Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <49b867c1-2b68-c034-9f60-b26a77ff326a@bytedance.com>
Date:   Mon, 16 Jan 2023 16:40:47 +0800
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (Macintosh; Intel Mac OS X 10.15; rv:102.0)
 Gecko/20100101 Thunderbird/102.5.1
Subject: Re: [PATCH 2/3] memblock: Make finding index faster when modify
 regions.
To:     Mike Rapoport <rppt@kernel.org>
Cc:     akpm@linux-foundation.org, linux-mm@kvack.org,
        linux-kernel@vger.kernel.org
References: <20230113082659.65276-1-zhangpeng.00@bytedance.com>
 <20230113082659.65276-3-zhangpeng.00@bytedance.com>
 <Y8QHehG1L+kuyqoR@kernel.org>
 <acbfecf4-e3ee-4df2-6685-6cefca38bde6@bytedance.com>
 <Y8T+qWlUT1KTeFNI@kernel.org>
From:   Peng Zhang <zhangpeng.00@bytedance.com>
In-Reply-To: <Y8T+qWlUT1KTeFNI@kernel.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


在 2023/1/16 15:37, Mike Rapoport 写道:
> On Mon, Jan 16, 2023 at 11:17:40AM +0800, Peng Zhang wrote:
>>
>>
>> 在 2023/1/15 22:02, Mike Rapoport 写道:
>>> Hi,
>>>
>>> On Fri, Jan 13, 2023 at 04:26:58PM +0800, Peng Zhang wrote:
>>>> We can use binary search to find the index to modify regions in
>>>> memblock_add_range() and memblock_isolate_range(). Because the
>>>> arrangement of regions is ordered. It may be faster when there are
>>>> many regions. So implemented a binary search and a new macro to walk
>>>> regions.
>>>
>>> Did you see a measurable speedup with this optimization?
>>> I'm not in favor of micro-optimizations that complicate code.
>>>
>> Thank you for your reply. I haven't measured this patch yet, theoretically
>> this small optimization might be difficult to observe.
>> If you think this patch complicates the code, you can ignore this patch.
>>
>> These three patches are independent and they can be applied independently.
>> The logic of the third patch is very simple. It will not complicate the
>> code. It is tested by the default configuration of qemu. The total number of
>> iterations of memblock_merge_regions() in the third patch is reduced from
>> more than one thousand to more than one hundred, this is only in the case of
>> a small number of regions. Can you consider the third patch?
> 
> Can you please send the numbers and show how did you obtained them?
> 
>> Sincerely yours,
>> Peng.
> 
I obtained the numbers like this:

void memblock_merge_regions(struct memblock_type *type) {
	static int iteration_count = 0;
	static int max_nr_regions = 0;

	max_nr_regions = max(max_nr_regions, (int)type->cnt);
	...
	while () {
		iteration_count++;
		...
	}
	pr_info("iteration_count: %d max_nr_regions %d", iteration_count, 
max_nr_regions);
}

Boot the kernel by qemu.
The folowing numbers is the last output.

master branch:
iteration_count: 1762 max_nr_regions 41

patched:
iteration_count: 182 max_nr_regions 41

If max_nr_regions is larger, the difference will be more.

Thanks.

Sincerely yours,
Peng