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;
Date:   Thu, 6 Oct 2022 15:09:15 -0400
From:   Eric Chanudet <echanude@redhat.com>
To:     Sebastian Andrzej Siewior <bigeasy@linutronix.de>
Cc:     Andy Gross <agross@kernel.org>,
        Bjorn Andersson <andersson@kernel.org>,
        Manivannan Sadhasivam <mani@kernel.org>,
        Jassi Brar <jassisinghbrar@gmail.com>,
        linux-arm-msm@vger.kernel.org, linux-rt-users@vger.kernel.org,
        linux-kernel@vger.kernel.org
Subject: Re: [PATCH] mailbox: qcom-ipcc: flag IRQ NO_THREAD
Message-ID: <20221006190915.2koau7vlnvinqk3b@echanude>
References: <20221003170849.383005-1-echanude@redhat.com>
 <Yzv/3vsIMtgu7noY@linutronix.de>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <Yzv/3vsIMtgu7noY@linutronix.de>
Precedence: bulk

On Tue, Oct 04, 2022 at 11:41:50AM +0200, Sebastian Andrzej Siewior wrote:
> On 2022-10-03 13:08:49 [-0400], Eric Chanudet wrote:
> > PREEMPT_RT forces qcom-ipcc's handler to be threaded with interrupts
> > enabled, which triggers a warning in __handle_irq_event_percpu():
> >     irq 173 handler irq_default_primary_handler+0x0/0x10 enabled interrupts
> >     WARNING: CPU: 0 PID: 77 at kernel/irq/handle.c:161 __handle_irq_event_percpu+0x4c4/0x4d0
> > 
> > Mark it IRQF_NO_THREAD to avoid running the handler in a threaded
> > context with threadirqs or PREEMPT_RT enabled.
> 
> The important bit of information is that this is a IRQ-multiplexer and
> such it must not be threaded. Otherwise its child-interrupts would be
> invoked from the thread handler which is not desired.
> This is noticed by PREEMPT_RT but also on a non-PREEMPT_RT kernel where
> `threadirqs' has been used.

Thank you, I will send a v2 shortly updating the description with the
above.

> Side note: Using request_irq() has the side effect that this interrupt
> pops-up in /proc/interrupts and the "child-interrupt", too. So you
> account two interrupts while "one" on the HW side occurred. 
> Maybe irq_set_chained_handler_and_data() would be better use.

I see.

> Either way, this addresses the problem,

-- 
Eric Chanudet