On Thu, Dec 10, 2015 at 06:01:59PM +0000, Juri Lelli wrote: > On 10/12/15 15:59, Mark Brown wrote: > > On Thu, Dec 10, 2015 at 02:15:04PM +0000, Dietmar Eggemann wrote: > > > On 23/11/15 14:28, Juri Lelli wrote: > > > > The new attribute shows up as: > > > > /sys/devices/system/cpu/cpu*/cpu_capacity > > > This sysfs interface is not really needed for arm or arm64. People can > > > build the dt blob if they want to change the values. Less code to carry > > > ... Let's focus on the core functionality, which is parsing cpu capacity > > > from dt file to task scheduler for heterogeneous systems. > > That does make the tuning process much more cumbersome - users have to > > rebuild and reboot to tweak the numbers rather than just tweaking the > > numbers and rerunning the benchmark (which seems like something people > > would want to automate). > IMHO, this is not a tuning interface. It is an alternative interface, > w.r.t. DTs, that we could use to provide default capacity values to the > kernel. I'm proposing both here as they make both sense to me. Then we > might dedice for which one to go (or if we need some other way) or to > keep both for flexibility. Kind of repeating what I said in the other mail but I'd say that any interface which provides a mechanism for setting a magic number that influences system performance is providing tuning. It's hard to see how else to describe it to be honest.
Attachment:
signature.asc
Description: PGP signature
