[RFC,net,0/1] Fix netdevim to correctly mark NAPI IDs

Message

Joe Damato March 29, 2025, midnight UTC

Greetings:

Sending this as an RFC because I'm not sure if this is net or net-next
material?

If users are using netdevsim out in the wild to test userland apps, this
might be an net material, but LMK.

If this is net material: I can resend without the cover letter and when
net-next re-opens I'll update the busy_poller.c test to check the NAPI
ID isn't zero.

If this net-next material: I'll wait until it reopens and send this
patch + an update to busy_poller.c as described above.

Thanks,
Joe

Joe Damato (1):
  netdevsim: Mark NAPI ID on skb in nsim_rcv

 drivers/net/netdevsim/netdev.c | 3 ++-
 1 file changed, 2 insertions(+), 1 deletion(-)


base-commit: 2ea396448f26d0d7d66224cb56500a6789c7ed07

Comments

Jakub Kicinski March 31, 2025, 8:36 p.m. UTC | #1

On Sat, 29 Mar 2025 00:00:28 +0000 Joe Damato wrote:
> If this net-next material: I'll wait until it reopens and send this
> patch + an update to busy_poller.c as described above.

Let's stick to net-next. Would it be possible / make sense to convert
the test to Python and move it to drivers/net ?

Joe Damato March 31, 2025, 10:32 p.m. UTC | #2

On Mon, Mar 31, 2025 at 01:36:15PM -0700, Jakub Kicinski wrote:
> On Sat, 29 Mar 2025 00:00:28 +0000 Joe Damato wrote:
> > If this net-next material: I'll wait until it reopens and send this
> > patch + an update to busy_poller.c as described above.
> 
> Let's stick to net-next. 

Sure, sounds good. I'll drop the fixes tag when I resend when
net-next is open, of course.

> Would it be possible / make sense to convert the test to Python
> and move it to drivers/net ?

Hmm. We could; I think originally the busy_poller.c test was added
because it was requested by Paolo for IRQ suspension and netdevsim
was the only option that I could find that supported NAPI IDs at the
time.

busy_poller.c itself seems more like a selftests/net thing since
it's testing some functionality of the core networking code.

Maybe mixing the napi_id != 0 test into busy_poller.c is the wrong
way to go at a higher level. Maybe there should be a test for
netdevsim itself that checks napi_id != 0 and that test would make
more sense under drivers/net vs mixing a check into busy_poller.c?

Jakub Kicinski March 31, 2025, 11:39 p.m. UTC | #3

On Mon, 31 Mar 2025 15:32:09 -0700 Joe Damato wrote:
> > Would it be possible / make sense to convert the test to Python
> > and move it to drivers/net ?  
> 
> Hmm. We could; I think originally the busy_poller.c test was added
> because it was requested by Paolo for IRQ suspension and netdevsim
> was the only option that I could find that supported NAPI IDs at the
> time.
> 
> busy_poller.c itself seems more like a selftests/net thing since
> it's testing some functionality of the core networking code.

I guess in my mind busy polling is tied to having IRQ-capable device.
Even if bulk of the logic resides in the core.

> Maybe mixing the napi_id != 0 test into busy_poller.c is the wrong
> way to go at a higher level. Maybe there should be a test for
> netdevsim itself that checks napi_id != 0 and that test would make
> more sense under drivers/net vs mixing a check into busy_poller.c?

Up to you. The patch make me wonder how many other corner cases / bugs
we may be missing in drivers. And therefore if we shouldn't flesh out
more device-related tests. But exercising the core code makes sense
in itself so no strong feelings.