2020-04-05 17:45:38

by Christophe Leroy

Subject: [RFC PATCH v2 13/13] powerpc/kernel: Do not use READ_ONCE() to access current thread_info flags

current is a volatile pointer hold by r2 register.

READ_ONCE() is not required.

Before: 327 cycles on null_syscall
Before: 325 cycles on null_syscall

Signed-off-by: Christophe Leroy <[email protected]>
---
arch/powerpc/kernel/syscall.c | 4 ++--
1 file changed, 2 insertions(+), 2 deletions(-)

diff --git a/arch/powerpc/kernel/syscall.c b/arch/powerpc/kernel/syscall.c
index 34fd66fd11a2..3d2f285b8a01 100644
--- a/arch/powerpc/kernel/syscall.c
+++ b/arch/powerpc/kernel/syscall.c
@@ -164,7 +164,7 @@ notrace unsigned long syscall_exit_prepare(unsigned long r3,
again:
#endif
local_irq_disable();
- ti_flags = READ_ONCE(*ti_flagsp);
+ ti_flags = current_thread_info()->flags;
while (unlikely(ti_flags & (_TIF_USER_WORK_MASK & ~_TIF_RESTORE_TM))) {
local_irq_enable();
if (ti_flags & _TIF_NEED_RESCHED) {
@@ -180,7 +180,7 @@ notrace unsigned long syscall_exit_prepare(unsigned long r3,
do_notify_resume(regs, ti_flags);
}
local_irq_disable();
- ti_flags = READ_ONCE(*ti_flagsp);
+ ti_flags = current_thread_info()->flags;
}

if (IS_ENABLED(CONFIG_PPC_BOOK3S) && IS_ENABLED(CONFIG_PPC_FPU)) {
--
2.25.0

2020-04-06 01:41:32

by Nicholas Piggin

Subject: Re: [RFC PATCH v2 13/13] powerpc/kernel: Do not use READ_ONCE() to access current thread_info flags

Christophe Leroy's on April 6, 2020 3:44 am:
> current is a volatile pointer hold by r2 register.
>
> READ_ONCE() is not required.
>
> Before: 327 cycles on null_syscall
> Before: 325 cycles on null_syscall

Patches against Michael's next-test branch have a few small changes
here. I need to look at the fallout on 64-bit code, although most
likely it's not measurable so if it saves you a few cycles we can
lean toward what ppc32 prefers.

Same with the compat test.

Thanks,
Nick