On Wed, Jun 25, 2025 at 5:24 PM Johannes Berg <johan...@sipsolutions.net> wrote: > > On Wed, 2025-06-25 at 12:27 +0200, Christophe Leroy wrote: > > > > Le 25/06/2025 à 11:52, Sabyrzhan Tasbolatov a écrit : > > > Replace the existing kasan_arch_is_ready() calls with kasan_enabled(). > > > Drop checks where the caller is already under kasan_enabled() condition. > > > > If I understand correctly, it means that KASAN won't work anymore > > between patch 2 and 9, because until the arch calls kasan_init_generic() > > kasan_enabled() will return false. > > > > The transition should be smooth and your series should remain bisectable. > > > > Or am I missing something ? > > > > Seems right to me, it won't work for architectures that define > kasan_arch_is_ready themselves I think? > > But since they have to literally #define it, could #ifdef on that > temporarily?
Thanks for catching it. You're right. I need to change the order of patches : - kasan: unify static kasan_flag_enabled across modes , then we should apply arch specific changes where we call kasan_init_generic in kasan_init. - kasan: replace kasan_arch_is_ready with kasan_enabled > > johannes
