Re: [PATCH 4/4] mm: migrate: Batch TLB flushing when unmapping pages for migration

[Mel Gorman]

On Thu, Apr 16, 2015 at 11:57:15AM -0700, Hugh Dickins wrote:
> > @@ -1098,6 +1098,8 @@ int migrate_pages(struct list_head *from, new_page_t get_new_page,
> >  	if (!swapwrite)
> >  		current->flags |= PF_SWAPWRITE;
> >  
> > +	alloc_tlb_ubc();
> > +
> >  	for(pass = 0; pass < 10 && retry; pass++) {
> >  		retry = 0;
> >  
> > @@ -1144,6 +1146,8 @@ out:
> >  	if (!swapwrite)
> >  		current->flags &= ~PF_SWAPWRITE;
> >  
> > +	try_to_unmap_flush();
> 
> This is the right place to aim to flush, but I think you have to make
> more changes before it is safe to do so here.
> 
> The putback_lru_page(page) in unmap_and_move() is commented "A page
> that has been migrated has all references removed and will be freed".
> 
> If you leave TLB flushing until after the page has been freed, then
> there's a risk that userspace will see, not the data it expects at
> whatever virtual address, but data placed in there by the next user
> of this freed page.
> 
> So you'll need to do a little restructuring first.
> 

Well spotted. I believe you are correct and it almost certainly applies to
patch 2 as well for similar reasons. It also impacts the maximum reasonable
batch size that can be managed while maintaing safety. I'll do the necessary
shuffling tomorrow or Monday.

-- 
Mel Gorman
SUSE Labs
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/