On Thu, May 25, 2023 at 11:50 AM Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote: > So it would probably improve on those numbers a bit more, but you'd > still have the fundamental race where *serial* duplicates end up > always wasting CPU effort and temporary vmalloc space. The known failed boots are with KASAN with a large number of CPUs, so the value in the mitigation would be to help those boot until userspace fixes it and we have enough time for propagation. But since it is not a full proof solution, it may seem like an odd thing to have in place later and this being lost as odd tribal knowledge. I'd be in favor of only applying the mitigation if we really are chasing userspace to fix this, and we'd be OK in later removing it after userspace gets this fixed / propagated. If we're going to have userspace fix this, who is volunteering? Luis
