Date:   Thu, 31 Mar 2022 20:19:02 +0800
From:   Leo Yan <leo.yan@...aro.org>
To:     Ali Saidi <alisaidi@...zon.com>
Cc:     Nick.Forrington@....com, acme@...nel.org,
        alexander.shishkin@...ux.intel.com, andrew.kilroy@....com,
        benh@...nel.crashing.org, german.gomez@....com,
        james.clark@....com, john.garry@...wei.com, jolsa@...nel.org,
        kjain@...ux.ibm.com, lihuafei1@...wei.com,
        linux-arm-kernel@...ts.infradead.org, linux-kernel@...r.kernel.org,
        linux-perf-users@...r.kernel.org, mark.rutland@....com,
        mathieu.poirier@...aro.org, mingo@...hat.com, namhyung@...nel.org,
        peterz@...radead.org, will@...nel.org
Subject: Re: [PATCH v4 2/4] perf arm-spe: Use SPE data source for neoverse
 cores

Hi Ali,

On Tue, Mar 29, 2022 at 02:32:14PM +0000, Ali Saidi wrote:

[...]

> > I still think we should consider to extend the memory levels to
> > demonstrate clear momory hierarchy on Arm archs, I personally like the
> > definitions for "PEER_CORE", "LCL_CLSTR", "PEER_CLSTR" and "SYS_CACHE",
> > though these cache levels are not precise like L1/L2/L3 levels, they can
> > help us to map very well for the cache topology on Arm archs and without
> > any confusion.  We could take this as an enhancement if you don't want
> > to bother the current patch set's upstreaming.
> 
> I'd like to do this in a separate patch, but I have one other proposal. The
> Neoverse cores L2 is strictly inclusive of the L1, so even if it's in the L1,
> it's also in the L2. Given that the Graviton systems and afaik the Ampere
> systems don't have any cache between the L2 and the SLC, thus anything from
> PEER_CORE, LCL_CLSTR, or PEER_CLSTR would hit in the L2, perhaps we
> should just set L2 for these cases? German, are you good with this for now? 

If we use a single cache level (no matterh it's L2 or ANY_CACHE) for
these data sources, it's hard for users to understand what's the cost
for the memory operations.  So here I suggested for these new cache
levels is not only about cache level, it's more about the information
telling the memory operation's cost.

[...]

> > Alternatively, I think it's good to pick up the patch series "perf c2c:
> > Sort cacheline with all loads" [1], rather than relying on HITM tag, the
> > patch series extends a new option "-d all" for perf c2c, so it displays
> > the suspecious false sharing cache lines based on load/store ops and
> > thread infos.  The main reason for holding on th patch set is due to we
> > cannot verify it with Arm SPE at that time point, as the time being Arm
> > SPE trace data was absent both store ops and data source packets.
> 
> Looking at examples I don't, at least from my system, data-source isn't set for
> stores, only for loads.

Ouch ...  If data source is not set for store operation, then all store
samples will absent cache level info.  Or should we set ANY_CACHE as
cache level for store operations?

> > I perfer to set PERF_MEM_SNOOP_HIT flag in this patch set and we can
> > upstream the patch series "perf c2c: Sort cacheline with all loads"
> > (only needs upstreaming patches 01, 02, 03, 10, 11, the rest patches
> > have been merged in the mainline kernel).
> > 
> > If this is fine for you, I can respin the patch series for "perf c2c".
> > Or any other thoughts?
> 
> I think this is a nice option to have in the tool-box, but from my point of
> view, I'd like someone who is familiar with c2c output on x86 to come to an
> arm64 system and be able to zero in on a ping-ponging line like they would
> otherwise. Highlighting a line that is moving between cores frequently which is
> likely in the exclusive state by tagging it an HITM accomplishes this and will
> make it easier to find these cases.  Your approach also has innaccurancies and
> wouldn't be able to differentiate between core X accessing a line a lot followed
> by core Y acessing a line alot vs the cores ping-ponging.  Yes, I agree that we
> will "overcount" HITM, but I don't think this is particularly bad and it does
> specifically highlight the core-2-core transfers that are likely a performance
> issue easily and it will result in easier identification of areas of false or
> true sharing and improve performance.

I don't want to block this patch set by this part, and either I don't
want to introduce any confusion for later users, especially I think
users who in later use this tool but it's hard for them to be aware any
assumptions in this discussion thread.  So two options would be fine
for me:

Option 1: if you and Arm mates can confirm that inaccuracy caused by
setting HITM is low (e.g. 2%-3% inaccuracy that introduced by directly
set HITM), I think this could be acceptable.  Otherwise, please
consider option 2.

Option 2: by default we set PERF_MEM_SNOOP_HIT flag since now actually
we have no info to support HITM.  Then use a new patch to add an extra
option (say '--coarse-hitm') for 'perf c2c' tool, a user can explictly
specify this option for 'perf c2c' command; when a user specifies this
option it means that the user understands and accepts inaccuracy by
forcing to use PERF_MEM_SNOOP_HITM flag.  I think you could refer to
the option '--stitch-lbr' for adding an option for 'perf c2c' tool.

Thanks,
Leo

Powered by blists - more mailing lists