On 31/10/2024 10:33, Will Deacon wrote: > Hi Jeff, > > Thanks for chiming in! > > On Wed, Oct 30, 2024 at 03:01:53PM -0700, Jeff Xu wrote: >> On Tue, Oct 29, 2024 at 7:46 AM Kevin Brodsky <kevin.brodsky@xxxxxxx> wrote: >>> TL;DR: reset POR_EL0 to "allow all" before writing the signal frame, >>> preventing spurious uaccess failures. > [...] > >>> @@ -924,12 +982,14 @@ SYSCALL_DEFINE0(rt_sigreturn) >>> if (!access_ok(frame, sizeof (*frame))) >>> goto badframe; >>> >>> - if (restore_sigframe(regs, frame)) >>> + if (restore_sigframe(regs, frame, &ua_state)) >>> goto badframe; >>> >>> if (restore_altstack(&frame->uc.uc_stack)) >>> goto badframe; >>> >> Do you need to move restore_altstack ahead of restore_sigframe? >> similar as x86 change [1], >> the discussion for this happened in [2] [3] >> >> [1] https://lore.kernel.org/lkml/20240802061318.2140081-5-aruna.ramakrishna@xxxxxxxxxx/ >> [2] https://lore.kernel.org/lkml/20240425210540.3265342-1-jeffxu@xxxxxxxxxxxx/ >> [3] https://lore.kernel.org/lkml/d0162c76c25bc8e1c876aebe8e243ff2e6862359.camel@xxxxxxxxx/ >> >>> + restore_user_access_state(&ua_state); > The POR isn't restored until here ^^^, so I _think_ restore_altstack() > is fine where it is. Kevin, can you confirm, please? Yes, that's correct, see my earlier reply [1]. - Kevin [1] https://lore.kernel.org/all/cd0e114d-57eb-4c90-bb6f-9abf0cc8920f@xxxxxxx/
