Hi,
The recent (changeset 7700) switch to use the architecture
specific skbuff allocation routine breaks with very large MTU sizes.
The below patch raises the MAX_SKBUFF_ORDER to 3. Could this be
applied to make it the default?
Thanks,
Phil
Signed-Off-by: Philip Auld <pauld@egenera.com>
diff -r 136b2d20dc81 linux-2.6-xen-sparse/arch/xen/kernel/skbuff.c
--- a/linux-2.6-xen-sparse/arch/xen/kernel/skbuff.c Wed Nov 9 15:08:37 2005
+++ b/linux-2.6-xen-sparse/arch/xen/kernel/skbuff.c Thu Nov 10 09:40:26 2005
@@ -17,7 +17,7 @@
/* Referenced in netback.c. */
/*static*/ kmem_cache_t *skbuff_cachep;
-#define MAX_SKBUFF_ORDER 2
+#define MAX_SKBUFF_ORDER 3
static kmem_cache_t *skbuff_order_cachep[MAX_SKBUFF_ORDER + 1];
static struct {
--
Philip R. Auld, Ph.D. Egenera, Inc.
Software Architect 165 Forest St.
(508) 858-2628 Marlboro, MA 01752
_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel
> The recent (changeset 7700) switch to use the > architecture specific skbuff allocation routine breaks with > very large MTU sizes. > The below patch raises the MAX_SKBUFF_ORDER to 3. Could this > be applied to make it the default?I suspect the only reason things worked for you before was that you were relying on swiotlb. (enabled if you have >2GB). I need to think through the implactions of changing MAX_SKBUFF_ORDER... Ian _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel
Hi Ian, Rumor has it that on Thu, Nov 10, 2005 at 09:08:44PM -0000 Ian Pratt said:> > > The recent (changeset 7700) switch to use the > > architecture specific skbuff allocation routine breaks with > > very large MTU sizes. > > The below patch raises the MAX_SKBUFF_ORDER to 3. Could this > > be applied to make it the default? > > I suspect the only reason things worked for you before was that you were > relying on swiotlb. (enabled if you have >2GB). I need to think through > the implactions of changing MAX_SKBUFF_ORDER...Can you explain the swiotlb part? The large MTUs are only on dom0. It worked before because alloc_skb used kmalloc and gave me pseudo-physical contigous regions, which is sufficient. Packets coming from and going to domUs would all be within the first page. But since dom0 is receiving packets from rdma I have to post packets large enough to cover anything that could be sent to dom0. And for compatibility within the system that''s about 17k, unfortunately. Cheers, Phil> > Ian-- Philip R. Auld, Ph.D. Egenera, Inc. Software Architect 165 Forest St. (508) 858-2628 Marlboro, MA 01752 _______________________________________________ Xen-devel mailing list Xen-devel@lists.xensource.com http://lists.xensource.com/xen-devel