Re: [PATCH 1/3] irq / PM: New driver interface for wakeup interrupts

From: Alan Stern <stern@rowland.harvard.edu>
To: "Rafael J. Wysocki" <rjw@rjwysocki.net>
Cc: Thomas Gleixner <tglx@linutronix.de>,
	Peter Zijlstra <peterz@infradead.org>,
	<linux-kernel@vger.kernel.org>,
	Linux PM list <linux-pm@vger.kernel.org>,
	Dmitry Torokhov <dtor@google.com>
Subject: Re: [PATCH 1/3] irq / PM: New driver interface for wakeup interrupts
Date: Thu, 31 Jul 2014 16:12:55 -0400 (EDT)	[thread overview]
Message-ID: <Pine.LNX.4.44L0.1407311552530.885-100000@iolanthe.rowland.org> (raw)
In-Reply-To: <2140085.bSS0bVjyxJ@vostro.rjw.lan>

On Thu, 31 Jul 2014, Rafael J. Wysocki wrote:

> And before we enter the wakeup handling slippery slope, let me make a note
> that this problem is bothering me quite a bit at the moment.  In my opinion
> we need to address it somehow regardless of the wakeup issues and I'm not sure
> if failing __setup_irq() when there's a mismatch (that is, there are existing
> actions for the given irq_desc and their IRQF_NO_SUSPEND settings are not
> consistent with the new one) is the right way to do that, because it may make
> things behave a bit randomly (it will always fail the second guy, but that need
> not be the one who's requested IRQF_NO_SUSPEND and it depends on the ordering
> between them).

Pardon me for sticking my nose into the middle of the conversation, but
here's what it looks like to me:

The entire no_irq phase of suspend/resume is starting to seem like a
mistake.  We should never have done it.  Alternatively, it might be
okay to disable _all_ interrupts during the no_irq phase provided they
are then _all_ enabled again before entering the sysdev and
platform-specific parts of suspend (or the final freeze).

As I understand it, the idea behind the no_irq phase was to make life 
easier for drivers.  They wouldn't have to worry about strange things 
cropping up while they're in the middle of powering down their devices 
or afterward.

Well, guess what?  It turns out that they do have to worry about it
after all.  Timers can still fire during suspend transitions, and if an
IRQ line is shared with a wakeup source then it won't be disabled.

The fact is, drivers should not rely on disabled interrupts to prevent
untimely interrupt requests or wakeups.  They should configure their
devices not to generate any interrupt requests at all, apart from
wakeup requests.  And their interrupt handlers shouldn't mind being
invoked for a shared IRQ, even after their devices are turned off.

Any driver that leaves its device capable of generating non-wakeup
interrupt requests during suspend, and relies on interrupts being
globally disabled to avoid problems, is most likely broken.  Yes, it
may be acceptable in cases where the IRQ line isn't shared, but it's
still a bad design.

Alan Stern