Re: [RFC v2 1/5] PM, Runtime, Add power_must_be_on flag

["Rafael J. Wysocki" <rjw@sisk.pl>]

On Saturday, May 05, 2012, huang ying wrote:
> On Sat, May 5, 2012 at 3:50 AM, Bjorn Helgaas <bhelgaas@google.com> wrote:
> > On Fri, May 4, 2012 at 2:13 AM, Huang Ying <ying.huang@intel.com> wrote:
> >> --- a/include/linux/pm.h
> >> +++ b/include/linux/pm.h
> >> @@ -536,6 +536,7 @@ struct dev_pm_info {
> >>        unsigned int            irq_safe:1;
> >>        unsigned int            use_autosuspend:1;
> >>        unsigned int            timer_autosuspends:1;
> >> +       unsigned int            power_must_be_on:1;
> >>        enum rpm_request        request;
> >>        enum rpm_status         runtime_status;
> >>        int                     runtime_error;
> >
> > It's a little weird to just add the field, with no users.  Would it
> > make sense to pull out the bits of patch 5 that use this and combine
> > them into a single smaller patch?
> 
> This patch is needed by some other subsystem too, such as ZPODD
> support in following patchset:
> 
> https://lkml.org/lkml/2012/3/28/23
> 
> The original plan is to merge this into linux-pm.git firstly, then
> merge various usage of this flag into various subsystem git tree.
> That will make cross-tree merging a little easier.  Is it possible?
> 
> > But see related comments there; it
> > might be safer to have a function that computes this whenever you need
> > it instead of caching the value.
> 
> This flag may be used (or calculated if implemented as a function)
> when device is in suspended state.  Reason is as follow from the
> changelog of this patch.
> 
> "
> In general, whether to put a device into power off state should be
> decided by the driver of the device, but for some buses, whether to
> put a device into power off state may be done by the parent of the
> device.  For example, a PCIe end point device may be put into power
> off state by the PCIe port connected to it.
> "
> 
> If all devices can calculate "power_must_be_on" when it is in
> suspended state, then this will be a design choice.
> 
> Hi, Rafael,
> 
> What do you think about the idea to replace .power_must_be_on with a
> function

That looks like a static characteristics of a device (ie. one that's not going
to change at run time), so it seems better to use a flag for it.

> (maybe add a new callback in pm_ops)?

Please don't do that.  We have many of them already.

Thanks,
Rafael