Re: perf: Support uncore in 4.9.112

["Jin, Yao" <yao.jin@linux.intel.com>]

On 7/19/2018 1:39 AM, Greg KH wrote:
> On Wed, Jul 18, 2018 at 10:09:44AM -0700, Andi Kleen wrote:
>> On Wed, Jul 18, 2018 at 11:29:50AM +0200, Greg KH wrote:
>>> On Wed, Jul 18, 2018 at 08:41:28AM +0800, Jin, Yao wrote:
>>>> Hi,
>>>>
>>>> The stable kernel 4.9.112 has supported Intel uncore feature in perf core.
>>>> While it also needs the perf tool supporting to let perf uncore feature
>>>> work.
>>>>
>>>> Following backport patches enables basic perf uncore feature in 4.9.112.
>>>>
>>>> For example, on skylake desktop,
>>>
>>> Why would anyone care about this on a "desktop" for 4.9?  No one should
>>> be using 4.9.y on a desktop anymore, it's over 2 years old, why would
>>> they expect any "new" hardware support to work for them?  Why can't they
>>
>> It's actually not new hardware support: Skylake is fairly old hardware
>> at this point.
> 
> So is 4.9.  I don't understand your point.  The hardware is obviously
> newer than 4.9 was, otherwise the support for it would already be in
> there, right?
> 
>>> just use 4.14.y or better yet. 4.17.y?  Desktops should NOT be using a 2
>>> year old kernel.
>>>
>>> Heck, servers shouldn't either, but that's a totally different rant.
>>
>> These chips are not only used in desktops but also in servers.
> 
> This was asked for with regards to desktops, so now I'm confused.
> Exactly who/what is going to be needing/wanting/using these changes?
> 

This patchset is not only regarding to desktop but also for servers. 
Sorry, my example in patch description brings confusion.

The patchset supports the server like Skylake, and it also supports some 
old servers, for example, Broadwell and Haswell.

4.9 kernel has uncore patches yet but at the perf tool side it doesn't 
have associated uncore patches, so actually the perf uncore feature 
doesn't work in 4.9.

We just want to enable the perf uncore feature in 4.9, it's especially 
useful for server performance analysis.

Thanks
Jin Yao

>>> However, for hardware that is newer than the base kernel version
>>> release, I have no sympathy.  Just use a newer kernel, right?
>>
>> We have customers which are on old kernels with new hardware.
> 
> That's obviously not a wise thing to do for lots of good reasons.  This
> exact example being a huge one (i.e. you can't go back in time and add
> support for hardware that was not out yet.)
> 
>> The backports happen either way. This is just an attempt to do it in a
>> coordinated fashion.
> 
> There was no coordination here, just a list of git commit ids.  Which is
> great, and all that is really needed, but I'm confused as to who is
> trying to coordinate with who?
> 
>>> What distro relies on a 4.9 kernel for brand new hardware that does not
>>> already support a newer kernel release for such hardware?
>>
>> None afaik, but there is a lot of Linux use beyond distros.
> 
> So no distro uses this, which makes me really wonder who would be the
> user of these backports.  And for how long?  Why are these people not
> moving to 4.14 already given that the published date for 4.9 end-of-life
> is getting very close.  Are you expecting to be rescued by Google again?
> 
> That can't be true as Android doesn't care about x86 servers :)
> 
> Please provide real solid information, including the answer to the other
> question I asked in my response, which was not included here for some
> reason.
> 
> thanks,
> 
> greg k-h
>