Received-SPF: pass (google.com: domain of linux-kernel+bounces-90467-linux.lists.archive=gmail.com@vger.kernel.org designates 147.75.199.223 as permitted sender) client-ip=147.75.199.223;
Feedback-ID: i56a14606:Fastmail
User-Agent: Cyrus-JMAP/3.11.0-alpha0-205-g4dbcac4545-fm-20240301.001-g4dbcac45
Precedence: bulk
MIME-Version: 1.0
Message-Id: <44216ff8-740b-45ca-9902-b52a04901655@app.fastmail.com>
In-Reply-To: <Zdc0M9Uf2zn63P0e@kekkonen.localdomain>
References: <20240219195807.517742-1-sakari.ailus@linux.intel.com>
 <20240219195807.517742-2-sakari.ailus@linux.intel.com>
 <MW5PR11MB5787130A75404600F47A7D9C8D562@MW5PR11MB5787.namprd11.prod.outlook.com>
 <Zdc0M9Uf2zn63P0e@kekkonen.localdomain>
Date: Mon, 04 Mar 2024 11:58:19 +0100
From: "Arnd Bergmann" <arnd@arndb.de>
To: "Sakari Ailus" <sakari.ailus@linux.intel.com>,
 "Wentong Wu" <wentong.wu@intel.com>
Cc: "Hans de Goede" <hdegoede@redhat.com>,
 "Tomas Winkler" <tomas.winkler@intel.com>,
 "Greg Kroah-Hartman" <gregkh@linuxfoundation.org>,
 "linux-kernel@vger.kernel.org" <linux-kernel@vger.kernel.org>
Subject: Re: [PATCH v2 1/3] mei: vsc: Call wake_up() in the threaded IRQ handler
Content-Type: text/plain

On Thu, Feb 22, 2024, at 12:46, Sakari Ailus wrote:
> Hi Wentong,
>
> On Thu, Feb 22, 2024 at 03:26:18AM +0000, Wu, Wentong wrote:
>> > From: Sakari Ailus <sakari.ailus@linux.intel.com>
>> > 
>> > The hard IRQ handler vsc_tp_irq() is called with a raw spinlock taken.
>> > wake_up() acquires a spinlock, a sleeping lock on PREEMPT_RT.
>> 
>> Does this mean we can't use wake_up() in isr? 
>
> Good question. A lot of callers currently do.

If driver has a traditional (non-threaded) handler, it should
always work fine: on non-PREEMPT_RT it can take the spinlock
and on PREEMPT_RT it automatically turns into a threaded
handler that can still call it.

> In this case, handle_edge_irq() takes the raw spinlock and acquiring the
> wake queue spinlock in wake_up() leads to sleeping IRQs disabled (see
> below).
>
> I don't think there's any harm in moving the wake_up() to the threaded
> handler.

It causes an extra bit of latency for the non-PREEMPT_RT case
because you now always have to go through two task switches.
This is probably fine if you don't care about latency. 

You can probably replace the open-coded completion (wait queue
head plus atomic) with a normal 'struct completion' and
call complete() in the isr instead. This one seems to only
take a raw spinlock already.

    Arnd