public inbox for linux-kernel@vger.kernel.org
 help / color / mirror / Atom feed
From: Boqun Feng <boqun.feng@gmail.com>
To: guoren@kernel.org
Cc: arnd@arndb.de, mark.rutland@arm.com, peterz@infradead.org,
	will@kernel.org, linux-arch@vger.kernel.org,
	linux-kernel@vger.kernel.org, linux-csky@vger.kernel.org,
	Guo Ren <guoren@linux.alibaba.com>
Subject: Re: [PATCH V3 1/3] csky: cmpxchg: Optimize with acquire & release
Date: Fri, 22 Apr 2022 11:20:38 +0800	[thread overview]
Message-ID: <YmIfBmisFVBu8Z5R@tardis> (raw)
In-Reply-To: <20220417083204.2048364-2-guoren@kernel.org>

[-- Attachment #1: Type: text/plain, Size: 5672 bytes --]

On Sun, Apr 17, 2022 at 04:32:02PM +0800, guoren@kernel.org wrote:
> From: Guo Ren <guoren@linux.alibaba.com>
> 
> Optimize arch_xchg|cmpxchg|cmpxchg_local with ASM acquire|release
> instructions instead of previous C based.
> 
> Important reference comment by Rutland:
> 8e86f0b409a4 ("arm64: atomics: fix use of acquire + release for
> full barrier semantics")
> 
> Link: https://lore.kernel.org/linux-riscv/CAJF2gTSAxpAi=LbAdu7jntZRUa=-dJwL0VfmDfBV5MHB=rcZ-w@mail.gmail.com/T/#m27a0f1342995deae49ce1d0e1f2683f8a181d6c3
> Signed-off-by: Guo Ren <guoren@linux.alibaba.com>
> Signed-off-by: Guo Ren <guoren@kernel.org>
> Cc: Mark Rutland <mark.rutland@arm.com>
> ---
>  arch/csky/include/asm/barrier.h | 11 +++---
>  arch/csky/include/asm/cmpxchg.h | 64 ++++++++++++++++++++++++++++++---
>  2 files changed, 67 insertions(+), 8 deletions(-)
> 
> diff --git a/arch/csky/include/asm/barrier.h b/arch/csky/include/asm/barrier.h
> index f4045dd53e17..fb63335ffa33 100644
> --- a/arch/csky/include/asm/barrier.h
> +++ b/arch/csky/include/asm/barrier.h
> @@ -37,17 +37,21 @@
>   * bar.brar
>   * bar.bwaw
>   */
> +#define ACQUIRE_FENCE		".long 0x8427c000\n"
> +#define RELEASE_FENCE		".long 0x842ec000\n"
> +#define FULL_FENCE		".long 0x842fc000\n"
> +
>  #define __bar_brw()	asm volatile (".long 0x842cc000\n":::"memory")
>  #define __bar_br()	asm volatile (".long 0x8424c000\n":::"memory")
>  #define __bar_bw()	asm volatile (".long 0x8428c000\n":::"memory")
>  #define __bar_arw()	asm volatile (".long 0x8423c000\n":::"memory")
>  #define __bar_ar()	asm volatile (".long 0x8421c000\n":::"memory")
>  #define __bar_aw()	asm volatile (".long 0x8422c000\n":::"memory")
> -#define __bar_brwarw()	asm volatile (".long 0x842fc000\n":::"memory")
> -#define __bar_brarw()	asm volatile (".long 0x8427c000\n":::"memory")
> +#define __bar_brwarw()	asm volatile (FULL_FENCE:::"memory")
> +#define __bar_brarw()	asm volatile (ACQUIRE_FENCE:::"memory")
>  #define __bar_bwarw()	asm volatile (".long 0x842bc000\n":::"memory")
>  #define __bar_brwar()	asm volatile (".long 0x842dc000\n":::"memory")
> -#define __bar_brwaw()	asm volatile (".long 0x842ec000\n":::"memory")
> +#define __bar_brwaw()	asm volatile (RELEASE_FENCE:::"memory")
>  #define __bar_brar()	asm volatile (".long 0x8425c000\n":::"memory")
>  #define __bar_brar()	asm volatile (".long 0x8425c000\n":::"memory")
>  #define __bar_bwaw()	asm volatile (".long 0x842ac000\n":::"memory")
> @@ -56,7 +60,6 @@
>  #define __smp_rmb()	__bar_brar()
>  #define __smp_wmb()	__bar_bwaw()
>  
> -#define ACQUIRE_FENCE		".long 0x8427c000\n"
>  #define __smp_acquire_fence()	__bar_brarw()
>  #define __smp_release_fence()	__bar_brwaw()
>  
> diff --git a/arch/csky/include/asm/cmpxchg.h b/arch/csky/include/asm/cmpxchg.h
> index d1bef11f8dc9..06c550448bf1 100644
> --- a/arch/csky/include/asm/cmpxchg.h
> +++ b/arch/csky/include/asm/cmpxchg.h
> @@ -64,15 +64,71 @@ extern void __bad_xchg(void);
>  #define arch_cmpxchg_relaxed(ptr, o, n) \
>  	(__cmpxchg_relaxed((ptr), (o), (n), sizeof(*(ptr))))
>  
> -#define arch_cmpxchg(ptr, o, n) 				\
> +#define __cmpxchg_acquire(ptr, old, new, size)			\
>  ({								\
> +	__typeof__(ptr) __ptr = (ptr);				\
> +	__typeof__(new) __new = (new);				\
> +	__typeof__(new) __tmp;					\
> +	__typeof__(old) __old = (old);				\
> +	__typeof__(*(ptr)) __ret;				\
> +	switch (size) {						\
> +	case 4:							\
> +		asm volatile (					\
> +		"1:	ldex.w		%0, (%3) \n"		\
> +		"	cmpne		%0, %4   \n"		\
> +		"	bt		2f       \n"		\
> +		"	mov		%1, %2   \n"		\
> +		"	stex.w		%1, (%3) \n"		\
> +		"	bez		%1, 1b   \n"		\
> +		ACQUIRE_FENCE					\
> +		"2:				 \n"		\
> +			: "=&r" (__ret), "=&r" (__tmp)		\
> +			: "r" (__new), "r"(__ptr), "r"(__old)	\
> +			:);					\
> +		break;						\
> +	default:						\
> +		__bad_xchg();					\
> +	}							\
> +	__ret;							\
> +})
> +
> +#define arch_cmpxchg_acquire(ptr, o, n) \
> +	(__cmpxchg_acquire((ptr), (o), (n), sizeof(*(ptr))))
> +
> +#define __cmpxchg(ptr, old, new, size)				\
> +({								\
> +	__typeof__(ptr) __ptr = (ptr);				\
> +	__typeof__(new) __new = (new);				\
> +	__typeof__(new) __tmp;					\
> +	__typeof__(old) __old = (old);				\
>  	__typeof__(*(ptr)) __ret;				\
> -	__smp_release_fence();					\
> -	__ret = arch_cmpxchg_relaxed(ptr, o, n);		\
> -	__smp_acquire_fence();					\
> +	switch (size) {						\
> +	case 4:							\
> +		asm volatile (					\
> +		"1:	ldex.w		%0, (%3) \n"		\
> +		"	cmpne		%0, %4   \n"		\
> +		"	bt		2f       \n"		\
> +		"	mov		%1, %2   \n"		\
> +		RELEASE_FENCE					\

FWIW, you probably need to make sure that a barrier instruction inside
an lr/sc loop is a good thing. IIUC, the execution time of a barrier
instruction is determined by the status of store buffers and invalidate
queues (and probably other stuffs), so it may increase the execution
time of the lr/sc loop, and make it unlikely to succeed. But this really
depends on how the arch executes these instructions.

Regards,
Boqun

> +		"	stex.w		%1, (%3) \n"		\
> +		"	bez		%1, 1b   \n"		\
> +		FULL_FENCE					\
> +		"2:				 \n"		\
> +			: "=&r" (__ret), "=&r" (__tmp)		\
> +			: "r" (__new), "r"(__ptr), "r"(__old)	\
> +			:);					\
> +		break;						\
> +	default:						\
> +		__bad_xchg();					\
> +	}							\
>  	__ret;							\
>  })
>  
> +#define arch_cmpxchg(ptr, o, n) \
> +	(__cmpxchg((ptr), (o), (n), sizeof(*(ptr))))
> +
> +#define arch_cmpxchg_local(ptr, o, n)				\
> +	(__cmpxchg_relaxed((ptr), (o), (n), sizeof(*(ptr))))
>  #else
>  #include <asm-generic/cmpxchg.h>
>  #endif
> -- 
> 2.25.1
> 

[-- Attachment #2: signature.asc --]
[-- Type: application/pgp-signature, Size: 488 bytes --]

  reply	other threads:[~2022-04-22  3:20 UTC|newest]

Thread overview: 6+ messages / expand[flat|nested]  mbox.gz  Atom feed  top
2022-04-17  8:32 [PATCH V3 0/3] csky: Optimize with acquire & release for atomic & cmpxchg guoren
2022-04-17  8:32 ` [PATCH V3 1/3] csky: cmpxchg: Optimize with acquire & release guoren
2022-04-22  3:20   ` Boqun Feng [this message]
2022-04-22  3:45     ` Guo Ren
2022-04-17  8:32 ` [PATCH V3 2/3] csky: atomic: Add custom atomic.h implementation guoren
2022-04-17  8:32 ` [PATCH V3 3/3] csky: atomic: Add conditional atomic operations' optimization guoren

Reply instructions:

You may reply publicly to this message via plain-text email
using any one of the following methods:

* Save the following mbox file, import it into your mail client,
  and reply-to-all from there: mbox

  Avoid top-posting and favor interleaved quoting:
  https://en.wikipedia.org/wiki/Posting_style#Interleaved_style

* Reply using the --to, --cc, and --in-reply-to
  switches of git-send-email(1):

  git send-email \
    --in-reply-to=YmIfBmisFVBu8Z5R@tardis \
    --to=boqun.feng@gmail.com \
    --cc=arnd@arndb.de \
    --cc=guoren@kernel.org \
    --cc=guoren@linux.alibaba.com \
    --cc=linux-arch@vger.kernel.org \
    --cc=linux-csky@vger.kernel.org \
    --cc=linux-kernel@vger.kernel.org \
    --cc=mark.rutland@arm.com \
    --cc=peterz@infradead.org \
    --cc=will@kernel.org \
    /path/to/YOUR_REPLY

  https://kernel.org/pub/software/scm/git/docs/git-send-email.html

* If your mail client supports setting the In-Reply-To header
  via mailto: links, try the mailto: link
Be sure your reply has a Subject: header at the top and a blank line before the message body.
This is a public inbox, see mirroring instructions
for how to clone and mirror all data and code used for this inbox