Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Date:   Sun, 15 Aug 2021 12:19:53 -0400
From:   Dennis Zhou <dennis@kernel.org>
To:     Junhua Huang <junhuahuangdream@163.com>
Cc:     dennis@kernel.org, tj@kernel.org, cl@linux.com, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org,
        Junhua Huang <huang.junhua@zte.com.cn>
Subject: Re: [PATCH] percpu: remove the unnecessary operation for bitmaps in
 finding suitable region
Message-ID: <YRk+qQ2Z7jqC8Pvl@fedora>
References: <20210815032018.192334-1-junhuahuangdream@163.com>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20210815032018.192334-1-junhuahuangdream@163.com>
Precedence: bulk

Hello,

On Sat, Aug 14, 2021 at 08:20:18PM -0700, Junhua Huang wrote:
> From: Junhua Huang <huang.junhua@zte.com.cn>
> 
> We use pcpu_next_fit_region to find the next fit region for alloc. Once
> return, the region fit the request and then break the loop, or get the
> next start offset from pcpu_is_populated. So there is no necessary to
> add bits to bit_off. At the same time, bits will set 0 in 
> pcpu_next_fit_region each time, so I think it is unnecessary to 
> set bits zero in pcpu_for_each_fit_region loops.
> 

I think you're right in this instance, but I don't think this change is
correct as a whole.

> Signed-off-by: Junhua Huang <huang.junhua@zte.com.cn>
> ---
>  mm/percpu.c | 2 --
>  1 file changed, 2 deletions(-)
> 
> diff --git a/mm/percpu.c b/mm/percpu.c
> index 7f2e0151c4e2..ec7a5d10f6c9 100644
> --- a/mm/percpu.c
> +++ b/mm/percpu.c
> @@ -489,7 +489,6 @@ static void pcpu_next_fit_region(struct pcpu_chunk *chunk, int alloc_bits,
>  	for (pcpu_next_fit_region((chunk), (alloc_bits), (align), &(bit_off), \
>  				  &(bits));				      \
>  	     (bit_off) < pcpu_chunk_map_bits((chunk));			      \
> -	     (bit_off) += (bits),					      \

Removing this makes the iterator by itself wrong. While correct in the
current usage, it could potentially be wrong for the next.

>  	     pcpu_next_fit_region((chunk), (alloc_bits), (align), &(bit_off), \
>  				  &(bits)))
>  
> @@ -1126,7 +1125,6 @@ static int pcpu_find_block_fit(struct pcpu_chunk *chunk, int alloc_bits,
>  			break;
>  
>  		bit_off = next_off;
> -		bits = 0;
>  	}
>  
>  	if (bit_off == pcpu_chunk_map_bits(chunk))
> -- 
> 2.25.1
> 
> 

Thanks,
Dennis