[PATCH 0/3] kvm-guest-drivers-linux: Fix GSO/partial csum support on older kernels

[PATCH 0/3] kvm-guest-drivers-linux: Fix GSO/partial csum support on older kernels

[Mark McLoughlin]

Hi,
        Here's a few patches to fix virtio_net GSO
and partial csum support under older kernels.

Cheers,
Mark.

[PATCH 1/3] Fix partial csum rx handling

[Mark McLoughlin]

When we receive a packet with a partial csum,
just set ip_summed to CHECKSUM_UNNECESSARY.

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
---
 hack-module.awk |    6 +++++-
 1 files changed, 5 insertions(+), 1 deletions(-)

diff --git a/hack-module.awk b/hack-module.awk
index 4bad1e7..b687951 100644
--- a/hack-module.awk
+++ b/hack-module.awk
@@ -114,7 +114,11 @@
 }
 
 /flags \& VIRTIO_NET_HDR_F_NEEDS_CSUM\)/ {
-    print "#ifndef COMPAT_csum_offset";
+    print "#ifdef COMPAT_csum_offset";
+    print "\tif (hdr->flags & VIRTIO_NET_HDR_F_NEEDS_CSUM) {";
+    print "\t\tskb->ip_summed = CHECKSUM_UNNECESSARY;";
+    print "\t}";
+    print "#else"
     need_endif_indent_brace = 1
 }
 
-- 
1.5.5.1

[PATCH 2/3] Fix partial csum tx handling

[Mark McLoughlin]

Older kernels didn't have the csum_start/csum_offset
skb members, but skb->csum did contain the offset into
the transport header of the csum field and csum->h
contained a pointer to the transport layer header.

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
---
 hack-module.awk |   13 ++++++++++---
 1 files changed, 10 insertions(+), 3 deletions(-)

diff --git a/hack-module.awk b/hack-module.awk
index b687951..d04c3b6 100644
--- a/hack-module.awk
+++ b/hack-module.awk
@@ -124,10 +124,17 @@
 
 /ip_summed == CHECKSUM_PARTIAL\)/ {
     print "#ifdef COMPAT_csum_offset";
-    print "\thdr->flags = 0;"
-    print "\thdr->csum_offset = hdr->csum_start = 0;"
+    print "\tif (skb->ip_summed == CHECKSUM_HW) {"
     print "#else"
-    need_endif_indent_brace = 1;
+    need_endif = 1;
+}
+
+/hdr->csum_start = skb->csum_start/ {
+    print "#ifdef COMPAT_csum_offset";
+    print "\t\thdr->csum_start = skb->h.raw - skb->data;";
+    print "\t\thdr->csum_offset = skb->csum;";
+    print "#else"
+    need_endif = 1;
 }
 
 /skb_transport_header/ {
-- 
1.5.5.1

[PATCH 3/3] Fix GSO tx handling

[Mark McLoughlin]

Use skb->h instead of skb->transport_header

skb->transport_header was introduced in 2.6.22

Signed-off-by: Mark McLoughlin <markmc@redhat.com>
Cc: Herbert Xu <herbert@gondor.apana.org.au>
---
 hack-module.awk |    2 +-
 1 files changed, 1 insertions(+), 1 deletions(-)

diff --git a/hack-module.awk b/hack-module.awk
index d04c3b6..a3275fa 100644
--- a/hack-module.awk
+++ b/hack-module.awk
@@ -139,7 +139,7 @@
 
 /skb_transport_header/ {
     print "#ifdef COMPAT_transport_header";
-    print "BUG();";
+    print "\t\thdr->hdr_len = skb->h.raw - skb->data;";
     print "#else";
     need_endif = 1;
 }
-- 
1.5.5.1

Re: [PATCH 1/3] Fix partial csum rx handling

[Herbert Xu]

On Tue, May 27, 2008 at 12:36:06PM +0100, Mark McLoughlin wrote:
> When we receive a packet with a partial csum,
> just set ip_summed to CHECKSUM_UNNECESSARY.
> 
> Signed-off-by: Mark McLoughlin <markmc@redhat.com>
> Cc: Herbert Xu <herbert@gondor.apana.org.au>

I'm not sure I follow the context of this patch.  Could you
explain what it's for?

Thanks,
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH 1/3] Fix partial csum rx handling

[Mark McLoughlin]

On Thu, 2008-05-29 at 16:50 +1000, Herbert Xu wrote:
> On Tue, May 27, 2008 at 12:36:06PM +0100, Mark McLoughlin wrote:
> > When we receive a packet with a partial csum,
> > just set ip_summed to CHECKSUM_UNNECESSARY.
> > 
> > Signed-off-by: Mark McLoughlin <markmc@redhat.com>
> > Cc: Herbert Xu <herbert@gondor.apana.org.au>
> 
> I'm not sure I follow the context of this patch.  Could you
> explain what it's for?

Yeah, sorry. These three patches are to kvm-guest-driver-linux's awk
script which munges the virtio drivers so that they can be built for
older kernels.

So, the three patches are what I needed to make csum/GSO handling work
correctly on the RHEL5 2.6.18 based kernel.

Cheers,
Mark.

Re: [PATCH 1/3] Fix partial csum rx handling

[Herbert Xu]

Hi Mark:

On Thu, May 29, 2008 at 08:58:23AM +0100, Mark McLoughlin wrote:
> 
> So, the three patches are what I needed to make csum/GSO handling work
> correctly on the RHEL5 2.6.18 based kernel.

Thanks for the explanation.
-- 
Visit Openswan at http://www.openswan.org/
Email: Herbert Xu ~{PmV>HI~} <herbert@gondor.apana.org.au>
Home Page: http://gondor.apana.org.au/~herbert/
PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt

Re: [PATCH 0/3] kvm-guest-drivers-linux: Fix GSO/partial csum support on older kernels

[Avi Kivity]

Mark McLoughlin wrote:
> Hi,
>         Here's a few patches to fix virtio_net GSO
> and partial csum support under older kernels.
>
>   

Applied all, thanks.  Sorry for the virtio-like latency in processing.

I don't much like the intense hackery involved in this.  The way I think 
it could be done is:

- hack virtio to use an API which is specific to kvm, but matches the 
current upstream API:

    s/net_func/virtio_compat_net_func/

  including data structures.

- define this API on top of the host kernel's real API.  For a recent 
enough kernel, that's a one-to-one mapping:

   virtio_compat_net_func() { return net_func(); }

  for older ones there's a more trickery involved.

  For kvm, this is a fairly successful strategy, but I imagine that for 
virtio-net this will be much, much, more difficult.

- write Documentation/stable_api_nonsense_nonsense.txt.


-- 
I have a truly marvellous patch that fixes the bug which this
signature is too narrow to contain.