Messages in this thread | | | Date | Wed, 20 May 2015 10:22:21 -0700 (PDT) | From | Vikas Shivappa <> | Subject | Re: [PATCH 3/7] x86/intel_rdt: Add support for cache bit mask management |
| |
On Mon, 18 May 2015, Thomas Gleixner wrote:
> On Mon, 18 May 2015, Vikas Shivappa wrote: >> On Fri, 15 May 2015, Thomas Gleixner wrote: >>>> +static inline bool intel_rdt_update_cpumask(int cpu) >>>> +{ >>>> + } >>> >>> You must be kidding. >> >> the rapl and cqm use similar code. You want me to keep a seperate package mask >> for this code which not would be that frequent at all ? > > You find for everything a place where you copied your stuff from > without thinking about it, right? > > Other people dessperately try to fix the cpu online times which are > more and more interesting the larger the systems become. So it might > be a good idea to come up with a proper fast implementation which can > be used everywhere instead of blindly copying code.
Ok , i can try to do this as a seperate patch after the cache allocation to get a support for faster implementation for traversing package and cpus in the packages which can be used by everyone. we would need to start from scratch with having packagemask_t equivalent to cpumask_t. hope that is fair ?
Thanks, Vikas
> >> for c-state transitions only idle_notifier gets called - so this is only when >> a new package is physically added. and we dont need anything for cache alloc >> for idle transitions. not really frequent ? > > Crap. It's called for every cpu which comes online and goes offline, > not for new packages. > > Thanks, > > tglx >
| |