Re: [PATCH 4/4] nohz: Set isolcpus when nohz_full is set

[Rik van Riel]

On 05/18/2015 10:22 AM, Mike Galbraith wrote:
> On Mon, 2015-05-18 at 10:07 -0400, Rik van Riel wrote:
>> On 05/17/2015 11:29 PM, Mike Galbraith wrote:
>>> On Sun, 2015-05-17 at 22:17 -0400, Rik van Riel wrote:
>>>> On 05/17/2015 01:30 AM, Mike Galbraith wrote:
>>>>
>>>>> Given that kernel initiated association to isolcpus, a user turning
>>>>> NO_HZ_FULL_ALL on had better not have much generic load to manage.  If
>>>>> he/she does not have CPUSETS enabled, or should Rik's patch rendering
>>>>> isolcpus immutable be merged, 
>>>>
>>>> My patch does not aim to make isolcpus immutable, it aims to make
>>>> isolcpus resistent to system management tools (like libvirt)
>>>> automatically undoing isolcpus the instant a cpuset with the default
>>>> cpus (inherited from the root group) is created.
>>>
>>> Aim or not, if cpusets is the sole modifier, it'll render isolcpus
>>> immutable, no?  Cpusets could grow an override to the override I
>>> suppose, to regain control of the resource it thinks it manages.
>>
>> The other way would be to make /sys/devices/system/cpu/isolcpus
>> (which Greg KH promised he would queue up for 4.2) writable.
> 
> Anything is better than override the override.  That's easy, but the
> changelog would have to be farmed out to a talented used car salesman.

For real time KVM, it is desirable to run the VCPU threads on
isolated CPUs as real time tasks.

Meanwhile, the emulator threads can run as normal tasks anywhere
on the system.

This means the /machine cpuset, which all guests live under,
needs to contain all the system's CPUs, both isolated and
non-isolated, otherwise it will be unable to have the VCPU
threads on isolated CPUs and the emulator threads on other
CPUs.

-- 
All rights reversed
--
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/