On 3/31/15 3:13 AM, Ingo Molnar wrote:
>
> * Peter Zijlstra <[email protected]> wrote:
>
>> On Mon, Mar 30, 2015 at 10:43:14AM +0200, Mike Galbraith wrote:
>>
>>> I think it would be a good thing if we can get away with it, but I
>>> also think you could safely bet your life that somebody will
>>> squeak.
>>
>> The thing I worry most about is that squeaking only happening 5
>> years later :/
>
> So lets start by keeping the sysctl thing with the very
> scheduler-internal names, but all zeroes and no effect of any change -
> i.e. a dead API in all but appearance. I don't think there's any
> legitimate use of those, beyond debugging, as we could change the
> internal implementation anymore and moot many of those flags.
>
> So lets trigger the squeaking that way. If any complaint comes in
> beyond 1-2 kernel releases then I don't think it's a regression, it
> turns into a feature request ...
Sorry to be so dense but I am not clear on what is acceptable and not.
As mentioned in a previous response, these are the scheduler related
files I am aware of:
- debufs file for sched_features (/sys/kernel/debug/sched_features)
- /proc/sys/kernel/sched_domain for tweaking scheduling parameters
- /proc/sched_debug - various internal stats
- /sys/devices/system/cpu entries. e.g., for cpu topology (physical
package id, core id, sibling cores and threads)
None of them show the sched_domain information which is the subject of
this patch.
Can someone clarify on the duplication concerns and such?
Thanks,
David
On Sun, 2015-04-05 at 20:10 -0600, David Ahern wrote:
>
> Can someone clarify on the duplication concerns and such?
My concern was eg you display flags in /sys, but to tweak flags etc,
you must visit /proc. Seems to me the missing info should either be
added to /proc (some under DEBUG?), or the lot should move to /sys.
I don't care all that much, making span etc visible anywhere is an win
over boot flag + big box (esp. + serial console).
-Mike