On Sun, Mar 27, 2005 at 12:42:56AM +1100, Nick Piggin wrote:
OK, thanks that would be good. You could well be right in your analysis.
May I suggest a possible avenue of investigation:
Yes, this patch seems to also be required, otherwise I see:
The above is with my fix to ARMs get_pgd_slow, which shows that we
accidentally freed the first entry in the L1 page table. With my
fix and your patch, low-vectored ARMs work again.
I don't think it'll be invasive to push my get_pgd_slow() fix before
these freepgt patches appear. For the record, this is the patch I'm
using at present. With a bit more effort, I could probably eliminate
pmd_alloc (and therefore the unnecessary spinlocking) here.