arm64: Mark functions using explicit register variables as '__always_inline'
authorWill Deacon <will@kernel.org>
Tue, 1 Oct 2019 10:43:13 +0000 (11:43 +0100)
committerWill Deacon <will@kernel.org>
Fri, 4 Oct 2019 10:15:42 +0000 (11:15 +0100)
As of ac7c3e4ff401 ("compiler: enable CONFIG_OPTIMIZE_INLINING forcibly"),
inline functions are no longer annotated with '__always_inline', which
allows the compiler to decide whether inlining is really a good idea or
not. Although this is a great idea on paper, the reality is that AArch64
GCC prior to 9.1 has been shown to get confused when creating an
out-of-line copy of a function passing explicit 'register' variables
into an inline assembly block:

  https://gcc.gnu.org/bugzilla/show_bug.cgi?id=91111

It's not clear whether this is specific to arm64 or not but, for now,
ensure that all of our functions using 'register' variables are marked
as '__always_inline' so that the old behaviour is effectively preserved.

Hopefully other architectures are luckier with their compilers.

Cc: Masahiro Yamada <yamada.masahiro@socionext.com>
Cc: Nicolas Saenz Julienne <nsaenzjulienne@suse.de>
Cc: Arnd Bergmann <arnd@arndb.de>
Cc: Russell King <linux@armlinux.org.uk>
Cc: Catalin Marinas <catalin.marinas@arm.com>
Cc: Nick Desaulniers <ndesaulniers@google.com>
Signed-off-by: Will Deacon <will@kernel.org>
arch/arm64/include/asm/atomic_lse.h

index c6bd87d..574808b 100644 (file)
@@ -321,7 +321,8 @@ static inline s64 __lse_atomic64_dec_if_positive(atomic64_t *v)
 }
 
 #define __CMPXCHG_CASE(w, sfx, name, sz, mb, cl...)                    \
-static inline u##sz __lse__cmpxchg_case_##name##sz(volatile void *ptr, \
+static __always_inline u##sz                                           \
+__lse__cmpxchg_case_##name##sz(volatile void *ptr,                     \
                                              u##sz old,                \
                                              u##sz new)                \
 {                                                                      \
@@ -362,7 +363,8 @@ __CMPXCHG_CASE(x,  ,  mb_, 64, al, "memory")
 #undef __CMPXCHG_CASE
 
 #define __CMPXCHG_DBL(name, mb, cl...)                                 \
-static inline long __lse__cmpxchg_double##name(unsigned long old1,     \
+static __always_inline long                                            \
+__lse__cmpxchg_double##name(unsigned long old1,                                \
                                         unsigned long old2,            \
                                         unsigned long new1,            \
                                         unsigned long new2,            \