Re: [PATCH net-next v2 0/2] flow_offload: add tc police parameters

On Thu, Feb 17, 2022 at 01:52:47PM +0200, Roi Dayan wrote:

On 2022-02-17 1:34 PM, Simon Horman wrote:

quoted

On Thu, Feb 17, 2022 at 08:28:01AM +0000, Jianbo Liu wrote:

quoted

As a preparation for more advanced police offload in mlx5 (e.g.,
jumping to another chain when bandwidth is not exceeded), extend the
flow offload API with more tc-police parameters. Adjust existing
drivers to reject unsupported configurations.

Hi,

I have a concern that
a) patch 1 introduces a facility that may break existing drivers; and
b) patch 2 then fixes this

I'd slightly prefer if the series was rearranged to avoid this problem.

...

Hi Simon,

It can't be rearranged as patch 2 can't compile without patch 1.
Patch 1 only adds more information passing to the driver.

The drivers functionality doesn't change. drivers today ignore
police information, like actions, and still being ignored after patch 1.

patch 2 updates the drivers to use that information instead of
ignoring it. so it fixes the drivers that without patch 1 can't be
fixed.

I think it would be possible, but...

On Thu, Feb 17, 2022 at 01:56:51PM +0200, Ido Schimmel wrote:

...

Not sure what you mean by the above. Patch #1 extends the flow offload
API with tc-police parameters that weren't communicated to drivers until
now. Drivers still ignore the new parameters after this patch. It is
only in patch #2 that these drivers reject configurations where the
parameters are set.

Therefore, the only breakage I see is the one that can happen after
patch #2: unaware user space that was installing actions that weren't
fully reflected to hardware.

If we want to be on the safe side, it is possible to remove the errors,
but keep the extack messages so that user space is at least somewhat
aware.

Yes, I see what you mean.
I'm now comfortable with the way this patchset is arranged.