On 15/12/19 15:43, Prarit Bhargava wrote:
>> Oh, I wasn't pointing out that it wasn't just lost, I
was pointing out
>> that NO_HZ_IDLE is not set because we run NO_HZ_FULL. We were one of
>> the first distros to do so, and it has worked well for us. I have a
>> fairly strong opinion about not dropping back to IDLE without good
>> reason.
>
> This wasn't a proposal to change anything here at all, sorry if that
> was the way it read. I was purely wondering, while digging through
> stuff around cpu idle, for the difference between arches.
>
> With the hit around NO_HZ_IDLE vs NO_HZ_FULL I dug some more and
> basically it seems the reason we don't have the later on the non
> x86_64 arches is because for some reason we unset
> VIRT_CPU_ACCOUNTING_GEN for all except x86_64, it looks to be
> historical, all our current architectures now look to support that
> option. Anyone aware of any reason we shouldn't use the
> NO_HZ_FULL/VIRT_CPU_ACCOUNTING_GEN as standard across all arches?
I don't know of a reason -- pbonzini? Have any input here?
No reason I know of, though note that VIRT_CPU_ACCOUNTING has nothing to
do with virtualization so my opinion may not count much. :)
IIRC there is no overhead for VIRT_CPU_ACCOUNTING unless context
tracking is enabled, is that correct? It should be all hidden behind a
static key.
Paolo