Re: [RFC] mm: fix a BUG, the page is allocated 2 times

[Vlastimil Babka]

On 10/12/2015 12:05 PM, Kirill A. Shutemov wrote:
> On Mon, Oct 12, 2015 at 03:58:51PM +0800, yalin wang wrote:
> > > On Oct 12, 2015, at 15:38, Vlastimil Babka <vbabka@xxxxxxx> wrote:
> > >
> > > On 10/12/2015 04:40 AM, yalin wang wrote:
> > > > Remove unlikely(order), because we are sure order is not zero if
> > > > code reach here, also add if (page == NULL), only allocate page again if
> > > > __rmqueue_smallest() failed or alloc_flags & ALLOC_HARDER == 0
> > >
> > > The second mentioned change is actually more important as it removes a memory leak! Thanks for catching this. The problem is in patch mm-page_alloc-reserve-pageblocks-for-high-order-atomic-allocations-on-demand.patch and seems to have been due to a change in the last submitted version to make sure the tracepoint is called.
> > >
> > > > Signed-off-by: yalin wang <yalin.wang2010@xxxxxxxxx>
> > > > ---
> > > >   mm/page_alloc.c | 6 +++---
> > > >   1 file changed, 3 insertions(+), 3 deletions(-)
> > > >
> > > > diff --git a/mm/page_alloc.c b/mm/page_alloc.c
> > > > index 0d6f540..de82e2c 100644
> > > > --- a/mm/page_alloc.c
> > > > +++ b/mm/page_alloc.c
> > > > @@ -2241,13 +2241,13 @@ struct page *buffered_rmqueue(struct zone *preferred_zone,
> > > >   		spin_lock_irqsave(&zone->lock, flags);
> > > >
> > > >   		page = NULL;
> > > > -		if (unlikely(order) && (alloc_flags & ALLOC_HARDER)) {
> > > > +		if (alloc_flags & ALLOC_HARDER) {
> > > >   			page = __rmqueue_smallest(zone, order, MIGRATE_HIGHATOMIC);
> > > >   			if (page)
> > > >   				trace_mm_page_alloc_zone_locked(page, order, migratetype);
> > > >   		}
> > > > -
> > > > -		page = __rmqueue(zone, order, migratetype, gfp_flags);
> > > > +		if (page == NULL)
> > >
> > > "if (!page)" is more common and already used below.
> > > We could skip the check for !page in case we don't go through the ALLOC_HARDER branch, but I guess it's not worth the goto, and hopefully the compiler is smart enough anywayâ
> > agree with your comments,
> > do i need send a new patch for this ?

I'd guess no need to, Andrew can edit the patch?

> Looks like a two patches to me: memory leak and removing always-true part
> of condifition.

Yeah but I'd expect both would be in the end folded into the buggy patch 
in -mm?

--
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/