* [PATCH qemu-kvm] Enable UFO on virtio-net and tap devices
@ 2009-10-07 21:50 Sridhar Samudrala
2009-10-08 10:07 ` Mark McLoughlin
0 siblings, 1 reply; 5+ messages in thread
From: Sridhar Samudrala @ 2009-10-07 21:50 UTC (permalink / raw)
To: avi, aliguori, kvm
linux 2.6.32 includes UDP fragmentation offload support in software.
So we can enable UFO on the host tap device if supported and allow setting
UFO on virtio-net in the guest.
This improves UDP stream performance significantly between guest to host
and inter-guest.
TUN_F_UFO is a new #define added to 2.6.32 kernel header file
include/linux/if_tun.h. Until this updated header file gets into distro
releases, i think we need to have this defined in qemu.
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
diff --git a/hw/virtio-net.c b/hw/virtio-net.c
index ce8e6cb..c73487d 100644
--- a/hw/virtio-net.c
+++ b/hw/virtio-net.c
@@ -150,7 +150,8 @@ static uint32_t virtio_net_get_features(VirtIODevice *vdev)
features |= (1 << VIRTIO_NET_F_HOST_TSO6);
features |= (1 << VIRTIO_NET_F_HOST_ECN);
features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
- /* Kernel can't actually handle UFO in software currently. */
+ features |= (1 << VIRTIO_NET_F_GUEST_UFO);
+ features |= (1 << VIRTIO_NET_F_HOST_UFO);
}
#endif
@@ -189,7 +190,8 @@ static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
(features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
(features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
(features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
- (features >> VIRTIO_NET_F_GUEST_ECN) & 1);
+ (features >> VIRTIO_NET_F_GUEST_ECN) & 1,
+ (features >> VIRTIO_NET_F_GUEST_UFO) & 1);
#endif
}
diff --git a/net.c b/net.c
index 8032ff8..1942e25 100644
--- a/net.c
+++ b/net.c
@@ -1528,8 +1528,13 @@ static int tap_probe_vnet_hdr(int fd)
}
#ifdef TUNSETOFFLOAD
+
+#ifndef TUN_F_UFO
+#define TUN_F_UFO 0x10
+#endif
+
static void tap_set_offload(VLANClientState *vc, int csum, int tso4, int tso6,
- int ecn)
+ int ecn, int ufo)
{
TAPState *s = vc->opaque;
unsigned int offload = 0;
@@ -1542,11 +1547,18 @@ static void tap_set_offload(VLANClientState *vc, int csum, int tso4, int tso6,
offload |= TUN_F_TSO6;
if ((tso4 || tso6) && ecn)
offload |= TUN_F_TSO_ECN;
+ if (ufo)
+ offload |= TUN_F_UFO;
}
- if (ioctl(s->fd, TUNSETOFFLOAD, offload) != 0)
- fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
- strerror(errno));
+ if (ioctl(s->fd, TUNSETOFFLOAD, offload) != 0) {
+ /* Try without UFO */
+ offload &= ~TUN_F_UFO;
+ if (ioctl(s->fd, TUNSETOFFLOAD, offload) != 0) {
+ fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
+ strerror(errno));
+ }
+ }
}
#endif /* TUNSETOFFLOAD */
@@ -1583,7 +1595,7 @@ static TAPState *net_tap_fd_init(VLANState *vlan,
s->vc->receive_raw = tap_receive_raw;
#ifdef TUNSETOFFLOAD
s->vc->set_offload = tap_set_offload;
- tap_set_offload(s->vc, 0, 0, 0, 0);
+ tap_set_offload(s->vc, 0, 0, 0, 0, 0);
#endif
tap_read_poll(s, 1);
snprintf(s->vc->info_str, sizeof(s->vc->info_str), "fd=%d", fd);
diff --git a/net.h b/net.h
index 925c67c..ac3701c 100644
--- a/net.h
+++ b/net.h
@@ -14,7 +14,7 @@ typedef ssize_t (NetReceive)(VLANClientState *, const uint8_t *, size_t);
typedef ssize_t (NetReceiveIOV)(VLANClientState *, const struct iovec *, int);
typedef void (NetCleanup) (VLANClientState *);
typedef void (LinkStatusChanged)(VLANClientState *);
-typedef void (SetOffload)(VLANClientState *, int, int, int, int);
+typedef void (SetOffload)(VLANClientState *, int, int, int, int, int);
struct VLANClientState {
NetReceive *receive;
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH qemu-kvm] Enable UFO on virtio-net and tap devices
2009-10-07 21:50 [PATCH qemu-kvm] Enable UFO on virtio-net and tap devices Sridhar Samudrala
@ 2009-10-08 10:07 ` Mark McLoughlin
2009-10-08 22:31 ` Sridhar Samudrala
0 siblings, 1 reply; 5+ messages in thread
From: Mark McLoughlin @ 2009-10-08 10:07 UTC (permalink / raw)
To: Sridhar Samudrala; +Cc: avi, aliguori, kvm
On Wed, 2009-10-07 at 14:50 -0700, Sridhar Samudrala wrote:
> linux 2.6.32 includes UDP fragmentation offload support in software.
> So we can enable UFO on the host tap device if supported and allow setting
> UFO on virtio-net in the guest.
Hmm, we really need to detect whether the host has tuntap UFO support
before advertising it to the guest. Maybe in net_tap_fd_init() we should
toggle TUN_F_UFO back and forth and check for EINVAL?
Cheers,
Mark.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH qemu-kvm] Enable UFO on virtio-net and tap devices
2009-10-08 10:07 ` Mark McLoughlin
@ 2009-10-08 22:31 ` Sridhar Samudrala
2009-10-09 7:11 ` Mark McLoughlin
0 siblings, 1 reply; 5+ messages in thread
From: Sridhar Samudrala @ 2009-10-08 22:31 UTC (permalink / raw)
To: Mark McLoughlin; +Cc: avi, aliguori, kvm
On Thu, 2009-10-08 at 11:07 +0100, Mark McLoughlin wrote:
> On Wed, 2009-10-07 at 14:50 -0700, Sridhar Samudrala wrote:
> > linux 2.6.32 includes UDP fragmentation offload support in software.
> > So we can enable UFO on the host tap device if supported and allow setting
> > UFO on virtio-net in the guest.
>
> Hmm, we really need to detect whether the host has tuntap UFO support
> before advertising it to the guest. Maybe in net_tap_fd_init() we should
> toggle TUN_F_UFO back and forth and check for EINVAL?
Sure. Here is an updated patch that checks for UFO support in host.
Thanks
Sridhar
Signed-off-by: Sridhar Samudrala <sri@us.ibm.com>
diff --git a/hw/virtio-net.c b/hw/virtio-net.c
index ce8e6cb..9561f34 100644
--- a/hw/virtio-net.c
+++ b/hw/virtio-net.c
@@ -150,7 +150,10 @@ static uint32_t virtio_net_get_features(VirtIODevice *vdev)
features |= (1 << VIRTIO_NET_F_HOST_TSO6);
features |= (1 << VIRTIO_NET_F_HOST_ECN);
features |= (1 << VIRTIO_NET_F_MRG_RXBUF);
- /* Kernel can't actually handle UFO in software currently. */
+ if (tap_has_ufo(host)) {
+ features |= (1 << VIRTIO_NET_F_GUEST_UFO);
+ features |= (1 << VIRTIO_NET_F_HOST_UFO);
+ }
}
#endif
@@ -189,7 +192,8 @@ static void virtio_net_set_features(VirtIODevice *vdev, uint32_t features)
(features >> VIRTIO_NET_F_GUEST_CSUM) & 1,
(features >> VIRTIO_NET_F_GUEST_TSO4) & 1,
(features >> VIRTIO_NET_F_GUEST_TSO6) & 1,
- (features >> VIRTIO_NET_F_GUEST_ECN) & 1);
+ (features >> VIRTIO_NET_F_GUEST_ECN) & 1,
+ (features >> VIRTIO_NET_F_GUEST_UFO) & 1);
#endif
}
diff --git a/net.c b/net.c
index 8032ff8..838e42c 100644
--- a/net.c
+++ b/net.c
@@ -1271,6 +1271,11 @@ void tap_using_vnet_hdr(void *opaque, int using_vnet_hdr)
{
}
+int tap_has_ufo(void *opaque)
+{
+ return 0;
+}
+
#else /* !defined(_WIN32) */
/* Maximum GSO packet size (64k) plus plenty of room for
@@ -1292,6 +1297,7 @@ typedef struct TAPState {
unsigned int write_poll : 1;
unsigned int has_vnet_hdr : 1;
unsigned int using_vnet_hdr : 1;
+ unsigned int has_ufo: 1;
} TAPState;
static int launch_script(const char *setup_script, const char *ifname, int fd);
@@ -1527,9 +1533,22 @@ static int tap_probe_vnet_hdr(int fd)
#endif
}
+int tap_has_ufo(void *opaque)
+{
+ VLANClientState *vc = opaque;
+ TAPState *s = vc->opaque;
+
+ return s ? s->has_ufo : 0;
+}
+
#ifdef TUNSETOFFLOAD
+
+#ifndef TUN_F_UFO
+#define TUN_F_UFO 0x10
+#endif
+
static void tap_set_offload(VLANClientState *vc, int csum, int tso4, int tso6,
- int ecn)
+ int ecn, int ufo)
{
TAPState *s = vc->opaque;
unsigned int offload = 0;
@@ -1542,11 +1561,18 @@ static void tap_set_offload(VLANClientState *vc, int csum, int tso4, int tso6,
offload |= TUN_F_TSO6;
if ((tso4 || tso6) && ecn)
offload |= TUN_F_TSO_ECN;
+ if (ufo)
+ offload |= TUN_F_UFO;
}
- if (ioctl(s->fd, TUNSETOFFLOAD, offload) != 0)
- fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
- strerror(errno));
+ if (ioctl(s->fd, TUNSETOFFLOAD, offload) != 0) {
+ /* Try without UFO */
+ offload &= ~TUN_F_UFO;
+ if (ioctl(s->fd, TUNSETOFFLOAD, offload) != 0) {
+ fprintf(stderr, "TUNSETOFFLOAD ioctl() failed: %s\n",
+ strerror(errno));
+ }
+ }
}
#endif /* TUNSETOFFLOAD */
@@ -1574,6 +1600,7 @@ static TAPState *net_tap_fd_init(VLANState *vlan,
int vnet_hdr)
{
TAPState *s;
+ unsigned int offload;
s = qemu_mallocz(sizeof(TAPState));
s->fd = fd;
@@ -1583,7 +1610,14 @@ static TAPState *net_tap_fd_init(VLANState *vlan,
s->vc->receive_raw = tap_receive_raw;
#ifdef TUNSETOFFLOAD
s->vc->set_offload = tap_set_offload;
- tap_set_offload(s->vc, 0, 0, 0, 0);
+
+ s->has_ufo = 0;
+ /* Check if tap supports UFO */
+ offload = TUN_F_CSUM | TUN_F_UFO;
+ if (ioctl(s->fd, TUNSETOFFLOAD, offload) == 0)
+ s->has_ufo = 1;
+
+ tap_set_offload(s->vc, 0, 0, 0, 0, 0);
#endif
tap_read_poll(s, 1);
snprintf(s->vc->info_str, sizeof(s->vc->info_str), "fd=%d", fd);
diff --git a/net.h b/net.h
index 925c67c..6bb6434 100644
--- a/net.h
+++ b/net.h
@@ -14,7 +14,7 @@ typedef ssize_t (NetReceive)(VLANClientState *, const uint8_t *, size_t);
typedef ssize_t (NetReceiveIOV)(VLANClientState *, const struct iovec *, int);
typedef void (NetCleanup) (VLANClientState *);
typedef void (LinkStatusChanged)(VLANClientState *);
-typedef void (SetOffload)(VLANClientState *, int, int, int, int);
+typedef void (SetOffload)(VLANClientState *, int, int, int, int, int);
struct VLANClientState {
NetReceive *receive;
@@ -92,6 +92,7 @@ void do_info_usernet(Monitor *mon);
int tap_has_vnet_hdr(void *opaque);
void tap_using_vnet_hdr(void *opaque, int using_vnet_hdr);
+int tap_has_ufo(void *opaque);
/* NIC info */
^ permalink raw reply related [flat|nested] 5+ messages in thread
* Re: [PATCH qemu-kvm] Enable UFO on virtio-net and tap devices
2009-10-08 22:31 ` Sridhar Samudrala
@ 2009-10-09 7:11 ` Mark McLoughlin
2009-10-13 15:35 ` Marcelo Tosatti
0 siblings, 1 reply; 5+ messages in thread
From: Mark McLoughlin @ 2009-10-09 7:11 UTC (permalink / raw)
To: Sridhar Samudrala; +Cc: avi, aliguori, kvm
On Thu, 2009-10-08 at 15:31 -0700, Sridhar Samudrala wrote:
> On Thu, 2009-10-08 at 11:07 +0100, Mark McLoughlin wrote:
> > On Wed, 2009-10-07 at 14:50 -0700, Sridhar Samudrala wrote:
> > > linux 2.6.32 includes UDP fragmentation offload support in software.
> > > So we can enable UFO on the host tap device if supported and allow setting
> > > UFO on virtio-net in the guest.
> >
> > Hmm, we really need to detect whether the host has tuntap UFO support
> > before advertising it to the guest. Maybe in net_tap_fd_init() we should
> > toggle TUN_F_UFO back and forth and check for EINVAL?
>
> Sure. Here is an updated patch that checks for UFO support in host.
Looks good to me, thanks
Acked-by: Mark McLoughlin <markmc@redhat.com>
Cheers,
Mark.
^ permalink raw reply [flat|nested] 5+ messages in thread
* Re: [PATCH qemu-kvm] Enable UFO on virtio-net and tap devices
2009-10-09 7:11 ` Mark McLoughlin
@ 2009-10-13 15:35 ` Marcelo Tosatti
0 siblings, 0 replies; 5+ messages in thread
From: Marcelo Tosatti @ 2009-10-13 15:35 UTC (permalink / raw)
To: Mark McLoughlin; +Cc: Sridhar Samudrala, avi, aliguori, kvm
On Fri, Oct 09, 2009 at 08:11:28AM +0100, Mark McLoughlin wrote:
> On Thu, 2009-10-08 at 15:31 -0700, Sridhar Samudrala wrote:
> > On Thu, 2009-10-08 at 11:07 +0100, Mark McLoughlin wrote:
> > > On Wed, 2009-10-07 at 14:50 -0700, Sridhar Samudrala wrote:
> > > > linux 2.6.32 includes UDP fragmentation offload support in software.
> > > > So we can enable UFO on the host tap device if supported and allow setting
> > > > UFO on virtio-net in the guest.
> > >
> > > Hmm, we really need to detect whether the host has tuntap UFO support
> > > before advertising it to the guest. Maybe in net_tap_fd_init() we should
> > > toggle TUN_F_UFO back and forth and check for EINVAL?
> >
> > Sure. Here is an updated patch that checks for UFO support in host.
>
> Looks good to me, thanks
>
> Acked-by: Mark McLoughlin <markmc@redhat.com>
Applied, thanks.
^ permalink raw reply [flat|nested] 5+ messages in thread
end of thread, other threads:[~2009-10-13 16:30 UTC | newest]
Thread overview: 5+ messages (download: mbox.gz follow: Atom feed
-- links below jump to the message on this page --
2009-10-07 21:50 [PATCH qemu-kvm] Enable UFO on virtio-net and tap devices Sridhar Samudrala
2009-10-08 10:07 ` Mark McLoughlin
2009-10-08 22:31 ` Sridhar Samudrala
2009-10-09 7:11 ` Mark McLoughlin
2009-10-13 15:35 ` Marcelo Tosatti
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox;
as well as URLs for NNTP newsgroup(s).