Re: Re: another must-fix: major PS/2 mouse problem

On Mon, 11 Aug 2003, Chris Friesen wrote:
> Donald Becker wrote:
> > On Tue, 29 Jul 2003, Chris Friesen wrote:
> 
> >>You might want to bump that up a little bit.  Querying carrier signal on 
> >>a tulip chip is 100usecs with interrupts off.
> 
> > My drivers, include the Tulip, don't read serial MII registers in the
> > interrupt service during noraml operation.  That would be a Very Bad
...
> > Serial MII management operations should only be done during timer ticks
> > and user-level ioctl() operations.
> 
> Yep, this is from user-level. Basically we were querying the driver 
> through ioctl() to check whether the card has carrier or not.  The board 
> manufacturer didn't bother hooking up the interrupt line on the phy so 
> we had to poll the phy for carrier loss detection.  As part of the 
> ioctl() the normal code turned off interrupts while doing the MII stuff.

> We didn't like that very much, so I made use of the fact that the clock 
> pulse for the MII connection was user-driven (and hence not 
> timing-sensitive) to allow interrupts to be on while querying the 
> carrier.  Seems to work fine.

Ahhh, yes, interrupts are disabled with a spin_lock_irqsave() in
mdio_read(), rather than being blocked by doing mdio_read() in the
interrupt service routine.

This is a relatively recently change to the driver.  Originally the
driver took care to not lock mdio_read(), but this was the easist way to
avoid problems on SMP machines.  There are better ways, but they depend
on the kernel version / SMP semanatics that have changed over time.


-- 
Donald Becker				[email protected]
Scyld Computing Corporation		http://www.scyld.com
914 Bay Ridge Road, Suite 220		Scyld Beowulf cluster system
Annapolis MD 21403			410-990-9993

_______________________________________________
vortex mailing list
[email protected]
http://www.scyld.com/mailman/listinfo/vortex