Correct interpretation of VF link-state=auto

Hi,

Regarding link-state attribute for a VF, 'man ip-link' says:
state auto|enable|disable - set the virtual link state as seen by the
specified VF. Setting to auto means a reflection of the PF link state,
enable lets the VF to communicate with other VFs on this host even if
the PF link state is down, disable causes the HW to drop any packets
sent by the VF.

However, I've seen that different interpretations are made about that
explanation, especially about "auto" configuration. It is not clear if
it should follow PF "physical link status" or PF "administrative link
status". With the latter, `ip set PF down` would put the VF down too,
but with the former you'd have to disconnect the physical port.

Thanks
-- 
Íñigo Huguet

On Tue, Jun 15, 2021 at 12:34 PM Íñigo Huguet [off-list ref] wrote:

Hi,

Regarding link-state attribute for a VF, 'man ip-link' says:
state auto|enable|disable - set the virtual link state as seen by the
specified VF. Setting to auto means a reflection of the PF link state,
enable lets the VF to communicate with other VFs on this host even if
the PF link state is down, disable causes the HW to drop any packets
sent by the VF.

However, I've seen that different interpretations are made about that
explanation, especially about "auto" configuration. It is not clear if
it should follow PF "physical link status" or PF "administrative link
status". With the latter, `ip set PF down` would put the VF down too,
but with the former you'd have to disconnect the physical port.

Thanks

Hello,

Sorry for bumping, anybody has any idea that could help?
-- 
Íñigo Huguet

On Tue, 15 Jun 2021 12:34:00 +0200 Íñigo Huguet wrote:

Hi,

Regarding link-state attribute for a VF, 'man ip-link' says:
state auto|enable|disable - set the virtual link state as seen by the
specified VF. Setting to auto means a reflection of the PF link state,
enable lets the VF to communicate with other VFs on this host even if
the PF link state is down, disable causes the HW to drop any packets
sent by the VF.

However, I've seen that different interpretations are made about that
explanation, especially about "auto" configuration. It is not clear if
it should follow PF "physical link status" or PF "administrative link
status". With the latter, `ip set PF down` would put the VF down too,
but with the former you'd have to disconnect the physical port.

Like all legacy SR-IOV networking the correct thing to do here is clear
as mud. I'd go for the link status of the PF netdev. If the netdev
cannot pass traffic (either for administrative or physical link reasons)
then VFs shouldn't talk either. But as I said, every vendor will have their
own interpretation, and different users may expect different things...

-----Original Message-----
From: Jakub Kicinski <kuba@kernel.org>
Sent: Friday, June 18, 2021 9:35 AM
To: Íñigo Huguet <redacted>
Cc: netdev@vger.kernel.org; Ivan Vecera <ivecera@redhat.com>; Edward Harold
Cree [off-list ref]; Dinan Gunawardena [off-list ref]; Pablo
Cascon [off-list ref]
Subject: Re: Correct interpretation of VF link-state=auto

On Tue, 15 Jun 2021 12:34:00 +0200 Íñigo Huguet wrote:

quoted

Hi,

Regarding link-state attribute for a VF, 'man ip-link' says:
state auto|enable|disable - set the virtual link state as seen by the
specified VF. Setting to auto means a reflection of the PF link state,
enable lets the VF to communicate with other VFs on this host even if
the PF link state is down, disable causes the HW to drop any packets
sent by the VF.

However, I've seen that different interpretations are made about that
explanation, especially about "auto" configuration. It is not clear if
it should follow PF "physical link status" or PF "administrative link
status". With the latter, `ip set PF down` would put the VF down too,
but with the former you'd have to disconnect the physical port.

Like all legacy SR-IOV networking the correct thing to do here is clear
as mud. I'd go for the link status of the PF netdev. If the netdev
cannot pass traffic (either for administrative or physical link reasons)
then VFs shouldn't talk either. But as I said, every vendor will have their
own interpretation, and different users may expect different things...

I like this interpretation too.. but I agree that it's unfortunately confusing and each vendor has done something different.. :(

On Fri, Jun 18, 2021 at 6:35 PM Jakub Kicinski [off-list ref] wrote:

Like all legacy SR-IOV networking the correct thing to do here is clear
as mud. I'd go for the link status of the PF netdev. If the netdev
cannot pass traffic (either for administrative or physical link reasons)
then VFs shouldn't talk either. But as I said, every vendor will have their
own interpretation, and different users may expect different things...

On Fri, Jun 18, 2021 at 9:10 PM Keller, Jacob E
[off-list ref] wrote:

I like this interpretation too.. but I agree that it's unfortunately confusing and each vendor has done something different.. :(

Thanks Jakub and Keller, at least now it's clear that it's not clear :P

Good enough info to move forward, for me.
-- 
Íñigo Huguet