Re: Sleeping function called from illegal context...

[Alan Cox]

On Sat, 2002-09-28 at 04:21, Andre Hedrick wrote:
> There is an issue of interrupt acknowledgement and when one can pre-empt.
> I would like to resolve the issue, but I need a global caller/notifier api
> from you so I can block IO in a safe spot on the 'data transfer' state
> bar.  Yeah, blah blah on underfined terms.
> 
> Some how I need to figure out how to address the pre-empt and keep the
> driver data stable.  Initially I would suggest throttling back on the
> request size to maybe 4k or 8k regardless.  I may not sound right but it
> will serve the purpose.

For things like old old broken PIO where interrupting the data stream
screws up the data thats actually already covered. Pre-empt does
actually do some things sensibly, and one of them is that when you hold
a lock or disable irq you also disable pre-empt. That means hdparm -u0
PIO interface code is still going to do the right thing

Reminds me though Robert (and Jeff)

drivers/net/8390.c still needs ei_start_xmit fixing

pre-emption should be disabled between

       /* Mask interrupts from the ethercard.
           SMP: We have to grab the lock here otherwise the IRQ handler

and
       disable_irq_nosync(dev->irq);

        spin_lock(&ei_local->page_lock);

So that we don't leave the IRQ disabled due to pre-emption

(that code is wonderfully deranged but its the only way to make 8390
based chips not screw up things like serial I/O on a SMP box)