Re: 2.4.20-rc1 - hang with processes stuck in D

[Trond Myklebust]

>>>>> " " == Jakob Oestergaard <[e-mail]> writes:

     > I suspected NFS problems (looks like someone re-wrote NFS
     > between 2.4.18 and 2.4.20-rc1) - but this is *not* the case.
     > The pauses happen on locally running processes as well.

     > It seems to correlate well with a remote host delivering a mail
     > (using maildir over NFS) - but this is not the only situation
     > in which it happens.

     > Everything using disk, both on NFS clients and locally running
     > processes, just pause. Five seconds after everything is like it
     > never happened.

If you are using HIGHMEM, then the stock 2.4.20-rc1 has a known issue
with an unbalanced kmap. Marcelo has already applied the following
patch in the latest bitkeeper update.

Cheers,
  Trond

# This is a BitKeeper generated patch for the following project:
# Project Name: Linux kernel tree
# This patch format is intended for GNU patch command version 2.5 or higher.
# This patch includes the following deltas:
#	           ChangeSet	1.774   -> 1.775  
#	    net/sunrpc/xdr.c	1.7     -> 1.8    
#
# The following is the BitKeeper ChangeSet Log
# --------------------------------------------
# 02/11/06	[e-mail]	1.775
# [PATCH] another kmap imbalance in 2.4.x/2.5.x RPC
# 
# >>>>> Andrew Ryan <[e-mail]> writes:
#      > So far so good on the crashes.  I'm able to get through a
#      > complete run of dbench using TCP mounts on 2.4.20rc1, which I
#      > haven't been able to do before this.
# 
# Marcelo, Linus
# 
#   We've uncovered yet another kmap imbalance in the new RPC code. This
# looks like it might be the last one (my debugging printks have been
# unable to unearth any more). One line fix + 4 line comment
# appended. Please apply to both 2.4.20-rc1 and 2.5.45...
# 
# Cheers,
#   Trond
# --------------------------------------------
#
diff -Nru a/net/sunrpc/xdr.c b/net/sunrpc/xdr.c
--- a/net/sunrpc/xdr.c	Fri Nov  8 19:42:24 2002
+++ b/net/sunrpc/xdr.c	Fri Nov  8 19:42:24 2002
@@ -244,6 +244,11 @@
 		pglen -= base;
 		base  += xdr->page_base;
 		ppage += base >> PAGE_CACHE_SHIFT;
+		/* Note: The offset means that the length of the first
+		 * page is really (PAGE_CACHE_SIZE - (base & ~PAGE_CACHE_MASK)).
+		 * In order to avoid an extra test inside the loop,
+		 * we bump pglen here, and just subtract PAGE_CACHE_SIZE... */
+		pglen += base & ~PAGE_CACHE_MASK;
 	}
 	for (;;) {
 		flush_dcache_page(*ppage);