Re: Commit 282d8998e997 (srcu: Prevent expedited GPs and blocking readers from consuming CPU) cause qemu boot slow

From: "zhangfei.gao@foxmail.com" <zhangfei.gao@foxmail.com>
To: Neeraj Upadhyay <quic_neeraju@quicinc.com>,
	Shameerali Kolothum Thodi  <shameerali.kolothum.thodi@huawei.com>,
	"paulmck@kernel.org" <paulmck@kernel.org>
Cc: Paolo Bonzini <pbonzini@redhat.com>,
	Zhangfei Gao <zhangfei.gao@linaro.org>,
	"linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>,
	"rcu@vger.kernel.org" <rcu@vger.kernel.org>,
	Lai Jiangshan <jiangshanlai@gmail.com>,
	Josh Triplett <josh@joshtriplett.org>,
	Mathieu Desnoyers <mathieu.desnoyers@efficios.com>,
	Matthew Wilcox <willy@infradead.org>,
	"mtosatti@redhat.com" <mtosatti@redhat.com>,
	Auger Eric <eric.auger@redhat.com>,
	"chenxiang (M)" <chenxiang66@hisilicon.com>
Subject: Re: Commit 282d8998e997 (srcu: Prevent expedited GPs and blocking readers from consuming CPU) cause qemu boot slow
Date: Tue, 14 Jun 2022 22:57:04 +0800	[thread overview]
Message-ID: <tencent_0783451107DBB16AC54D4269800D941D0708@qq.com> (raw)
In-Reply-To: <88780123-32a3-ab34-cd3d-01e4a3421f03@quicinc.com>


On 2022/6/14 下午10:14, Neeraj Upadhyay wrote:
>
>
> On 6/14/2022 7:33 PM, zhangfei.gao@foxmail.com wrote:
>>
>>
>> On 2022/6/14 下午8:19, Neeraj Upadhyay wrote:
>>>
>>>>
>>>> 5.18-rc4 based               ~8sec
>>>>
>>>> 5.19-rc1                     ~2m43sec
>>>>
>>>> 5.19-rc1+fix1                 ~19sec
>>>>
>>>> 5.19-rc1-fix2                 ~19sec
>>>>
>>>
>>> If you try below diff on top of either 5.19-rc1+fix1 or 
>>> 5.19-rc1-fix2 ; does it show any difference in boot time?
>>>
>>> --- a/kernel/rcu/srcutree.c
>>> +++ b/kernel/rcu/srcutree.c
>>> @@ -706,7 +706,7 @@ static void srcu_schedule_cbs_snp(struct 
>>> srcu_struct *ssp, struct srcu_node *snp
>>>   */
>>>  static void srcu_gp_end(struct srcu_struct *ssp)
>>>  {
>>> -       unsigned long cbdelay;
>>> +       unsigned long cbdelay = 1;
>>>         bool cbs;
>>>         bool last_lvl;
>>>         int cpu;
>>> @@ -726,7 +726,9 @@ static void srcu_gp_end(struct srcu_struct *ssp)
>>>         spin_lock_irq_rcu_node(ssp);
>>>         idx = rcu_seq_state(ssp->srcu_gp_seq);
>>>         WARN_ON_ONCE(idx != SRCU_STATE_SCAN2);
>>> -       cbdelay = !!srcu_get_delay(ssp);
>>> +       if (ULONG_CMP_LT(READ_ONCE(ssp->srcu_gp_seq), 
>>> READ_ONCE(ssp->srcu_gp_seq_needed_exp)))
>>> +               cbdelay = 0;
>>> +
>>>         WRITE_ONCE(ssp->srcu_last_gp_end, ktime_get_mono_fast_ns());
>>>
>> Test here:
>> qemu: https://github.com/qemu/qemu/tree/stable-6.1
>> kernel: 
>> https://github.com/Linaro/linux-kernel-uadk/tree/uacce-devel-5.19-srcu-test
>> (in case test patch not clear, push in git tree)
>>
>> Hardware: aarch64
>>
>> 1. 5.18-rc6
>> real    0m8.402s
>> user    0m3.015s
>> sys     0m1.102s
>>
>> 2. 5.19-rc1
>> real    2m41.433s
>> user    0m3.097s
>> sys     0m1.177s
>>
>> 3. 5.19-rc1 + fix1 from Paul
>> real    2m43.404s
>> user    0m2.880s
>> sys     0m1.214s
>>
>> 4. 5.19-rc1 + fix2: fix1 + Remove "if (!jbase)" block
>> real    0m15.262s
>> user    0m3.003s
>> sys     0m1.033s
>>
>> When build kernel in the meantime, load time become longer.
>>
>> 5. 5.19-rc1 + fix3: fix1 + SRCU_MAX_NODELAY_PHASE 1000000
>> real    0m15.215s
>> user    0m2.942s
>> sys    0m1.172s
>>
>> 6. 5.19-rc1 + fix4: fix1 + Neeraj's change of srcu_gp_end　
>> real    1m23.936s
>> user    0m2.969s
>> sys    0m1.181s
>>
>
7. 5.19-rc1 + fix5: fix4 + Remove "if (!jbase)" block

real 0m11.418s

user 0m3.031s

sys 0m1.067s


8. 5.19-rc1 + fix 6: fix4 + SRCU_MAX_NODELAY_PHASE 1000000

real 0m11.154s ~12s

user 0m2.919s

sys 0m1.064s


Thanks

> Thanks for this data. Can you please share below test combo also?
>
> 7. 5.19-rc1 + fix5: fix2 + Neeraj's change of srcu_gp_end
>
>
> 8. 5.19-rc1 + fix6: fix3 + Neeraj's change of srcu_gp_end
>
>
> Thanks
> Neeraj
>
>> More test details: https://docs.qq.com/doc/DRXdKalFPTVlUbFN5
>>
>> Thanks
>>