Re: kernel bug(VM_BUG_ON_PAGE) with 3.18.13 in mm/migrate.c

[Mel Gorman <mgorman@suse.de>]

On Thu, May 28, 2015 at 11:38:36AM -0700, Jovi Zhangwei wrote:
> Hi Mel,
> 
> On Thu, May 28, 2015 at 5:00 AM, Mel Gorman <mgorman@suse.de> wrote:
> > On Wed, May 27, 2015 at 11:05:33AM -0700, Jovi Zhangwei wrote:
> >> Hi,
> >>
> >> I got below kernel bug error in our 3.18.13 stable kernel.
> >> "kernel BUG at mm/migrate.c:1661!"
> >>
> >> Source code:
> >>
> >> 1657    static int numamigrate_isolate_page(pg_data_t *pgdat, struct page *page)
> >> 1658   {
> >> 1659            int page_lru;
> >> 1660
> >> 1661           VM_BUG_ON_PAGE(compound_order(page) &&
> >> !PageTransHuge(page), page);
> >>
> >> It's easy to trigger the error by run tcpdump in our system.(not sure
> >> it will easily be reproduced in another system)
> >> "sudo tcpdump -i bond0.100 'tcp port 4242' -c 100000000000 -w 4242.pcap"
> >>
> >> Any comments for this bug would be great appreciated. thanks.
> >>
> >
> > What sort of compound page is it? What sort of VMA is it in? hugetlbfs
> > pages should never be tagged for NUMA migrate and never enter this
> > path. Transparent huge pages are handled properly so I'm wondering
> > exactly what type of compound page this is and what mapped it into
> > userspace.
> >
> Thanks for your reply.
> 
> After reading net/packet/af_packet.c:alloc_one_pg_vec_page, I found
> there indeed have compound page maped into userspace.
> 

Ok, it's clear now. Thanks very much.

> I sent a patch for this issue(you may received it), but not sure it's
> right to fix,
> feel free to update it or use your own patch.
> 

It avoids the problem but it's not the best fix because a lot of useless
overhead has been incurred for a page that can never be migrated. Can you
try the following instead please?

---8<---

sched, numa: Do not hint for NUMA balancing on VM_MIXEDMAP mappings

Jovi Zhangwei reported the following problem

  Below kernel vm bug can be triggered by tcpdump which mmaped a lot of pages
  with GFP_COMP flag.

  [Mon May 25 05:29:33 2015] page:ffffea0015414000 count:66 mapcount:1 mapping:          (null) index:0x0
  [Mon May 25 05:29:33 2015] flags: 0x20047580004000(head)
  [Mon May 25 05:29:33 2015] page dumped because: VM_BUG_ON_PAGE(compound_order(page) && !PageTransHuge(page))
  [Mon May 25 05:29:33 2015] ------------[ cut here ]------------
  [Mon May 25 05:29:33 2015] kernel BUG at mm/migrate.c:1661!
  [Mon May 25 05:29:33 2015] invalid opcode: 0000 [#1] SMP

Compound pages cannot be migrated and it was not expected that such pages
be marked for NUMA balancing. This did not take into account that drivers
such as net/packet/af_packet.c may insert compound pages into userspace
with vm_insert_page. This patch tells the NUMA balancing protection scanner
to skip all VM_MIXEDMAP mappings which avoids the possibility that compound
pages are marked for migration.

Signed-off-by: Mel Gorman <mgorman@suse.de>
Reported-by: Jovi Zhangwei <jovi@cloudflare.com>
---
 kernel/sched/fair.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/kernel/sched/fair.c b/kernel/sched/fair.c
index 241213be507c..486d00c408b0 100644
--- a/kernel/sched/fair.c
+++ b/kernel/sched/fair.c
@@ -2166,7 +2166,7 @@ void task_numa_work(struct callback_head *work)
 	}
 	for (; vma; vma = vma->vm_next) {
 		if (!vma_migratable(vma) || !vma_policy_mof(vma) ||
-			is_vm_hugetlb_page(vma)) {
+			is_vm_hugetlb_page(vma) || (vma->vm_flags & VM_MIXEDMAP)) {
 			continue;
 		}