On Wed, Nov 07, 2012 at 04:11:46PM -0800, Andrew Morton wrote:
> On Thu, 08 Nov 2012 09:32:18 +1030
> Rusty Russell <ru...@rustcorp.com.au> wrote:
> 
> > Rafael Aquini <aqu...@redhat.com> writes:
> > > + * virtballoon_migratepage - perform the balloon page migration on 
> > > behalf of
> > > + *                            a compation thread.     (called under page 
> > > lock)
> > 
> > > + if (!mutex_trylock(&vb->balloon_lock))
> > > +         return -EAGAIN;
> > 
> > Erk, OK...
> 
> Not really.  As is almost always the case with a trylock, it needs a
> comment explaining why we couldn't use the far superior mutex_lock(). 
> Data: this reader doesn't know!
>

That was just to alleviate balloon_lock contention if we're migrating pages
concurrently with balloon_fill() or balloon_leak(), as it's easier to retry
the page migration later (in the contended case).

 
> > > + /* balloon's page migration 1st step  -- inflate "newpage" */
> > > + spin_lock_irqsave(&vb_dev_info->pages_lock, flags);
> > > + balloon_page_insert(newpage, mapping, &vb_dev_info->pages);
> > > + vb_dev_info->isolated_pages--;
> > > + spin_unlock_irqrestore(&vb_dev_info->pages_lock, flags);
> > > + vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
> > > + set_page_pfns(vb->pfns, newpage);
> > > + tell_host(vb, vb->inflate_vq);
> > 
> > tell_host does wait_event(), so you can't call it under the page_lock.
> > Right?
> 
> Sleeping inside lock_page() is OK.  More problematic is that GFP_KERNEL
> allocation.  iirc it _should_ be OK.  Core VM uses trylock_page() and
> the filesystems shouldn't be doing a synchronous lock_page() in the
> pageout path.  But I suspect it isn't a well-tested area.


The locked page under migration is not contended by any other FS / core VM path,
as it is already isolated by compaction to a private migration page list.
OTOH, there's a chance of another parallel compaction thread hitting this page
while scanning page blocks for isolation, but that path can be considered safe
as it uses trylock_page()



> 
> > You probably get away with it because current qemu will service you
> > immediately.  You could spin here in this case for the moment.
> > 
> > There's a second call to tell_host():
> > 
> > > + /*
> > > +  * balloon's page migration 2nd step -- deflate "page"
> > > +  *
> > > +  * It's safe to delete page->lru here because this page is at
> > > +  * an isolated migration list, and this step is expected to happen here
> > > +  */
> > > + balloon_page_delete(page);
> > > + vb->num_pfns = VIRTIO_BALLOON_PAGES_PER_PAGE;
> > > + set_page_pfns(vb->pfns, page);
> > > + tell_host(vb, vb->deflate_vq);
> > 
> > The first one can be delayed, the second one can be delayed if the host
> > didn't ask for VIRTIO_BALLOON_F_MUST_TELL_HOST (qemu doesn't).
> > 
> > We could implement a proper request queue for these, and return -EAGAIN
> > if the queue fills.  Though in practice, it's not important (it might
> > help performance).
> 
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majord...@vger.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Reply via email to