Re: [PATCH -mm] swsusp: Improve handling of highmem

[Pavel Machek <pavel@ucw.cz>]

Hi!

> > > Currently swsusp saves the contents of highmem pages by copying them to the
> > > normal zone which is quite inefficient  (eg. it requires two normal pages to be
> > > used for saving one highmem page).  This may be improved by using highmem
> > > for saving the contents of saveable highmem pages.
> > ...
> > >  include/linux/suspend.h |    9 
> > >  kernel/power/power.h    |    2 
> > >  kernel/power/snapshot.c |  841 ++++++++++++++++++++++++++++++++++++------------
> > >  kernel/power/swap.c     |    2 
> > >  kernel/power/swsusp.c   |   53 +--
> > >  kernel/power/user.c     |    2 
> > >  mm/vmscan.c             |    3 
> > 
> > Well, I just hoped that highmem would quietly die out...
> > 
> > ...
> > > +static struct page *alloc_image_page(gfp_t gfp_mask) {
> > > +	struct page *page;
> > 
> > { should go on new line.
> 
> Ah, yes, thanks.  I'll fix this later if you don't mind.

Ok.

> > >  	memory_bm_position_reset(orig_bm);
> > >  	memory_bm_position_reset(copy_bm);
> > >  	do {
> > >  		pfn = memory_bm_next_pfn(orig_bm);
> > > -		if (likely(pfn != BM_END_OF_MAP)) {
1) ##############################################
> > > -			struct page *page;
> > > -			void *src;
> > > -
> > > -			page = pfn_to_page(pfn);
> > > -			src = page_address(page);
> > > -			page = pfn_to_page(memory_bm_next_pfn(copy_bm));
> > > -			copy_data_page(page_address(page), src);
> > > -		}
> > > +		if (likely(pfn != BM_END_OF_MAP))
2) ##############################################
> > > +			copy_data_page(memory_bm_next_pfn(copy_bm), pfn);
> > >  	} while (pfn != BM_END_OF_MAP);
3) ####################################
> > >  }
> > 
> > While(1) and "if (pfn != BM_END_OF_MAP) { ...break; } ? Why do you
> > need to test pfn != BM_END_OF_MAP *three* times?
> 
> Why?  It's two times, and I don't like while(1) loops, really.

I see three of them, and while(1) loop is probably best solution
here... "Too scared of break; so I copy conditions"? ;-).

> > > @@ -1233,6 +1233,9 @@ out:
> > >  	}
> > >  	if (!all_zones_ok) {
> > >  		cond_resched();
> > > +
> > > +		try_to_freeze();
> > > +
> > >  		goto loop_again;
> > >  	}
> > 
> > What is this?
> 
> :-)
> 
> This is needed because during the resume there likely are no free pages in the
> highmem zone which makes kswapd spin forever, but we have to freeze it before
> the image is restored.

Could that happen with current code, too? (I'm trying to sense if this
should be merged now as a separate patch).
								Pavel
-- 
(english) http://www.livejournal.com/~pavelmachek
(cesky, pictures) http://atrey.karlin.mff.cuni.cz/~pavel/picture/horses/blog.html