On Tue, 2013-01-08 at 18:14 -0800, Eric Dumazet wrote: > On Tue, 2013-01-08 at 23:23 +0000, Eric Wong wrote: > > Mel Gorman <[email protected]> wrote: > > > Please try the following patch. However, even if it works the benefit of > > > capture may be so marginal that partially reverting it and simplifying > > > compaction.c is the better decision. > > > > I already got my VM stuck on this one. I had two twosleepy instances, > > 2774 was the one that got stuck (also confirmed by watching top). > > > > Btw, have you been able to reproduce this on your end? > > > > I think the easiest reproduction on my 2-core VM is by running 2 > > twosleepy processes and doing the following to dirty a lot of pages: > > Given the persistent sk_stream_wait_memory() traces I suspect a plain > TCP bug, triggered by some extra wait somewhere. > > Please mm guys don't spend too much time right now, I'll try to > reproduce the problem. > > Don't be confused by sk_stream_wait_memory() name. > A thread is stuck here because TCP stack is failing to wake it. >
Hmm, it seems sk_filter() can return -ENOMEM because skb has the pfmemalloc() set. It seems nobody really tested this stuff under memory stress. Mel, it looks like you are the guy who could fix this, after all ;) One TCP socket keeps retransmitting an SKB via loopback, and TCP stack drops the packet again and again. commit c93bdd0e03e848555d144eb44a1f275b871a8dd5 Author: Mel Gorman <[email protected]> Date: Tue Jul 31 16:44:19 2012 -0700 netvm: allow skb allocation to use PFMEMALLOC reserves Change the skb allocation API to indicate RX usage and use this to fall back to the PFMEMALLOC reserve when needed. SKBs allocated from the reserve are tagged in skb->pfmemalloc. If an SKB is allocated from the reserve and the socket is later found to be unrelated to page reclaim, the packet is dropped so that the memory remains available for page reclaim. Network protocols are expected to recover from this packet loss. [[email protected]: Ideas taken from various patches] [[email protected]: Use static branches, coding style corrections] [[email protected]: Avoid unnecessary cast, fix !CONFIG_NET build] Signed-off-by: Mel Gorman <[email protected]> Acked-by: David S. Miller <[email protected]> Cc: Neil Brown <[email protected]> Cc: Peter Zijlstra <[email protected]> Cc: Mike Christie <[email protected]> Cc: Eric B Munson <[email protected]> Cc: Eric Dumazet <[email protected]> Cc: Sebastian Andrzej Siewior <[email protected]> Cc: Mel Gorman <[email protected]> Cc: Christoph Lameter <[email protected]> Signed-off-by: Andrew Morton <[email protected]> Signed-off-by: Linus Torvalds <[email protected]> -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
