Re: 2.4 amiserial

[Roman Z. <zi...@fh...>]

Hi,

On Mon, 18 Dec 2000, Michel D=E4nzer wrote:

> > Linux more or less stops serial i/o when doing something with the disk.=
 I
> > have been told I can't compare these (but i just did ;). Is it due to
> > Linux's way of handling interrupts, or PCC<->m68k interaction, or somet=
hing
> > else?
>=20
> The m68k is not involved with Linux/APUS, except maybe for the interrupt
> emulation, where I imagine at least part of the problem might be.

The m68k cpu isn't doing anything right now. It's possible to dedicate the
m68k cpu to a certain task (e.g. serial interrupt handling), but that
would require a special driver. We would also need changes to the
bootloader to keep the m68k cpu alive, so it can be reactivated. Right now
just tell it to shut up with no interface to get it back.

> Besides, I seem to remember reading something about large parts of the Li=
nux
> IDE code running with interrupts off. That might explain why it seems to =
work
> much better for my with my 603e using SCSI.

There might be other parts involved here, e.g. the different mmu handling,
where parts probably run with mmu disabled (I have to check that...).
These parts are not used with 604.
On the other hand the ide driver does quite a lot at interrupt level, the
ide driver transfers uses pio mode to transfer the data (which also
involves quite a lot of mmu table lookups), where the scsi driver
probably uses dma.
We could test that, but I need an adventurous soul with a backup for
that. :) We can replace the irq spinlocks in the ide interrupt with the
nonirq spinlock version. As long as you use the modified driver only with
a single ide interface and on a single cpu system, it should be safe to
use. Anyone wants to try a patch for that? :-)

bye, Roman