On Thu, 18 Aug 2022, Andy Shevchenko wrote: > On Tue, Aug 16, 2022 at 3:11 PM Ilpo Järvinen > <ilpo.jarvinen@xxxxxxxxxxxxxxx> wrote: > > > > ->set_termios() callchains input old ktermios (mainly used as fallback > > and feature compares). It is discarded right after returning from the > > calls and should therefore not be modified by drivers because any such > > change will get lost (instead, the changes must be applied to the new > > ktermios). While doing this patchset, I found a few such issues that > > were fixed earlier. > > > > Now enforce old ktermios constness. Another goodie is that "get" named > > functions in tty_baudrate.c that previously mucked with the old > > ktermios can no longer have such side-effects. I found out that the > > ktermios adjustments made were dead-code for all in-tree archs anyway. > > Reviewed-by: Andy Shevchenko <andy.shevchenko@xxxxxxxxx> > for all, but patch 3. > > I'm not sure we can blindly use old_termios settings, is there any > guarantee that old_termios _always_ has a correct baud rate settings? Old_termios is just the previous termios the port was using. If the baudrate in termios was invalid already by then, it's another issue that should be fixed (but I cannot see how that could occur assuming the validation works). How could it get wrong baud rate settings if the kernel sets (old_)termios (earlier) through these same paths (which validated it)? -- i.
