On Tuesday 27 May 2008 22:03:32 Liu Yu wrote: > > > > Did you discover this through inspection, or do you have a > > guest that triggers this bug? I don't believe there is large > > page support for e500 Linux userspace (and the kernel > > mappings don't ever need invalidation)? > > > > I discovered it through inspection. As recently I'm porting your code for E500. > E500 support the instruction tlbiva, so I have to read this part of your code carefully. Ah, OK. I will commit your patch soon, but I do need to update the tree and clean up a number of half-working patches we tried in order to get good performance for a demo. :) > Yes. E500 doesn't have large mapping in userspace. Neither does 44x, right? Yes, but there is a patch that increases the base page size from 4K to 64K. We've played with this on the host to reduce the number of host TLB misses when emulating the large guest kernel mapping, and eventually we should try it in the guest too. Unfortunately the 64K patch has another issue: it dramatically increases memory allocations. On our 256MB boards we could boot ELDK fine, but actually ran out of memory trying to boot Ubuntu (on the host). -- Hollis Blanchard IBM Linux Technology Center -- To unsubscribe from this list: send the line "unsubscribe kvm-ppc" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
