Re: CVS commit: src/sys/dev/usb

[Nick Hudson <nick.hudson%gmx.co.uk@localhost>]

On 24/06/2019 10:40, Ryota Ozaki wrote:
> On Mon, Jun 24, 2019 at 6:27 PM matthew green <mrg%eterna.com.au@localhost> wrote:
> > > Only KERNEL_LOCK (and some splsoftnet) is required for the network stack
> > > now.  Remaining splnets are for network drivers.  (softnet_lock is also required
> > > in some cases but it's another story...)
> >
> > great!  i studied the code and i couldn't find any issues
> > in any of the relevant paths, so i'm glad to hear it's
> > supposed to be like this.
> >
> > for one particular case (ether_ioctl) how is this diff:
> >
> > - * Common ioctls for Ethernet interfaces.  Note, we must be
> > - * called at splnet().
> > + * Common ioctls for Ethernet interfaces.
> > + *
> > + * Non IFEF_MPSAFE drivers must call this function at at least called
> > + * at splsoftnet().
> >
> > or should they also be with kernel lock?
>
> Yes.
>
> Also I think splnet is still needed for ether_ioctl for sure because
> it has to care not only the network stack but also network drivers.

If a driver is made MP safe and regardless of if it is marked
IFEF_MPSAFE then I think the splnet part can be dropped.  That is,
struct ifnet is either KERNEL_LOCK / IFNET_LOCK AND driver mutex
protected.  The driver mutexes provide appropriate mutex exclusion
between threads and interrupts.

Would you agree?

Thanks,
Nick