Avoid the open-coding of the accounted time acquisition in
runtime_active|suspend_time_show() and make them call
pm_runtime_active|suspended_time() instead.

Note that this change also indirectly avoids holding dev->power.lock
around the do_div() computation and the sprintf() call which is an
additional improvement.

Signed-off-by: Ulf Hansson <ulf.hansson@linaro.org>
[ rjw: Changelog ]
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

All device objects in the driver model contain fields that control the
handling of various power management activities. However, it's not
always useful. There are few instances where pseudo devices are added
to the model just to take advantage of many other features like
kobjects, udev events, and so on. One such example is cpu devices and
their caches.

The sysfs for the cpu caches are managed by adding devices with cpu
as the parent in cpu_device_create() when secondary cpu is brought
online. Generally when the secondary CPUs are hotplugged back in as part
of resume from suspend-to-ram, we call cpu_device_create() from the cpu
hotplug state machine while the cpu device associated with that CPU is
not yet ready to be resumed as the device_resume() call happens bit
later. It's not really needed to set the flag is_prepared for cpu
devices as they are mostly pseudo device and hotplug framework deals
with state machine and not managed through the cpu device.

This often results in annoying warning when resuming:
Enabling non-boot CPUs ...
CPU1: Booted secondary processor
 cache: parent cpu1 should not be sleeping
CPU1 is up
CPU2: Booted secondary processor
 cache: parent cpu2 should not be sleeping
CPU2 is up
.... and so on.

So in order to fix these kind of errors, we could just completely avoid
doing any power management related initialisations and operations if
they are not used by these devices.

Add no_pm flags to indicate that the device doesn't require any sort of
PM activities and all of them can be completely skipped. We can use the
same flag to also avoid adding not used *power* sysfs entries for these
devices. For now, lets use this for cpu cache devices.

Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Sudeep Holla <sudeep.holla@arm.com>
Tested-by: Eugeniu Rosca <erosca@de.adit-jv.com>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Replace jiffies-based accounting for runtime_active_time and
runtime_suspended_time with ktime-based accounting. This makes the
runtime debug counters inline with genpd and other PM subsytems which
use ktime-based accounting.

Timekeeping is initialized before driver_init(). It's only at that time
that PM-runtime can be enabled.

Signed-off-by: Thara Gopinath <thara.gopinath@linaro.org>
[switch from ktime to raw nsec]
Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

PM-runtime uses the timer infrastructure for autosuspend. This implies
that the minimum time before autosuspending a device is in the range
of 1 tick included to 2 ticks excluded
 -On arm64 this means between 4ms and 8ms with default jiffies
  configuration
 -And on arm, it is between 10ms and 20ms

These values are quite high for embedded systems which sometimes want
the duration to be in the range of 1 ms.

It is possible to switch autosuspend over to using hrtimers to get
finer granularity for short durations and take advantage of slack to
retain some margins and get long timeouts with minimum wakeups.

On an arm64 platform that uses 1ms for autosuspending timeout of its
GPU, idle power is reduced by 10% with hrtimer.

The latency impact on arm64 hikey octo cores is:
 - mark_last_busy: from 1.11 us to 1.25 us
 - rpm_suspend: from 15.54 us to 15.38 us
[Only the code path of rpm_suspend() that starts hrtimer has been
measured.]

arm64 image (arm64 default defconfig) decreases by around 3KB
with following details:

$ size vmlinux-timer
   text	   data	    bss	    dec	    hex	filename
12034646	6869268	 386840	19290754	1265a82	vmlinux

$ size vmlinux-hrtimer
   text	   data	    bss	    dec	    hex	filename
12030550	6870164	 387032	19287746	1264ec2	vmlinux

The latency impact on arm 32bits snowball dual cores is :
 - mark_last_busy: from 0.31 us usec to 0.77 us
 - rpm_suspend: from 6.83 us to 6.67 usec

The increase of the image for snowball platform that I used for
testing performance impact, is neglictable (244B).

$ size vmlinux-timer
   text	   data	    bss	    dec	    hex	filename
7157961	2119580	 264120	9541661	 91981d	build-ux500/vmlinux

size vmlinux-hrtimer
   text	   data	    bss	    dec	    hex	filename
7157773	2119884	 264248	9541905	 919911	vmlinux-hrtimer

And arm 32bits image (multi_v7_defconfig) increases by around 1.7KB
with following details:

$ size vmlinux-timer
   text	   data	    bss	    dec	    hex	filename
13304443	6803420	 402768	20510631	138f7a7	vmlinux

$ size vmlinux-hrtimer
   text	   data	    bss	    dec	    hex	filename
13304299	6805276	 402768	20512343	138fe57	vmlinux

Signed-off-by: Vincent Guittot <vincent.guittot@linaro.org>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>

Middle-layer code doing suspend-time optimizations for devices with
the DPM_FLAG_SMART_SUSPEND flag set (currently, the PCI bus type and
the ACPI PM domain) needs to make the core skip ->thaw_early and
->thaw callbacks for those devices in some cases and it sets the
power.direct_complete flag for them for this purpose.

However, it turns out that setting power.direct_complete outside of
the PM core is a bad idea as it triggers an excess invocation of
pm_runtime_enable() in device_resume().

For this reason, provide a helper to clear power.is_late_suspended
and power.is_suspended to be invoked by the middle-layer code in
question instead of setting power.direct_complete and make that code
call the new helper.

Fixes: c4b65157aeef (PCI / PM: Take SMART_SUSPEND driver flag into account)
Fixes: 05087360fd7a (ACPI / PM: Take SMART_SUSPEND driver flag into account)
Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>

Define and document a new driver flag, DPM_FLAG_LEAVE_SUSPENDED, to
instruct the PM core and middle-layer (bus type, PM domain, etc.)
code that it is desirable to leave the device in runtime suspend
after system-wide transitions to the working state (for example,
the device may be slow to resume and it may be better to avoid
resuming it right away).

Generally, the middle-layer code involved in the handling of the
device is expected to indicate to the PM core whether or not the
device may be left in suspend with the help of the device's
power.may_skip_resume status bit.  That has to happen in the "noirq"
phase of the preceding system suspend (or analogous) transition.
The middle layer is then responsible for handling the device as
appropriate in its "noirq" resume callback which is executed
regardless of whether or not the device may be left suspended, but
the other resume callbacks (except for ->complete) will be skipped
automatically by the core if the device really can be left in
suspend.

The additional power.must_resume status bit introduced for the
implementation of this mechanisn is used internally by the PM core
to track the requirement to resume the device (which may depend on
its children etc).

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>

Make the PCI bus type take DPM_FLAG_SMART_SUSPEND into account in its
system-wide PM callbacks and make sure that all code that should not
run in parallel with pci_pm_runtime_resume() is executed in the "late"
phases of system suspend, freeze and poweroff transitions.

[Note that the pm_runtime_suspended() check in pci_dev_keep_suspended()
is an optimization, because if is not passed, all of the subsequent
checks may be skipped and some of them are much more overhead in
general.]

Also use the observation that if the device is in runtime suspend
at the beginning of the "late" phase of a system-wide suspend-like
transition, its state cannot change going forward (runtime PM is
disabled for it at that time) until the transition is over and the
subsequent system-wide PM callbacks should be skipped for it (as
they generally assume the device to not be suspended), so add checks
for that in pci_pm_suspend_late/noirq(), pci_pm_freeze_late/noirq()
and pci_pm_poweroff_late/noirq().

Moreover, if pci_pm_resume_noirq() or pci_pm_restore_noirq() is
called during the subsequent system-wide resume transition and if
the device was left in runtime suspend previously, its runtime PM
status needs to be changed to "active" as it is going to be put
into the full-power state, so add checks for that too to these
functions.

In turn, if pci_pm_thaw_noirq() runs after the device has been
left in runtime suspend, the subsequent "thaw" callbacks need
to be skipped for it (as they may not work correctly with a
suspended device), so set the power.direct_complete flag for the
device then to make the PM core skip those callbacks.

In addition to the above add a core helper for checking if
DPM_FLAG_SMART_SUSPEND is set and the device runtime PM status is
"suspended" at the same time, which is done quite often in the new
code (and will be done elsewhere going forward too).

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>

Define and document a SMART_SUSPEND flag to instruct bus types and PM
domains that the system suspend callbacks provided by the driver can
cope with runtime-suspended devices, so from the driver's perspective
it should be safe to leave devices in runtime suspend during system
suspend.

Setting that flag may also cause middle-layer code (bus types,
PM domains etc.) to skip invocations of the ->suspend_late and
->suspend_noirq callbacks provided by the driver if the device
is in runtime suspend at the beginning of the "late" phase of
the system-wide suspend transition, in which case the driver's
system-wide resume callbacks may be invoked back-to-back with
its ->runtime_suspend callback, so the driver has to be able to
cope with that too.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>

The motivation for this change is to provide a way to work around
a problem with the direct-complete mechanism used for avoiding
system suspend/resume handling for devices in runtime suspend.

The problem is that some middle layer code (the PCI bus type and
the ACPI PM domain in particular) returns positive values from its
system suspend ->prepare callbacks regardless of whether the driver's
->prepare returns a positive value or 0, which effectively prevents
drivers from being able to control the direct-complete feature.
Some drivers need that control, however, and the PCI bus type has
grown its own flag to deal with this issue, but since it is not
limited to PCI, it is better to address it by adding driver flags at
the core level.

To that end, add a driver_flags field to struct dev_pm_info for flags
that can be set by device drivers at the probe time to inform the PM
core and/or bus types, PM domains and so on on the capabilities and/or
preferences of device drivers.  Also add two static inline helpers
for setting that field and testing it against a given set of flags
and make the driver core clear it automatically on driver remove
and probe failures.

Define and document two PM driver flags related to the direct-
complete feature: NEVER_SKIP and SMART_PREPARE that can be used,
respectively, to indicate to the PM core that the direct-complete
mechanism should never be used for the device and to inform the
middle layer code (bus types, PM domains etc) that it can only
request the PM core to use the direct-complete mechanism for
the device (by returning a positive value from its ->prepare
callback) if it also has been requested by the driver.

While at it, make the core check pm_runtime_suspended() when
setting power.direct_complete so that it doesn't need to be
checked by ->prepare callbacks.

Signed-off-by: Rafael J. Wysocki <rafael.j.wysocki@intel.com>
Acked-by: Greg Kroah-Hartman <gregkh@linuxfoundation.org>
Acked-by: Bjorn Helgaas <bhelgaas@google.com>
Reviewed-by: Ulf Hansson <ulf.hansson@linaro.org>