On Wed, Oct 16, 2024 at 8:50 PM Ard Biesheuvel <ardb@xxxxxxxxxx> wrote: > Yes, so we rely on the fact that changes to the vmalloc area and > changes to the associated shadow mappings always occur in combination, > right? Yes otherwise it is pretty much the definition of a KASAN violation. Mostly it "just works" because all low-level operations emitted by the compiler and all memcpy() (etc) are patched to do any memory access in tandem, this vmalloc_seq-thing was a big confusion for me. I'll send out the revised patches so people can test! Yours, Linus Walleij
