Jason R Thorpe wrote:

> On Wed, Jun 21, 2000 at 03:47:25PM +1000, Simon Burge wrote:
> 
>  > I've got a program that's basically:
>  > 
>  > 	u_int32_t vgaIObase;
>  > 
>  > 	alpha_pci_io_enable(1);
>  > 	vgaIOBase = (inb(0x3cc) & 0x01) ? 0x3D0 : 0x3B0;
>  > 
>  > and it does roughly:
>  > 
>  > 	mmap(size = 0x40000000, addr = 0x8580000000) = 0x1603ec000
>  > 	alpha_pci_io_swiz: abw->abw_addr = 0x1603ec000
>  > 	alpha_pci_io_swiz: port = 0x1603f3980
>  > 	alpha_pci_io_swiz_inb: return (*port && 0xff);
>  > 
>  > and at the last line the machine hangs hard and needs a reset.
> 
> hm.  That's ... odd.  The swizzle code in libalpha certainly appears to
> have done the correct thing.  I guess the only issue is if mmap mapped
> the correct pages underneath that user VA.

Didn't think of checking that.  I'll check it out.

>  > If I add a kernel printf to show the K0SEG version (0x8580007980) I at
>  > least get a number back (0x3000067) but I've yet to figure out if the
>  > number is actually correct, although it being odd means that SuperProbe
>  > would think that the display is color and not mono which is right.
>  > 
>  > >From what I understand, while I'm just probing the I/O space I don't
>  > need to map anything in with alpha_pci_mem_map() yet, right?
> 
> No, that's correct.
> 
> Anyway, please get your mmap changes committed to the trunk, and I'll
> see what I can do to help you debug this.

Will do later today after I CVS update - my source tree is a couple of
days old.

Ta,
Simon.