleep.c	2004-02-15
> > > > 16:35:30.030146520 +0100 @@ -342,6 +342,12 @@
> > > >
> > > >  	} while (!in_value);
> > > >
> > > > +	/* Enable non_wakeup_gpes as long interrupts are disabled */
> > > > +	status = acpi_hw_enable_non_wakeup_gpes ();
> > > > +	if (ACPI_FAILURE (status)) {
> > > > +		return_ACPI_STATUS (status);
> > > > +	}
> > > > +
> > > >  	return_ACPI_STATUS (AE_OK);
> > >
> > > First, this path is *never reached* for S3 and S4, so you break somehow
> > > S3 and S4 in fact...
> >
> > hmm, where do we jump after wakeup from S3/S4 ?
> >
> > > Second, the path for which you remove the enable gpes is supposed to be
> > > run with interrupt *dissabled*.  If that is not the case, then it is the
> > > real bug in fact.
> >
> > as you can see from log (or in drivers/acpi/sleep/main.c)
> > acpi_leave_sleep_state called with interrupts enabled.
> > and you are right: acpi_leave_sleep_state _must_ be called with interrupts
> > disabled, so no wonder why sleep/resume fail on many laptops :(


-- 
Huw Rogers <count0-tC47gz4GrgtWk0Htik3J/w@public.gmane.org>



-------------------------------------------------------
SF.Net is sponsored by: Speed Start Your Linux Apps Now.
Build and deploy apps & Web services for Linux with
a free DVD software kit from IBM. Click Now!
http://ads.osdn.com/?ad_id=1356&alloc_id=3438&op=click