On Mon, Jun 15, 2015 at 05:04:38PM +0200, Stefan Sperling wrote:
> After suspend/resume I've found an iwm task sleeping forever. ps showed:
>
> 0 7029 0 0 -22 0 0 0 iwmau2 DK ?? 0:00.00 (iwmns)
>
> In this state, the driver awaits a notification from the firmware
> which it won't ever get after the system has resumed from sleep.
> The task running the state change to 'auth' state should abort itself.
>
> The diff below should fix this particular (hard to reproduce) problem.
> I don't particularly enjoy adding more hacks like this but this is at
> least consistent with how this driver currently operates.
PLease disregard for now. This is not a very good fix.
I'd like to attempt a broader approach.
>
> Index: if_iwm.c
> ===================================================================
> RCS file: /cvs/src/sys/dev/pci/if_iwm.c,v
> retrieving revision 1.45
> diff -u -p -r1.45 if_iwm.c
> --- if_iwm.c 15 Jun 2015 08:06:11 -0000 1.45
> +++ if_iwm.c 15 Jun 2015 11:59:54 -0000
> @@ -4984,7 +4984,7 @@ iwm_auth(struct iwm_softc *sc)
> struct iwm_node *in = (void *)ic->ic_bss;
> uint32_t duration;
> uint32_t min_duration;
> - int error;
> + int error, generation;
>
> in->in_assoc = 0;
>
> @@ -5025,6 +5025,7 @@ iwm_auth(struct iwm_softc *sc)
> 100 + in->in_ni.ni_intval);
> iwm_mvm_protect_session(sc, in, duration, min_duration, 500);
>
> + generation = sc->sc_generation;
> while (sc->sc_auth_prot != 2) {
> /*
> * well, meh, but if the kernel is sleeping for half a
> @@ -5038,7 +5039,9 @@ iwm_auth(struct iwm_softc *sc)
> sc->sc_auth_prot = 0;
> return EAUTH;
> }
> - tsleep(&sc->sc_auth_prot, 0, "iwmau2", 0);
> + tsleep(&sc->sc_auth_prot, 0, "iwmau2", hz);
> + if (generation != sc->sc_generation)
> + return (ENXIO);
> }
>
> return 0;
