On Tue, 23 Nov 2021 at 17:40, Richard Henderson
<[email protected]> wrote:
>
> We don't need a register that can live across the syscall;
> we only need a register that can live until the syscall.

What about the case where:
 * we execute the sc instruction (r11 trashed)
 * the syscall is one that from the host kernel point of
   view is restartable
 * the kernel arranges to restart the syscall by rewinding the
   PC to point to the start of the 'sc' instruction
 * our rewind_if_in_safe_syscall() rewinds PC further to
   point at safe_syscall_start
 * we want to use r11 again, but it was trashed in step 1
?

Put another way, this patch is effectively a revert of
commit 5d9f3ea081721, which was a fix to an observed bug.

-- PMM

Reply via email to