Am Montag, 27. Oktober 2014, 11:47:41 schrieb Felipe Balbi:

On Mon, Oct 27, 2014 at 05:41:03PM +0100, Johan Hovold wrote:

quoted

On Mon, Oct 27, 2014 at 11:38:40AM -0500, Felipe Balbi wrote:

quoted

On Mon, Oct 27, 2014 at 04:26:46PM +0000, Romain Perier wrote:

quoted

As discussed on the mailing list, it makes more sense to rename this
property to "system-power-controller". Problem being that the word
"source" usually tends to be used for inputs and that is out of
control of the OS. The poweroff capability is an output which simply
turns the system-power off. Also, this property might be used by
drivers which power-off the system and power back on subsequent RTC
alarms. This seems to suggest to remove "poweroff" from the property
name and to choose "system-power-controller" as the more generic
name. This patchs adds the required renaming changes and defines an
helper function which is compatible with both properties, the old one
prefixed by a vendor name and the new one without any prefix.

I think you still need to support poweroff-source since it has been
released on a stable kernel. Perhaps add a warning message telling users
it's deprecated and asking them to switch over to
system-power-controller ? Still, simply removing it isn't very nice.

No, Romain sent a patch that replaced "<vendor>,system-power-controller"
with "poweroff-source". It's now in Mark's tree (for v3.19), and this
series "reverts" to the old name minus the vendor-prefix.

oh, so poweroff-source isn't in Linus' tree yet ? (/me goes grep)

Then it should be fine. My bad.

Many of the other comments are still valid because even though
poweroff-source isn't in mainline yet, this series still creates
bisection points which are broken. The best solution would be to drop
all those patches from Mark's tree. Read, not revert, drop.

There have never been any users of the poweroff-source. The act8846 in the 
radxarock would have been the first, but I held off with the dts patch as the 
naming issue came up at the same time.

So I guess if Romain keeps the renaming together there shouldn't be any other 
bad bisection points?