Re: [linux-pm] [RFC PATCH 2/4] ACPI: introduce the mechanism to save/restore ACPI NVS memory

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Thursday, 10 of July 2008, Zhang Rui wrote:
> On Wed, 2008-07-09 at 22:26 +0200, Rafael J. Wysocki wrote:
> > On Wednesday, 9 of July 2008, Henrique de Moraes Holschuh wrote:
> > > On Tue, 08 Jul 2008, Pavel Machek wrote:
> > > > > >  static int acpi_hibernation_begin(void)
> > > > > >  {
> > > > > >  	acpi_target_sleep_state = ACPI_STATE_S4;
> > > > > >  
> > > > > > -	return 0;
> > > > > > +	/* allocate pages for ACPI NVS memory before swsusp_shrink_memory */
> > > > > > +	return acpi_allocate_nvs_pages();
> > > > > 
> > > > > I think we shouldn't abort hibernation because of that.
> > > > > 
> > > > > This may be an emergency hibernation due to critical battery status and we
> > > > > surely don't want to about that.
> > > > 
> > > > I disagree here. If spec says 'you must save this', and we don't, we
> > > > are asking for subtle, dangerous, and very hard to repoduce problems.
> > > 
> > > Indeed.  Especially after we start doing the right thing, and vendors WILL
> > > expect us to do the right thing (save the NVS memory) *always*, not just
> > > "almost always".
> > > 
> > > If you don't want the hibernate path to fail because of OOM, have whatever
> > > memory it needs pre-allocated, set aside and protected at startup so that it
> > > will always be there when you need to hibernate.
> > 
> > Okay, so be it.
> 
> You mean allocate enough pages during startup and never release them?

No, please leave that as is.

> > 
> > However, speaking of vendors, I'd like us to be able to switch that off using
> > an 'acpi_sleep=' option (s4_no_nvs comes to mind) and/or blacklist broken
> > BIOSes.
> 
> sound good, I'll do that. :)

OK

Thanks,
Rafael