On 10 June 2014 23:42, Tomasz Figa <tomasz.figa@xxxxxxxxx> wrote: > On 10.06.2014 23:27, Greg Kroah-Hartman wrote: >> On Tue, Jun 10, 2014 at 11:27:45PM +0200, Rafael J. Wysocki wrote: >>> On Tuesday, June 10, 2014 02:53:26 PM Ulf Hansson wrote: >>>> On 10 June 2014 14:11, Rafael J. Wysocki <rafael@xxxxxxxxxx> wrote: >>>>> On Tue, Jun 10, 2014 at 12:51 PM, Ulf Hansson <ulf.hansson@xxxxxxxxxx> wrote: >>>>>> From: Tomasz Figa <t.figa@xxxxxxxxxxx> >>>>>> >>>>>> On a number of platforms, devices are part of controllable power >>>>>> domains, which need to be enabled before such devices can be accessed >>>>>> and may be powered down when the device is idle to save some power. >>>>>> This means that on systems that support power domain control using >>>>>> generic power domains subsystem, it is necessary to add device to its >>>>>> power domain before binding a driver to it and remove it from its power >>>>>> domain after its driver is unbound to make sure that an unused device >>>>>> does not affect power domain state. >>>>>> >>>>>> Since this is not limited to particular busses and specific >>>>>> archs/platforms, >>>>> >>>>> Actually, this isn't correrct. It is limited to the platforms that >>>>> use Device Trees now. >>>> >>>> Correct, we should update the commit message/docs. >>>> >>>>> >>>>> Moreover, it is not consistent with the way we add devices to the ACPI PM >>>>> domain, which is the ACPI counterpart of this. >>>> >>>> I am not sure why you think consistency for ACPI is important here. >>>> ACPI PM will still be able to handle it's domain/device registering as >>>> before. There are even other pm_domains that don't use genpd which >>>> need to handle this themselves. >>> >>> My point is that doing things like that in different places for different >>> firmware interfaces is confusing and likely to lead to coding mistakes in >>> the future. >>> >>>> Or are you saying that you prefer bus notifiers in favour of making >>>> use of the driver core for this matter? >>> >>> Well, please grep for acpi_dev_pm_attach() and see where it is done. >>> Surely not in drivers/base/dd.c. Also I'm not sure why you're talking >>> about bus notifiers in this context. >>> >>>> Shouldn't the driver core handle most of the common things for a device >>>> driver? >>> >>> Common, yes. Platform-specific, no. >>> >>>> Let's compare how the pinctrls are being managed in the driver core, for >>>> example. >>> >>> pinctrl has Device Trees support only at the moment (as far as firmware >>> interfaces go) and quite frankly I'm not sure if/how we'll need to change >>> it to cover ACPI as well. >>> >>> But for power domains, please keep that stuff away from dd.c. That is, >>> unless Greg specifically disagrees with me and decides to apply this >>> patch regardless. :-) >> >> Nope, no disagreement from me toward you at all here, keep up the good >> work :) > > OK, so proposed solution is to put this in: > > - platform_drv_probe(), > - spi_drv_probe(), > - i2c_device_probe(), > - amba_probe(), > > ... > > - and any other bus type, which can have devices instantiated from DT. > > If this is what you mean, I still think putting this in dd is cleaner > and more scalable, but I'm not going to insist, as I believe you have > good reasons to prefer this approach over current one. I am not so sure we should ignore scalability. Currently there are SoCs those that uses bus notifiers to handle the device registration for it's pm_domain. When each of them converts to use bus' ->probe() function instead, they will have their own version of acpi_dev_pm_attach(), and the scalability becomes an issue. An option could be to invent a common "pm_domain_attach_dev() API", which will check what kind of power domain the device may reside in. In principle invoke acpi_dev_pm_attach() and if no attachment happens, try genpd_bind_domain(). Kind regards Ulf Hansson -- To unsubscribe from this list: send the line "unsubscribe devicetree" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
