releases/6.7.12/loongarch-change-__my_cpu_offset-definition-to-avoid.patch

   1 From 0be97594f41cf83eccc512a5a625d84178a8feeb Mon Sep 17 00:00:00 2001
   2 From: Sasha Levin <sashal@kernel.org>
   3 Date: Tue, 19 Mar 2024 15:50:34 +0800
   4 Subject: LoongArch: Change __my_cpu_offset definition to avoid
   5  mis-optimization
   6
   7 From: Huacai Chen <chenhuacai@loongson.cn>
   8
   9 [ Upstream commit c87e12e0e8c1241410e758e181ca6bf23efa5b5b ]
  10
  11 From GCC commit 3f13154553f8546a ("df-scan: remove ad-hoc handling of
  12 global regs in asms"), global registers will no longer be forced to add
  13 to the def-use chain. Then current_thread_info(), current_stack_pointer
  14 and __my_cpu_offset may be lifted out of the loop because they are no
  15 longer treated as "volatile variables".
  16
  17 This optimization is still correct for the current_thread_info() and
  18 current_stack_pointer usages because they are associated to a thread.
  19 However it is wrong for __my_cpu_offset because it is associated to a
  20 CPU rather than a thread: if the thread migrates to a different CPU in
  21 the loop, __my_cpu_offset should be changed.
  22
  23 Change __my_cpu_offset definition to treat it as a "volatile variable",
  24 in order to avoid such a mis-optimization.
  25
  26 Cc: stable@vger.kernel.org
  27 Reported-by: Xiaotian Wu <wuxiaotian@loongson.cn>
  28 Reported-by: Miao Wang <shankerwangmiao@gmail.com>
  29 Signed-off-by: Xing Li <lixing@loongson.cn>
  30 Signed-off-by: Hongchen Zhang <zhanghongchen@loongson.cn>
  31 Signed-off-by: Rui Wang <wangrui@loongson.cn>
  32 Signed-off-by: Huacai Chen <chenhuacai@loongson.cn>
  33 Signed-off-by: Sasha Levin <sashal@kernel.org>
  34 ---
  35  arch/loongarch/include/asm/percpu.h | 7 ++++++-
  36  1 file changed, 6 insertions(+), 1 deletion(-)
  37
  38 diff --git a/arch/loongarch/include/asm/percpu.h b/arch/loongarch/include/asm/percpu.h
  39 index 9b36ac003f890..8f290e5546cf7 100644
  40 --- a/arch/loongarch/include/asm/percpu.h
  41 +++ b/arch/loongarch/include/asm/percpu.h
  42 @@ -29,7 +29,12 @@ static inline void set_my_cpu_offset(unsigned long off)
  43         __my_cpu_offset = off;
  44         csr_write64(off, PERCPU_BASE_KS);
  45  }
  46 -#define __my_cpu_offset __my_cpu_offset
  47 +
  48 +#define __my_cpu_offset                                        \
  49 +({                                                     \
  50 +       __asm__ __volatile__("":"+r"(__my_cpu_offset)); \
  51 +       __my_cpu_offset;                                \
  52 +})
  53
  54  #define PERCPU_OP(op, asm_op, c_op)                                    \
  55  static __always_inline unsigned long __percpu_##op(void *ptr,          \
  56 --
  57 2.43.0
  58