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: <5efc39c5-865b-1c0f-e22d-dc79c67dbcf2@efficios.com>
Date:   Fri, 31 Mar 2023 19:56:09 -0400
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.8.0
Subject: Re: [RFC PATCH] sched: Introduce per-mm/cpu concurrency id state
Content-Language: en-US
To:     Aaron Lu <aaron.lu@intel.com>
Cc:     Peter Zijlstra <peterz@infradead.org>, linux-kernel@vger.kernel.org
References: <20230330230911.228720-1-mathieu.desnoyers@efficios.com>
 <20230331083830.GA186694@ziqianlu-desk2>
From:   Mathieu Desnoyers <mathieu.desnoyers@efficios.com>
In-Reply-To: <20230331083830.GA186694@ziqianlu-desk2>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 2023-03-31 04:38, Aaron Lu wrote:
> On Thu, Mar 30, 2023 at 07:09:11PM -0400, Mathieu Desnoyers wrote:
> 
>>   void sched_mm_cid_exit_signals(struct task_struct *t)
>>   {
>>   	struct mm_struct *mm = t->mm;
>> -	unsigned long flags;
>> +	struct rq *rq = this_rq();
> 
> Got many below messages due to the above line:
> 
> [   19.294089] BUG: using smp_processor_id() in preemptible [00000000] code: kworker/u449:0/1621
> 
>> +	struct rq_flags rf;
>>   
>>   	if (!mm)
>>   		return;
>> -	local_irq_save(flags);
>> +	rq_lock_irqsave(rq, &rf);
>>   	mm_cid_put(mm, t->mm_cid);
>>   	t->mm_cid = -1;
>>   	t->mm_cid_active = 0;
>> -	local_irq_restore(flags);
>> +	rq_unlock_irqrestore(rq, &rf);
>>   }
>>   
>>   void sched_mm_cid_before_execve(struct task_struct *t)
>>   {
>>   	struct mm_struct *mm = t->mm;
>> -	unsigned long flags;
>> +	struct rq *rq = this_rq();
> 
> Also here;
> 
>> +	struct rq_flags rf;
>>   
>>   	if (!mm)
>>   		return;
>> -	local_irq_save(flags);
>> +	rq_lock_irqsave(rq, &rf);
>>   	mm_cid_put(mm, t->mm_cid);
>>   	t->mm_cid = -1;
>>   	t->mm_cid_active = 0;
>> -	local_irq_restore(flags);
>> +	rq_unlock_irqrestore(rq, &rf);
>>   }
>>   
>>   void sched_mm_cid_after_execve(struct task_struct *t)
>>   {
>>   	struct mm_struct *mm = t->mm;
>> -	unsigned long flags;
>> +	struct rq *rq = this_rq();
> 
> And here.
> 
>> +	struct rq_flags rf;
>>   
>>   	if (!mm)
>>   		return;
>> -	local_irq_save(flags);
>> +	rq_lock_irqsave(rq, &rf);
>>   	t->mm_cid = mm_cid_get(mm);
>>   	t->mm_cid_active = 1;
>> -	local_irq_restore(flags);
>> +	rq_unlock_irqrestore(rq, &rf);
>>   	rseq_set_notify_resume(t);
>>   }
> 
> I used below diff to get rid of these messages without understanding the
> purpose of these functions:

I'll fold this fix into the next round, thanks!

Mathieu

> 
> diff --git a/kernel/sched/core.c b/kernel/sched/core.c
> index f07b87d155bd..7194c29f3c91 100644
> --- a/kernel/sched/core.c
> +++ b/kernel/sched/core.c
> @@ -11444,45 +11444,57 @@ void sched_mm_cid_migrate_to(struct rq *dst_rq, struct task_struct *t, int src_c
>   void sched_mm_cid_exit_signals(struct task_struct *t)
>   {
>   	struct mm_struct *mm = t->mm;
> -	struct rq *rq = this_rq();
>   	struct rq_flags rf;
> +	struct rq *rq;
>   
>   	if (!mm)
>   		return;
> +
> +	preempt_disable();
> +	rq = this_rq();
>   	rq_lock_irqsave(rq, &rf);
>   	mm_cid_put(mm, t->mm_cid);
>   	t->mm_cid = -1;
>   	t->mm_cid_active = 0;
>   	rq_unlock_irqrestore(rq, &rf);
> +	preempt_enable();
>   }
>   
>   void sched_mm_cid_before_execve(struct task_struct *t)
>   {
>   	struct mm_struct *mm = t->mm;
> -	struct rq *rq = this_rq();
>   	struct rq_flags rf;
> +	struct rq *rq;
>   
>   	if (!mm)
>   		return;
> +
> +	preempt_disable();
> +	rq = this_rq();
>   	rq_lock_irqsave(rq, &rf);
>   	mm_cid_put(mm, t->mm_cid);
>   	t->mm_cid = -1;
>   	t->mm_cid_active = 0;
>   	rq_unlock_irqrestore(rq, &rf);
> +	preempt_enable();
>   }
>   
>   void sched_mm_cid_after_execve(struct task_struct *t)
>   {
>   	struct mm_struct *mm = t->mm;
> -	struct rq *rq = this_rq();
>   	struct rq_flags rf;
> +	struct rq *rq;
>   
>   	if (!mm)
>   		return;
> +
> +	preempt_disable();
> +	rq = this_rq();
>   	rq_lock_irqsave(rq, &rf);
>   	t->mm_cid = mm_cid_get(mm);
>   	t->mm_cid_active = 1;
>   	rq_unlock_irqrestore(rq, &rf);
> +	preempt_enable();
>   	rseq_set_notify_resume(t);
>   }
>   

-- 
Mathieu Desnoyers
EfficiOS Inc.
https://www.efficios.com