Re: [PATCH] PM / Hibernate: Feed the wathdog when creating snapshot

[Chen Yu]

Hi Andrew,
On Mon, Aug 21, 2017 at 01:26:00PM -0700, Andrew Morton wrote:
> On Mon, 21 Aug 2017 23:08:18 +0800 Chen Yu <yu.c.chen@xxxxxxxxx> wrote:
> 
> > There is a problem that when counting the pages for creating
> > the hibernation snapshot will take significant amount of
> > time, especially on system with large memory. Since the counting
> > job is performed with irq disabled, this might lead to NMI lockup.
> > The following warning were found on a system with 1.5TB DRAM:
> > 
> > ...
> > 
> > It has taken nearly 20 seconds(2.10GHz CPU) thus the NMI lockup
> > was triggered. In case the timeout of the NMI watch dog has been
> > set to 1 second, a safe interval should be 6590003/20 = 320k pages
> > in theory. However there might also be some platforms running at a
> > lower frequency, so feed the watchdog every 100k pages.
> > 
> > ...
> >
> > --- a/mm/page_alloc.c
> > +++ b/mm/page_alloc.c
> > @@ -2531,9 +2532,12 @@ void drain_all_pages(struct zone *zone)
> >  
> >  #ifdef CONFIG_HIBERNATION
> >  
> > +/* Touch watchdog for every WD_INTERVAL_PAGE pages. */
> > +#define WD_INTERVAL_PAGE	(100*1024)
> > +
> >  void mark_free_pages(struct zone *zone)
> >  {
> > -	unsigned long pfn, max_zone_pfn;
> > +	unsigned long pfn, max_zone_pfn, page_num = 0;
> >  	unsigned long flags;
> >  	unsigned int order, t;
> >  	struct page *page;
> > @@ -2548,6 +2552,9 @@ void mark_free_pages(struct zone *zone)
> >  		if (pfn_valid(pfn)) {
> >  			page = pfn_to_page(pfn);
> >  
> > +			if (!((page_num++) % WD_INTERVAL_PAGE))
> > +				touch_nmi_watchdog();
> > +
> >  			if (page_zone(page) != zone)
> >  				continue;
> >  
> > @@ -2561,8 +2568,11 @@ void mark_free_pages(struct zone *zone)
> >  			unsigned long i;
> >  
> >  			pfn = page_to_pfn(page);
> > -			for (i = 0; i < (1UL << order); i++)
> > +			for (i = 0; i < (1UL << order); i++) {
> > +				if (!((page_num++) % WD_INTERVAL_PAGE))
> > +					touch_nmi_watchdog();
> >  				swsusp_set_page_free(pfn_to_page(pfn + i));
> > +			}
> >  		}
> >  	}
> >  	spin_unlock_irqrestore(&zone->lock, flags);
> 
> hm, is it really worth all the WD_INTERVAL_PAGE stuff? 
> touch_nmi_watchdog() is pretty efficient and calling it once-per-page
> may not have a measurable effect.
>
We have version 1 of patch to feed the dog once-per-page. And we thought
it might look more elegant if we feed the dog every N pages.
> And if we're really concerned about the performance impact it would be
> better to make WD_INTERVAL_PAGE a power of 2 (128*1024?) to avoid the
> modulus operation.
> 
Ok, I'll change the interval to 128*1024 then.
Thanks,
	Yu