net/iavf: initialize large VF setting at startup
Checks
Commit Message
lv_enabled is used to remember if large VF setting is configured, but
its value is not initialized at startup. This can lead to a different
configuration regarding the initial configuration of the PF.
For example, a first start is done with 8 rx and 8 tx queues. Large VF
is not needed and reta size is 64.
A second start is done with 20 rx and tx queues. Large VF is required
and reta size is 256.
A third start is done with 2 rx and tx queues. Large VF is not needed
but reta size is 256 as the PF has been configured during the second
start.
In order to have a consistent behavior regarding reta size whatever
the configuration of the PF (may be changed by a previous boot) the
lv_enabled must be set properly at the init phase.
Signed-off-by: Ghalem Boudour <ghalem.boudour@6wind.com>
Signed-off-by: Thibaut Collet <thibaut.collet@6wind.com>
Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
---
drivers/net/iavf/iavf_ethdev.c | 2 ++
1 file changed, 2 insertions(+)
Comments
> -----Original Message-----
> From: Ghalem Boudour <ghalem.boudour@6wind.com>
> Sent: Thursday, February 3, 2022 5:04 PM
> To: Wu, Jingjing <jingjing.wu@intel.com>; Xing, Beilei <beilei.xing@intel.com>
> Cc: dev@dpdk.org; Ghalem Boudour <ghalem.boudour@6wind.com>;
> Thibaut Collet <thibaut.collet@6wind.com>; Olivier Matz
> <olivier.matz@6wind.com>
> Subject: [PATCH] net/iavf: initialize large VF setting at startup
>
> lv_enabled is used to remember if large VF setting is configured, but its value
> is not initialized at startup. This can lead to a different configuration
> regarding the initial configuration of the PF.
> For example, a first start is done with 8 rx and 8 tx queues. Large VF is not
> needed and reta size is 64.
> A second start is done with 20 rx and tx queues. Large VF is required and reta
> size is 256.
> A third start is done with 2 rx and tx queues. Large VF is not needed but reta
> size is 256 as the PF has been configured during the second start.
>
> In order to have a consistent behavior regarding reta size whatever the
> configuration of the PF (may be changed by a previous boot) the lv_enabled
> must be set properly at the init phase.
>
> Signed-off-by: Ghalem Boudour <ghalem.boudour@6wind.com>
> Signed-off-by: Thibaut Collet <thibaut.collet@6wind.com>
> Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
> ---
> drivers/net/iavf/iavf_ethdev.c | 2 ++
> 1 file changed, 2 insertions(+)
>
> diff --git a/drivers/net/iavf/iavf_ethdev.c b/drivers/net/iavf/iavf_ethdev.c
> index 79397f15e54f..09410be4b712 100644
> --- a/drivers/net/iavf/iavf_ethdev.c
> +++ b/drivers/net/iavf/iavf_ethdev.c
> @@ -2276,6 +2276,8 @@ iavf_init_vf(struct rte_eth_dev *dev)
> PMD_INIT_LOG(ERR, "unable to allocate rss_lut
> memory");
> goto err_rss;
> }
> + if (vf->vsi_res->num_queue_pairs >
> IAVF_MAX_NUM_QUEUES_DFLT)
> + vf->lv_enabled = true;
Why you put this operation in the part of " VIRTCHNL_VF_OFFLOAD_RSS_PF"?
> }
>
> if (vf->vf_res->vf_cap_flags &
> VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC) {
> --
> 2.30.2
Hi Ting,
The purpose of the correction is to ensure that the rss configuration
is aligned with the number of queues
allocated on the current boot and not impacted by a previous
configuration done in a previous boot.
The rss configuration is only available if VIRTCHNL_VF_OFFLOAD_RSS_PF
is present and that is the reason
that we have set the operation in part of VIRTCHNL_VF_OFFLOAD_RSS_PF.
Nevertheless this operation can be moved outside to be more generic
and also covers iavf_config_irq_map.
On Wed, Feb 9, 2022 at 3:42 AM Xu, Ting <ting.xu@intel.com> wrote:
>
> > -----Original Message-----
> > From: Ghalem Boudour <ghalem.boudour@6wind.com>
> > Sent: Thursday, February 3, 2022 5:04 PM
> > To: Wu, Jingjing <jingjing.wu@intel.com>; Xing, Beilei <beilei.xing@intel.com>
> > Cc: dev@dpdk.org; Ghalem Boudour <ghalem.boudour@6wind.com>;
> > Thibaut Collet <thibaut.collet@6wind.com>; Olivier Matz
> > <olivier.matz@6wind.com>
> > Subject: [PATCH] net/iavf: initialize large VF setting at startup
> >
> > lv_enabled is used to remember if large VF setting is configured, but its value
> > is not initialized at startup. This can lead to a different configuration
> > regarding the initial configuration of the PF.
> > For example, a first start is done with 8 rx and 8 tx queues. Large VF is not
> > needed and reta size is 64.
> > A second start is done with 20 rx and tx queues. Large VF is required and reta
> > size is 256.
> > A third start is done with 2 rx and tx queues. Large VF is not needed but reta
> > size is 256 as the PF has been configured during the second start.
> >
> > In order to have a consistent behavior regarding reta size whatever the
> > configuration of the PF (may be changed by a previous boot) the lv_enabled
> > must be set properly at the init phase.
> >
> > Signed-off-by: Ghalem Boudour <ghalem.boudour@6wind.com>
> > Signed-off-by: Thibaut Collet <thibaut.collet@6wind.com>
> > Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
> > ---
> > drivers/net/iavf/iavf_ethdev.c | 2 ++
> > 1 file changed, 2 insertions(+)
> >
> > diff --git a/drivers/net/iavf/iavf_ethdev.c b/drivers/net/iavf/iavf_ethdev.c
> > index 79397f15e54f..09410be4b712 100644
> > --- a/drivers/net/iavf/iavf_ethdev.c
> > +++ b/drivers/net/iavf/iavf_ethdev.c
> > @@ -2276,6 +2276,8 @@ iavf_init_vf(struct rte_eth_dev *dev)
> > PMD_INIT_LOG(ERR, "unable to allocate rss_lut
> > memory");
> > goto err_rss;
> > }
> > + if (vf->vsi_res->num_queue_pairs >
> > IAVF_MAX_NUM_QUEUES_DFLT)
> > + vf->lv_enabled = true;
>
> Why you put this operation in the part of " VIRTCHNL_VF_OFFLOAD_RSS_PF"?
>
> > }
> >
> > if (vf->vf_res->vf_cap_flags &
> > VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC) {
> > --
> > 2.30.2
>
> -----Original Message-----
> From: Ghalem Boudour <ghalem.boudour@6wind.com>
> Sent: Thursday, February 10, 2022 3:27 AM
> To: Xu, Ting <ting.xu@intel.com>
> Cc: Wu, Jingjing <jingjing.wu@intel.com>; Xing, Beilei <beilei.xing@intel.com>;
> dev@dpdk.org; Thibaut Collet <thibaut.collet@6wind.com>; Olivier Matz
> <olivier.matz@6wind.com>
> Subject: Re: [PATCH] net/iavf: initialize large VF setting at startup
>
> Hi Ting,
>
> The purpose of the correction is to ensure that the rss configuration is
> aligned with the number of queues allocated on the current boot and not
> impacted by a previous configuration done in a previous boot.
> The rss configuration is only available if VIRTCHNL_VF_OFFLOAD_RSS_PF is
> present and that is the reason that we have set the operation in part of
> VIRTCHNL_VF_OFFLOAD_RSS_PF.
>
> Nevertheless this operation can be moved outside to be more generic and
> also covers iavf_config_irq_map.
>
> On Wed, Feb 9, 2022 at 3:42 AM Xu, Ting <ting.xu@intel.com> wrote:
> >
> > > -----Original Message-----
> > > From: Ghalem Boudour <ghalem.boudour@6wind.com>
> > > Sent: Thursday, February 3, 2022 5:04 PM
> > > To: Wu, Jingjing <jingjing.wu@intel.com>; Xing, Beilei
> > > <beilei.xing@intel.com>
> > > Cc: dev@dpdk.org; Ghalem Boudour <ghalem.boudour@6wind.com>;
> Thibaut
> > > Collet <thibaut.collet@6wind.com>; Olivier Matz
> > > <olivier.matz@6wind.com>
> > > Subject: [PATCH] net/iavf: initialize large VF setting at startup
> > >
> > > lv_enabled is used to remember if large VF setting is configured,
> > > but its value is not initialized at startup. This can lead to a
> > > different configuration regarding the initial configuration of the PF.
> > > For example, a first start is done with 8 rx and 8 tx queues. Large
> > > VF is not needed and reta size is 64.
> > > A second start is done with 20 rx and tx queues. Large VF is
> > > required and reta size is 256.
> > > A third start is done with 2 rx and tx queues. Large VF is not
> > > needed but reta size is 256 as the PF has been configured during the
> second start.
> > >
> > > In order to have a consistent behavior regarding reta size whatever
> > > the configuration of the PF (may be changed by a previous boot) the
> > > lv_enabled must be set properly at the init phase.
> > >
> > > Signed-off-by: Ghalem Boudour <ghalem.boudour@6wind.com>
> > > Signed-off-by: Thibaut Collet <thibaut.collet@6wind.com>
> > > Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
> > > ---
> > > drivers/net/iavf/iavf_ethdev.c | 2 ++
> > > 1 file changed, 2 insertions(+)
> > >
> > > diff --git a/drivers/net/iavf/iavf_ethdev.c
> > > b/drivers/net/iavf/iavf_ethdev.c index 79397f15e54f..09410be4b712
> > > 100644
> > > --- a/drivers/net/iavf/iavf_ethdev.c
> > > +++ b/drivers/net/iavf/iavf_ethdev.c
> > > @@ -2276,6 +2276,8 @@ iavf_init_vf(struct rte_eth_dev *dev)
> > > PMD_INIT_LOG(ERR, "unable to allocate rss_lut
> > > memory");
> > > goto err_rss;
> > > }
> > > + if (vf->vsi_res->num_queue_pairs >
> > > IAVF_MAX_NUM_QUEUES_DFLT)
> > > + vf->lv_enabled = true;
> >
> > Why you put this operation in the part of "
> VIRTCHNL_VF_OFFLOAD_RSS_PF"?
> >
> > > }
> > >
> > > if (vf->vf_res->vf_cap_flags &
> > > VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC) {
> > > --
> > > 2.30.2
> >
I think it is OK to move outside to the generic vf initialization since it is not only bound with RSS, what's your opinion?
I agree with you.
Will submit a new revision.
Thanks
On Thu, Feb 10, 2022 at 3:43 AM Xu, Ting <ting.xu@intel.com> wrote:
>
> > -----Original Message-----
> > From: Ghalem Boudour <ghalem.boudour@6wind.com>
> > Sent: Thursday, February 10, 2022 3:27 AM
> > To: Xu, Ting <ting.xu@intel.com>
> > Cc: Wu, Jingjing <jingjing.wu@intel.com>; Xing, Beilei <beilei.xing@intel.com>;
> > dev@dpdk.org; Thibaut Collet <thibaut.collet@6wind.com>; Olivier Matz
> > <olivier.matz@6wind.com>
> > Subject: Re: [PATCH] net/iavf: initialize large VF setting at startup
> >
> > Hi Ting,
> >
> > The purpose of the correction is to ensure that the rss configuration is
> > aligned with the number of queues allocated on the current boot and not
> > impacted by a previous configuration done in a previous boot.
> > The rss configuration is only available if VIRTCHNL_VF_OFFLOAD_RSS_PF is
> > present and that is the reason that we have set the operation in part of
> > VIRTCHNL_VF_OFFLOAD_RSS_PF.
> >
> > Nevertheless this operation can be moved outside to be more generic and
> > also covers iavf_config_irq_map.
> >
> > On Wed, Feb 9, 2022 at 3:42 AM Xu, Ting <ting.xu@intel.com> wrote:
> > >
> > > > -----Original Message-----
> > > > From: Ghalem Boudour <ghalem.boudour@6wind.com>
> > > > Sent: Thursday, February 3, 2022 5:04 PM
> > > > To: Wu, Jingjing <jingjing.wu@intel.com>; Xing, Beilei
> > > > <beilei.xing@intel.com>
> > > > Cc: dev@dpdk.org; Ghalem Boudour <ghalem.boudour@6wind.com>;
> > Thibaut
> > > > Collet <thibaut.collet@6wind.com>; Olivier Matz
> > > > <olivier.matz@6wind.com>
> > > > Subject: [PATCH] net/iavf: initialize large VF setting at startup
> > > >
> > > > lv_enabled is used to remember if large VF setting is configured,
> > > > but its value is not initialized at startup. This can lead to a
> > > > different configuration regarding the initial configuration of the PF.
> > > > For example, a first start is done with 8 rx and 8 tx queues. Large
> > > > VF is not needed and reta size is 64.
> > > > A second start is done with 20 rx and tx queues. Large VF is
> > > > required and reta size is 256.
> > > > A third start is done with 2 rx and tx queues. Large VF is not
> > > > needed but reta size is 256 as the PF has been configured during the
> > second start.
> > > >
> > > > In order to have a consistent behavior regarding reta size whatever
> > > > the configuration of the PF (may be changed by a previous boot) the
> > > > lv_enabled must be set properly at the init phase.
> > > >
> > > > Signed-off-by: Ghalem Boudour <ghalem.boudour@6wind.com>
> > > > Signed-off-by: Thibaut Collet <thibaut.collet@6wind.com>
> > > > Reviewed-by: Olivier Matz <olivier.matz@6wind.com>
> > > > ---
> > > > drivers/net/iavf/iavf_ethdev.c | 2 ++
> > > > 1 file changed, 2 insertions(+)
> > > >
> > > > diff --git a/drivers/net/iavf/iavf_ethdev.c
> > > > b/drivers/net/iavf/iavf_ethdev.c index 79397f15e54f..09410be4b712
> > > > 100644
> > > > --- a/drivers/net/iavf/iavf_ethdev.c
> > > > +++ b/drivers/net/iavf/iavf_ethdev.c
> > > > @@ -2276,6 +2276,8 @@ iavf_init_vf(struct rte_eth_dev *dev)
> > > > PMD_INIT_LOG(ERR, "unable to allocate rss_lut
> > > > memory");
> > > > goto err_rss;
> > > > }
> > > > + if (vf->vsi_res->num_queue_pairs >
> > > > IAVF_MAX_NUM_QUEUES_DFLT)
> > > > + vf->lv_enabled = true;
> > >
> > > Why you put this operation in the part of "
> > VIRTCHNL_VF_OFFLOAD_RSS_PF"?
> > >
> > > > }
> > > >
> > > > if (vf->vf_res->vf_cap_flags &
> > > > VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC) {
> > > > --
> > > > 2.30.2
> > >
>
> I think it is OK to move outside to the generic vf initialization since it is not only bound with RSS, what's your opinion?
@@ -2276,6 +2276,8 @@ iavf_init_vf(struct rte_eth_dev *dev)
PMD_INIT_LOG(ERR, "unable to allocate rss_lut memory");
goto err_rss;
}
+ if (vf->vsi_res->num_queue_pairs > IAVF_MAX_NUM_QUEUES_DFLT)
+ vf->lv_enabled = true;
}
if (vf->vf_res->vf_cap_flags & VIRTCHNL_VF_OFFLOAD_RX_FLEX_DESC) {