One of the roles of the PM core is to prevent different PM callbacks
executed for the same device object from racing with each other.
Unfortunately, after commit e8665002477f0278f84f898145b1f141ba26ee26
(PM: Allow pm_runtime_suspend() to succeed during system suspend)
runtime PM callbacks may be executed concurrently with system
suspend/resume callbacks for the same device.

The main reason for commit e8665002477f0278f84f898145b1f141ba26ee26
was that some subsystems and device drivers wanted to use runtime PM
helpers, pm_runtime_suspend() and pm_runtime_put_sync() in
particular, for carrying out the suspend of devices in their
.suspend() callbacks.  However, as it's been determined recently,
there are multiple reasons not to do so, inlcuding:

 * The caller really doesn't control the runtime PM usage counters,
   because user space can access them through sysfs and effectively
   block runtime PM.  That means using pm_runtime_suspend() or
   pm_runtime_get_sync() to suspend devices during system suspend
   may or may not work.

 * If a driver calls pm_runtime_suspend() from its .suspend()
   callback, it causes the subsystem's .runtime_suspend() callback to
   be executed, which leads to the call sequence:

   subsys->suspend(dev)
      driver->suspend(dev)
         pm_runtime_suspend(dev)
            subsys->runtime_suspend(dev)

   recursive from the subsystem's point of view.  For some subsystems
   that may actually work (e.g. the platform bus type), but for some
   it will fail in a rather spectacular fashion (e.g. PCI).  In each
   case it means a layering violation.

 * Both the subsystem and the driver can provide .suspend_noirq()
   callbacks for system suspend that can do whatever the
   .runtime_suspend() callbacks do just fine, so it really isn't
   necessary to call pm_runtime_suspend() during system suspend.

 * The runtime PM's handling of wakeup devices is usually different
   from the system suspend's one, so .runtime_suspend() may simply be
   inappropriate for system suspend.

 * System suspend is supposed to work even if CONFIG_PM_RUNTIME is
   unset.

 * The runtime PM workqueue is frozen before system suspend, so if
   whatever the driver is going to do during system suspend depends
   on it, that simply won't work.

Still, there is a good reason to allow pm_runtime_resume() to
succeed during system suspend and resume (for instance, some
subsystems and device drivers may legitimately use it to ensure that
their devices are in full-power states before suspending them).
Moreover, there is no reason to prevent runtime PM callbacks from
being executed in parallel with the system suspend/resume .prepare()
and .complete() callbacks and the code removed by commit
e8665002477f0278f84f898145b1f141ba26ee26 went too far in this
respect.  On the other hand, runtime PM callbacks, including
.runtime_resume(), must not be executed during system suspend's
"late" stage of suspending devices and during system resume's "early"
device resume stage.

Taking all of the above into consideration, make the PM core
acquire a runtime PM reference to every device and resume it if
there's a runtime PM resume request pending right before executing
the subsystem-level .suspend() callback for it.  Make the PM core
drop references to all devices right after executing the
subsystem-level .resume() callbacks for them.  Additionally,
make the PM core disable the runtime PM framework for all devices
during system suspend, after executing the subsystem-level .suspend()
callbacks for them, and enable the runtime PM framework for all
devices during system resume, right before executing the
subsystem-level .resume() callbacks for them.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Kevin Hilman <khilman@ti.com>

The naming convention used by commit 7538e3db6e015e890825fbd9f86599b
(PM: Add support for device power domains), which introduced the
struct dev_power_domain type for representing device power domains,
evidently confuses some developers who tend to think that objects
of this type must correspond to "power domains" as defined by
hardware, which is not the case.  Namely, at the kernel level, a
struct dev_power_domain object can represent arbitrary set of devices
that are mutually dependent power management-wise and need not belong
to one hardware power domain.  To avoid that confusion, rename struct
dev_power_domain to struct dev_pm_domain and rename the related
pointers in struct device and struct pm_clk_notifier_block from
pwr_domain to pm_domain.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Kevin Hilman <khilman@ti.com>

The PM core doesn't handle suspend failures correctly when it comes to
asynchronously suspended devices.  These devices are moved onto the
dpm_suspended_list as soon as the corresponding async thread is
started up, and they remain on the list even if they fail to suspend
or the sleep transition is cancelled before they get suspended.  As a
result, when the PM core unwinds the transition, it tries to resume
the devices even though they were never suspended.

This patch (as1474) fixes the problem by adding a new "is_suspended"
flag to dev_pm_info.  Devices are resumed only if the flag is set.

[rjw:
 * Moved the dev->power.is_suspended check into device_resume(),
   because we need to complete dev->power.completion and clear
   dev->power.is_prepared too for devices whose
   dev->power.is_suspended flags are unset.
 * Fixed __device_suspend() to avoid setting dev->power.is_suspended
   if async_error is different from zero.]

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: stable@kernel.org

This patch (as1473) renames the "in_suspend" field in struct
dev_pm_info to "is_prepared", in preparation for an upcoming change.
The new name is more descriptive of what the field really means.

Signed-off-by: Alan Stern <stern@rowland.harvard.edu>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Cc: stable@kernel.org

If device drivers allocate substantial amounts of memory (above 1 MB)
in their hibernate .freeze() callbacks (or in their legacy suspend
callbcks during hibernation), the subsequent creation of hibernate
image may fail due to the lack of memory.  This is the case, because
the drivers' .freeze() callbacks are executed after the hibernate
memory preallocation has been carried out and the preallocated amount
of memory may be too small to cover the new driver allocations.
Unfortunately, the drivers' .prepare() callbacks also are executed
after the hibernate memory preallocation has completed, so they are
not suitable for allocating additional memory either.  Thus the only
way a driver can safely allocate memory during hibernation is to use
a hibernate/suspend notifier.  However, the notifiers are called
before the freezing of user space and the drivers wanting to use them
for allocating additional memory may not know how much memory needs
to be allocated at that point.

To let device drivers overcome this difficulty rework the hibernation
sequence so that the memory preallocation is carried out after the
drivers' .prepare() callbacks have been executed, so that the
.prepare() callbacks can be used for allocating additional memory
to be used by the drivers' .freeze() callbacks.  Update documentation
to match the new behavior of the code.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>

Change the PM core's behavior related to power domains in such a way
that, if a power domain is defined for a given device, its callbacks
will be executed instead of and not in addition to the device
subsystem's PM callbacks.

The idea behind the initial implementation of power domains handling
by the PM core was that power domain callbacks would be executed in
addition to subsystem callbacks, so that it would be possible to
extend the subsystem callbacks by using power domains.  It turns out,
however, that this wouldn't be really convenient in some important
situations.

For example, there are systems in which power can only be removed
from entire power domains.  On those systems it is not desirable to
execute device drivers' PM callbacks until it is known that power is
going to be removed from the devices in question, which means that
they should be executed by power domain callbacks rather then by
subsystem (e.g. bus type) PM callbacks, because subsystems generally
have no information about what devices belong to which power domain.
Thus, for instance, if the bus type in question is the platform bus
type, its PM callbacks generally should not be called in addition to
power domain callbacks, because they run device drivers' callbacks
unconditionally if defined.

While in principle the default subsystem PM callbacks, or a subset of
them, may be replaced with different functions, it doesn't seem
correct to do so, because that would change the subsystem's behavior
with respect to all devices in the system, regardless of whether or
not they belong to any power domains.  Thus, the only remaining
option is to make power domain callbacks take precedence over
subsystem callbacks.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Grant Likely <grant.likely@secretlab.ca>
Acked-by: Kevin Hilman <khilman@ti.com>

It turns out that some PCI devices are only found to be
wakeup-capable during registration, in which case, when
device_set_wakeup_capable() is called, device_is_registered() already
returns 'true' for the given device, but dpm_sysfs_add() hasn't been
called for it yet.  This leads to situations in which the device's
power.can_wakeup flag is not set as requested because of failing
wakeup_sysfs_add() and its wakeup-related sysfs files are not
created, although they should be present.  This is a post-2.6.38
regression introduced by commit cb8f51bdadb7969139c2e39c2defd4cde98c1
(PM: Do not create wakeup sysfs files for devices that cannot wake
up).

To work around this problem initialize the device's power.entry
field to an empty list head and make device_set_wakeup_capable()
check if it is still empty before attempting to add the devices
wakeup-related sysfs files with wakeup_sysfs_add().  Namely, if
power.entry is still empty at this point, device_pm_add() hasn't been
called yet for the device and its wakeup-related files will be
created later, so device_set_wakeup_capable() doesn't have to create
them.

Reported-and-tested-by: Tino Keitel <tino.keitel@tikei.de>
Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>

Xen save/restore is going to use hibernate device callbacks for
quiescing devices and putting them back to normal operations and it
would need to select CONFIG_HIBERNATION for this purpose.  However,
that also would cause the hibernate interfaces for user space to be
enabled, which might confuse user space, because the Xen kernels
don't support hibernation.  Moreover, it would be wasteful, as it
would make the Xen kernels include a substantial amount of code that
they would never use.

To address this issue introduce new power management Kconfig option
CONFIG_HIBERNATE_CALLBACKS, such that it will only select the code
that is necessary for the hibernate device callbacks to work and make
CONFIG_HIBERNATION select it.  Then, Xen save/restore will be able to
select CONFIG_HIBERNATE_CALLBACKS without dragging the entire
hibernate code along with it.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Tested-by: Shriram Rajagopalan <rshriram@cs.ubc.ca>

The code handling system-wide power transitions (eg. suspend-to-RAM)
can in theory execute callbacks provided by the device's bus type,
device type and class in each phase of the power transition.  In
turn, the runtime PM core code only calls one of those callbacks at
a time, preferring bus type callbacks to device type or class
callbacks and device type callbacks to class callbacks.

It seems reasonable to make them both behave in the same way in that
respect.  Moreover, even though a device may belong to two subsystems
(eg. bus type and device class) simultaneously, in practice power
management callbacks for system-wide power transitions are always
provided by only one of them (ie. if the bus type callbacks are
defined, the device class ones are not and vice versa).  Thus it is
possible to modify the code handling system-wide power transitions
so that it follows the core runtime PM code (ie. treats the
subsystem callbacks as mutually exclusive).

On the other hand, the core runtime PM code will choose to execute,
for example, a runtime suspend callback provided by the device type
even if the bus type's struct dev_pm_ops object exists, but the
runtime_suspend pointer in it happens to be NULL.  This is confusing,
because it may lead to the execution of callbacks from different
subsystems during different operations (eg. the bus type suspend
callback may be executed during runtime suspend of the device, while
the device type callback will be executed during system suspend).

Make all of the power management code treat subsystem callbacks in
a consistent way, such that:
(1) If the device's type is defined (eg. dev->type is not NULL)
    and its pm pointer is not NULL, the callbacks from dev->type->pm
    will be used.
(2) If dev->type is NULL or dev->type->pm is NULL, but the device's
    class is defined (eg. dev->class is not NULL) and its pm pointer
    is not NULL, the callbacks from dev->class->pm will be used.
(3) If dev->type is NULL or dev->type->pm is NULL and dev->class is
    NULL or dev->class->pm is NULL, the callbacks from dev->bus->pm
    will be used provided that both dev->bus and dev->bus->pm are
    not NULL.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Acked-by: Kevin Hilman <khilman@ti.com>
Reasoning-sounds-sane-to: Grant Likely <grant.likely@secretlab.ca>
Acked-by: Greg Kroah-Hartman <gregkh@suse.de>

The platform bus type is often used to handle Systems-on-a-Chip (SoC)
where all devices are represented by objects of type struct
platform_device.  In those cases the same "platform" device driver
may be used with multiple different system configurations, but the
actions needed to put the devices it handles into a low-power state
and back into the full-power state may depend on the design of the
given SoC.  The driver, however, cannot possibly include all the
information necessary for the power management of its device on all
the systems it is used with.  Moreover, the device hierarchy in its
current form also is not suitable for representing this kind of
information.

The patch below attempts to address this problem by introducing
objects of type struct dev_power_domain that can be used for
representing power domains within a SoC.  Every struct
dev_power_domain object provides a sets of device power
management callbacks that can be used to perform what's needed for
device power management in addition to the operations carried out by
the device's driver and subsystem.

Namely, if a struct dev_power_domain object is pointed to by the
pwr_domain field in a struct device, the callbacks provided by its
ops member will be executed in addition to the corresponding
callbacks provided by the device's subsystem and driver during all
power transitions.

Signed-off-by: Rafael J. Wysocki <rjw@sisk.pl>
Tested-and-acked-by: Kevin Hilman <khilman@ti.com>