linux-kernel - Re: [PATCH] softirq: Remove raise_softirq from tasklet_action

lists.openwall.net		lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC
Open Source and information security mailing list archives
Hash Suite: Windows password security audit tool. GUI, reports in PDF.
[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Date:   Wed, 23 Feb 2022 21:20:36 +0530
From:   Mukesh Ojha <quic_mojha@...cinc.com>
To:     Frederic Weisbecker <frederic@...nel.org>
CC:     <linux-kernel@...r.kernel.org>, <peterz@...radead.org>,
        <tglx@...utronix.de>, <paulmck@...nel.org>, <will@...nel.org>,
        <dave@...olabs.net>
Subject: Re: [PATCH] softirq: Remove raise_softirq from
 tasklet_action_common()


On 2/21/2022 4:48 PM, Mukesh Ojha wrote:
>
> On 2/9/2022 4:34 AM, Frederic Weisbecker wrote:
>> On Sat, Feb 05, 2022 at 06:43:25PM +0530, Mukesh Ojha wrote:
>>> Think about a scenario when all other cores are in suspend
>>> and one core is only running ksoftirqd and it is because
>>> some client has invoked tasklet_hi_schedule() only once
>>> during that phase.
>>>
>>> tasklet_action_common() handles that softirq and marks the
>>> same softirq as pending again. And due to that core keeps
>>> running the softirq handler [1] forever and it is not able to
>>> go to suspend.
>>>
>>> We can get rid of raising softirq from tasklet handler.
>>>
>>> [1]
>>> <idle>-0    [003]   13058.769081:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13058.769085: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13058.769087:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13058.769091:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13058.769094: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13058.769097:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13058.769100:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13058.769103: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13058.769106:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13058.769109:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13059.058923:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> ...
>>> ..
>>> ..
>>> ..
>>>
>>> <idle>-0    [003]   13059.058951:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13059.058954: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13059.058957:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13059.058960:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13059.058963: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13059.058966:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13059.058969:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13059.058972: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13059.058975:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13059.058978:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13059.058981: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13059.058984:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13059.058987:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13059.058990: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13059.058993:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>> <idle>-0    [003]   13059.058996:  softirq_entry vec=0  
>>> action=HI_SOFTIRQ
>>> <idle>-0     [003]  13059.059000: softirq_raise: vec=0 
>>> [action=HI_SOFTIRQ]
>>> <idle>-0    [003]   13059.059002:  softirq_exit   vec=0 
>>> action=HI_SOFTIRQ
>>>
>>> Signed-off-by: Mukesh Ojha <quic_mojha@...cinc.com>
>>> ---
>>>   kernel/softirq.c | 1 -
>>>   1 file changed, 1 deletion(-)
>>>
>>> diff --git a/kernel/softirq.c b/kernel/softirq.c
>>> index 41f4709..d3e6fb9 100644
>>> --- a/kernel/softirq.c
>>> +++ b/kernel/softirq.c
>>> @@ -795,7 +795,6 @@ static void tasklet_action_common(struct 
>>> softirq_action *a,
>>>           t->next = NULL;
>>>           *tl_head->tail = t;
>>>           tl_head->tail = &t->next;
>>> -        __raise_softirq_irqoff(softirq_nr);
>>>           local_irq_enable();
>> That requeue happens when the tasklet is already executing on some 
>> other CPU
>> or when it has been disabled through tasklet_disable().
>>
>> So you can't just remove that line or you'll break everything.
>>
>> It would be nice to identify which tasklet keeps being requeued. Is 
>> it because
>> something called tasklet_disable() to it and never called back 
>> tasklet_enable() ?
>
> Hi @Frederic,
>
> Thanks for the reply.
> Suppose a scenario where a tasklet is scheduled/queued from one client 
> and before running of tasklet handler, same tasklet gets
> disabled from some other cpu.
> In this scenario, while the handlers runs it will be keep on marking 
> the softirq pending even though tasklet is disabled.
> Tasklet will be enabled but after coming out of low power mode.
> Will it look to be valid case ?


Never mind, we should call tasklet_kill() followed by tasklet_disable().
I suspect, the issue is in client code, some race is setting schedule 
bit and marking the softirq pending  even after doing tasklet_kill => 
tasklet_disable in cleanup path.

-Mukesh


>
> -Mukesh
>
>
>>
>> Thanks.