Re: [PATCH -mm] mm,numa,mprotect: always continue after finding a stable thp page

[Mel Gorman <mgorman@suse.de>]

On Fri, Mar 07, 2014 at 03:09:23PM +0000, Mel Gorman wrote:
> On Fri, Mar 07, 2014 at 02:06:50PM +0000, Mel Gorman wrote:
> > On Thu, Mar 06, 2014 at 05:52:47PM -0500, Rik van Riel wrote:
> > > On 03/06/2014 05:31 PM, Rik van Riel wrote:
> > > >On Thu, 06 Mar 2014 16:12:28 -0500
> > > >Sasha Levin <sasha.levin@oracle.com> wrote:
> > > >
> > > >>While fuzzing with trinity inside a KVM tools guest running latest -next kernel I've hit the
> > > >>following spew. This seems to be introduced by your patch "mm,numa: reorganize change_pmd_range()".
> > > >
> > > >That patch should not introduce any functional changes, except for
> > > >the VM_BUG_ON that catches the fact that we fell through to the 4kB
> > > >pte handling code, despite having just handled a THP pmd...
> > > >
> > > >Does this patch fix the issue?
> > > >
> > > >Mel, am I overlooking anything obvious? :)
> > > >
> > > >---8<---
> > > >
> > > >Subject: mm,numa,mprotect: always continue after finding a stable thp page
> > > >
> > > >When turning a thp pmds into a NUMA one, change_huge_pmd will
> > > >return 0 when the pmd already is a NUMA pmd.
> > > 
> > > I did miss something obvious.  In this case, the code returns 1.
> > > 
> > > >However, change_pmd_range would fall through to the code that
> > > >handles 4kB pages, instead of continuing on to the next pmd.
> > > 
> > > Maybe the case that I missed is when khugepaged is in the
> > > process of collapsing pages into a transparent huge page?
> > > 
> > > If the virtual CPU gets de-scheduled by the host for long
> > > enough, it would be possible for khugepaged to run on
> > > another virtual CPU, and turn the pmd into a THP pmd,
> > > before that VM_BUG_ON test.
> > > 
> > > I see that khugepaged takes the mmap_sem for writing in the
> > > collapse code, and it looks like task_numa_work takes the
> > > mmap_sem for reading, so I guess that may not be possible?
> > > 
> > 
> > mmap_sem will prevent a parallel collapse but what prevents something
> > like the following?
> > 
> > 							do_huge_pmd_wp_page
> > change_pmd_range
> > if (!pmd_trans_huge(*pmd) && pmd_none_or_clear_bad(pmd))
> > 	continue;
> > 							pmdp_clear_flush(vma, haddr, pmd);
> > if (pmd_trans_huge(*pmd)) {
> > 	.... path not taken ....
> > }
> > 							page_add_new_anon_rmap(new_page, vma, haddr);
> > 							set_pmd_at(mm, haddr, pmd, entry);
> > VM_BUG_ON(pmd_trans_huge(*pmd));
> > 
> > We do not hold the page table lock during the pmd_trans_huge check and we
> > do not recheck it under PTF lock in change_pte_range()
> > 
> 
> This is a completely untested prototype. It rechecks pmd_trans_huge
> under the lock and falls through if it hit a parallel split. It's not
> perfect because it could decide to fall through just because there was
> no prot_numa work to do but it's for illustration purposes. Secondly,
> I noted that you are calling invalidate for every pmd range. Is that not
> a lot of invalidations? We could do the same by just tracking the address
> of the first invalidation.
> 

And there were other minor issues. This is still untested but Sasha,
can you try it out please? I discussed this with Rik on IRC for a bit and
reckon this should be sufficient if the correct race has been identified.

The race can only really happen for prot_numa updates and it's ok to bail on
those updates if a race occurs because all we miss is a few hinting faults.
That simplifies the patch considerably but throw in some comments to
explain it

diff --git a/mm/mprotect.c b/mm/mprotect.c
index 2afc40e..72061a2 100644
--- a/mm/mprotect.c
+++ b/mm/mprotect.c
@@ -46,6 +46,17 @@ static unsigned long change_pte_range(struct vm_area_struct *vma, pmd_t *pmd,
 	unsigned long pages = 0;
 
 	pte = pte_offset_map_lock(mm, pmd, addr, &ptl);
+
+	/*
+	 * For a prot_numa update we only hold mmap_sem for read so there is a
+	 * potential race with faulting where a pmd was temporarily none so
+	 * recheck it under the lock and bail if we race
+	 */
+	if (prot_numa && unlikely(pmd_trans_huge(*pmd))) {
+		pte_unmap_unlock(pte, ptl);
+		return 0;
+	}
+
 	arch_enter_lazy_mmu_mode();
 	do {
 		oldpte = *pte;
@@ -141,12 +152,13 @@ static inline unsigned long change_pmd_range(struct vm_area_struct *vma,
 						pages += HPAGE_PMD_NR;
 						nr_huge_updates++;
 					}
+
+					/* huge pmd was handled */
 					continue;
 				}
 			}
 			/* fall through, the trans huge pmd just split */
 		}
-		VM_BUG_ON(pmd_trans_huge(*pmd));
 		this_pages = change_pte_range(vma, pmd, addr, next, newprot,
 				 dirty_accountable, prot_numa);
 		pages += this_pages;

--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"dont@kvack.org"> email@kvack.org </a>