On Wed, Jan 09, 2002 at 12:08:47AM -0700, Jason Gunthorpe wrote:
> I just noticed that in setup.c there is this little bit:
>
> s = read_32bit_cp0_register(CP0_STATUS);
> s &= ~(ST0_CU1|ST0_CU2|ST0_CU3|ST0_KX|ST0_SX|ST0_FR);
> s |= ST0_CU0;
> write_32bit_cp0_register(CP0_STATUS, s);
>
> And it doesn't mask off ST0_UX - is this an oversight? With my RM7K the
> kernel is called with ST0_UX set, and since it doesn't clear it the XTLB
> handler is called - which faults things..
On all firmware I've made the 32-bit kernel run on it is invoked with UX
cleared so I just didn't bother to clear it myself.
> So, would this patch be appropriate in general:
>
> --- setup.c 2001/12/02 11:34:38 1.96
> +++ setup.c 2002/01/09 08:05:43
> @@ -558,7 +558,7 @@
>
> /* Disable coprocessors and set FPU for 16 FPRs */
> s = read_32bit_cp0_register(CP0_STATUS);
> - s &= ~(ST0_CU1|ST0_CU2|ST0_CU3|ST0_KX|ST0_SX|ST0_FR);
> + s &= ~(ST0_CU1|ST0_CU2|ST0_CU3|ST0_UX|ST0_KX|ST0_SX|ST0_FR);
> s |= ST0_CU0;
> write_32bit_cp0_register(CP0_STATUS, s);
>
> or is it better to make the xtlb handler work in the 32 bit case?
No, your patch is the right thing. Enabeling UX would also permit the
use of 64-bit instructions and that wouldn't work on the 32-bit kernel.
Ralf
|