lkml.org 
[lkml]   [2014]   [Mar]   [5]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: [PATCH 31/48] uv: Replace __get_cpu_var


On 3/5/2014 1:57 PM, Christoph Lameter wrote:
> The driver seems to use local64_t to define a single static instance of a
> counter and then seems to think that it is safe to increment the counter
> from multiple processors using local64_inc and friends. Common
> misunderstanding and a reason why I wanted the this_cpu operations.
>
> The counters seem to be exported via module parameters.. So I guess we
> need to define these per cpu and then sum them up when they need to be
> displayed.
>
> Dimitri?
>
> Maybe lets move this outside of this patchset.
>

Hi Christoph,

I haven't had much chance yet to look over your proposed changes but
FYI, the counters are strictly feedback to insure that there are not
unhandled NMI events from the perf subsystem. The exact count is
irrelevant. IOW, counts in the double or triple digits is okay,
counts > 100,000 is definitely not okay (there are multiple millions
of perf events every 'perf top' refresh.)

I'm not sure if this alters how you want to approach the changes.

Thanks,
Mike


\
 
 \ /
  Last update: 2014-03-06 04:41    [W:0.119 / U:0.236 seconds]
©2003-2020 Jasper Spaans|hosted at Digital Ocean and TransIP|Read the blog|Advertise on this site