[PATCH V3 2/5] riscv: atomic: Optimize acquire and release for AMO operations
Guo Ren
guoren at kernel.org
Thu Apr 21 20:43:02 PDT 2022
Ping Boqun & Daniel & Andrea,
Have you any comments on the patch? This revert 0123f4d76ca6
("riscv/spinlock: Strengthen implementations with fences").
But I think it's considerable because reducing the fence would gain
benefits in performance in our hardware.
In RISC-V ISA manual
- .aq: If the aq bit is set, then no later memory operations
in this RISC-V hart can be observed to take place
before the AMO.
- .rl: If the rl bit is set, then other RISC-V harts will not
observe the AMO before memory accesses preceding the
AMO in this RISC-V hart.
- .aqrl: Setting both the aq and the rl bit on an AMO makes the
sequence sequentially consistent, meaning that it cannot
be reordered with earlier or later memory operations
from the same hart.
On Wed, Apr 20, 2022 at 10:44 PM <guoren at kernel.org> wrote:
>
> From: Guo Ren <guoren at linux.alibaba.com>
>
> Current acquire & release implementations from atomic-arch-
> fallback.h are using __atomic_acquire/release_fence(), it cause
> another extra "fence r, rw/fence rw,w" instruction after/before
> AMO instruction. RISC-V AMO instructions could combine acquire
> and release in the instruction self which could reduce a fence
> instruction. Here is from RISC-V ISA 10.4 Atomic Memory
> Operations:
>
> To help implement multiprocessor synchronization, the AMOs
> optionally provide release consistency semantics.
> - .aq: If the aq bit is set, then no later memory operations
> in this RISC-V hart can be observed to take place
> before the AMO.
> - .rl: If the rl bit is set, then other RISC-V harts will not
> observe the AMO before memory accesses preceding the
> AMO in this RISC-V hart.
> - .aqrl: Setting both the aq and the rl bit on an AMO makes the
> sequence sequentially consistent, meaning that it cannot
> be reordered with earlier or later memory operations
> from the same hart.
>
> Signed-off-by: Guo Ren <guoren at linux.alibaba.com>
> Signed-off-by: Guo Ren <guoren at kernel.org>
> Cc: Palmer Dabbelt <palmer at dabbelt.com>
> Cc: Mark Rutland <mark.rutland at arm.com>
> Cc: Andrea Parri <parri.andrea at gmail.com>
> Cc: Dan Lustig <dlustig at nvidia.com>
> ---
> arch/riscv/include/asm/atomic.h | 64 ++++++++++++++++++++++++++++++++
> arch/riscv/include/asm/cmpxchg.h | 12 ++----
> 2 files changed, 68 insertions(+), 8 deletions(-)
>
> diff --git a/arch/riscv/include/asm/atomic.h b/arch/riscv/include/asm/atomic.h
> index ac9bdf4fc404..20ce8b83bc18 100644
> --- a/arch/riscv/include/asm/atomic.h
> +++ b/arch/riscv/include/asm/atomic.h
> @@ -99,6 +99,30 @@ c_type arch_atomic##prefix##_fetch_##op##_relaxed(c_type i, \
> return ret; \
> } \
> static __always_inline \
> +c_type arch_atomic##prefix##_fetch_##op##_acquire(c_type i, \
> + atomic##prefix##_t *v) \
> +{ \
> + register c_type ret; \
> + __asm__ __volatile__ ( \
> + " amo" #asm_op "." #asm_type ".aq %1, %2, %0" \
> + : "+A" (v->counter), "=r" (ret) \
> + : "r" (I) \
> + : "memory"); \
> + return ret; \
> +} \
> +static __always_inline \
> +c_type arch_atomic##prefix##_fetch_##op##_release(c_type i, \
> + atomic##prefix##_t *v) \
> +{ \
> + register c_type ret; \
> + __asm__ __volatile__ ( \
> + " amo" #asm_op "." #asm_type ".rl %1, %2, %0" \
> + : "+A" (v->counter), "=r" (ret) \
> + : "r" (I) \
> + : "memory"); \
> + return ret; \
> +} \
> +static __always_inline \
> c_type arch_atomic##prefix##_fetch_##op(c_type i, atomic##prefix##_t *v) \
> { \
> register c_type ret; \
> @@ -118,6 +142,18 @@ c_type arch_atomic##prefix##_##op##_return_relaxed(c_type i, \
> return arch_atomic##prefix##_fetch_##op##_relaxed(i, v) c_op I; \
> } \
> static __always_inline \
> +c_type arch_atomic##prefix##_##op##_return_acquire(c_type i, \
> + atomic##prefix##_t *v) \
> +{ \
> + return arch_atomic##prefix##_fetch_##op##_acquire(i, v) c_op I; \
> +} \
> +static __always_inline \
> +c_type arch_atomic##prefix##_##op##_return_release(c_type i, \
> + atomic##prefix##_t *v) \
> +{ \
> + return arch_atomic##prefix##_fetch_##op##_release(i, v) c_op I; \
> +} \
> +static __always_inline \
> c_type arch_atomic##prefix##_##op##_return(c_type i, atomic##prefix##_t *v) \
> { \
> return arch_atomic##prefix##_fetch_##op(i, v) c_op I; \
> @@ -140,22 +176,38 @@ ATOMIC_OPS(sub, add, +, -i)
>
> #define arch_atomic_add_return_relaxed arch_atomic_add_return_relaxed
> #define arch_atomic_sub_return_relaxed arch_atomic_sub_return_relaxed
> +#define arch_atomic_add_return_acquire arch_atomic_add_return_acquire
> +#define arch_atomic_sub_return_acquire arch_atomic_sub_return_acquire
> +#define arch_atomic_add_return_release arch_atomic_add_return_release
> +#define arch_atomic_sub_return_release arch_atomic_sub_return_release
> #define arch_atomic_add_return arch_atomic_add_return
> #define arch_atomic_sub_return arch_atomic_sub_return
>
> #define arch_atomic_fetch_add_relaxed arch_atomic_fetch_add_relaxed
> #define arch_atomic_fetch_sub_relaxed arch_atomic_fetch_sub_relaxed
> +#define arch_atomic_fetch_add_acquire arch_atomic_fetch_add_acquire
> +#define arch_atomic_fetch_sub_acquire arch_atomic_fetch_sub_acquire
> +#define arch_atomic_fetch_add_release arch_atomic_fetch_add_release
> +#define arch_atomic_fetch_sub_release arch_atomic_fetch_sub_release
> #define arch_atomic_fetch_add arch_atomic_fetch_add
> #define arch_atomic_fetch_sub arch_atomic_fetch_sub
>
> #ifndef CONFIG_GENERIC_ATOMIC64
> #define arch_atomic64_add_return_relaxed arch_atomic64_add_return_relaxed
> #define arch_atomic64_sub_return_relaxed arch_atomic64_sub_return_relaxed
> +#define arch_atomic64_add_return_acquire arch_atomic64_add_return_acquire
> +#define arch_atomic64_sub_return_acquire arch_atomic64_sub_return_acquire
> +#define arch_atomic64_add_return_release arch_atomic64_add_return_release
> +#define arch_atomic64_sub_return_release arch_atomic64_sub_return_release
> #define arch_atomic64_add_return arch_atomic64_add_return
> #define arch_atomic64_sub_return arch_atomic64_sub_return
>
> #define arch_atomic64_fetch_add_relaxed arch_atomic64_fetch_add_relaxed
> #define arch_atomic64_fetch_sub_relaxed arch_atomic64_fetch_sub_relaxed
> +#define arch_atomic64_fetch_add_acquire arch_atomic64_fetch_add_acquire
> +#define arch_atomic64_fetch_sub_acquire arch_atomic64_fetch_sub_acquire
> +#define arch_atomic64_fetch_add_release arch_atomic64_fetch_add_release
> +#define arch_atomic64_fetch_sub_release arch_atomic64_fetch_sub_release
> #define arch_atomic64_fetch_add arch_atomic64_fetch_add
> #define arch_atomic64_fetch_sub arch_atomic64_fetch_sub
> #endif
> @@ -178,6 +230,12 @@ ATOMIC_OPS(xor, xor, i)
> #define arch_atomic_fetch_and_relaxed arch_atomic_fetch_and_relaxed
> #define arch_atomic_fetch_or_relaxed arch_atomic_fetch_or_relaxed
> #define arch_atomic_fetch_xor_relaxed arch_atomic_fetch_xor_relaxed
> +#define arch_atomic_fetch_and_acquire arch_atomic_fetch_and_acquire
> +#define arch_atomic_fetch_or_acquire arch_atomic_fetch_or_acquire
> +#define arch_atomic_fetch_xor_acquire arch_atomic_fetch_xor_acquire
> +#define arch_atomic_fetch_and_release arch_atomic_fetch_and_release
> +#define arch_atomic_fetch_or_release arch_atomic_fetch_or_release
> +#define arch_atomic_fetch_xor_release arch_atomic_fetch_xor_release
> #define arch_atomic_fetch_and arch_atomic_fetch_and
> #define arch_atomic_fetch_or arch_atomic_fetch_or
> #define arch_atomic_fetch_xor arch_atomic_fetch_xor
> @@ -186,6 +244,12 @@ ATOMIC_OPS(xor, xor, i)
> #define arch_atomic64_fetch_and_relaxed arch_atomic64_fetch_and_relaxed
> #define arch_atomic64_fetch_or_relaxed arch_atomic64_fetch_or_relaxed
> #define arch_atomic64_fetch_xor_relaxed arch_atomic64_fetch_xor_relaxed
> +#define arch_atomic64_fetch_and_acquire arch_atomic64_fetch_and_acquire
> +#define arch_atomic64_fetch_or_acquire arch_atomic64_fetch_or_acquire
> +#define arch_atomic64_fetch_xor_acquire arch_atomic64_fetch_xor_acquire
> +#define arch_atomic64_fetch_and_release arch_atomic64_fetch_and_release
> +#define arch_atomic64_fetch_or_release arch_atomic64_fetch_or_release
> +#define arch_atomic64_fetch_xor_release arch_atomic64_fetch_xor_release
> #define arch_atomic64_fetch_and arch_atomic64_fetch_and
> #define arch_atomic64_fetch_or arch_atomic64_fetch_or
> #define arch_atomic64_fetch_xor arch_atomic64_fetch_xor
> diff --git a/arch/riscv/include/asm/cmpxchg.h b/arch/riscv/include/asm/cmpxchg.h
> index 12debce235e5..1af8db92250b 100644
> --- a/arch/riscv/include/asm/cmpxchg.h
> +++ b/arch/riscv/include/asm/cmpxchg.h
> @@ -52,16 +52,14 @@
> switch (size) { \
> case 4: \
> __asm__ __volatile__ ( \
> - " amoswap.w %0, %2, %1\n" \
> - RISCV_ACQUIRE_BARRIER \
> + " amoswap.w.aq %0, %2, %1\n" \
> : "=r" (__ret), "+A" (*__ptr) \
> : "r" (__new) \
> : "memory"); \
> break; \
> case 8: \
> __asm__ __volatile__ ( \
> - " amoswap.d %0, %2, %1\n" \
> - RISCV_ACQUIRE_BARRIER \
> + " amoswap.d.aq %0, %2, %1\n" \
> : "=r" (__ret), "+A" (*__ptr) \
> : "r" (__new) \
> : "memory"); \
> @@ -87,16 +85,14 @@
> switch (size) { \
> case 4: \
> __asm__ __volatile__ ( \
> - RISCV_RELEASE_BARRIER \
> - " amoswap.w %0, %2, %1\n" \
> + " amoswap.w.rl %0, %2, %1\n" \
> : "=r" (__ret), "+A" (*__ptr) \
> : "r" (__new) \
> : "memory"); \
> break; \
> case 8: \
> __asm__ __volatile__ ( \
> - RISCV_RELEASE_BARRIER \
> - " amoswap.d %0, %2, %1\n" \
> + " amoswap.d.rl %0, %2, %1\n" \
> : "=r" (__ret), "+A" (*__ptr) \
> : "r" (__new) \
> : "memory"); \
> --
> 2.25.1
>
--
Best Regards
Guo Ren
ML: https://lore.kernel.org/linux-csky/
More information about the linux-riscv
mailing list