Re: [PATCH] igb: don't reuse pages with pfmemalloc flag

[Roman Gushchin <klamm@yandex-team.ru>]

Thank you!

Probably we should add it to stable trees too?

--
Regards,
Roman

22.10.2014, 22:30, "Jeff Kirsher" <jeffrey.t.kirsher@intel.com>:
> On Wed, 2014-10-22 at 17:50 +0400, Roman Gushchin wrote:
>>  Incoming packet is dropped silently by sk_filter(), if the skb was
>>  allocated from pfmemalloc reserves and the corresponding socket is
>>  not marked with the SOCK_MEMALLOC flag.
>>
>>  Igb driver allocates pages for DMA with __skb_alloc_page(), which
>>  calls alloc_pages_node() with the __GFP_MEMALLOC flag. So, in case
>>  of OOM condition, igb can get pages with pfmemalloc flag set.
>>
>>  If an incoming packet hits the pfmemalloc page and is large enough
>>  (small packets are copying into the memory, allocated with
>>  netdev_alloc_skb_ip_align(), so they are not affected), it will be
>>  dropped.
>>
>>  This behavior is ok under high memory pressure, but the problem is
>>  that the igb driver reuses these mapped pages. So, packets are still
>>  dropping even if all memory issues are gone and there is a plenty
>>  of free memory.
>>
>>  In my case, some TCP sessions hang on a small percentage (< 0.1%)
>>  of machines days after OOMs.
>>
>>  Fix this by avoiding reuse of such pages.
>>
>>  Signed-off-by: Roman Gushchin <klamm@yandex-team.ru>
>>  ---
>>   drivers/net/ethernet/intel/igb/igb_main.c | 6 +++++-
>>   1 file changed, 5 insertions(+), 1 deletion(-)
>
> Thanks Roman, I have added you patch to my queue.

------------------------------------------------------------------------------
_______________________________________________
E1000-devel mailing list
E1000-devel@lists.sourceforge.net
https://lists.sourceforge.net/lists/listinfo/e1000-devel
To learn more about Intel&#174; Ethernet, visit http://communities.intel.com/community/wired