[PATCH] make MAX_SKBUFF_ORDER page size independent

[PATCH] make MAX_SKBUFF_ORDER page size independent

[Isaku Yamahata]

[-- Attachment #1: Type: text/plain, Size: 228 bytes --]


make MAX_SKBUFF_ORDER page size independent.
Presumably the original value 4 comes from 64KB = (4KB << 4) which is
ip packet maximum size.
However page size is not always 4KB. e.g. IA64 default page size is 16KB.

-- 
yamahata

[-- Attachment #2: 11771:93590d295f38_max_skbuff_order.patch --]
[-- Type: text/plain, Size: 976 bytes --]

# HG changeset patch
# User yamahata@valinux.co.jp
# Date 1159851774 -32400
# Node ID 93590d295f381e50cd95a161379da3ec9129fdfe
# Parent  121ea4bab59bed2bbcb6fee321d49d138954fcfd
make MAX_SKBUFF_ORDER page size independent.
Presumably the original value 4 comes from 64KB = (4KB << 4) which is
ip packet maximum size.
However page size is not always 4KB. e.g. IA64 default page size is 16KB.
PATCHNAME: max_skbuff_order

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>

diff -r 121ea4bab59b -r 93590d295f38 linux-2.6-xen-sparse/drivers/xen/core/skbuff.c
--- a/linux-2.6-xen-sparse/drivers/xen/core/skbuff.c	Tue Oct 03 13:46:08 2006 +0900
+++ b/linux-2.6-xen-sparse/drivers/xen/core/skbuff.c	Tue Oct 03 14:02:54 2006 +0900
@@ -18,7 +18,7 @@
 /*static*/ kmem_cache_t *skbuff_cachep;
 EXPORT_SYMBOL(skbuff_cachep);
 
-#define MAX_SKBUFF_ORDER 4
+#define MAX_SKBUFF_ORDER (16 - PAGE_SHIFT)
 static kmem_cache_t *skbuff_order_cachep[MAX_SKBUFF_ORDER + 1];
 
 static struct {

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: [Xen-ia64-devel] [PATCH] make MAX_SKBUFF_ORDER page size independent

[Tristan Gingold]

Le Mardi 03 Octobre 2006 07:06, Isaku Yamahata a écrit :
> make MAX_SKBUFF_ORDER page size independent.
> Presumably the original value 4 comes from 64KB = (4KB << 4) which is
> ip packet maximum size.
> However page size is not always 4KB. e.g. IA64 default page size is 16KB.
I suppose it fixes the fedora bug ?

I wonder if this patch works when pagesize is 64KB ?  Maybe force 1 as the 
minimum value ?

Tristan.

Re: Re: [Xen-ia64-devel] [PATCH] make MAX_SKBUFF_ORDER page size independent

[Keir Fraser]

On 3/10/06 7:43 am, "Tristan Gingold" <Tristan.Gingold@bull.net> wrote:

> Le Mardi 03 Octobre 2006 07:06, Isaku Yamahata a écrit :
>> make MAX_SKBUFF_ORDER page size independent.
>> Presumably the original value 4 comes from 64KB = (4KB << 4) which is
>> ip packet maximum size.
>> However page size is not always 4KB. e.g. IA64 default page size is 16KB.
> I suppose it fixes the fedora bug ?
> 
> I wonder if this patch works when pagesize is 64KB ?  Maybe force 1 as the
> minimum value ?

We want to move to the copying receive path as default asap by the way, you
might want to check that it works with ia64 (boot a domU with extra Linux
cmdline option "xennet.rx_copy" and check you get a boot message saying that
a net device eth0 has been created with copying receive path).

Also since our network interfaces now support scatter-gather (thanks to
Herbert Xu) I expect to get rid of our special skbuff allocators, at least
for oridnary alloc_skb and dev_alloc_skb calls. This should happen early in
the 3.0.4 development cycle.

 -- Keir

Re: Re: [Xen-ia64-devel] [PATCH] make MAX_SKBUFF_ORDER page size independent

[Tristan Gingold]

Le Mardi 03 Octobre 2006 08:59, Keir Fraser a écrit :
> On 3/10/06 7:43 am, "Tristan Gingold" <Tristan.Gingold@bull.net> wrote:
> > Le Mardi 03 Octobre 2006 07:06, Isaku Yamahata a écrit :
> >> make MAX_SKBUFF_ORDER page size independent.
> >> Presumably the original value 4 comes from 64KB = (4KB << 4) which is
> >> ip packet maximum size.
> >> However page size is not always 4KB. e.g. IA64 default page size is
> >> 16KB.
> >
> > I suppose it fixes the fedora bug ?
> >
> > I wonder if this patch works when pagesize is 64KB ?  Maybe force 1 as
> > the minimum value ?
>
> We want to move to the copying receive path as default asap by the way, you
> might want to check that it works with ia64 (boot a domU with extra Linux
> cmdline option "xennet.rx_copy" and check you get a boot message saying
> that a net device eth0 has been created with copying receive path).
This option has net yet been merged in xen-ia64-unstable.hg; I will try ASAP.

> Also since our network interfaces now support scatter-gather (thanks to
> Herbert Xu) I expect to get rid of our special skbuff allocators, at least
> for oridnary alloc_skb and dev_alloc_skb calls. This should happen early in
> the 3.0.4 development cycle.
Good news.

Tristan.

Re: Re: [Xen-ia64-devel] [PATCH] make MAX_SKBUFF_ORDER page size independent

[Isaku Yamahata]

[-- Attachment #1: Type: text/plain, Size: 531 bytes --]


On Tue, Oct 03, 2006 at 08:43:34AM +0200, Tristan Gingold wrote:
> Le Mardi 03 Octobre 2006 07:06, Isaku Yamahata a écrit :
> > make MAX_SKBUFF_ORDER page size independent.
> > Presumably the original value 4 comes from 64KB = (4KB << 4) which is
> > ip packet maximum size.
> > However page size is not always 4KB. e.g. IA64 default page size is 16KB.
> I suppose it fixes the fedora bug ?
> 
> I wonder if this patch works when pagesize is 64KB ?  Maybe force 1 as the 
> minimum value ?

fixed.

-- 
yamahata

[-- Attachment #2: 11713:8ba0111bb95f_max_skbuff_order.patch --]
[-- Type: text/plain, Size: 1041 bytes --]

# HG changeset patch
# User yamahata@valinux.co.jp
# Date 1159867300 -32400
# Node ID 8ba0111bb95fbd2193a5b4e59fad6cde91ceda3c
# Parent  6e7cc23ab18c23fdb2a8b4d0a7778c4144851818
make MAX_SKBUFF_ORDER page size independent.
Presumably the original value 4 comes from 64KB = (4KB << 4) which is
ip packet maximum size.
However page size is not always 4KB. e.g. IA64 default page size is 16KB.
PATCHNAME: max_skbuff_order

Signed-off-by: Isaku Yamahata <yamahata@valinux.co.jp>

diff -r 6e7cc23ab18c -r 8ba0111bb95f linux-2.6-xen-sparse/drivers/xen/core/skbuff.c
--- a/linux-2.6-xen-sparse/drivers/xen/core/skbuff.c	Mon Oct 02 21:53:07 2006 -0600
+++ b/linux-2.6-xen-sparse/drivers/xen/core/skbuff.c	Tue Oct 03 18:21:40 2006 +0900
@@ -18,7 +18,11 @@
 /*static*/ kmem_cache_t *skbuff_cachep;
 EXPORT_SYMBOL(skbuff_cachep);
 
-#define MAX_SKBUFF_ORDER 4
+#if PAGE_SHIFT < 16
+#define MAX_SKBUFF_ORDER (16 - PAGE_SHIFT)
+#else
+#define MAX_SKBUFF_ORDER 1
+#endif
 static kmem_cache_t *skbuff_order_cachep[MAX_SKBUFF_ORDER + 1];
 
 static struct {

[-- Attachment #3: Type: text/plain, Size: 138 bytes --]

_______________________________________________
Xen-devel mailing list
Xen-devel@lists.xensource.com
http://lists.xensource.com/xen-devel

Re: Re: [Xen-ia64-devel] [PATCH] make MAX_SKBUFF_ORDER page size independent

[Alex Williamson]

On Tue, 2006-10-03 at 10:28 +0200, Tristan Gingold wrote:
> Le Mardi 03 Octobre 2006 08:59, Keir Fraser a écrit :
> > We want to move to the copying receive path as default asap by the way, you
> > might want to check that it works with ia64 (boot a domU with extra Linux
> > cmdline option "xennet.rx_copy" and check you get a boot message saying
> > that a net device eth0 has been created with copying receive path).
> This option has net yet been merged in xen-ia64-unstable.hg; I will try ASAP.

   Seems to work for me on xen-unstable.hg.  I merged this code into
xen-ia64-unstable.hg and had to make one little tweak to get it going
there, but it looks good now.  Thanks,

	Alex

-- 
Alex Williamson                             HP Open Source & Linux Org.