On Sun, 30 Jun 2002, Zwane Mwaikambo wrote:
> On Sun, 30 Jun 2002, Bartlomiej Zolnierkiewicz wrote:
>
> > > (1) ide-taskfile.c: ide_do_drive_cmd(..., ide_preempt) holds channel
> > > lock. Do not reacquire. NMI watchdog triggered by just booting
> > > computer with IDE cdrom.
> >
> > Mentioned in 95 changelog.
> > Already fixed in my tree, but thanks anyway.
>
> Hmm i just spent some time last night trying to go through possible
> paths for ide_do_drive_cmd to come up with a solution for that one, do you
> use some sort of SCM so that i can keep track of whats been covered?
Unfortunately no, I have only dialup...
> > Attached patch is next ide-clean patch pre-patch ;), just not to duplicate
> > efforts. Changelog is also included. As always use with care, standard
> > disclaimer apply.
>
> Thanks
>
> > And final note: I think that previous locking (2.4.x but ch->lock instead
> > of global io_request_lock) was well tuned and almost 100% correct.
> > Recent changes just made it worse (sorry Martin :) ).
> > Now even if we add unmasking IRQs with disabling currently handled IRQ, it
> > will be less friendlier to shared PCI interrupts (especially in PIO it
> > will be overkill to disable shared IRQ for handling PIO intr!),
> > so I want to revert to previous scheme...
>
> Agreed there, thanks again for the patches.
>
> Zwane Mwaikambo
>
> --
> http://function.linuxpower.ca
I will also forward You my reply to Petr, it shows my (correct?)
understanding of previous vs. actual IDE locking...
If you find any errors in thinking please let my now :)
Greets.
--
Bartlomiej