Re: [PATCH net] net: Use NAPI_* in test_bit when stopping napi kthread

[Jakub Kicinski <kuba@kernel.org>]

On Thu, 11 Sep 2025 07:31:11 -0700 Samiullah Khawaja wrote:
> > > Is this basically addressing the bug that Martin run into?  
> > Not really. That one was because the busy polling bit remained set
> > during kthread stop. Basically in this function when we unset
> > STATE_THREADED, we also needed to unset STATE_THREADED_BUSY_POLL.
> >
> > @@ -7000,7 +7002,8 @@ static void napi_stop_kthread(struct napi_struct *napi)
> >                  */
> >                 if ((val & NAPIF_STATE_SCHED_THREADED) ||
> >                     !(val & NAPIF_STATE_SCHED)) {
> > -                       new = val & (~NAPIF_STATE_THREADED);
> > +                       new = val & (~(NAPIF_STATE_THREADED |
> > +                                      NAPIF_STATE_THREADED_BUSY_POLL));
> >  
> Just to add to my last email: I did find this test_bit issue while
> working on the fix for that other problem.

I see, makes sense. I was curious whether your previous posting would
work with just this fix. I guess we'll wait until next week 'cause
this fix didn't make it to today's PR in time :(