Netronome NFP PMD support in MoonGen #195
Comments
|
Also, can the driver be enabled / linked by default, or alternatively be enabled by a build.sh flag like was done for the Mellanox cards with --mlx4 and --mlx5 ? |
|
I've added it to be linked in by default, but I couldn't test it. I did find some server with some netronome NIC (pci device id 0x4000) and wanted to test it to add some config flags to make sure statistics etc are reported correctly. But I couldn't get it to work because I don't have access to the necessary libraries/firmware. |
|
Hey, |
|
Hi,
19ee:4000 refers to physical functions o the NFP. The NFP PMD unfortunately
does not support these devices until DPDK 17.11 or backports thereof.
If MoonGen uses DPDK 17.11 or later, or an earlier version with appropriate
backports, this should work out of the box, otherwise only device 6003
(virtual functions) are supported.
…On Tue, Jul 24, 2018 at 7:20 AM, Kash ***@***.***> wrote:
Hey,
I'm facing a similar issue. When I use MoonGen with the Netronome 4000 it
does not detect the device. I'm using today's master build. Any help from
your side is appreciated.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#195 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AS__NZW4CTpzf5QqoR9eIXmyjdjQNLjbks5uJq6ngaJpZM4P6R5f>
.
|
|
We just got a few new netronome nics, so I can check that later :)
… Am 24.07.2018 um 08:26 schrieb Netronome-moongen ***@***.***>:
Hi,
19ee:4000 refers to physical functions o the NFP. The NFP PMD unfortunately
does not support these devices until DPDK 17.11 or backports thereof.
If MoonGen uses DPDK 17.11 or later, or an earlier version with appropriate
backports, this should work out of the box, otherwise only device 6003
(virtual functions) are supported.
On Tue, Jul 24, 2018 at 7:20 AM, Kash ***@***.***> wrote:
> Hey,
> I'm facing a similar issue. When I use MoonGen with the Netronome 4000 it
> does not detect the device. I'm using today's master build. Any help from
> your side is appreciated.
>
> —
> You are receiving this because you authored the thread.
> Reply to this email directly, view it on GitHub
> <#195 (comment)>,
> or mute the thread
> <https://github.com/notifications/unsubscribe-auth/AS__NZW4CTpzf5QqoR9eIXmyjdjQNLjbks5uJq6ngaJpZM4P6R5f>
> .
>
—
You are receiving this because you commented.
Reply to this email directly, view it on GitHub, or mute the thread.
|
|
One other thing to note:
IIRC the pci.ids file in MoonGen only lists the base PCI device ID of 6000,
which is used for Netronome NFP's. The device will be detected as an
Unknown device or an Unknown Physical function, depending o how MoonGen
detects / interprets the device type.
On Tue, Jul 24, 2018 at 9:34 AM, Paul Emmerich <[email protected]>
wrote:
… We just got a few new netronome nics, so I can check that later :)
> Am 24.07.2018 um 08:26 schrieb Netronome-moongen <
***@***.***>:
>
> Hi,
>
> 19ee:4000 refers to physical functions o the NFP. The NFP PMD
unfortunately
> does not support these devices until DPDK 17.11 or backports thereof.
>
> If MoonGen uses DPDK 17.11 or later, or an earlier version with
appropriate
> backports, this should work out of the box, otherwise only device 6003
> (virtual functions) are supported.
>
>
>
> On Tue, Jul 24, 2018 at 7:20 AM, Kash ***@***.***> wrote:
>
> > Hey,
> > I'm facing a similar issue. When I use MoonGen with the Netronome 4000
it
> > does not detect the device. I'm using today's master build. Any help
from
> > your side is appreciated.
> >
> > —
> > You are receiving this because you authored the thread.
> > Reply to this email directly, view it on GitHub
> > <#195 (comment)
>,
> > or mute the thread
> > <https://github.com/notifications/unsubscribe-auth/AS__
NZW4CTpzf5QqoR9eIXmyjdjQNLjbks5uJq6ngaJpZM4P6R5f>
> > .
> >
> —
> You are receiving this because you commented.
> Reply to this email directly, view it on GitHub, or mute the thread.
—
You are receiving this because you authored the thread.
Reply to this email directly, view it on GitHub
<#195 (comment)>,
or mute the thread
<https://github.com/notifications/unsubscribe-auth/AS__NZJufvljEnhAR71_QaKVySUC41xOks5uJs31gaJpZM4P6R5f>
.
|
|
Hey, Thanks for pointing out the difference in using the PF and VF. I was using the PF all this while and it was not working. I tried using the VF and it detects the ports. I tried to compile MoonGen with dpdk-17.11.1 but got some build error. I have not looked into it yet. |
|
I'll upgrade to some DPDK LTS release soonish as we are also currently facing a few issues with newer NICs |
|
Regarding pci.ids: we just use the version from http://pci-ids.ucw.cz/ Try to get your devices added there, we don't want to maintain our own list of PCI devices (we did it in the past and it was too annoying). See http://pci-ids.ucw.cz/mods/?action=help?help=newitem This will also fix it in other standard tools such as lspci |
|
Thanks Paul! Looking forward to the update. |
Fixes emmericp#195. Signed-off-by: Paul Chaignon <[email protected]>
The Netronome NFP PMD is part of the dpdk with MoonGen.
However, enabling the NFP PMD via the dpdk common_base config file does not link it into MoonGen / libmoon
Enabling the NFP PMD does compile librte_pmd_nfp.
Steps to verify:
clone latest MoonGen from github
git submodule update --init --recursive
Edit libmoon/deps/dpdk/config/common_base to enable NFP PMD
run build.sh
If I then try to run MoonGen on a Netronome device, I get "Found 0 usable devices"
(The huge tables were set up with setup_hugetlbfs.sh, the dpdk-conf.lua file was set up to include the proper PCI addresses in the white list, and the interfaces were bound to igb_uio
If I then manually link in librte_pmd_nfp by editing CMakeists.txt to add rte_pmd_nfp to target_link_librries (not the right way to do it, bt proves the point) and run build.sh again, the NFP intefaces work
So basically, enabling the NFP PMD sgould also link the driver into MoonGen
Expect: Enabling NFP PMD to also automatically link librte_pmd_nfp into libmoon / MoonGen
See: driver compiled but not linked
The text was updated successfully, but these errors were encountered: