On Wed, 9 Apr 2014 14:22:42 +0600 Roman Mamedov <r...@romanrm.net> wrote:
> On Wed, 9 Apr 2014 10:57:22 +0300 > Siarhei Siamashka <siarhei.siamas...@gmail.com> wrote: > > > > (there needs to be a power-saving alternative to 'performance' that's > > > selectable without recompiling the kernel), > > > > Are you retracting your objection against 'interactive' then? > > The main objection is that it's not in the mainline, and I couldn't find any > sign it's ever going to be. 'Interactive' seems to be just one more of those > "cute embedded nonsense hacks"[1]. But "that's just, like, my opinion, man"; I > don't think I am in any position to deny inclusion of patches into > linux-sunxi. Well, you are bikeshedding this defconfig stuff since a long time ago, effectively blocking any attempts to cleanup the cpufreq mess :-) So now we have no cpufreq for sun5i, the 'ondemand' default governor for sun4i and the 'fantasy' default governor for sun7i. A stalemate? But indeed, you are even not using these defconfigs yourself, but offering some of your own: https://romanrm.net/a10/kernel So any possible changes to linux-sunxi defconfigs do not affect you in any way. > The major mainline power-saving alternative to 'performance' is 'ondemand'. But 'ondemand' is total crap, at least in the sunxi-3.4 kernel. And IMHO it does not make any sense to permit this crap to be exposed as the out-of-the-box experience for users. It is not like the availability of the 'ondemand' governor is mandatory. The users can always configure and compile their kernels themselves, or use pre-built kernels provided by people like you. Also the sets of frequencies and voltages for cpufreq are specified in the fex files. We can't be sure that all these operating points are really configured properly for all the sunxi devices. If one of the intermediate rarely used frequencies is not configured right (has wrong voltage for example), then the users may also have some very hard to debug reliability problems. Again, if we are not gaining much from the 'ondemand' governor in terms of power consumption reduction, then there is no point to potentially compromise reliability. There are some reports about mysterious deadlocks on some devices. Let's take the 'ondemand' governor out of the picture and see if this helps. > Another one is 'conservative', yes, but it seems to be directly contrary to > your goal, that is to maintain desktop interactivity in all situations. > Also "ondemand does a better job on all workloads"[2]. Yeah, maybe it did in 2008, on some other hardware. But it is 2014 now, and we mostly care about sunxi hardware with the currently available kernels. Our first hand experience does matter. > > Or what about the other alternatives? > > There's currently about 30 of them; knock yourself out[3] :) > Some of which with names like Smoothass, Brazilianwax and Lulzactive sound > totally rad, I bet they're light-years ahead of this 'interactive', we should > add those into the kernel. If you really want to talk about this, the tests of all these governors in some practical usage scenarios are very much welcome. > > [1] > http://www.cnx-software.com/2014/02/01/cute-embedded-nonsense-hacks-nouveau-driver-for-tegra-k1-and-android-defaults-to-art/ > > [2] http://comments.gmane.org/gmane.linux.redhat.fedora.kernel/1360 > > [3] http://rootzwiki.com/topic/40336-cpu-governors-explained/ > -- Best regards, Siarhei Siamashka -- You received this message because you are subscribed to the Google Groups "linux-sunxi" group. To unsubscribe from this group and stop receiving emails from it, send an email to linux-sunxi+unsubscr...@googlegroups.com. For more options, visit https://groups.google.com/d/optout.