[dpdk-users] eBPF in DPDK

[dpdk-users] eBPF in DPDK

[Avi Cohen]

Hi All,
can someone refer to the flow of a packet that is processed  in eBPF
program in the kernel . and then forwarded to DPDK-app in user space.
this was integrated into DPDK 18.05   but not documented. (
https://doc.dpdk.org/guides-18.05/prog_guide/bpf_lib.html)
what is the PMD type for this flow ? new socket ? what is the penalty ?
i thought to do some traffic filtering  there.
pros/cons ?
Best Regards
Avi

Re: [dpdk-users] eBPF in DPDK

[Stephen Hemminger]

On Mon, 8 Apr 2019 17:25:55 +0300
Avi Cohen <acohen@ves.io> wrote:

> Hi All,
> can someone refer to the flow of a packet that is processed  in eBPF
> program in the kernel . and then forwarded to DPDK-app in user space.
> this was integrated into DPDK 18.05   but not documented. (
> https://doc.dpdk.org/guides-18.05/prog_guide/bpf_lib.html)
> what is the PMD type for this flow ? new socket ? what is the penalty ?
> i thought to do some traffic filtering  there.
> pros/cons ?
> Best Regards
> Avi


Unlike Linux which has a predetermined set of steps in processing packets.
DPDK is a raw toolkit where the processing is composed totally by the
network application.

Therefore DPDK has a raw eBPF library, it is up to the application to install it where
it wants in the packet processing. One example is done via rx and tx callbacks
as shown in the rte_bpf_ethdev.h.

Re: [dpdk-users] eBPF in DPDK

[Avi Cohen]

Stephen
I did not understand. Is this eBPF dpdk feature can run in the kernel or
userspace ?
Regards avi

בתאריך יום ב׳, 8 באפר׳ 2019, 18:09, מאת Stephen Hemminger ‏<
stephen@networkplumber.org>:

> On Mon, 8 Apr 2019 17:25:55 +0300
> Avi Cohen <acohen@ves.io> wrote:
>
> > Hi All,
> > can someone refer to the flow of a packet that is processed  in eBPF
> > program in the kernel . and then forwarded to DPDK-app in user space.
> > this was integrated into DPDK 18.05   but not documented. (
> > https://doc.dpdk.org/guides-18.05/prog_guide/bpf_lib.html)
> > what is the PMD type for this flow ? new socket ? what is the penalty ?
> > i thought to do some traffic filtering  there.
> > pros/cons ?
> > Best Regards
> > Avi
>
>
> Unlike Linux which has a predetermined set of steps in processing packets.
> DPDK is a raw toolkit where the processing is composed totally by the
> network application.
>
> Therefore DPDK has a raw eBPF library, it is up to the application to
> install it where
> it wants in the packet processing. One example is done via rx and tx
> callbacks
> as shown in the rte_bpf_ethdev.h.
>

Re: [dpdk-users] eBPF in DPDK

[Stephen Hemminger]

On Mon, 8 Apr 2019 21:16:53 +0300
Avi Cohen <acohen@ves.io> wrote:

> Stephen
> I did not understand. Is this eBPF dpdk feature can run in the kernel or
> userspace ?
> Regards avi
> 
> בתאריך יום ב׳, 8 באפר׳ 2019, 18:09, מאת Stephen Hemminger ‏<
> stephen@networkplumber.org>:  
> 
> > On Mon, 8 Apr 2019 17:25:55 +0300
> > Avi Cohen <acohen@ves.io> wrote:
> >  
> > > Hi All,
> > > can someone refer to the flow of a packet that is processed  in eBPF
> > > program in the kernel . and then forwarded to DPDK-app in user space.
> > > this was integrated into DPDK 18.05   but not documented. (
> > > https://doc.dpdk.org/guides-18.05/prog_guide/bpf_lib.html)
> > > what is the PMD type for this flow ? new socket ? what is the penalty ?
> > > i thought to do some traffic filtering  there.
> > > pros/cons ?
> > > Best Regards
> > > Avi  
> >
> >
> > Unlike Linux which has a predetermined set of steps in processing packets.
> > DPDK is a raw toolkit where the processing is composed totally by the
> > network application.
> >
> > Therefore DPDK has a raw eBPF library, it is up to the application to
> > install it where
> > it wants in the packet processing. One example is done via rx and tx
> > callbacks
> > as shown in the rte_bpf_ethdev.h.
> >  

DPDK is all in userspace. The eBPF library is just another facility
like LPM lookup, hash library and ACL.

Re: [dpdk-users] eBPF in DPDK

[Avi Cohen]

I  know DPDK is in userspace. But i know there is also some talkings about
AF_XDP , in which xdp program is running in kernel and then w zero copy can
be fwd to dpdk app via the PMD AF_XDP

בתאריך יום ב׳, 8 באפר׳ 2019, 21:31, מאת Stephen Hemminger ‏<
stephen@networkplumber.org>:

> On Mon, 8 Apr 2019 21:16:53 +0300
> Avi Cohen <acohen@ves.io> wrote:
>
> > Stephen
> > I did not understand. Is this eBPF dpdk feature can run in the kernel or
> > userspace ?
> > Regards avi
> >
> > בתאריך יום ב׳, 8 באפר׳ 2019, 18:09, מאת Stephen Hemminger ‏<
> > stephen@networkplumber.org>:
> >
> > > On Mon, 8 Apr 2019 17:25:55 +0300
> > > Avi Cohen <acohen@ves.io> wrote:
> > >
> > > > Hi All,
> > > > can someone refer to the flow of a packet that is processed  in eBPF
> > > > program in the kernel . and then forwarded to DPDK-app in user space.
> > > > this was integrated into DPDK 18.05   but not documented. (
> > > > https://doc.dpdk.org/guides-18.05/prog_guide/bpf_lib.html)
> > > > what is the PMD type for this flow ? new socket ? what is the
> penalty ?
> > > > i thought to do some traffic filtering  there.
> > > > pros/cons ?
> > > > Best Regards
> > > > Avi
> > >
> > >
> > > Unlike Linux which has a predetermined set of steps in processing
> packets.
> > > DPDK is a raw toolkit where the processing is composed totally by the
> > > network application.
> > >
> > > Therefore DPDK has a raw eBPF library, it is up to the application to
> > > install it where
> > > it wants in the packet processing. One example is done via rx and tx
> > > callbacks
> > > as shown in the rte_bpf_ethdev.h.
> > >
>
> DPDK is all in userspace. The eBPF library is just another facility
> like LPM lookup, hash library and ACL.
>