| Message ID | 1524241477-8456-1-git-send-email-pablo.cascon@netronome.com |
|---|---|
| State | Accepted |
| Delegated to: | Ian Stokes |
| Headers | show |
| Series | [ovs-dev,v3,1/1] netdev-dpdk: remove enabling scatter for jumbo RX support | expand |
On Fri, Apr 20, 2018 at 05:24:37PM +0100, Pablo Cascón wrote: > Currently to RX jumbo packets fails for NICs not supporting scatter. > Scatter is not strictly needed for jumbo RX support. This change fixes > the issue by only enabling scatter for NICs known to need it to > support jumbo RX. Add a quirk for "igb" while the PMD is fixed. > > Reported-by: Louis Peens <louis.peens@netronome.com> > Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> > Reviewed-by: Simon Horman <simon.horman@netronome.com> Hi Ian, this is a pretty important fix for us. I would be most grateful if you could find some time to review it. > --- > > Changelog: > v3->v2: > - only check for driver_name > - tested with "nfp" and not with "igb" > > > lib/netdev-dpdk.c | 12 +++++++++--- > 1 file changed, 9 insertions(+), 3 deletions(-) > > diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c > index ee39cbe..02ed85b 100644 > --- a/lib/netdev-dpdk.c > +++ b/lib/netdev-dpdk.c > @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) > int diag = 0; > int i; > struct rte_eth_conf conf = port_conf; > + struct rte_eth_dev_info info; > > - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly > - * enabled. */ > + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly > + * enabling scatter to support jumbo RX. Note: PMDs are not > + * required to set the offload capabilities and so is not reliable > + * info, only the driver_name is after testing the PMD/NIC */ > if (dev->mtu > ETHER_MTU) { > - conf.rxmode.enable_scatter = 1; > + rte_eth_dev_info_get(dev->port_id, &info); > + if (!strcmp(info.driver_name, "igb")) { > + conf.rxmode.enable_scatter = 1; > + } > } > > conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & > -- > 2.7.4 > > _______________________________________________ > dev mailing list > dev@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-dev
> On Fri, Apr 20, 2018 at 05:24:37PM +0100, Pablo Cascón wrote: > > Currently to RX jumbo packets fails for NICs not supporting scatter. > > Scatter is not strictly needed for jumbo RX support. This change fixes > > the issue by only enabling scatter for NICs known to need it to > > support jumbo RX. Add a quirk for "igb" while the PMD is fixed. > > > > Reported-by: Louis Peens <louis.peens@netronome.com> > > Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> > > Reviewed-by: Simon Horman <simon.horman@netronome.com> > > Hi Ian, > > this is a pretty important fix for us. > I would be most grateful if you could find some time to review it. Hi Simon, I started looking at it yesterday and finishing some testing specific to IGB today, will follow up once I have completed it. Thanks Ian > > > --- > > > > Changelog: > > v3->v2: > > - only check for driver_name > > - tested with "nfp" and not with "igb" > > > > > > lib/netdev-dpdk.c | 12 +++++++++--- > > 1 file changed, 9 insertions(+), 3 deletions(-) > > > > diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index > > ee39cbe..02ed85b 100644 > > --- a/lib/netdev-dpdk.c > > +++ b/lib/netdev-dpdk.c > > @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, > int n_rxq, int n_txq) > > int diag = 0; > > int i; > > struct rte_eth_conf conf = port_conf; > > + struct rte_eth_dev_info info; > > > > - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be > explicitly > > - * enabled. */ > > + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly > > + * enabling scatter to support jumbo RX. Note: PMDs are not > > + * required to set the offload capabilities and so is not reliable > > + * info, only the driver_name is after testing the PMD/NIC */ > > if (dev->mtu > ETHER_MTU) { > > - conf.rxmode.enable_scatter = 1; > > + rte_eth_dev_info_get(dev->port_id, &info); > > + if (!strcmp(info.driver_name, "igb")) { > > + conf.rxmode.enable_scatter = 1; > > + } > > } > > > > conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & > > -- > > 2.7.4 > > > > _______________________________________________ > > dev mailing list > > dev@openvswitch.org > > https://mail.openvswitch.org/mailman/listinfo/ovs-dev
On 26 April 2018 at 12:44, Stokes, Ian <ian.stokes@intel.com> wrote: > > On Fri, Apr 20, 2018 at 05:24:37PM +0100, Pablo Cascón wrote: > > > Currently to RX jumbo packets fails for NICs not supporting scatter. > > > Scatter is not strictly needed for jumbo RX support. This change fixes > > > the issue by only enabling scatter for NICs known to need it to > > > support jumbo RX. Add a quirk for "igb" while the PMD is fixed. > > > > > > Reported-by: Louis Peens <louis.peens@netronome.com> > > > Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> > > > Reviewed-by: Simon Horman <simon.horman@netronome.com> > > > > Hi Ian, > > > > this is a pretty important fix for us. > > I would be most grateful if you could find some time to review it. > > Hi Simon, > > I started looking at it yesterday and finishing some testing specific to > IGB today, will follow up once I have completed it. > Great, thanks Ian. > Thanks > Ian > > > > > > --- > > > > > > Changelog: > > > v3->v2: > > > - only check for driver_name > > > - tested with "nfp" and not with "igb" > > > > > > > > > lib/netdev-dpdk.c | 12 +++++++++--- > > > 1 file changed, 9 insertions(+), 3 deletions(-) > > > > > > diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index > > > ee39cbe..02ed85b 100644 > > > --- a/lib/netdev-dpdk.c > > > +++ b/lib/netdev-dpdk.c > > > @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk > *dev, > > int n_rxq, int n_txq) > > > int diag = 0; > > > int i; > > > struct rte_eth_conf conf = port_conf; > > > + struct rte_eth_dev_info info; > > > > > > - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be > > explicitly > > > - * enabled. */ > > > + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly > > > + * enabling scatter to support jumbo RX. Note: PMDs are not > > > + * required to set the offload capabilities and so is not reliable > > > + * info, only the driver_name is after testing the PMD/NIC */ > > > if (dev->mtu > ETHER_MTU) { > > > - conf.rxmode.enable_scatter = 1; > > > + rte_eth_dev_info_get(dev->port_id, &info); > > > + if (!strcmp(info.driver_name, "igb")) { > > > + conf.rxmode.enable_scatter = 1; > > > + } > > > } > > > > > > conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & > > > -- > > > 2.7.4 > > > > > > _______________________________________________ > > > dev mailing list > > > dev@openvswitch.org > > > https://mail.openvswitch.org/mailman/listinfo/ovs-dev >
> Currently to RX jumbo packets fails for NICs not supporting scatter. > Scatter is not strictly needed for jumbo RX support. This change fixes the > issue by only enabling scatter for NICs known to need it to support jumbo > RX. Add a quirk for "igb" while the PMD is fixed. > Thanks for the v3 Pablo, comments inline. > Reported-by: Louis Peens <louis.peens@netronome.com> > Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> > Reviewed-by: Simon Horman <simon.horman@netronome.com> > --- > > Changelog: > v3->v2: > - only check for driver_name > - tested with "nfp" and not with "igb" > > > lib/netdev-dpdk.c | 12 +++++++++--- > 1 file changed, 9 insertions(+), 3 deletions(-) > > diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index ee39cbe..02ed85b > 100644 > --- a/lib/netdev-dpdk.c > +++ b/lib/netdev-dpdk.c > @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, > int n_rxq, int n_txq) > int diag = 0; > int i; > struct rte_eth_conf conf = port_conf; > + struct rte_eth_dev_info info; > > - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be > explicitly > - * enabled. */ > + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly > + * enabling scatter to support jumbo RX. Note: PMDs are not > + * required to set the offload capabilities and so is not reliable > + * info, only the driver_name is after testing the PMD/NIC */ > if (dev->mtu > ETHER_MTU) { > - conf.rxmode.enable_scatter = 1; > + rte_eth_dev_info_get(dev->port_id, &info); > + if (!strcmp(info.driver_name, "igb")) { For igb devices condition above will never hit, the driver name in info will be reported as 'net_e1000_igb'. I'm seeing a bit of a list growing for other drivers that support the current behavior/requirement for scatter with Jumbo frames. Through testing today I saw that Scatter must be set for ixgbe SRIOV functions also, "net_ixgbe_vf". From code inspection em devices "net_e1000_em" would have to be included also. I would think that 'net_e1000_igb_vf' will be required also when set_mtu support is introduced in the future (quite likely), but for the moment that’s not relevant. I'm wondering if there's an alternative solution to checking driver names as we could end up with an extensive list. Would have to look at this. Ian > + conf.rxmode.enable_scatter = 1; > + } > } > > conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & > -- > 2.7.4 > > _______________________________________________ > dev mailing list > dev@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-dev
On 04/26/2018 04:26 PM, Stokes, Ian wrote: >> Currently to RX jumbo packets fails for NICs not supporting scatter. >> Scatter is not strictly needed for jumbo RX support. This change fixes the >> issue by only enabling scatter for NICs known to need it to support jumbo >> RX. Add a quirk for "igb" while the PMD is fixed. >> > > Thanks for the v3 Pablo, comments inline. > >> Reported-by: Louis Peens <louis.peens@netronome.com> >> Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> >> Reviewed-by: Simon Horman <simon.horman@netronome.com> >> --- >> >> Changelog: >> v3->v2: >> - only check for driver_name >> - tested with "nfp" and not with "igb" >> >> >> lib/netdev-dpdk.c | 12 +++++++++--- >> 1 file changed, 9 insertions(+), 3 deletions(-) >> >> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index ee39cbe..02ed85b >> 100644 >> --- a/lib/netdev-dpdk.c >> +++ b/lib/netdev-dpdk.c >> @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, >> int n_rxq, int n_txq) >> int diag = 0; >> int i; >> struct rte_eth_conf conf = port_conf; >> + struct rte_eth_dev_info info; >> >> - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be >> explicitly >> - * enabled. */ >> + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly >> + * enabling scatter to support jumbo RX. Note: PMDs are not >> + * required to set the offload capabilities and so is not reliable >> + * info, only the driver_name is after testing the PMD/NIC */ >> if (dev->mtu > ETHER_MTU) { >> - conf.rxmode.enable_scatter = 1; >> + rte_eth_dev_info_get(dev->port_id, &info); >> + if (!strcmp(info.driver_name, "igb")) { > > For igb devices condition above will never hit, the driver name in info will be reported as 'net_e1000_igb'. > > I'm seeing a bit of a list growing for other drivers that support the current behavior/requirement for scatter with Jumbo frames. > > Through testing today I saw that Scatter must be set for ixgbe SRIOV functions also, "net_ixgbe_vf". > > From code inspection em devices "net_e1000_em" would have to be included also. > > I would think that 'net_e1000_igb_vf' will be required also when set_mtu support is introduced in the future (quite likely), but for the moment that’s not relevant. > > I'm wondering if there's an alternative solution to checking driver names as we could end up with an extensive list. Would have to look at this. > Did I hear someone say "alternative solution"? ;-) ---8<--- That's why I think it's better to make the exception for "nfp" at present. It works for nfp, it works for Intel and it works for any other NICs that currently work. When OVS is updated to a future DPDK release where all the PMDs support setting/not setting this flag, then I agree any workaround should be made for NICs that are not properly reporting their status, or have quirks. ---8<--- https://mail.openvswitch.org/pipermail/ovs-dev/2018-April/346256.html Kevin. > Ian > >> + conf.rxmode.enable_scatter = 1; >> + } >> } >> >> conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & >> -- >> 2.7.4 >> >> _______________________________________________ >> dev mailing list >> dev@openvswitch.org >> https://mail.openvswitch.org/mailman/listinfo/ovs-dev > _______________________________________________ > dev mailing list > dev@openvswitch.org > https://mail.openvswitch.org/mailman/listinfo/ovs-dev >
> On 04/26/2018 04:26 PM, Stokes, Ian wrote: > >> Currently to RX jumbo packets fails for NICs not supporting scatter. > >> Scatter is not strictly needed for jumbo RX support. This change > >> fixes the issue by only enabling scatter for NICs known to need it to > >> support jumbo RX. Add a quirk for "igb" while the PMD is fixed. > >> > > > > Thanks for the v3 Pablo, comments inline. > > > >> Reported-by: Louis Peens <louis.peens@netronome.com> > >> Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> > >> Reviewed-by: Simon Horman <simon.horman@netronome.com> > >> --- > >> > >> Changelog: > >> v3->v2: > >> - only check for driver_name > >> - tested with "nfp" and not with "igb" > >> > >> > >> lib/netdev-dpdk.c | 12 +++++++++--- > >> 1 file changed, 9 insertions(+), 3 deletions(-) > >> > >> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index > >> ee39cbe..02ed85b > >> 100644 > >> --- a/lib/netdev-dpdk.c > >> +++ b/lib/netdev-dpdk.c > >> @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk > >> *dev, int n_rxq, int n_txq) > >> int diag = 0; > >> int i; > >> struct rte_eth_conf conf = port_conf; > >> + struct rte_eth_dev_info info; > >> > >> - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be > >> explicitly > >> - * enabled. */ > >> + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly > >> + * enabling scatter to support jumbo RX. Note: PMDs are not > >> + * required to set the offload capabilities and so is not reliable > >> + * info, only the driver_name is after testing the PMD/NIC */ > >> if (dev->mtu > ETHER_MTU) { > >> - conf.rxmode.enable_scatter = 1; > >> + rte_eth_dev_info_get(dev->port_id, &info); > >> + if (!strcmp(info.driver_name, "igb")) { > > > > For igb devices condition above will never hit, the driver name in info > will be reported as 'net_e1000_igb'. > > > > I'm seeing a bit of a list growing for other drivers that support the > current behavior/requirement for scatter with Jumbo frames. > > > > Through testing today I saw that Scatter must be set for ixgbe SRIOV > functions also, "net_ixgbe_vf". > > > > From code inspection em devices "net_e1000_em" would have to be included > also. > > > > I would think that 'net_e1000_igb_vf' will be required also when set_mtu > support is introduced in the future (quite likely), but for the moment > that’s not relevant. > > > > I'm wondering if there's an alternative solution to checking driver > names as we could end up with an extensive list. Would have to look at > this. > > > > Did I hear someone say "alternative solution"? ;-) > > ---8<--- > That's why I think it's better to make the exception for "nfp" at present. > It works for nfp, it works for Intel and it works for any other NICs that > currently work. > > When OVS is updated to a future DPDK release where all the PMDs support > setting/not setting this flag, then I agree any workaround should be made > for NICs that are not properly reporting their status, or have quirks. > ---8<--- > > https://mail.openvswitch.org/pipermail/ovs-dev/2018-April/346256.html > +1 I do think this is a simpler solution until we have the offload API fully supported in DPDK for the relevant devices. @Pablo: Do you want the solution backported to previous OVS versions also? The "nfp" approach would be ok I think but as flagged in previous mail, listing igb, ixgbe etc would be a different story as the behavior of these changed between DPDK releases. Just something we should be aware of. Ian > Kevin. > > > Ian > > > >> + conf.rxmode.enable_scatter = 1; > >> + } > >> } > >> > >> conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & > >> -- > >> 2.7.4 > >> > >> _______________________________________________ > >> dev mailing list > >> dev@openvswitch.org > >> https://mail.openvswitch.org/mailman/listinfo/ovs-dev > > _______________________________________________ > > dev mailing list > > dev@openvswitch.org > > https://mail.openvswitch.org/mailman/listinfo/ovs-dev > >
On 26/04/18 19:41, Stokes, Ian wrote: >> On 04/26/2018 04:26 PM, Stokes, Ian wrote: >>>> Currently to RX jumbo packets fails for NICs not supporting scatter. >>>> Scatter is not strictly needed for jumbo RX support. This change >>>> fixes the issue by only enabling scatter for NICs known to need it to >>>> support jumbo RX. Add a quirk for "igb" while the PMD is fixed. >>>> >>> Thanks for the v3 Pablo, comments inline. >>> >>>> Reported-by: Louis Peens <louis.peens@netronome.com> >>>> Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> >>>> Reviewed-by: Simon Horman <simon.horman@netronome.com> >>>> --- >>>> >>>> Changelog: >>>> v3->v2: >>>> - only check for driver_name >>>> - tested with "nfp" and not with "igb" >>>> >>>> >>>> lib/netdev-dpdk.c | 12 +++++++++--- >>>> 1 file changed, 9 insertions(+), 3 deletions(-) >>>> >>>> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index >>>> ee39cbe..02ed85b >>>> 100644 >>>> --- a/lib/netdev-dpdk.c >>>> +++ b/lib/netdev-dpdk.c >>>> @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk >>>> *dev, int n_rxq, int n_txq) >>>> int diag = 0; >>>> int i; >>>> struct rte_eth_conf conf = port_conf; >>>> + struct rte_eth_dev_info info; >>>> >>>> - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be >>>> explicitly >>>> - * enabled. */ >>>> + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly >>>> + * enabling scatter to support jumbo RX. Note: PMDs are not >>>> + * required to set the offload capabilities and so is not reliable >>>> + * info, only the driver_name is after testing the PMD/NIC */ >>>> if (dev->mtu > ETHER_MTU) { >>>> - conf.rxmode.enable_scatter = 1; >>>> + rte_eth_dev_info_get(dev->port_id, &info); >>>> + if (!strcmp(info.driver_name, "igb")) { >>> For igb devices condition above will never hit, the driver name in info >> will be reported as 'net_e1000_igb'. >>> I'm seeing a bit of a list growing for other drivers that support the >> current behavior/requirement for scatter with Jumbo frames. >>> Through testing today I saw that Scatter must be set for ixgbe SRIOV >> functions also, "net_ixgbe_vf". >>> From code inspection em devices "net_e1000_em" would have to be included >> also. >>> I would think that 'net_e1000_igb_vf' will be required also when set_mtu >> support is introduced in the future (quite likely), but for the moment >> that’s not relevant. >>> I'm wondering if there's an alternative solution to checking driver >> names as we could end up with an extensive list. Would have to look at >> this. >> Did I hear someone say "alternative solution"? ;-) >> >> ---8<--- >> That's why I think it's better to make the exception for "nfp" at present. >> It works for nfp, it works for Intel and it works for any other NICs that >> currently work. >> >> When OVS is updated to a future DPDK release where all the PMDs support >> setting/not setting this flag, then I agree any workaround should be made >> for NICs that are not properly reporting their status, or have quirks. >> ---8<--- >> >> https://mail.openvswitch.org/pipermail/ovs-dev/2018-April/346256.html >> > +1 > > I do think this is a simpler solution until we have the offload API fully supported in DPDK for the relevant devices. OK fine, reluctantly agree. Let me share a tested v4 that makes the exception for "nfp". > > @Pablo: Do you want the solution backported to previous OVS versions also? The "nfp" approach would be ok I think but as flagged in previous mail, listing igb, ixgbe etc would be a different story as the behavior of these changed between DPDK releases. Yes, backport would be good please. The behavior of "nfp" has not changed (scatter is not needed for jumbo across releases) > > Just something we should be aware of. > > Ian > >> Kevin. >> >>> Ian >>> >>>> + conf.rxmode.enable_scatter = 1; >>>> + } >>>> } >>>> >>>> conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & >>>> -- >>>> 2.7.4 >>>> >>>> _______________________________________________ >>>> dev mailing list >>>> dev@openvswitch.org >>>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev >>> _______________________________________________ >>> dev mailing list >>> dev@openvswitch.org >>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev >>>
> On 26/04/18 19:41, Stokes, Ian wrote: > >> On 04/26/2018 04:26 PM, Stokes, Ian wrote: > >>>> Currently to RX jumbo packets fails for NICs not supporting scatter. > >>>> Scatter is not strictly needed for jumbo RX support. This change > >>>> fixes the issue by only enabling scatter for NICs known to need it > >>>> to support jumbo RX. Add a quirk for "igb" while the PMD is fixed. > >>>> > >>> Thanks for the v3 Pablo, comments inline. > >>> > >>>> Reported-by: Louis Peens <louis.peens@netronome.com> > >>>> Signed-off-by: Pablo Cascón <pablo.cascon@netronome.com> > >>>> Reviewed-by: Simon Horman <simon.horman@netronome.com> > >>>> --- > >>>> > >>>> Changelog: > >>>> v3->v2: > >>>> - only check for driver_name > >>>> - tested with "nfp" and not with "igb" > >>>> > >>>> > >>>> lib/netdev-dpdk.c | 12 +++++++++--- > >>>> 1 file changed, 9 insertions(+), 3 deletions(-) > >>>> > >>>> diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index > >>>> ee39cbe..02ed85b > >>>> 100644 > >>>> --- a/lib/netdev-dpdk.c > >>>> +++ b/lib/netdev-dpdk.c > >>>> @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk > >>>> *dev, int n_rxq, int n_txq) > >>>> int diag = 0; > >>>> int i; > >>>> struct rte_eth_conf conf = port_conf; > >>>> + struct rte_eth_dev_info info; > >>>> > >>>> - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be > >>>> explicitly > >>>> - * enabled. */ > >>>> + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly > >>>> + * enabling scatter to support jumbo RX. Note: PMDs are not > >>>> + * required to set the offload capabilities and so is not > reliable > >>>> + * info, only the driver_name is after testing the PMD/NIC */ > >>>> if (dev->mtu > ETHER_MTU) { > >>>> - conf.rxmode.enable_scatter = 1; > >>>> + rte_eth_dev_info_get(dev->port_id, &info); > >>>> + if (!strcmp(info.driver_name, "igb")) { > >>> For igb devices condition above will never hit, the driver name in > >>> info > >> will be reported as 'net_e1000_igb'. > >>> I'm seeing a bit of a list growing for other drivers that support > >>> the > >> current behavior/requirement for scatter with Jumbo frames. > >>> Through testing today I saw that Scatter must be set for ixgbe SRIOV > >> functions also, "net_ixgbe_vf". > >>> From code inspection em devices "net_e1000_em" would have to be > >>> included > >> also. > >>> I would think that 'net_e1000_igb_vf' will be required also when > >>> set_mtu > >> support is introduced in the future (quite likely), but for the > >> moment that’s not relevant. > >>> I'm wondering if there's an alternative solution to checking driver > >> names as we could end up with an extensive list. Would have to look > >> at this. > >> Did I hear someone say "alternative solution"? ;-) > >> > >> ---8<--- > >> That's why I think it's better to make the exception for "nfp" at > present. > >> It works for nfp, it works for Intel and it works for any other NICs > >> that currently work. > >> > >> When OVS is updated to a future DPDK release where all the PMDs > >> support setting/not setting this flag, then I agree any workaround > >> should be made for NICs that are not properly reporting their status, > or have quirks. > >> ---8<--- > >> > >> https://mail.openvswitch.org/pipermail/ovs-dev/2018-April/346256.html > >> > > +1 > > > > I do think this is a simpler solution until we have the offload API > fully supported in DPDK for the relevant devices. > > OK fine, reluctantly agree. Let me share a tested v4 that makes the > exception for "nfp". > > > > @Pablo: Do you want the solution backported to previous OVS versions > also? The "nfp" approach would be ok I think but as flagged in previous > mail, listing igb, ixgbe etc would be a different story as the behavior of > these changed between DPDK releases. > > Yes, backport would be good please. The behavior of "nfp" has not changed > (scatter is not needed for jumbo across releases) > Thanks for your work on this Pablo, look forward to the v4. Thanks Ian > > > > Just something we should be aware of. > > > > Ian > > > >> Kevin. > >> > >>> Ian > >>> > >>>> + conf.rxmode.enable_scatter = 1; > >>>> + } > >>>> } > >>>> > >>>> conf.rxmode.hw_ip_checksum = (dev->hw_ol_features & > >>>> -- > >>>> 2.7.4 > >>>> > >>>> _______________________________________________ > >>>> dev mailing list > >>>> dev@openvswitch.org > >>>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev > >>> _______________________________________________ > >>> dev mailing list > >>> dev@openvswitch.org > >>> https://mail.openvswitch.org/mailman/listinfo/ovs-dev > >>>
diff --git a/lib/netdev-dpdk.c b/lib/netdev-dpdk.c index ee39cbe..02ed85b 100644 --- a/lib/netdev-dpdk.c +++ b/lib/netdev-dpdk.c @@ -694,11 +694,17 @@ dpdk_eth_dev_queue_setup(struct netdev_dpdk *dev, int n_rxq, int n_txq) int diag = 0; int i; struct rte_eth_conf conf = port_conf; + struct rte_eth_dev_info info; - /* For some NICs (e.g. Niantic), scatter_rx mode needs to be explicitly - * enabled. */ + /* Quirk: as of DPDK 17.11.1 igb's PMD requires explicitly + * enabling scatter to support jumbo RX. Note: PMDs are not + * required to set the offload capabilities and so is not reliable + * info, only the driver_name is after testing the PMD/NIC */ if (dev->mtu > ETHER_MTU) { - conf.rxmode.enable_scatter = 1; + rte_eth_dev_info_get(dev->port_id, &info); + if (!strcmp(info.driver_name, "igb")) { + conf.rxmode.enable_scatter = 1; + } } conf.rxmode.hw_ip_checksum = (dev->hw_ol_features &