Hello, Sergey Bugaev, le dim. 05 mars 2023 23:12:42 +0300, a ecrit: > On Wed, Mar 1, 2023 at 9:44 AM Luca Dariz <l...@orpolo.org> wrote: > > Il 28/02/23 15:14, Sergey Bugaev ha scritto: > > > - nothing else is clobbered, in particular not rflags (or is the > > > "reserved" half of rflags clobbered?) and not the registers that > > > contain args > > > > if we follow the usual calling conventions, the registers containing > > args are clobbered. In fact, in the code I set them to 0 before sysret, > > to avoid the risk of them containing sensitive information from the > > syscall execution. > > Hmm, but if I do thread_get_state () while the thread is blocked > inside a syscall, I will see the original (pre-syscall) values of the > registers, right? This is important for inspecting the RPC being done, > see SYSCALL_EXAMINE and MSG_EXAMINE in glibc.
Indeed, we would want to see the values. > It would be easier if the registers were preserved after the syscall > too, but if that's not the case we should be able to work around that > in userspace. Specifically, trampoline.c wants to modify the (on-stack > for i386) values of 'option' and 'timeout' of a mach_msg call that > another thread is making, so that once INTR_MSG_TRAP returns, > _hurd_intr_rpc_mach_msg () will see the modified values. We could > maybe do the same on x86_64 by just modifying the corresponding > registers, but then we need to differentiate the registers having been > modified and now holding the new value, and having been just clobbered > by the real syscall. Indeed, it would be simpler for userland to just preserve the values. I.e. save them and restore them into i386_saved_state, like the callee-saved registers. Samuel
