[dpdk-dev] [PATCH 0/2] virtio fix false offload claims

[dpdk-dev] [PATCH 0/2] virtio fix false offload claims

[Stephen Hemminger]

While doing code for Hyper-V, noticed that the virtio driver was
confused about receive versus transmit offloads.  The virtio
checksum offload is L4 (TCP/UDP) only, not IPv4. Also, TSO
and LRO are not the same.

This may break some program that was assuming it was getting offloads
that it wasn't.

Stephen Hemminger (2):
  virtio: don't falsely claim to do IP checksum
  virtio: don't claim to support LRO

 drivers/net/virtio/virtio_ethdev.c | 30 +++++-------------------------
 1 file changed, 5 insertions(+), 25 deletions(-)

-- 
2.11.0

[dpdk-dev] [PATCH 1/2] virtio: don't falsely claim to do IP checksum

[Stephen Hemminger]

The virtio driver is confused about the meaning of the ip_checksum
flag.  In DPDK, ip_checksum means the hardware is capable of checking
the Layer 3 IP checksum.  But KVM/QEMU does not do that. The flag
VIRTIO_NET_F_GUEST_CSUM controls whether the receive side does
Layer 4 (TCP/UDP) checksum offload.

Fix by erroring out any requests to do IP checksum.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
---
 drivers/net/virtio/virtio_ethdev.c | 16 +++++++---------
 1 file changed, 7 insertions(+), 9 deletions(-)

diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
index 5c826f47740c..ca607904fa37 100644
--- a/drivers/net/virtio/virtio_ethdev.c
+++ b/drivers/net/virtio/virtio_ethdev.c
@@ -1664,8 +1664,13 @@ virtio_dev_configure(struct rte_eth_dev *dev)
 
 	PMD_INIT_LOG(DEBUG, "configure");
 	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
-	if (rxmode->hw_ip_checksum)
-		req_features |= (1ULL << VIRTIO_NET_F_GUEST_CSUM);
+
+	/* Virtio does L4 checksum but not L3! */
+	if (rxmode->hw_ip_checksum) {
+		PMD_DRV_LOG(NOTICE,
+			    "virtio does not support IP checksum");
+		return -ENOTSUP;
+	}
 	if (rxmode->enable_lro)
 		req_features |=
 			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
@@ -1678,13 +1683,6 @@ virtio_dev_configure(struct rte_eth_dev *dev)
 			return ret;
 	}
 
-	if (rxmode->hw_ip_checksum &&
-		!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_CSUM)) {
-		PMD_DRV_LOG(NOTICE,
-			"rx ip checksum not available on this host");
-		return -ENOTSUP;
-	}
-
 	if (rxmode->enable_lro &&
 		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
 			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
-- 
2.11.0

[dpdk-dev] [PATCH 2/2] virtio: don't claim to support LRO

[Stephen Hemminger]

The current virtio supports Transmit Segmentation Offload, but
does not really support Large Receive Offload. The driver was confusing
the two offloads.

Signed-off-by: Stephen Hemminger <stephen@networkplumber.org>
---
 drivers/net/virtio/virtio_ethdev.c | 22 ++--------------------
 1 file changed, 2 insertions(+), 20 deletions(-)

diff --git a/drivers/net/virtio/virtio_ethdev.c b/drivers/net/virtio/virtio_ethdev.c
index ca607904fa37..00a3122780ba 100644
--- a/drivers/net/virtio/virtio_ethdev.c
+++ b/drivers/net/virtio/virtio_ethdev.c
@@ -1659,11 +1659,8 @@ virtio_dev_configure(struct rte_eth_dev *dev)
 {
 	const struct rte_eth_rxmode *rxmode = &dev->data->dev_conf.rxmode;
 	struct virtio_hw *hw = dev->data->dev_private;
-	uint64_t req_features;
-	int ret;
 
 	PMD_INIT_LOG(DEBUG, "configure");
-	req_features = VIRTIO_PMD_DEFAULT_GUEST_FEATURES;
 
 	/* Virtio does L4 checksum but not L3! */
 	if (rxmode->hw_ip_checksum) {
@@ -1671,23 +1668,10 @@ virtio_dev_configure(struct rte_eth_dev *dev)
 			    "virtio does not support IP checksum");
 		return -ENOTSUP;
 	}
-	if (rxmode->enable_lro)
-		req_features |=
-			(1ULL << VIRTIO_NET_F_GUEST_TSO4) |
-			(1ULL << VIRTIO_NET_F_GUEST_TSO6);
-
-	/* if request features changed, reinit the device */
-	if (req_features != hw->req_guest_features) {
-		ret = virtio_init_device(dev, req_features);
-		if (ret < 0)
-			return ret;
-	}
 
-	if (rxmode->enable_lro &&
-		(!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4) ||
-			!vtpci_with_feature(hw, VIRTIO_NET_F_GUEST_TSO4))) {
+	if (rxmode->enable_lro) {
 		PMD_DRV_LOG(NOTICE,
-			"lro not available on this host");
+			    "virtio does not support Large Receive Offload");
 		return -ENOTSUP;
 	}
 
@@ -1913,8 +1897,6 @@ virtio_dev_info_get(struct rte_eth_dev *dev, struct rte_eth_dev_info *dev_info)
 	}
 	tso_mask = (1ULL << VIRTIO_NET_F_GUEST_TSO4) |
 		(1ULL << VIRTIO_NET_F_GUEST_TSO6);
-	if ((host_features & tso_mask) == tso_mask)
-		dev_info->rx_offload_capa |= DEV_RX_OFFLOAD_TCP_LRO;
 
 	dev_info->tx_offload_capa = 0;
 	if (hw->guest_features & (1ULL << VIRTIO_NET_F_CSUM)) {
-- 
2.11.0

Re: [dpdk-dev] [PATCH 0/2] virtio fix false offload claims

[Yuanhan Liu]

On Fri, Jul 07, 2017 at 12:52:48PM -0700, Stephen Hemminger wrote:
> While doing code for Hyper-V, noticed that the virtio driver was
> confused about receive versus transmit offloads.  The virtio
> checksum offload is L4 (TCP/UDP) only, not IPv4. Also, TSO
> and LRO are not the same.
> 
> This may break some program that was assuming it was getting offloads
> that it wasn't.

Applied to dpdk-next-virtio.

And I think they should be backported to stable releases, thus,

    Cc: stable@dpdk.org

Thanks.

	--yliu
> 
> Stephen Hemminger (2):
>   virtio: don't falsely claim to do IP checksum
>   virtio: don't claim to support LRO
> 
>  drivers/net/virtio/virtio_ethdev.c | 30 +++++-------------------------
>  1 file changed, 5 insertions(+), 25 deletions(-)
> 
> -- 
> 2.11.0

Re: [dpdk-dev] [PATCH 0/2] virtio fix false offload claims

[Olivier MATZ]

Hello,

On Sat, Jul 08, 2017 at 11:12:22AM +0800, Yuanhan Liu wrote:
> On Fri, Jul 07, 2017 at 12:52:48PM -0700, Stephen Hemminger wrote:
> > While doing code for Hyper-V, noticed that the virtio driver was
> > confused about receive versus transmit offloads.  The virtio
> > checksum offload is L4 (TCP/UDP) only, not IPv4. Also, TSO
> > and LRO are not the same.
> > 
> > This may break some program that was assuming it was getting offloads
> > that it wasn't.
> 
> Applied to dpdk-next-virtio.
> 
> And I think they should be backported to stable releases, thus,
> 
>     Cc: stable@dpdk.org
> 
> Thanks.
> 
> 	--yliu
> > 
> > Stephen Hemminger (2):
> >   virtio: don't falsely claim to do IP checksum
> >   virtio: don't claim to support LRO
> > 
> >  drivers/net/virtio/virtio_ethdev.c | 30 +++++-------------------------
> >  1 file changed, 5 insertions(+), 25 deletions(-)
> > 
> > -- 
> > 2.11.0

I think these 2 commits break the virtio offload, which can be tested as
described in this test plan:
http://dpdk.org/ml/archives/dev/2016-October/048092.html

First, about checksum: the description of rxmode->hw_ip_checksum is:

     hw_ip_checksum   : 1, /**< IP/UDP/TCP checksum offload enable. */

So, while I agree the name is not well chosen, it is valid to set it
for virtio to enable L4 checksum.

Then about LRO: setting rxmode->enable_lro is a way to tell the host that the
guest is ok to receive tso packets. From the guest point of view, it is like
enabling lro on a physical driver. Again, it is valid and useful to do this.

Before removing these features, it would have been nice to have a quick look at
the commits that introduced them.

Olivier

Re: [dpdk-dev] [PATCH 0/2] virtio fix false offload claims

[Stephen Hemminger]

On Wed, 23 Aug 2017 11:30:26 +0200
Olivier MATZ <olivier.matz@6wind.com> wrote:

> Hello,
> 
> On Sat, Jul 08, 2017 at 11:12:22AM +0800, Yuanhan Liu wrote:
> > On Fri, Jul 07, 2017 at 12:52:48PM -0700, Stephen Hemminger wrote:  
> > > While doing code for Hyper-V, noticed that the virtio driver was
> > > confused about receive versus transmit offloads.  The virtio
> > > checksum offload is L4 (TCP/UDP) only, not IPv4. Also, TSO
> > > and LRO are not the same.
> > > 
> > > This may break some program that was assuming it was getting offloads
> > > that it wasn't.  
> > 
> > Applied to dpdk-next-virtio.
> > 
> > And I think they should be backported to stable releases, thus,
> > 
> >     Cc: stable@dpdk.org
> > 
> > Thanks.
> > 
> > 	--yliu  
> > > 
> > > Stephen Hemminger (2):
> > >   virtio: don't falsely claim to do IP checksum
> > >   virtio: don't claim to support LRO
> > > 
> > >  drivers/net/virtio/virtio_ethdev.c | 30 +++++-------------------------
> > >  1 file changed, 5 insertions(+), 25 deletions(-)
> > > 
> > > -- 
> > > 2.11.0  
> 
> I think these 2 commits break the virtio offload, which can be tested as
> described in this test plan:
> http://dpdk.org/ml/archives/dev/2016-October/048092.html
> 
> First, about checksum: the description of rxmode->hw_ip_checksum is:
> 
>      hw_ip_checksum   : 1, /**< IP/UDP/TCP checksum offload enable. */
> 
> So, while I agree the name is not well chosen, it is valid to set it
> for virtio to enable L4 checksum.
> 
> Then about LRO: setting rxmode->enable_lro is a way to tell the host that the
> guest is ok to receive tso packets. From the guest point of view, it is like
> enabling lro on a physical driver. Again, it is valid and useful to do this.
> 
> Before removing these features, it would have been nice to have a quick look at
> the commits that introduced them.

I am ok with keeping LRO as long as the documentation changed.  And virtio
driver did some enforcement. 

For checksums, the hw_ip_checksum flag either needs to be more fine grain (IP, UDP, TCP)
which would be best, or virtio would have to check IP checksum in software.

Re: [dpdk-dev] [PATCH 0/2] virtio fix false offload claims

[Olivier MATZ]

On Wed, Aug 23, 2017 at 08:31:35AM -0700, Stephen Hemminger wrote:
> On Wed, 23 Aug 2017 11:30:26 +0200
> Olivier MATZ <olivier.matz@6wind.com> wrote:
> 
> > Hello,
> > 
> > On Sat, Jul 08, 2017 at 11:12:22AM +0800, Yuanhan Liu wrote:
> > > On Fri, Jul 07, 2017 at 12:52:48PM -0700, Stephen Hemminger wrote:  
> > > > While doing code for Hyper-V, noticed that the virtio driver was
> > > > confused about receive versus transmit offloads.  The virtio
> > > > checksum offload is L4 (TCP/UDP) only, not IPv4. Also, TSO
> > > > and LRO are not the same.
> > > > 
> > > > This may break some program that was assuming it was getting offloads
> > > > that it wasn't.  
> > > 
> > > Applied to dpdk-next-virtio.
> > > 
> > > And I think they should be backported to stable releases, thus,
> > > 
> > >     Cc: stable@dpdk.org
> > > 
> > > Thanks.
> > > 
> > > 	--yliu  
> > > > 
> > > > Stephen Hemminger (2):
> > > >   virtio: don't falsely claim to do IP checksum
> > > >   virtio: don't claim to support LRO
> > > > 
> > > >  drivers/net/virtio/virtio_ethdev.c | 30 +++++-------------------------
> > > >  1 file changed, 5 insertions(+), 25 deletions(-)
> > > > 
> > > > -- 
> > > > 2.11.0  
> > 
> > I think these 2 commits break the virtio offload, which can be tested as
> > described in this test plan:
> > http://dpdk.org/ml/archives/dev/2016-October/048092.html
> > 
> > First, about checksum: the description of rxmode->hw_ip_checksum is:
> > 
> >      hw_ip_checksum   : 1, /**< IP/UDP/TCP checksum offload enable. */
> > 
> > So, while I agree the name is not well chosen, it is valid to set it
> > for virtio to enable L4 checksum.
> > 
> > Then about LRO: setting rxmode->enable_lro is a way to tell the host that the
> > guest is ok to receive tso packets. From the guest point of view, it is like
> > enabling lro on a physical driver. Again, it is valid and useful to do this.
> > 
> > Before removing these features, it would have been nice to have a quick look at
> > the commits that introduced them.
> 
> I am ok with keeping LRO as long as the documentation changed.  And virtio
> driver did some enforcement. 
> 
> For checksums, the hw_ip_checksum flag either needs to be more fine grain (IP, UDP, TCP)
> which would be best, or virtio would have to check IP checksum in software.
> 

For checksum, yes, the rxconf should be more fine-grained and renamed.

But apart from the name which is confusing, it was not wrong.
Setting hw_ip_checksum=1 means: "allow the driver to return packets with
checksums flags != unknown". These flags are good,bad,unknown,none for
both l3 and l4. So virtio driver always return unknown for l3, and
none|unknown|good|bad for l4, depending on what the host passed.

Re: [dpdk-dev] [PATCH 0/2] virtio fix false offload claims

[Stephen Hemminger]

On Wed, 23 Aug 2017 18:14:44 +0200
Olivier MATZ <olivier.matz@6wind.com> wrote:

> On Wed, Aug 23, 2017 at 08:31:35AM -0700, Stephen Hemminger wrote:
> > On Wed, 23 Aug 2017 11:30:26 +0200
> > Olivier MATZ <olivier.matz@6wind.com> wrote:
> >   
> > > Hello,
> > > 
> > > On Sat, Jul 08, 2017 at 11:12:22AM +0800, Yuanhan Liu wrote:  
> > > > On Fri, Jul 07, 2017 at 12:52:48PM -0700, Stephen Hemminger wrote:    
> > > > > While doing code for Hyper-V, noticed that the virtio driver was
> > > > > confused about receive versus transmit offloads.  The virtio
> > > > > checksum offload is L4 (TCP/UDP) only, not IPv4. Also, TSO
> > > > > and LRO are not the same.
> > > > > 
> > > > > This may break some program that was assuming it was getting offloads
> > > > > that it wasn't.    
> > > > 
> > > > Applied to dpdk-next-virtio.
> > > > 
> > > > And I think they should be backported to stable releases, thus,
> > > > 
> > > >     Cc: stable@dpdk.org
> > > > 
> > > > Thanks.
> > > > 
> > > > 	--yliu    
> > > > > 
> > > > > Stephen Hemminger (2):
> > > > >   virtio: don't falsely claim to do IP checksum
> > > > >   virtio: don't claim to support LRO
> > > > > 
> > > > >  drivers/net/virtio/virtio_ethdev.c | 30 +++++-------------------------
> > > > >  1 file changed, 5 insertions(+), 25 deletions(-)
> > > > > 
> > > > > -- 
> > > > > 2.11.0    
> > > 
> > > I think these 2 commits break the virtio offload, which can be tested as
> > > described in this test plan:
> > > http://dpdk.org/ml/archives/dev/2016-October/048092.html
> > > 
> > > First, about checksum: the description of rxmode->hw_ip_checksum is:
> > > 
> > >      hw_ip_checksum   : 1, /**< IP/UDP/TCP checksum offload enable. */
> > > 
> > > So, while I agree the name is not well chosen, it is valid to set it
> > > for virtio to enable L4 checksum.
> > > 
> > > Then about LRO: setting rxmode->enable_lro is a way to tell the host that the
> > > guest is ok to receive tso packets. From the guest point of view, it is like
> > > enabling lro on a physical driver. Again, it is valid and useful to do this.
> > > 
> > > Before removing these features, it would have been nice to have a quick look at
> > > the commits that introduced them.  
> > 
> > I am ok with keeping LRO as long as the documentation changed.  And virtio
> > driver did some enforcement. 
> > 
> > For checksums, the hw_ip_checksum flag either needs to be more fine grain (IP, UDP, TCP)
> > which would be best, or virtio would have to check IP checksum in software.
> >   
> 
> For checksum, yes, the rxconf should be more fine-grained and renamed.
> 
> But apart from the name which is confusing, it was not wrong.
> Setting hw_ip_checksum=1 means: "allow the driver to return packets with
> checksums flags != unknown". These flags are good,bad,unknown,none for
> both l3 and l4. So virtio driver always return unknown for l3, and
> none|unknown|good|bad for l4, depending on what the host passed.

You are right, I forgot that there now is way to indicate unknown
for checksums. Before that it was limited good/bad.

Let's revert both these patches, and update the documentation?