Message ID | 1539188552-129083-3-git-send-email-tiago.lam@intel.com |
---|---|
State | Changes Requested |
Delegated to: | Ian Stokes |
Headers | show |
Series | Support multi-segment mbufs | expand |
> From: Mark Kavanagh <mark.b.kavanagh@intel.com> > > dp_packets are created using xmalloc(); in the case of OvS-DPDK, it's > possible the the resultant mbuf portion of the dp_packet contains random > data. For some mbuf fields, specifically those related to multi-segment > mbufs and/or offload features, random values may cause unexpected > behaviour, should the dp_packet's contents be later copied to a DPDK mbuf. > It is critical therefore, that these fields should be initialized to 0. > > This patch ensures that the following mbuf fields are initialized to > appropriate values on creation of a new dp_packet: > - ol_flags=0 > - nb_segs=1 > - tx_offload=0 > - packet_type=0 > - next=NULL > > Adapted from an idea by Michael Qiu <qiudayu@chinac.com>: > https://patchwork.ozlabs.org/patch/777570/ > > Co-authored-by: Tiago Lam <tiago.lam@intel.com> > > Signed-off-by: Mark Kavanagh <mark.b.kavanagh@intel.com> > Signed-off-by: Tiago Lam <tiago.lam@intel.com> > Acked-by: Eelco Chaudron <echaudro@redhat.com> > Acked-by: Flavio Leitner <fbl@sysclose.org> Thanks Tiago, This will need a rebase. One minor comment below. > --- > lib/dp-packet.h | 9 +++++---- > 1 file changed, 5 insertions(+), 4 deletions(-) > > diff --git a/lib/dp-packet.h b/lib/dp-packet.h index ba91e58..b948fe1 > 100644 > --- a/lib/dp-packet.h > +++ b/lib/dp-packet.h > @@ -625,14 +625,15 @@ dp_packet_mbuf_rss_flag_reset(struct dp_packet *p > OVS_UNUSED) } > > /* This initialization is needed for packets that do not come > - * from DPDK interfaces, when vswitchd is built with --with-dpdk. > - * The DPDK rte library will still otherwise manage the mbuf. > - * We only need to initialize the mbuf ol_flags. */ > + * from DPDK interfaces, when vswitchd is built with --with-dpdk. */ > static inline void > dp_packet_mbuf_init(struct dp_packet *p OVS_UNUSED) { #ifdef > DPDK_NETDEV > - p->mbuf.ol_flags = 0; > + struct rte_mbuf *mbuf = &(p->mbuf); > + mbuf->ol_flags = mbuf->tx_offload = mbuf->packet_type = 0; > + mbuf->nb_segs = 1; Is it just for ease of access that you cast a pointer to mbuf? Just looking at the other rte specific functions implemented in the file, they access mbuf via p->mbuf rather than the cast. I would like to keep implementation uniform across functions if possible. Thanks Ian > + mbuf->next = NULL; > #endif > } > > -- > 2.7.4
Hi Tiago, Ian, On Tue, Oct 16, 2018 at 02:28:56PM +0000, Stokes, Ian wrote: > > From: Mark Kavanagh <mark.b.kavanagh@intel.com> > > > > dp_packets are created using xmalloc(); in the case of OvS-DPDK, it's > > possible the the resultant mbuf portion of the dp_packet contains random > > data. For some mbuf fields, specifically those related to multi-segment > > mbufs and/or offload features, random values may cause unexpected > > behaviour, should the dp_packet's contents be later copied to a DPDK mbuf. > > It is critical therefore, that these fields should be initialized to 0. > > > > This patch ensures that the following mbuf fields are initialized to > > appropriate values on creation of a new dp_packet: > > - ol_flags=0 > > - nb_segs=1 > > - tx_offload=0 > > - packet_type=0 > > - next=NULL > > > > Adapted from an idea by Michael Qiu <qiudayu@chinac.com>: > > https://patchwork.ozlabs.org/patch/777570/ > > > > Co-authored-by: Tiago Lam <tiago.lam@intel.com> > > > > Signed-off-by: Mark Kavanagh <mark.b.kavanagh@intel.com> > > Signed-off-by: Tiago Lam <tiago.lam@intel.com> > > Acked-by: Eelco Chaudron <echaudro@redhat.com> > > Acked-by: Flavio Leitner <fbl@sysclose.org> > > Thanks Tiago, > > This will need a rebase. One minor comment below. > > --- > > lib/dp-packet.h | 9 +++++---- > > 1 file changed, 5 insertions(+), 4 deletions(-) > > > > diff --git a/lib/dp-packet.h b/lib/dp-packet.h index ba91e58..b948fe1 > > 100644 > > --- a/lib/dp-packet.h > > +++ b/lib/dp-packet.h > > @@ -625,14 +625,15 @@ dp_packet_mbuf_rss_flag_reset(struct dp_packet *p > > OVS_UNUSED) } > > > > /* This initialization is needed for packets that do not come > > - * from DPDK interfaces, when vswitchd is built with --with-dpdk. > > - * The DPDK rte library will still otherwise manage the mbuf. > > - * We only need to initialize the mbuf ol_flags. */ > > + * from DPDK interfaces, when vswitchd is built with --with-dpdk. */ > > static inline void > > dp_packet_mbuf_init(struct dp_packet *p OVS_UNUSED) { #ifdef > > DPDK_NETDEV > > - p->mbuf.ol_flags = 0; > > + struct rte_mbuf *mbuf = &(p->mbuf); > > + mbuf->ol_flags = mbuf->tx_offload = mbuf->packet_type = 0; > > + mbuf->nb_segs = 1; > > Is it just for ease of access that you cast a pointer to mbuf? > > Just looking at the other rte specific functions implemented in the > file, they access mbuf via p->mbuf rather than the cast. I would > like to keep implementation uniform across functions if possible. This seems to be needed but not related to the multi seg, so I wonder if this patch could be posted as a standalone fix. The goal is to apply it sooner and reduce the patchset size. What do you think? Thanks, fbl > > Thanks > Ian > > > + mbuf->next = NULL; > > #endif > > } > > > > -- > > 2.7.4
> Hi Tiago, Ian, > > On Tue, Oct 16, 2018 at 02:28:56PM +0000, Stokes, Ian wrote: > > > From: Mark Kavanagh <mark.b.kavanagh@intel.com> > > > > > > dp_packets are created using xmalloc(); in the case of OvS-DPDK, > > > it's possible the the resultant mbuf portion of the dp_packet > > > contains random data. For some mbuf fields, specifically those > > > related to multi-segment mbufs and/or offload features, random > > > values may cause unexpected behaviour, should the dp_packet's contents > be later copied to a DPDK mbuf. > > > It is critical therefore, that these fields should be initialized to > 0. > > > > > > This patch ensures that the following mbuf fields are initialized to > > > appropriate values on creation of a new dp_packet: > > > - ol_flags=0 > > > - nb_segs=1 > > > - tx_offload=0 > > > - packet_type=0 > > > - next=NULL > > > > > > Adapted from an idea by Michael Qiu <qiudayu@chinac.com>: > > > https://patchwork.ozlabs.org/patch/777570/ > > > > > > Co-authored-by: Tiago Lam <tiago.lam@intel.com> > > > > > > Signed-off-by: Mark Kavanagh <mark.b.kavanagh@intel.com> > > > Signed-off-by: Tiago Lam <tiago.lam@intel.com> > > > Acked-by: Eelco Chaudron <echaudro@redhat.com> > > > Acked-by: Flavio Leitner <fbl@sysclose.org> > > > > Thanks Tiago, > > > > This will need a rebase. One minor comment below. > > > --- > > > lib/dp-packet.h | 9 +++++---- > > > 1 file changed, 5 insertions(+), 4 deletions(-) > > > > > > diff --git a/lib/dp-packet.h b/lib/dp-packet.h index > > > ba91e58..b948fe1 > > > 100644 > > > --- a/lib/dp-packet.h > > > +++ b/lib/dp-packet.h > > > @@ -625,14 +625,15 @@ dp_packet_mbuf_rss_flag_reset(struct dp_packet > > > *p > > > OVS_UNUSED) } > > > > > > /* This initialization is needed for packets that do not come > > > - * from DPDK interfaces, when vswitchd is built with --with-dpdk. > > > - * The DPDK rte library will still otherwise manage the mbuf. > > > - * We only need to initialize the mbuf ol_flags. */ > > > + * from DPDK interfaces, when vswitchd is built with --with-dpdk. > > > + */ > > > static inline void > > > dp_packet_mbuf_init(struct dp_packet *p OVS_UNUSED) { #ifdef > > > DPDK_NETDEV > > > - p->mbuf.ol_flags = 0; > > > + struct rte_mbuf *mbuf = &(p->mbuf); > > > + mbuf->ol_flags = mbuf->tx_offload = mbuf->packet_type = 0; > > > + mbuf->nb_segs = 1; > > > > Is it just for ease of access that you cast a pointer to mbuf? > > > > Just looking at the other rte specific functions implemented in the > > file, they access mbuf via p->mbuf rather than the cast. I would like > > to keep implementation uniform across functions if possible. > > This seems to be needed but not related to the multi seg, so I wonder if > this patch could be posted as a standalone fix. The goal is to apply it > sooner and reduce the patchset size. > > What do you think? > +1, I think patches 1-3 could be applied independent of the series itself. There's been a few comments on the latest revision, Tiago if you get a chance to send another revision of these separate to the series we could look to merge as part of the pull request this week. Thanks Ian > Thanks, > fbl > > > > > > Thanks > > Ian > > > > > + mbuf->next = NULL; > > > #endif > > > } > > > > > > -- > > > 2.7.4
On 01/11/2018 19:30, Stokes, Ian wrote: >> Hi Tiago, Ian, >> >> On Tue, Oct 16, 2018 at 02:28:56PM +0000, Stokes, Ian wrote: >>>> From: Mark Kavanagh <mark.b.kavanagh@intel.com> >>>> >>>> dp_packets are created using xmalloc(); in the case of OvS-DPDK, >>>> it's possible the the resultant mbuf portion of the dp_packet >>>> contains random data. For some mbuf fields, specifically those >>>> related to multi-segment mbufs and/or offload features, random >>>> values may cause unexpected behaviour, should the dp_packet's contents >> be later copied to a DPDK mbuf. >>>> It is critical therefore, that these fields should be initialized to >> 0. >>>> >>>> This patch ensures that the following mbuf fields are initialized to >>>> appropriate values on creation of a new dp_packet: >>>> - ol_flags=0 >>>> - nb_segs=1 >>>> - tx_offload=0 >>>> - packet_type=0 >>>> - next=NULL >>>> >>>> Adapted from an idea by Michael Qiu <qiudayu@chinac.com>: >>>> https://patchwork.ozlabs.org/patch/777570/ >>>> >>>> Co-authored-by: Tiago Lam <tiago.lam@intel.com> >>>> >>>> Signed-off-by: Mark Kavanagh <mark.b.kavanagh@intel.com> >>>> Signed-off-by: Tiago Lam <tiago.lam@intel.com> >>>> Acked-by: Eelco Chaudron <echaudro@redhat.com> >>>> Acked-by: Flavio Leitner <fbl@sysclose.org> >>> >>> Thanks Tiago, >>> >>> This will need a rebase. One minor comment below. >>>> --- >>>> lib/dp-packet.h | 9 +++++---- >>>> 1 file changed, 5 insertions(+), 4 deletions(-) >>>> >>>> diff --git a/lib/dp-packet.h b/lib/dp-packet.h index >>>> ba91e58..b948fe1 >>>> 100644 >>>> --- a/lib/dp-packet.h >>>> +++ b/lib/dp-packet.h >>>> @@ -625,14 +625,15 @@ dp_packet_mbuf_rss_flag_reset(struct dp_packet >>>> *p >>>> OVS_UNUSED) } >>>> >>>> /* This initialization is needed for packets that do not come >>>> - * from DPDK interfaces, when vswitchd is built with --with-dpdk. >>>> - * The DPDK rte library will still otherwise manage the mbuf. >>>> - * We only need to initialize the mbuf ol_flags. */ >>>> + * from DPDK interfaces, when vswitchd is built with --with-dpdk. >>>> + */ >>>> static inline void >>>> dp_packet_mbuf_init(struct dp_packet *p OVS_UNUSED) { #ifdef >>>> DPDK_NETDEV >>>> - p->mbuf.ol_flags = 0; >>>> + struct rte_mbuf *mbuf = &(p->mbuf); >>>> + mbuf->ol_flags = mbuf->tx_offload = mbuf->packet_type = 0; >>>> + mbuf->nb_segs = 1; >>> >>> Is it just for ease of access that you cast a pointer to mbuf? >>> >>> Just looking at the other rte specific functions implemented in the >>> file, they access mbuf via p->mbuf rather than the cast. I would like >>> to keep implementation uniform across functions if possible. >> >> This seems to be needed but not related to the multi seg, so I wonder if >> this patch could be posted as a standalone fix. The goal is to apply it >> sooner and reduce the patchset size. >> >> What do you think? >> > > +1, I think patches 1-3 could be applied independent of the series itself. > > There's been a few comments on the latest revision, Tiago if you get a chance to send another revision of these separate to the series we could look to merge as part of the pull request this week. > Sounds good. Just sent a new series with the first three patches only - it addresses the latest comments in v11 of the multi-segment mbuf series. I've also kept the ACK's, let me know if that's not OK. Thanks, Tiago. > Thanks > Ian >> Thanks, >> fbl >> >> >>> >>> Thanks >>> Ian >>> >>>> + mbuf->next = NULL; >>>> #endif >>>> } >>>> >>>> -- >>>> 2.7.4 >
diff --git a/lib/dp-packet.h b/lib/dp-packet.h index ba91e58..b948fe1 100644 --- a/lib/dp-packet.h +++ b/lib/dp-packet.h @@ -625,14 +625,15 @@ dp_packet_mbuf_rss_flag_reset(struct dp_packet *p OVS_UNUSED) } /* This initialization is needed for packets that do not come - * from DPDK interfaces, when vswitchd is built with --with-dpdk. - * The DPDK rte library will still otherwise manage the mbuf. - * We only need to initialize the mbuf ol_flags. */ + * from DPDK interfaces, when vswitchd is built with --with-dpdk. */ static inline void dp_packet_mbuf_init(struct dp_packet *p OVS_UNUSED) { #ifdef DPDK_NETDEV - p->mbuf.ol_flags = 0; + struct rte_mbuf *mbuf = &(p->mbuf); + mbuf->ol_flags = mbuf->tx_offload = mbuf->packet_type = 0; + mbuf->nb_segs = 1; + mbuf->next = NULL; #endif }