Message-ID: <CAMbhsRQAw2+as13NVvWkh7ojyRs+jNpFist2gmiiEY52U=o84A@mail.gmail.com>
Date:	Thu, 15 Mar 2012 16:37:19 -0700
From:	Colin Cross <ccross@...roid.com>
To:	linux-kernel@...r.kernel.org
Cc:	linux-arm-kernel@...ts.infradead.org,
	linux-pm@...ts.linux-foundation.org, Kevin Hilman <khilman@...com>,
	Len Brown <len.brown@...el.com>,
	Trinabh Gupta <g.trinabh@...il.com>,
	Arjan van de Ven <arjan@...ux.intel.com>,
	Deepthi Dharwar <deepthi@...ux.vnet.ibm.com>,
	Greg Kroah-Hartman <gregkh@...e.de>,
	Kay Sievers <kay.sievers@...y.org>,
	Santosh Shilimkar <santosh.shilimkar@...com>,
	Daniel Lezcano <daniel.lezcano@...aro.org>,
	Amit Kucheria <amit.kucheria@...aro.org>,
	Lorenzo Pieralisi <lorenzo.pieralisi@....com>,
	Colin Cross <ccross@...roid.com>, Rob Lee <rob.lee@...aro.org>
Subject: Re: [PATCHv2 0/5] coupled cpuidle state support

On Wed, Mar 14, 2012 at 11:29 AM, Colin Cross <ccross@...roid.com> wrote:
> On some ARM SMP SoCs (OMAP4460, Tegra 2, and probably more), the
> cpus cannot be independently powered down, either due to
> sequencing restrictions (on Tegra 2, cpu 0 must be the last to
> power down), or due to HW bugs (on OMAP4460, a cpu powering up
> will corrupt the gic state unless the other cpu runs a work
> around).  Each cpu has a power state that it can enter without
> coordinating with the other cpu (usually Wait For Interrupt, or
> WFI), and one or more "coupled" power states that affect blocks
> shared between the cpus (L2 cache, interrupt controller, and
> sometimes the whole SoC).  Entering a coupled power state must
> be tightly controlled on both cpus.
>
> The easiest solution to implementing coupled cpu power states is
> to hotplug all but one cpu whenever possible, usually using a
> cpufreq governor that looks at cpu load to determine when to
> enable the secondary cpus.  This causes problems, as hotplug is an
> expensive operation, so the number of hotplug transitions must be
> minimized, leading to very slow response to loads, often on the
> order of seconds.
>
> This patch series implements an alternative solution, where each
> cpu will wait in the WFI state until all cpus are ready to enter
> a coupled state, at which point the coupled state function will
> be called on all cpus at approximately the same time.
>
> Once all cpus are ready to enter idle, they are woken by an smp
> cross call.  At this point, there is a chance that one of the
> cpus will find work to do, and choose not to enter suspend.  A
> final pass is needed to guarantee that all cpus will call the
> power state enter function at the same time.  During this pass,
> each cpu will increment the ready counter, and continue once the
> ready counter matches the number of online coupled cpus.  If any
> cpu exits idle, the other cpus will decrement their counter and
> retry.
>
> To use coupled cpuidle states, a cpuidle driver must:
>
>   Set struct cpuidle_device.coupled_cpus to the mask of all
>   coupled cpus, usually the same as cpu_possible_mask if all cpus
>   are part of the same cluster.  The coupled_cpus mask must be
>   set in the struct cpuidle_device for each cpu.
>
>   Set struct cpuidle_device.safe_state to a state that is not a
>   coupled state.  This is usually WFI.
>
>   Set CPUIDLE_FLAG_COUPLED in struct cpuidle_state.flags for each
>   state that affects multiple cpus.
>
>   Provide a struct cpuidle_state.enter function for each state
>   that affects multiple cpus.  This function is guaranteed to be
>   called on all cpus at approximately the same time.  The driver
>   should ensure that the cpus all abort together if any cpu tries
>   to abort once the function is called.
>
> This series has been tested by implementing a test cpuidle state
> that uses the parallel barrier helper function to verify that
> all cpus call the function at the same time.
>
> This patch set has a few disadvantages over the hotplug governor,
> but I think they are all fairly minor:
>   * Worst-case interrupt latency can be increased.  If one cpu
>     receives an interrupt while the other is spinning in the
>     ready_count loop, the second cpu will be stuck with
>     interrupts off until the first cpu finished processing
>     its interrupt and exits idle.  This will increase the worst
>     case interrupt latency by the worst-case interrupt processing
>     time, but should be very rare.
>   * Interrupts are processed while still inside pm_idle.
>     Normally, interrupts are only processed at the very end of
>     pm_idle, just before it returns to the idle loop.  Coupled
>     states requires processing interrupts inside
>     cpuidle_enter_state_coupled in order to distinguish between
>     the smp_cross_call from another cpu that is now idle and an
>     interrupt that should cause idle to exit.
>     I don't see a way to fix this without either being able to
>     read the next pending irq from the interrupt chip, or
>     querying the irq core for which interrupts were processed.
>   * Since interrupts are processed inside cpuidle, the next
>     timer event could change.  The new timer event will be
>     handled correctly, but the idle state decision made by
>     the governor will be out of date, and will not be revisited.
>     The governor select function could be called again every time,
>     but this could lead to a lot of work being done by an idle
>     cpu if the other cpu was mostly busy.
>
> v2:
>   * removed the coupled lock, replacing it with atomic counters
>   * added a check for outstanding pokes before beginning the
>     final transition to avoid extra wakeups
>   * made the cpuidle_coupled struct completely private
>   * fixed kerneldoc comment formatting
>   * added a patch with a helper function for resynchronizing
>     cpus after aborting idle
>   * added a patch (not for merging) to add trace events for
>     verification and performance testing

I forgot to mention, this patch series is on v3.3-rc7, and will
conflict with the cpuidle timekeeping patches.  If those go in first
(which is likely), I will rework this series on top of it.  I left it
on v3.3-rc7 now to make testing easier.
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@...r.kernel.org
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/

Powered by blists - more mailing lists