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;
From:   Thomas Gleixner <tglx@linutronix.de>
To:     Hillf Danton <hdanton@sina.com>
Cc:     Thomas Pfaff <tpfaff@pcs.com>, linux-kernel@vger.kernel.org,
        Marc Zyngier <maz@kernel.org>
Subject: Re: [PATCH v3] irq/core: synchronize irq_thread startup
In-Reply-To: <20220503004209.4670-1-hdanton@sina.com>
References: <552fe7b4-9224-b183-bb87-a8f36d335690@pcs.com>
 <20220502160106.4587-1-hdanton@sina.com>
 <20220503004209.4670-1-hdanton@sina.com>
Date:   Tue, 03 May 2022 09:38:01 +0200
Message-ID: <87pmkvkqw6.ffs@tglx>
MIME-Version: 1.0
Content-Type: text/plain
Precedence: bulk

Hillf,

On Tue, May 03 2022 at 08:42, Hillf Danton wrote:
> On Mon, 02 May 2022 21:24:45 +0200 Thomas Gleixner  wrote:
>> On Tue, May 03 2022 at 00:01, Hillf Danton wrote:
>> > +	if (!waked)
>> > +		wake_threads_waitq(desc);
>> 
>> That's a guarantee to make desc->threads_active go negative in the case
>> that the thread was never woken by a hard interrupt handler. IOW, you
>> created a new problem which did not exist before.
>
> The count of active threads would not drop below zero with the change,
> given the comment in __irq_wake_thread(). It is incremented before
> wakeup.

There is no guarantee that the wake-up happens via __irq_wake_thread().
kthread_stop() does a wake-up too, but that obviously _cannot_ increment
the active counter because it does not know about it at all.

   create_thread()
      thread_fn()
        woken = false;
        wait_for_wakeup_or_stop()       <- Stop is set, no interrupt happened
                                        <- ergo woken == false   
        if (!woken)
      	  wake_threads_waitq(desc)
            atomic_dec_and_test(..)   <- underflow

>> The problem discussed here is not a problem in irq_thread(), it's a
>> problem of not reaching this function in the first place. See the on
>> point analysis in Thomas Pfaffs patch.
>
> Well why is the count above zero without wakeup? IOW why is there imbalance
> in count if the irq thread never gets a CPU to run?

Look at kthread() and the condition under which threadfn() is invoked.

Thanks,

        tglx