Tim Waugh (email@example.com)
Sat, 18 Apr 1998 16:45:10 +0100 (BST)
On Sat, 18 Apr 1998, Andrea Arcangeli wrote:
> You are perfectly right. We are currently using the waitlist only for
> device that are sleeping in parport_claim_or_block() while we can use it
> for all device and run the parport_claim() that own the port inside
> parport_release() for all devices (as it' s now done only for the device
> sleeping in ..claim_or_block()). The use of the waitlist will allow a real
> round-robin scheduling between devices.
> This patch will change the waitlist usage making it very smarter.
I think this is probably the way things should be done, yes. However, it
won't be done that way in 2.1 because of the freeze. I would like to only
send real bug-fixes to Linus.
So, for example, if the "if (no-one claimed) schedule();" patch I sent to
this list not long ago makes sense, I would rather send Linus that than
introduce more complexity to the wait_queues. In 2.3 maybe we can do
things like that.
> For devices that are not sleeping in parport_claim_or_block() and that
> provides a wakeup callback we could assume that the parport_claim() is run
> from the callback, but since it' s so simple change that I updated all
> (I hope ;-) pardevice to the new regime.
> If somebody will use the old wakeup callback style, recalling
> parport_claim() inside the wakeup callback, this will not broke. Only the
> guy will see a ton of messages like this:
> parport0: ppa already owner
I'm less keen on this aspect. I think we should think more carefully
about _why_ someone would be using the non-blocking interface in the first
> I also removed the sleep in parport_pc since I really don' t think it can
I've had zero feedback on this yet, so I'll leave it in for now. If
people can compile without the sleep and let me know if they have
success/failure, that would be good.
> This patch also automatically fix the LP_PREEMPT() define in lp.h that
> currently is wrong (I noticed it thinking about your last email).
I've only skim-read your patch - please can you explain the problem with
the macro, and how it is fixed?
-- To unsubscribe, send mail to: firstname.lastname@example.org --
-- with the single word "unsubscribe" in the body of the message. --
This archive was generated by hypermail 2.0b3 on Wed 30 Dec 1998 - 10:17:38 EST