Re: Atari TT

[Michael Schmitz <schmitz@mail.biophys.uni-duesseldorf.de>]

Thorsten,

> 1) CONFIG_MODULES=n
> 2) Cannot do that because, without the network, we cannot load
>    any modules because that’s where they are…

See my reply to Alan - vector 112 has got nothing to do with MFP. 
 
> >Try swapping this so it looks like such:
> 
> I’ll try with Alan’s patch first, unless you say either one’s
> better.

Try Alan's - mine is just cosmetic cleanup. 
 
> >Solution - use stram_pool=512k (or whatever size the frame buffer requires;
> >use a video mode to minimize framebuffer size). If you're already doing that
> >- minimize the kernel until you've got enough free ST-RAM. How large is the
> >kernel once it's been uncompressed? 
> 
> We suggested that already. I’m now making even smaller kernels,
> for testing…

The kernel should be small enough to fit into the ST-RAM, try the -s option.
If you then run into trouble with the EtherNEC driver (interrupt for stopped
card; that's what made me leave it out of the kernel) I'll add the
atari_mfptimer_startup() function to enable timer D chained interrupts as
well as the geenrating timer only once the first such interrupt has been
requested. It'll look somethink like

static unsigned int atari_mfptimer_startup(struct irq_data *data)
{
        int mfp_num = data->irq - IRQ_MFP_TIMER1;

	/* on startup of first timer D interrupt, start timer */
        if (!stmfp_base.int_mask) {
        	/* prepare timer D data for use as poll interrupt */
        	/* set Timer D data Register - needs to be > 0 */
        	st_mfp.tim_dt_d = 254;  /* < 100 Hz */
        	/* start timer D, div = 1:100 */
        	st_mfp.tim_ct_cd = (st_mfp.tim_ct_cd & 0xf0) | 0x6;
        	atari_enable_irq(IRQ_MFP_TIMD);
        }
        stmfp_base.int_mask |= 1 << mfp_num;
	return 0;
}

and remove the above block tweaking st_mfp.tim_* from atari_init_IRQ(). If
you feel creative. Not even compile tested, so beware. 
 
> >That's called stram_pool these days. -s should be used anyway, otherwise the
> >ST-RAM is not available for use by the kernlel _at all_. Stupid discontigmem
> >memory model (someone please implement sparsemem. Andreas? Roman?). 
> 
> Ouch.

Not sure sparsemem will do the trick - Geert did mention it at some state so
I'd thought I'd pas it on. I get a headache from looking at arch/m68k/mm. 

Cheers,

	Michael