Messages in this thread | | | Date | Fri, 8 May 2015 13:55:57 -0700 (PDT) | From | Vikas Shivappa <> | Subject | Re: [PATCH 4/7] x86/intel_rdt: Implement scheduling support for Intel RDT |
| |
On Fri, 8 May 2015, Peter Zijlstra wrote:
> On Thu, May 07, 2015 at 04:15:41PM -0700, Vikas Shivappa wrote: >>> No. >>> >>> 1) two arch hooks right after one another is FAIL >>> 1a) just 'fix' the existing hook >>> 2) current is cheap and easily obtainable without passing it as >>> an argument >> >> will fix to just use an existing hook in finish_task_switch and >> current(get_current) since the stack would already be changed .. >> >> Thanks, >> Vikas >> >>> 3) why do you need the hook in the first place? >>> 3a) why can't you put this in __switch_to()? This is very much x86 only >>> code. > > ^ please also answer 3, why can't this go in __swtich_to()?
perf uses similar(#1a) hook to update its MSRs (including for cache monitoring ). Also since switch_to is for registers state and stack , may be a safer option to use it in the finish_arch_switch? Thats kind of why we had it there at first but had a seperate hook.
#define finish_arch_switch(prev) \ do { \ intel_rdt_sched_in(); \ } while (0)
Hpa , any comments ?
>
| |