On 15 November 2013 10:29, Bjørn Mork <bjorn@xxxxxxx> wrote: > Ulf Hansson <ulf.hansson@xxxxxxxxxx> writes: > >> I think it is kind of strange to give provision to userspace to >> control a "request based" power management feature in the kernel. Me >> personally can't think of any good use case, but I comes from the >> embedded/ARM world so I might not have the full picture. > > Because resuming a device inevitably adds latency, and that only > userspace can decide the latency vs power policy for any given device, > system and user? That is a valid point. Though we have pm_qos, should not that be the input for these kind of requirement for a driver? In other words, if the latency requirement can't be met for a device while re-activating it, it should not enter runtime suspend state. > > There are also other reasons, mostly related to device bugs. But the > above is the most important in my view. Without that control knob I > don't think I would have enabled runtime PM in any network driver while > the netif is running. The additional latency is noticable. > > Anyway, this is a pointless discussion. The power/control ABI is in use > so many places that it cannot be changed. I'd suggest googling "break > userspace" before even suggesting anything like that again. > I agree, and has never been a suggestion from my side either, more like a statement that I don't like the sysfs for runtime PM. :-) Kind regards Ulf Hansson > > Bjørn -- To unsubscribe from this list: send the line "unsubscribe linux-pci" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
