On Friday, July 31, 2015 11:39:07 AM Viresh Kumar wrote: > On 30-07-15, 20:53, Rafael J. Wysocki wrote: > > Well, on ACPI systems we actually do probe CPU devices. We have a processor > > driver there that binds to CPU devices and the cpufreq driver is just a > > frontend to that. > > Hmm, maybe I need to look at that in detail.. > > > So question is what prevents DT-based systems from doing it analogously. > > Don't have an answer to it yet. > > > Now, even if you use a fake platform device for that (I'm sure there are > > reasons for doing that, but I'd very much like them to be explained), > > The other reason apart from the EPROBE_DEFER thing was to identify the > right driver for a platform. For multiplatform kernels, there can be > multiple cpufreq drivers present in the kernel and there was no other > way to identify the right driver platform wants to probe. > > > then > > all of the information on dependencies should already be available to the > > ->probe callback of that device's driver, so it can check them before > > registering the cpufreq interface, can't it? > > That's what we try to do today for cpufreq-dt, for example. But that > has to be done for every possible policy the system can have as all > might have separate resources to allocate. For cpufreq-dt, we do it > only for cpu0 today, and assume others will work as well if cpu0 can. > > The real deal is that we need a probe() per policy here, for which > init() fitted well :) > > > Essentially, what you're suggesting to do is something like: Make the > > ->probe > > of one device's driver register a subsys interface for a specific bus type > > and check what ->add_dev of that interface returns for each device on that > > bus and if that is -EPROBE_DEFER, return it as its own return value. Do you > > honestly think this is a good design? > > No. I don't really thing so. That's why I was asking for suggestions > to do it proper. Maybe processor driver is the way to look for, I will > investigate further on that. > > But until the time that is done, and I expect that to take some time, > can't we check the return value of ->add_dev()?
As I said, either do it everywhere, or do it nowhere (in which case it can be void). Doing it in one place only is plain confusing and generally incorrect. Thanks, Rafael -- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to [email protected] More majordomo info at http://vger.kernel.org/majordomo-info.html Please read the FAQ at http://www.tux.org/lkml/
