lkml.org 
[lkml]   [2012]   [Jun]   [20]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
    /
    Date
    SubjectRe: [PATCH 0/4] cputime: Virtual cputime accounting small cleanups and consolidation
    From
    2012/6/21 Peter Zijlstra <peterz@infradead.org>:
    > On Tue, 2012-06-19 at 15:43 +0200, Frederic Weisbecker wrote:
    >>
    >> I wish we could do more vtime cputime accounting consolidation
    >> but archs do the things pretty differently although I bet the
    >> behaviour could be more unified.
    >>
    > Yes.. so s390,ia64 use thread_info, ppc uses their paca (arch private
    > precursor to per-cpu data).
    >
    > So I understand why s390,ia64 want the sched hook, but I don't see why
    > ppc would need it, their account_process_tick() can fold whatever they
    > need on the tick.

    I think in any case you need to flush the time on a descheduling task otherwise
    its pending time will be accounted later to the next task when it
    receives an irq.

    So I fear we still need that sched switch hook even with per cpu data. This
    may be a simple account_system_vtime() call.

    > So I think reworking s390,ia64 to use per-cpu storage should get rid of
    > this switch hook altogether.
    >
    >
    > Now everybody using VIRT_CPU_ACCOUNTING has the
    > syscall/__irq_{enter,exit} hooks and uses 64bit cputime_t.
    >
    > IRQ_TIME_ACCOUNTING only has the __irq_{enter,exit} hook and uses
    > unsigned long cputime_t, it adds cputime_one_jiffy every tick when the
    > per-cpu counter is ahead of the cputime.
    >
    >
    > We could merge both and do away with the 64bit cputime thing by keeping
    > a (2nd) per-cpu kernel_cpustat which we fill with optional syscall/irq
    > hooks and have account_process_tick() first check if any of the
    > fine-grained fields overflow and if not, fall back to the regular tick
    > accounting (much like an extended irqtime_account_process_tick).
    >
    >
    > This would merge the fine-grain and tick based code-paths and do away
    > with the whole cputime_t mess, it would also merge the VIRT and IRQ
    > paths and make Frederic's optional syscall accounting trivial.
    >
    > Or am I missing something obvious here? -- its late after all.

    No that looks pretty sensible.

    > I'll try and write some code tomorrow.

    Thanks!


    \
     
     \ /
      Last update: 2012-06-21 03:21    [W:0.025 / U:0.712 seconds]
    ©2003-2016 Jasper Spaans. hosted at Digital OceanAdvertise on this site