Le 06/04/2020 à 03:29, Nicholas Piggin a écrit :
Christophe Leroy's on April 6, 2020 3:44 am:
When r3 is not modified, reload it from regs->orig_r3 to free
volatile registers. This avoids a stack frame for the likely part
of syscall_call_exception()
Before : 353 cycles on null_syscall
After : 347 cycles on null_syscall
[...]
Signed-off-by: Christophe Leroy <christophe.le...@c-s.fr>
---
arch/powerpc/kernel/syscall.c | 2 ++
1 file changed, 2 insertions(+)
diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
index 69d75fc4a5eb..630c423e089a 100644
--- a/arch/powerpc/kernel/syscall.c
+++ b/arch/powerpc/kernel/syscall.c
@@ -91,6 +91,8 @@ notrace long system_call_exception(long r3, long r4, long r5,
} else if (unlikely(r0 >= NR_syscalls)) {
return -ENOSYS;
+ } else {
+ r3 = regs->orig_gpr3;
}
So this just gives enough volatiles to avoid spilling to stack? I wonder
about other various options here if they would cause a spill anyway.
Interesting optimisation, it would definitely need a comment. Would be
nice if we had a way to tell the compiler that a local can be reloaded
from a particular address.
Ok, comment added.
Christophe
