Linux Virtualization - Jan 2014 - [PATCH net-next 1/3] net: allow > 0 order atomic page alloc in skb_page_frag_refill

On Thu, 2014-01-02 at 16:56 -0800, Eric Dumazet wrote:
> 
> My suggestion is to use a recent kernel, and/or eventually backport the
> mm fixes if any.
> 
> order-3 allocations should not reclaim 2GB out of 8GB.
> 
> There is a reason PAGE_ALLOC_COSTLY_ORDER exists and is 3
Hmm... it looks like I missed __GFP_NORETRY



diff --git a/net/core/sock.c b/net/core/sock.c
index 5393b4b719d7..5f42a4d70cb2 100644
--- a/net/core/sock.c
+++ b/net/core/sock.c
@@ -1872,7 +1872,7 @@ bool skb_page_frag_refill(unsigned int sz, struct
page_frag *pfrag, gfp_t prio)
 		gfp_t gfp = prio;
 
 		if (order)
-			gfp |= __GFP_COMP | __GFP_NOWARN;
+			gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY;
 		pfrag->page = alloc_pages(gfp, order);
 		if (likely(pfrag->page)) {
 			pfrag->offset = 0;

On Thu, Jan 2, 2014 at 8:26 PM, Eric Dumazet <eric.dumazet at gmail.com>
wrote:
> On Thu, 2014-01-02 at 16:56 -0800, Eric Dumazet wrote:
>
>>
>> My suggestion is to use a recent kernel, and/or eventually backport the
>> mm fixes if any.
>>
>> order-3 allocations should not reclaim 2GB out of 8GB.
>>
>> There is a reason PAGE_ALLOC_COSTLY_ORDER exists and is 3
Sorry 2GB cache out of 8GB phys, ~1GB gets reclaimed. Regardless the
reclaimation of cache is minor compared to the compaction event that
precedes it, I haven't spotted something addressing that yet -
isolate_migratepages_range ()/compact_checklock_irqsave(). If even
more of memory was unmoveable, the compaction routines would be hit
even harder as reclaimation wouldn't do anything - mm would have to
get very very smart about unmoveable pages being freed and just fail
allocations/oom kill if nothing has changed without running through
compaction/reclaim fruitlessly.

I guess this is a bit of a tangent since what I'm saying proves the
patch from Michael doesn't make this behavior worse.
>
> Hmm... it looks like I missed __GFP_NORETRY
>
>
>
> diff --git a/net/core/sock.c b/net/core/sock.c
> index 5393b4b719d7..5f42a4d70cb2 100644
> --- a/net/core/sock.c
> +++ b/net/core/sock.c
> @@ -1872,7 +1872,7 @@ bool skb_page_frag_refill(unsigned int sz, struct
page_frag *pfrag, gfp_t prio)
>                 gfp_t gfp = prio;
>
>                 if (order)
> -                       gfp |= __GFP_COMP | __GFP_NOWARN;
> +                       gfp |= __GFP_COMP | __GFP_NOWARN | __GFP_NORETRY;
>                 pfrag->page = alloc_pages(gfp, order);
>                 if (likely(pfrag->page)) {
>                         pfrag->offset = 0;
>
>
>
Yes this seems like it will make the situation better, but one send()
may still cause a direct_compact and direct_reclaim() cycle to happen,
followed immediately by another direct_compact() if direct_reclaim()
didn't free an order-3. Now have all cpu's doing a send(), you can
still get heavy spinlock contention in the routines described above.
The major change I see here is that allocations > order-0 used to be
rare, now it's on every send().

I can try your patch to see how much things improve.

-Debabrata

>> On Thu, 2014-01-02 at 16:56 -0800, Eric Dumazet wrote:
>>
>> Hmm... it looks like I missed __GFP_NORETRY
>>
>>
>>
>> diff --git a/net/core/sock.c b/net/core/sock.c
>> index 5393b4b719d7..5f42a4d70cb2 100644
>> --- a/net/core/sock.c
>> +++ b/net/core/sock.c
>> @@ -1872,7 +1872,7 @@ bool skb_page_frag_refill(unsigned int sz, struct
page_frag *pfrag, gfp_t prio)
>>                 gfp_t gfp = prio;
>>
>>                 if (order)
>> -                       gfp |= __GFP_COMP | __GFP_NOWARN;
>> +                       gfp |= __GFP_COMP | __GFP_NOWARN |
__GFP_NORETRY;
>>                 pfrag->page = alloc_pages(gfp, order);
>>                 if (likely(pfrag->page)) {
>>                         pfrag->offset = 0;
>>
>>
>>
There is another patch needed (looks like good stable fixes):
diff --git a/net/core/skbuff.c b/net/core/skbuff.c
index 06e72d3..d42d48c 100644
--- a/net/core/skbuff.c
+++ b/net/core/skbuff.c
@@ -378,7 +378,7 @@ refill:
                        gfp_t gfp = gfp_mask;

                        if (order)
-                               gfp |= __GFP_COMP | __GFP_NOWARN;
+                               gfp |= __GFP_COMP | __GFP_NOWARN |
__GFP_NORETRY;
                        nc->frag.page = alloc_pages(gfp, order);
                        if (likely(nc->frag.page))
                                break;

This reduces the really pathological compact/reclaim behavior but
doesn't fix it. Actually it still really quite bad because the whole
thing loops until it gets to order-0 so it's effectively trying the
allocation 4 times anyway. I typically see non-zero order allocations
very rarely without these two pieces of code. I hotpatched a running
system to get results from this quickly. Even setting the max order to
order-1 I still see bad behavior. If anything this behavior should be
conditional until this is ironed out.

Performance data: http://pastebin.ubuntu.com/6687527/

-Debabrata