On Wed, Feb 05, 2020 at 05:17:11PM -0800, Andy Lutomirski wrote: > On Wed, Feb 5, 2020 at 2:39 PM Kristen Carlson Accardi > <kristen@xxxxxxxxxxxxxxx> wrote: > > > > At boot time, find all the function sections that have separate .text > > sections, shuffle them, and then copy them to new locations. Adjust > > any relocations accordingly. > > > > > + sort(base, num_syms, sizeof(int), kallsyms_cmp, kallsyms_swp); > > Hah, here's a huge bottleneck. Unless you are severely > memory-constrained, never do a sort with an expensive swap function > like this. Instead allocate an array of indices that starts out as > [0, 1, 2, ...]. Sort *that* where the swap function just swaps the > indices. Then use the sorted list of indices to permute the actual > data. The result is exactly one expensive swap per item instead of > one expensive swap per swap. I think there are few places where memory-vs-speed need to be examined. I remain surprised about how much memory the entire series already uses (58MB in my local tests), but I suspect this is likely dominated by the two factors: a full copy of the decompressed kernel, and that the "allocator" in the image doesn't really implement free(): https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git/tree/include/linux/decompress/mm.h#n55 -- Kees Cook
