mii_bus problem

[Hans Petter Selasky]

On Friday 12 January 2007 14:37, Hans Petter Selasky wrote:
> Hi,
>
> Can someone here explain from where I can call "mii_pollstat()". I read
> somewhere on the internet that I cannot call this function from the so
> called "tick" routine, where "mii_tick()" is called. I currently have the
> following code:
>
> static void
> aue_cfg_tick(struct aue_softc *sc,
>              struct aue_config_copy *cc, u_int16_t refcount)
> {
>         printf("%s:%d\n", __FUNCTION__, __LINE__);
>
>         struct ifnet * ifp = sc->sc_ifp;
>         struct mii_data * mii = GET_MII(sc);
>
>         if ((cc == NULL) ||
>             (ifp == NULL) ||
>             (mii == NULL)) {
>             /* not ready */
>             return;
>         }
>
>         mii_tick(mii);
>
>         mii_pollstat(mii);
>
>         if ((sc->sc_flags & AUE_FLAG_WAIT_LINK) &&
>             (mii->mii_media_status & IFM_ACTIVE) &&
>             (IFM_SUBTYPE(mii->mii_media_active) != IFM_NONE)) {
>             sc->sc_flags &= ~AUE_FLAG_WAIT_LINK;
>         }
>
>         sc->sc_media_active = mii->mii_media_active;
>         sc->sc_media_status = mii->mii_media_status;
>
>         /* start stopped transfers, if any */
>
>         aue_start_transfers(sc);
>
>         return;
> }
>
>
> But after this function returns, the kernel corrupts, and it pagefaults in
> the "swi6 taskqueue". Commenting out "mii_pollstat()" fixes the kernel
> panic.

After some debugging I found out that the kernel corrupted because there was 
not a pointer to "struct ifnet" at the beginning of the softc. I see that 
some parts of the MII code uses an "EVIL HACK" to get the "struct ifnet" 
pointer. This should be fixed.

Now calling mii_pollstat() works.

--HPS