Message-ID: <ced10a5a-d5d9-ef45-a454-706fff89a672@bytedance.com>
Date:   Mon, 6 Feb 2023 12:18:00 +0000
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.4.2
Subject: Re: [External] Re: [PATCH v6 00/11] Parallel CPU bringup for x86_64
Content-Language: en-US
To:     David Woodhouse <dwmw2@infradead.org>, Russ Anderson <rja@hpe.com>
Cc:     tglx@linutronix.de, arjan@linux.intel.com, mingo@redhat.com,
        bp@alien8.de, dave.hansen@linux.intel.com, hpa@zytor.com,
        x86@kernel.org, pbonzini@redhat.com, paulmck@kernel.org,
        linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        rcu@vger.kernel.org, mimoja@mimoja.de, hewenliang4@huawei.com,
        thomas.lendacky@amd.com, seanjc@google.com, pmenzel@molgen.mpg.de,
        fam.zheng@bytedance.com, punit.agrawal@bytedance.com,
        simon.evans@bytedance.com, liangma@liangbit.com
References: <20230202215625.3248306-1-usama.arif@bytedance.com>
 <20230205191734.GA6027@hpe.com>
 <3b626936ccc4d5123cb9113378bec6f77182def3.camel@infradead.org>
From:   Usama Arif <usama.arif@bytedance.com>
In-Reply-To: <3b626936ccc4d5123cb9113378bec6f77182def3.camel@infradead.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk


On 06/02/2023 08:28, David Woodhouse wrote:
> On Sun, 2023-02-05 at 13:17 -0600, Russ Anderson wrote:
>>
>> Gave the v6 patchset a try on a system with 1920 logocal cpus
>> (sixteen 60 core Sapphire Rapids sockets with Hyperthreadding
>> enabled).
>>
>> Without the patchset it took 71 seconds to start all the cpus.
>> With the v6 patchset it took 14 seconds to start all the cpus,
>> a reduction of 57 seconds.  That is impressive.
>>
>> Full boot, to root login prompt, without patches takes 223 seconds.
>> This patchset reduces the full boot time by 57 seconds, a 25%
>> reduction.
> 
> Nice; thanks for testing.
> 
> Is that with just the "part1" patch series which has been posted, or
> also with the 'parallel part 2' still taking shape in the tree at
> https://git.infradead.org/users/dwmw2/linux.git/shortlog/refs/heads/parallel-6.2-rc6
> 
> I believe Usama said the second phase of parallelism didn't really help
> much in terms of overall timing? Confirming that *without* all the
> debug prints would be interesting. And we can look for what still
> *could* be made parallel.

I think it would be interesting to get the numbers for such a big 
machine for 3 cases: part1, part1+reuse timer calibration and part1+part2.

Russ mentioned testing v6, so I guess the above numbers are for 
part1+reuse timer calibration.

For my machine the smpboot times were:

No patches: 700ms
part 1:100ms
part1+reuse timer calibration: 30ms
part1+part2: 30ms

Thanks,
Usama