On Thu, 19 Mar 2015, Steven Rostedt wrote:
> On Thu, 19 Mar 2015 15:16:25 -0700
> Linus Torvalds <[email protected]> wrote:
>
> > So I don't think the ring-buffer change is necessarily _wrong_, but if
> > this is a performance issue, why don't we just fix it up for the
> > generic case rather than for just one user?
>
> I totally agree with your analysis, but it's up to Christoph to come up
> with an answer to your questions.
Something beyond: Do not use this_cpu_* when preemption is already
off but use __this_cpu_*?
On Tue, 19 May 2015 10:35:32 -0500 (CDT)
Christoph Lameter <[email protected]> wrote:
> On Thu, 19 Mar 2015, Steven Rostedt wrote:
>
> > On Thu, 19 Mar 2015 15:16:25 -0700
> > Linus Torvalds <[email protected]> wrote:
> >
> > > So I don't think the ring-buffer change is necessarily _wrong_, but if
> > > this is a performance issue, why don't we just fix it up for the
> > > generic case rather than for just one user?
> >
> > I totally agree with your analysis, but it's up to Christoph to come up
> > with an answer to your questions.
>
> Something beyond: Do not use this_cpu_* when preemption is already
> off but use __this_cpu_*?
I think the question was, why exactly does the generic this_cpu_read()
require disabling preemption? What breaks if it is not disabled?
-- Steve
On Tue, 19 May 2015, Steven Rostedt wrote:
> On Tue, 19 May 2015 10:35:32 -0500 (CDT)
> Christoph Lameter <[email protected]> wrote:
>
> > On Thu, 19 Mar 2015, Steven Rostedt wrote:
> >
> > > On Thu, 19 Mar 2015 15:16:25 -0700
> > > Linus Torvalds <[email protected]> wrote:
> > >
> > > > So I don't think the ring-buffer change is necessarily _wrong_, but if
> > > > this is a performance issue, why don't we just fix it up for the
> > > > generic case rather than for just one user?
> > >
> > > I totally agree with your analysis, but it's up to Christoph to come up
> > > with an answer to your questions.
> >
> > Something beyond: Do not use this_cpu_* when preemption is already
> > off but use __this_cpu_*?
>
> I think the question was, why exactly does the generic this_cpu_read()
> require disabling preemption? What breaks if it is not disabled?
Ok I answered that before. There is a cache line population/eviction
issue (because it may happen on the wrong cache) but basically I think
its mostly there for symmetries sake. We could remove
this_cpu_read/write and let __this_cpu_read()/__this_cpu_write be used in non
preemptible contexts.