>>> On 10.03.16 at 01:08, <vikr...@codeaurora.org> wrote: > I believe there is a bug in setup_frametable_mappings() where the function > allocates pages for pagetables via alloc_boot_pages() but does not zero them > out. This results in a crash on Qualcomm systems when prefetching is enabled > since the processor is free to prefetch a location for which it finds a > 'valid' > translation in the page tables, but actually holds uninitialized memory. > > A memset to 0 of the variable 'second' for the nr_second pages that were > allocated fixes the problem. > > I didn't see a fix in master branch but may have missed a bug report on the > mailing list.
Why don't you submit your patch then? That said, I'm getting the impression that all three uses of alloc_boot_pages() in arm/mm.c have this problem, not just the one you mention. Jan _______________________________________________ Xen-devel mailing list Xen-devel@lists.xen.org http://lists.xen.org/xen-devel
