more j820 hacking, not many fruits

[Oleg Gusev]

On Friday 13 May 2005 08:14, Faré wrote:
>
> Hopefully, we'd get 2.6 running at some time (who
> am I kidding?).

There is no other way.

>
> Looks like there's a race condition
> somewhere, probably some conflict (bad locking/masking/reserving)
> regarding interrupts or DMA, or maybe is trying something too
> PC-specific (hopefully, the incompatibility doesn't lie in the
> firmware). 

These cards work rock solid on zaurus/ipaq/etc.
Must be our buggy sa1101 pcmcia/irq support, most likely
some timing issue.

> I should probably resell this card,
> and try to get an older one... or buy a new computer.

Do you have a notebook ? You can switch on the debugging
info and trace the functions called.

>
> As long as I'm in the
> ramdisk, it runs fine, and I remember that's a symptom we've had in
> the past, but I'm incapable of determining what to fix. 

SA1101/PCMCIA/IRQ/IDE.

> Maybe I should 
> try to mix-and-match parts of our hh.org based kernel with the vrs1
> kernel until I locate the error. Sigh. Any taker? I'm getting tired of
> this.

We should split our CVS into "small" patches that make sence
in general, and get rid of the things that don't belong three.
Fully modularize the drivers. I don't remember if these
patches are committed or lost, but i could load/unload even the
keyboard driver from the serial console.
I have actually added some of "our" modifications to the 2.4 hh.org CVS tree.
We can and will do it with 2.6 in cooperation with the hh.org guys, where
the changes to other parts of the kernel are needed.

>
> Kernel driver debugging is a truly horrible experience.
> Compile-reboot-crash cycles. 

printk is you friend. Welcome to the monolithic kernel world :)

> Sigh. I'm getting tired of Linux, 
> actually.

Then switch to the porting of GNU HURD ;-)
Or to wince for the really exciting experience.

 Oleg.