Re: [PATCH] perf stat: Support per-cluster aggregation

[Jonathan Cameron <Jonathan.Cameron@Huawei.com>]

On Fri, 24 Mar 2023 12:24:22 +0000
Jonathan Cameron <Jonathan.Cameron@Huawei.com> wrote:

> On Fri, 24 Mar 2023 10:34:33 +0800
> Jie Zhan <zhanjie9@hisilicon.com> wrote:
> 
> > On 13/03/2023 16:59, Yicong Yang wrote:  
> > > From: Yicong Yang <yangyicong@hisilicon.com>
> > >
> > > Some platforms have 'cluster' topology and CPUs in the cluster will
> > > share resources like L3 Cache Tag (for HiSilicon Kunpeng SoC) or L2
> > > cache (for Intel Jacobsville). Currently parsing and building cluster
> > > topology have been supported since [1].
> > >
> > > perf stat has already supported aggregation for other topologies like
> > > die or socket, etc. It'll be useful to aggregate per-cluster to find
> > > problems like L3T bandwidth contention or imbalance.
> > >
> > > This patch adds support for "--per-cluster" option for per-cluster
> > > aggregation. Also update the docs and related test. The output will
> > > be like:
> > >
> > > [root@localhost tmp]# perf stat -a -e LLC-load --per-cluster -- sleep 5
> > >
> > >   Performance counter stats for 'system wide':
> > >
> > > S56-D0-CLS158    4      1,321,521,570      LLC-load
> > > S56-D0-CLS594    4        794,211,453      LLC-load
> > > S56-D0-CLS1030    4             41,623      LLC-load
> > > S56-D0-CLS1466    4             41,646      LLC-load
> > > S56-D0-CLS1902    4             16,863      LLC-load
> > > S56-D0-CLS2338    4             15,721      LLC-load
> > > S56-D0-CLS2774    4             22,671      LLC-load
> > > [...]
> > >
> > > [1] commit c5e22feffdd7 ("topology: Represent clusters of CPUs within a die")
> > >
> > > Signed-off-by: Yicong Yang <yangyicong@hisilicon.com>    
> > 
> > An end user may have to check sysfs to figure out what CPUs those 
> > cluster IDs account for.
> > 
> > Any better method to show the mapping between CPUs and cluster IDs?  
> 
> The cluster code is capable of using the ACPI_PPTT_ACPI_PROCESSOR_ID field
> if valid for the cluster level of PPTT.
> 
> The numbers in the example above look like offsets into the PPTT table
> so I think the PPTT table is missing that information.
> 
> Whilst not a great description anyway (it's just an index), the UUID
> that would be in there can convey more info on which cluster this is.
> 
> 
> > 
> > Perhaps adding a conditional cluster id (when there are clusters) in the 
> > "--per-core" output may help.  
> 
> That's an interesting idea.  You'd want to include the other levels
> if doing that.  So whenever you do a --per-xxx it also provides the
> cluster / die / node / socket etc as relevant 'above' the level of xxx
> Fun is that node and die can flip which would make this tricky to do.

Ignore me on this.  I'd not looked at the patch closely when I wrote
this.  Clearly a lot of this information is already provided - the
suggestion was to consider adding cluster to that mix which makes
sense to me.

Jonathan

> 
> Jonathan
> 
> > 
> > Apart form that, this works well on my aarch64.
> > 
> > Tested-by: Jie Zhan <zhanjie9@hisilicon.com>  
> 
> > 
> >   
>