Re: [PATCH 0/8] Suspend block api (version 7)

[Arve Hjønnevåg]

2010/5/19 Rafael J. Wysocki <rjw@xxxxxxx>:
> On Wednesday 19 May 2010, Arve Hjønnevåg wrote:
>> 2010/5/18 Rafael J. Wysocki <rjw@xxxxxxx>:
>> > On Wednesday 19 May 2010, Arve Hjønnevåg wrote:
>> >> 2010/5/18 Rafael J. Wysocki <rjw@xxxxxxx>:
>> >> > On Wednesday 19 May 2010, Arve Hjønnevåg wrote:
>> >> >> 2010/5/18 Rafael J. Wysocki <rjw@xxxxxxx>:
>> >> >> > On Tuesday 18 May 2010, Arve Hjønnevåg wrote:
>> >> >> >> 2010/5/18 Rafael J. Wysocki <rjw@xxxxxxx>:
>> >> >> >> > On Tuesday 18 May 2010, Arve Hjønnevåg wrote:
>> >> >> >> >> 2010/5/17 Rafael J. Wysocki <rjw@xxxxxxx>:
>> >> >> >> >> > On Monday 17 May 2010, Arve Hjønnevåg wrote:
>> >> >> >> >> >> 2010/5/14 Rafael J. Wysocki <rjw@xxxxxxx>:
>> >> >> >> >> >> > On Friday 14 May 2010, Arve Hjønnevåg wrote:
>> >> >> >> >> >> >> This patch series adds a suspend-block api that provides the same
>> >> >> >> >> >> >> functionality as the android wakelock api. This version has some
>> >> >> >> >> >> >> changes from, or requested by, Rafael. The most notable changes are:
>> >> >> >> >> >> >> - DEFINE_SUSPEND_BLOCKER and suspend_blocker_register have been added
>> >> >> >> >> >> >>   for statically allocated suspend blockers.
>> >> >> >> >> >> >> - suspend_blocker_destroy is now called suspend_blocker_unregister
>> >> >> >> >> >> >> - The user space mandatory _INIT ioctl has been replaced with an
>> >> >> >> >> >> >>   optional _SET_NAME ioctl.
>> >> >> >> >> >> >>
>> >> >> >> >> >> >> I kept the ack and reviewed by tags on two of the patches even though
>> >> >> >> >> >> >> there were a few cosmetic changes.
>> >> >> >> >> >> >
>> >> >> >> >> >> > Thanks for the patches, I think they are in a pretty good shape now.
>> >> >> >> >> >> >
>> >> >> >> >> >> > That said, I'd like the changelogs to be a bit more descriptive, at least for
>> >> >> >> >> >> > patch [1/8].  I think it should explain (in a few words) what the purpose of
>> >> >> >> >> >> > the feature is and what problems it solves that generally a combination of
>> >> >> >> >> >> > runtime PM and cpuidle is not suitable for in your opinion.  IOW, why you
>> >> >> >> >> >> > think we need that feature.
>> >> >> >> >> >> >
>> >> >> >> >> >>
>> >> >> >> >> >> How about:
>> >> >> >> >> >>
>> >> >> >> >> >> PM: Add opportunistic suspend support.
>> >> >> >> >> >
>> >> >> >> >> > "PM: Opportunistic suspend support" would be sufficient IMO.
>> >> >> >> >> >
>> >> >> >> >> > Now, I'd start with the motivation.  Like "Power management features present
>> >> >> >> >> > in the current mainline kernel are insufficient to get maximum possible energy
>> >> >> >> >> > savings on some platforms, such as Android, because ..." (here go explanations
>> >> >> >> >> > why this is the case in your opinion).
>> >> >> >> >> >
>> >> >> >> >> > Next, "To allow Android and similar platforms to save more energy than they
>> >> >> >> >> > currently can save using the mainline kernel, introduce a mechanism by which
>> >> >> >> >> > the system is automatically suspended (i.e. put into a system-wide sleep state)
>> >> >> >> >> > whenever it's not doing useful work, called opportunistic suspend".
>> >> >> >> >> >
>> >> >> >> >> > "For this purpose introduce the suspend blockers framework allowing the
>> >> >> >> >> > kernel's power management subsystem to decide when it is desirable to suspend
>> >> >> >> >> > the system (i.e. when useful work is not being done).  Add an API that ..."
>> >> >> >> >> >
>> >> >> >> >>
>> >> >> >> >> PM: Opportunistic suspend support.
>> >> >> >> >>
>> >> >> >> >> Power management features present in the current mainline kernel are
>> >> >> >> >> insufficient to get maximum possible energy savings on some platforms,
>> >> >> >> >> such as Android, because low power states can only safely be entered
>> >> >> >> >> from idle.
>> >> >> >> >
>> >> >> >> > Do you mean CPU low-power states or system low-power states here?
>> >> >> >> >
>> >> >> >> I think either.
>> >> >> >
>> >> >> > The statement is not true for system low-power states (or sleep states),
>> >> >> > because generally (ie. on many platforms) they aren't entered from idle.
>> >> >> > Suspend is for entering them.
>> >> >> >
>> >> >> I don't think that makes my statement false. If you can only safely
>> >> >> enter low power states from idle, but some low power states cannot be
>> >> >> entered from idle, then it follows that you cannot safely enter those
>> >> >> low power states.
>> >> >
>> >> > Define "safely", then.
>> >> >
>> >> OK. Is this clearer:
>> >
>> > Not really.
>> >
>> >> Power management features present in the current mainline kernel are
>> >> insufficient to get maximum possible energy savings on some platforms,
>> >> such as Android, because some wakeup events must work at all times
>> >> which means that low power states can only safely be entered from idle.
>> >
>> > It's not clear what you mean by "some wakeup events must work at all times"
>> > and what "safely" means.
>> >
>> >> Suspend, in its current form, cannot be used, since wakeup events that
>> >> occur right after initiating suspend will not be processed until another
>> >> possibly unrelated event wake up the system again.
>> >
>> > Well, I _guess_ I know what you're trying to say, but I'm not sure if that's
>> > the core of the problem the patchset is supposed to address.
>> >
>> > The problem, as I see it, is really two-fold.
>> >
>> > First, to save as much energy as reasonably possible you need to put everything
>> > except for memory (ie. I/O devices and CPUs) into low-power states and let that
>> > stay in these low-power states for as long as reasonably possible, right?
>>
>> Memory should also be in a low power state, but yes, everything should
>> be in the lowest power possible state for as long as possible.
>>
>> >
>> > Second, you need the system to _always_ respond to some _specific_ events
>> > regardless of its current state, correct?
>> >
>> > Now, at present, we can put I/O devices and the CPU into low-power states
>> > at the same time in two ways: (1) by using runtime PM and cpuidle and (2) by
>> > suspending the system.  Unfortunately, none of these approaches is ideal,
>> > because (1) is vulnerable to periodic timers and polling apps (and CPU-bound
>> > apps) and (2) doesn't guarantee that all events you care about will be
>> > responded to.
>> >
>> > The proposed solution is to use suspend with an additional mechanism that
>> > will prevent the events that have to be responded to from being lost.  This
>> > additional mechanism is suspend blockers.
>> >
>> > Is this correct?
>>
>> Yes.
>>
>> >
>> >> >> > So, there is a difference and the explanation can go like this: "... to achieve
>> >> >> > maximum energy savings one has to put all I/O devices and the CPU into
>> >> >> > low-power states, but the CPU can only be put into a low-power state from
>> >> >> > idle.  This, in turn, means that the CPU leaves the low-power state on
>> >> >> > interrupts triggered by periodic timers and user space processes that use
>> >> >> > polling.  It turns out, however, that many of these events causing the CPU to
>> >> >> > leave the low-power state aren't essential for the desired system functionality
>> >> >> > and from the power management point of view it would be better if they didn't
>> >> >> > occur".
>> >> >>
>> >> >> This only explain why we still want to use suspend on systems that can
>> >> >> enter the same power states from idle and suspend.
>> >> >
>> >> > Sorry, I'm confused now.  Isn't that why you do that on Android?
>> >> >
>> >>
>> >> Yes, but it is not the only reason to use opportunistic support. It is
>> >> way more important to have opportunistic suspend support if the
>> >> hardware can enter lower power states from suspend than idle.
>> >
>> > Sure, but you can add something like "There also are systems where some
>> > devices cannot be put into low-power states without suspending the entire
>> > system (or the low-power states available to them without suspending the
>> > entire system are substantially shallower than the low-power states they are
>> > put into when the entire system is suspended), so the system have to be
>> > suspended as a whole to achieve the maximum energy savings".
>> >
>>
>> Does this work for you:
>
> Well, almost.
>
>> PM: Opportunistic suspend support.
>>
>> Power management features present in the current mainline kernel are
>> insufficient to get maximum possible energy savings on some platforms,
>> such as Android,
>
> I'd finish the first sentence here.
>
>> because the system must always respond to certain
>> events. Suspend, in its current form, cannot be used on these
>> platforms,
>
> This doesn't seem to be correct.  Surely you can suspend these systems, but
> you can't address the problem at hand using suspend as is.
>
> I'd say (in the second sentence) "The problem is that to save maximum amount of
> energy all system hardware components need to be in the lowest-power states
> available for as long as reasonably possible, but at the same time the system
> must always respond to certain events, regardless of the current state of the
> hardware.
>
> The first goal can be achieved either by using device runtime PM and cpuidle to
> put all hardware into low-power states, transparently from the user space point
> of view, or by suspending the whole system.  However, system suspend, in its
> current form, does not guarantee that the events of interest will always be
> responded to, since wakeup events ..."
>
>> since wakeup events (events that wake the CPU from idle and
>> the system from suspend) that occur right after initiating suspend
>> will not be processed until another possibly unrelated event wake up
>
> "... wakes up ..." I guess.
>
>> the system again.
>>
>> On hardware where idle can enter the same power state as suspend, idle
>> combined with runtime PM can be used, but periodic wakeups increase
>> the average power consumption. Suspending the system also reduces the
>> harm caused by apps that never go idle. There also are systems where
>> some devices cannot be put into low-power states without suspending
>> the entire system (or the low-power states available to them without
>> suspending the entire system are substantially shallower than the
>> low-power states they are put into when the entire system is
>> suspended), so the system have to be suspended as a whole to achieve
>> the maximum energy savings.
>>
>> To allow Android and similar platforms to save more energy than they
>> currently can save using the mainline kernel, introduce a mechanism by
>> which the system is automatically suspended (i.e. put into a
>> system-wide sleep state) whenever it's not doing useful work, called
>> opportunistic suspend.
>
> I'd say "... whenever it's not doing work that's immediately useful to the
> user, called  ..." or something like this.
>
> The rest is fine by me.
>

PM: Opportunistic suspend support.

Power management features present in the current mainline kernel are
insufficient to get maximum possible energy savings on some platforms,
such as Android.  The problem is that to save maximum amount of energy
all system hardware components need to be in the lowest-power states
available for as long as reasonably possible, but at the same time the
system must always respond to certain events, regardless of the
current state of the hardware.

The first goal can be achieved either by using device runtime PM and
cpuidle to put all hardware into low-power states, transparently from
the user space point of view, or by suspending the whole system.
However, system suspend, in its current form, does not guarantee that
the events of interest will always be responded to, since wakeup
events (events that wake the CPU from idle and the system from
suspend) that occur right after initiating suspend will not be
processed until another possibly unrelated event wakes the system up
again.

On hardware where idle can enter the same power state as suspend, idle
combined with runtime PM can be used, but periodic wakeups increase
the average power consumption. Suspending the system also reduces the
harm caused by apps that never go idle. There also are systems where
some devices cannot be put into low-power states without suspending
the entire system (or the low-power states available to them without
suspending the entire system are substantially shallower than the
low-power states they are put into when the entire system is
suspended), so the system has to be suspended as a whole to achieve
the maximum energy savings.

To allow Android and similar platforms to save more energy than they
currently can save using the mainline kernel, introduce a mechanism by
which the system is automatically suspended (i.e. put into a
system-wide sleep state) whenever it's not doing work that's
immediately useful to the user, called opportunistic suspend.

For this purpose introduce the suspend blockers framework allowing the
kernel's power management subsystem to decide when it is desirable to
suspend the system (i.e. when the system is not doing anything the
user really cares about at the moment and therefore it may be
suspended). Add an API that that drivers can use to block
opportunistic suspend. This is needed to avoid losing wakeup events
that occur right after suspend is initiated.

Add /sys/power/policy that selects the behavior of /sys/power/state.
After setting the policy to opportunistic, writes to /sys/power/state
become non-blocking requests that specify which suspend state to enter
when no suspend blockers are active. A special state, "on", stops the
process by activating the "main" suspend blocker.

-- 
Arve Hjønnevåg
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/