2019-05-23 13:11:19

by Mukesh Ojha

[permalink] [raw]
Subject: Perf: Preserving the event across CPU hotunplug/plug and Creation of an event on offine CPU

Hi Peter/All,

This is regarding the discussion happen in the past about
https://lkml.org/lkml/2018/2/15/1324

Where the exact ask is to allow preserving and creation of events on a
offline CPU, so that when the CPU
comes online it will start counting.

I had a look at your patch too and resolve crash during while trying to
create an event on an offline cpu.

In your patch,  you seem to disable event when cpu goes offline which is
exactly deleting the event
from the pmu and add when it comes online, it seems to  work.

But, For the purpose of allowing the creation of event while CPU is
offline is not able to count event while
CPU coming online, for that i did some change, that did work.

Also, I have query about the events which gets destroyed while CPU is
offline and we need to remove them
once cpu comes online right ? As Raghavendra also queried the same in
the above thread.

Don't we need  a list where we maintain the events which gets destroyed
while CPU is dead ?
and clean it  up when CPU comes online ?

Thanks.
Mukesh



2019-05-28 16:19:04

by Mukesh Ojha

[permalink] [raw]
Subject: Re: Perf: Preserving the event across CPU hotunplug/plug and Creation of an event on offine CPU

Friendly Ping.


On 5/23/2019 6:39 PM, Mukesh Ojha wrote:
> Hi Peter/All,
>
> This is regarding the discussion happen in the past about
> https://lkml.org/lkml/2018/2/15/1324
>
> Where the exact ask is to allow preserving and creation of events on a
> offline CPU, so that when the CPU
> comes online it will start counting.
>
> I had a look at your patch too and resolve crash during while trying
> to create an event on an offline cpu.
>
> In your patch,  you seem to disable event when cpu goes offline which
> is exactly deleting the event
> from the pmu and add when it comes online, it seems to  work.
>
> But, For the purpose of allowing the creation of event while CPU is
> offline is not able to count event while
> CPU coming online, for that i did some change, that did work.
>
> Also, I have query about the events which gets destroyed while CPU is
> offline and we need to remove them
> once cpu comes online right ? As Raghavendra also queried the same in
> the above thread.
>
> Don't we need  a list where we maintain the events which gets
> destroyed while CPU is dead ?
> and clean it  up when CPU comes online ?
>
> Thanks.
> Mukesh
>
>