On Sun, Apr 08, 2007 at 09:20:31PM +0400, Sergei Shtylyov wrote:
> > static struct plat_serial8250_port pcit_cplus_data[] = {
> >- PORT(0x3f8, 4),
> >+ PORT(0x3f8, 0),
> > PORT(0x2f8, 3),
> > PORT(0x3e8, 4),
> > PORT(0x2e8, 3),
>
> Hm, what is that -- UART #1 without IRQ?
workaround for not fully working interrupts on UART1. IRQ 0 means
polling. Read the source.
> > static struct resource sni_io_resource = {
> >- .start = 0x00001000UL,
> >+ .start = 0x00000000UL,
> > .end = 0x03bfffffUL,
> >- .name = "PCIT IO MEM",
> >+ .name = "PCIT IO",
> > .flags = IORESOURCE_IO,
> > };
>
> Why us this necessary, only beacuse compatible peripherals are behind
> PCI?
> EISA is behind PCI as well, yet you're setting PCIBIOS_MIN_IO to 0x9000.
> Does this all really make sense? :-/
it does, how about reading the PCI code ?
EISA IO address space is 0x0000 - 0xffff, so this IO addresses need to
be forwarded by the PCI host bridge. PCIBIOS_MIN_IO is for the PCI
address assignment code, and tells this code to start allocating IO
space starting at 0x9000. This is needed because the pci eisa code
will use n + 0x1000 as EISA slot base addresses, which gives 0x8000
for the 8th (last) slot. So it's IMHO a good idea to avoid collisions
between EISA and PCI for IO space.
> This is certainly *not* a PCI or [E]ISA resource. It's decoded by the
> *host* bridge.
so ? It's an IO address no device should use, because it won't work.
Therefore mark it busy. That's all the code does.
> > .start = 0xcfc,
> > .end = 0xcff,
> > .name = "PCI config data",
>
> Well, why not just join them into one?
what's your point ? This stuff is all about giving some hints and
avoiding address assignment collisions. I could just drop the whole
table and nothing will change, because the PCI code doesn't assign
IO addresses below 0x9000. Fine with me, but I think it doesn't hurt
to know, what IO addresses are used for some stuff.
Thomas.
--
Crap can work. Given enough thrust pigs will fly, but it's not necessary a
good idea. [ RFC1925, 2.3 ]
