mbox series

[RFC,v2,0/5] Split driver specific commands out of testpmd

Message ID 20220518194649.1868574-1-david.marchand@redhat.com (mailing list archive)
Headers
Series Split driver specific commands out of testpmd |

Message

David Marchand May 18, 2022, 7:46 p.m. UTC 
  Hello,

Following TB decision and recent discussions on the driver specific
commands in testpmd, here is a proposal on how the split could be done.

For now, this series simply moves the testpmd code in the driver
directory. The driver specific testpmd code is still compiled as part of
testpmd compilation via a global meson testpmd_driver_sources list.

TODO:
- ixgbe bypass commands in testpmd are "dead" code since switch to meson,
  as the RTE_LIBRTE_IXGBE_BYPASS define is not set while compiling testpmd.
  I am tempted to simply drop those, since no one complained about issue
  for the last two years. For now, this series reinstates them.

- this series keeps the command names as is. We could consolidate with a
  clear prefix so that users directly know they are exerting a special
  feature that makes no sense on other hw.
  For this, I had in mind introducing a "driver" top level prefix,
  where drivers would hook their specific stuff. For example:
  testpmd> driver ixgbe set vf split drop (port_id) (vf_id) (on|off)

- the documentation of those commands is left untouched, we should
  probably move any existing doc into the driver documentation itself.
  testpmd documentation could then have a generic
  "Driver specific commands" section pointing at all other driver docs.

Opinions?

Comments

Andrew Rybchenko May 20, 2022, 7:04 a.m. UTC | #1 
On 5/18/22 22:46, David Marchand wrote:
> Hello,
> 
> Following TB decision and recent discussions on the driver specific
> commands in testpmd, here is a proposal on how the split could be done.
> 
> For now, this series simply moves the testpmd code in the driver
> directory. The driver specific testpmd code is still compiled as part of
> testpmd compilation via a global meson testpmd_driver_sources list.

I like the approach suggested in the RFC. It is simple and it
makes absolutely clear who is responsible for corresponding
testpmd code maintenance.

> 
> TODO:
> - ixgbe bypass commands in testpmd are "dead" code since switch to meson,
>    as the RTE_LIBRTE_IXGBE_BYPASS define is not set while compiling testpmd.
>    I am tempted to simply drop those, since no one complained about issue
>    for the last two years. For now, this series reinstates them.

IMHO two years is sufficient period to say that it is unused
and remove it.

> 
> - this series keeps the command names as is. We could consolidate with a
>    clear prefix so that users directly know they are exerting a special
>    feature that makes no sense on other hw.
>    For this, I had in mind introducing a "driver" top level prefix,
>    where drivers would hook their specific stuff. For example:
>    testpmd> driver ixgbe set vf split drop (port_id) (vf_id) (on|off)

Good idea and I think that it should be fixed in a separate
follow up patches.

> 
> - the documentation of those commands is left untouched, we should
>    probably move any existing doc into the driver documentation itself.
>    testpmd documentation could then have a generic
>    "Driver specific commands" section pointing at all other driver docs.

Yes, it would be good to move documentation as well.

> 
> Opinions?
> 
>