Re: [RFC] Add "rpm_not_supported" flag

From: "Rafael J. Wysocki" <rjw@rjwysocki.net>
To: Alan Stern <stern@rowland.harvard.edu>
Cc: Greg Kroah-Hartman <gregkh@linuxfoundation.org>,
	Allen Yu <alleny@nvidia.com>, Pavel Machek <pavel@ucw.cz>,
	Len Brown <len.brown@intel.com>,
	Dan Williams <dan.j.williams@intel.com>,
	Linux-pm mailing list <linux-pm@vger.kernel.org>,
	Kernel development list <linux-kernel@vger.kernel.org>
Subject: Re: [RFC] Add "rpm_not_supported" flag
Date: Thu, 03 Jul 2014 23:16:31 +0200	[thread overview]
Message-ID: <1703139.e6mBtKGhYW@vostro.rjw.lan> (raw)
In-Reply-To: <Pine.LNX.4.44L0.1407021002380.874-100000@iolanthe.rowland.org>

On Wednesday, July 02, 2014 10:27:06 AM Alan Stern wrote:
> On Wed, 2 Jul 2014, Rafael J. Wysocki wrote:
> 
> > > Then you have no other objections to the patch?
> > 
> > My concern still is that it will be confusing, because people won't read the
> > documentation carefully enough and will confuse "runtime PM never used" with
> > "hardware can't do PM".  I'm not sure how to make that more clear, though.
> 
> I could emphasize that distinction a little more strongly in the 
> documentation.
> 
> > Also we have the no_callbacks flag and I wonder if/how it is related to the
> > new one.  Do we still need both?
> 
> They mean different things.  The no_callbacks flag is used when we want 
> the PM core to think the device can be in RPM_SUSPENDED at times (it is 
> "logically suspended").  rpm_not_supported is used when we want the PM 
> core to think the device must always be in RPM_ACTIVE.
> 
> > In addition to that, I think that "hardware can't do PM" should apply to the
> > handling of system suspend resume too.
> 
> Maybe.  For the use case Dan Williams and I are working on, it doesn't 
> matter; for other cases it might matter.  That's why I named the flag 
> "rpm_not_supported" -- it applies specifically to runtime PM, not 
> system PM.
> 
> Here's a brief summary of the story behind this patch...
> 
> At one point, I suggested to Dan that instead of doing something
> special for these devices, we could simply have the runtime_suspend()
> routine always return -EBUSY.  He didn't like that idea because then
> the user would see the device was never powering down but would have no
> idea why.  The rpm_not_supported flag provides this information to the
> user by causing the power/runtime_status attribute to say "not
> supported".  (Although to be entirely fair, we could just put a message
> in the kernel log during probe if the hardware doesn't support runtime
> suspend.)
> 
> Instead, Dan introduced a messy PM QoS mechanism in commit
> e3d105055525.  I didn't like that approach, but Greg merged it before I
> objected.
> 
> Do you have any suggestions?

I need some more time to think about that.  I'm on vacation till Monday,
I should be able to get to this by then.  I hope that's not a problem.

Rafael