Re: [PARPORT] Re: 2.1.96 patches [my not new regime]

Andrea Arcangeli (
Mon, 20 Apr 1998 21:36:02 +0200 (CEST)

On Fri, 17 Apr 1998, Tim Waugh wrote:

>On Thu, 16 Apr 1998, Andrea Arcangeli wrote:
>> No. You missed the underlined parport_claim() before wake_up the other
>> waiter:
>But it won't do for drivers who are using parport_claim and registering a
>wakeup (instead of using parport_claim_or_block), for whatever reason.

Arghh, asking me that I trusted that now a device that uses
parport_claim() and register a wakeup callback does not use the wait list
;-). This is false as Phil reminded me. The point is this:

        /* If this is the first time we tried to claim the port, register an
           interest. This is only allowed for devices sleeping in
           parport_claim_or_block(), or those with a wakeup function. */
        if (dev->waiting & 2 || dev->wakeup) {
                spin_lock_irqsave (&port->lock, flags);
                if (port->cad == NULL) {
                        /* The port got released in the meantime. */
                        spin_unlock_irqrestore (&port->lock, flags);

Your patch to pxx_yield() is _not_ needed. My latest parport patch is not
needed too, it can be considered a clean up, but we can sure live without
that too (feel free to apply it if you like since it is sure still right
and sounds more nice to me). The only good thing it does is that it allow
devices that don' t use parport_claim_or_block() or don't provide wakeup()
to use the waitlist too, but since there are not such kind of devices out
there (and I think there never will be one) this don't make real
differences now. We should instead increase timeslice to HZ/5 since it' s
probably this that given me a better scheduling since a ppa cycle probably
take more than 10/HZ sec.

So please take a look at my patch and apply it if you like (it pratically
don' t change nothing) but I can understand how the scheduling works more
simply (probably since _I_ wrote it ;-).

This is not more an issue I think.

Andrea[s] Arcangeli

-- To unsubscribe, send mail to: --
-- with the single word "unsubscribe" in the body of the message. --

This archive was generated by hypermail 2.0b3 on Wed 30 Dec 1998 - 10:17:39 EST