Received-SPF: pass (google.com: domain of linux-kernel+bounces-21120-linux.lists.archive=gmail.com@vger.kernel.org designates 2604:1380:45e3:2400::1 as permitted sender) client-ip=2604:1380:45e3:2400::1;
Feedback-ID: i56a14606:Fastmail
User-Agent: Cyrus-JMAP/3.9.0-alpha0-1364-ga51d5fd3b7-fm-20231219.001-ga51d5fd3
Precedence: bulk
MIME-Version: 1.0
Message-Id: <1633a5e7-213e-4621-bdc5-5cc056da9d99@app.fastmail.com>
In-Reply-To: <tencent_13A0B6B4A3136E46CC448874232A9F956006@qq.com>
References: <tencent_13A0B6B4A3136E46CC448874232A9F956006@qq.com>
Date: Tue, 09 Jan 2024 17:43:00 +0100
From: "Arnd Bergmann" <arnd@arndb.de>
To: "Yangyu Chen" <cyy@cyyself.name>
Cc: "Christoph Hellwig" <hch@lst.de>,
 "Alexander Potapenko" <glider@google.com>,
 "Mike Frysinger" <vapier@gentoo.org>,
 Linux-Arch <linux-arch@vger.kernel.org>, linux-kernel@vger.kernel.org
Subject: Re: [PATCH] asm-generic: flush icache only when vma->vm_flags has VM_EXEC set
Content-Type: text/plain

On Tue, Jan 9, 2024, at 15:45, Yangyu Chen wrote:
> For some ISAs like RISC-V, which may not support bus broadcast-based
> icache flushing instructions, it's necessary to send IPIs to all of the
> CPUs in the system to flush the icache. This process can be expensive for
> these ISAs and introduce disturbances during performance profiling.
> Limiting the icache flush to occur only when the vma->vm_flags has VM_EXEC
> can help minimize the frequency of these operations.
>
> Signed-off-by: Yangyu Chen <cyy@cyyself.name>
> ---
>  include/asm-generic/cacheflush.h | 3 ++-
>  1 file changed, 2 insertions(+), 1 deletion(-)
>
> diff --git a/include/asm-generic/cacheflush.h 
> b/include/asm-generic/cacheflush.h
> index 84ec53ccc450..729d51536575 100644
> --- a/include/asm-generic/cacheflush.h
> +++ b/include/asm-generic/cacheflush.h
> @@ -102,7 +102,8 @@ static inline void flush_cache_vunmap(unsigned long 
> start, unsigned long end)
>  	do { \
>  		instrument_copy_to_user((void __user *)dst, src, len); \
>  		memcpy(dst, src, len); \
> -		flush_icache_user_page(vma, page, vaddr, len); \
> +		if (vma->vm_flags & VM_EXEC) \
> +			flush_icache_user_page(vma, page, vaddr, len); \
>  	} while (0)
>  #endif

This is not my normal area of expertise, but I wonder if you can't
just do this the same way as alpha and openrisc by having the
condition in architecture specific code.

     Arnd