Re: [PATCH v2 3/3] perf vendor events amd: update Zen1 events to V2

[Kim Phillips <kim.phillips@amd.com>]

On 2/27/20 2:00 PM, Vijay Thakkar wrote:
>> OSRR for AMD Family 17h processors, Models 00h-2Fh, 56255 Rev 3.03 - July, 2018
> I have included this for v3 that I will submit later, including all the
> changes for the FPU counters. Sorry, I messed up copy-pasting the text
> and forgot to change the trailing pipe number.

Please also look at addressing the review comments for patch 2 of 3.

>> and their counts don't seem to match up very well when running
>> various workloads.  The microarchitecture is likely to have changed
>> in this area from families prior to 17h, so a MAB alloc can likely
>> count different events than what is presumed here: a Data cache
>> load/store/prefetch miss.
>>
>> I think it's safer to just leave the PPR text "LS MAB Allocates
>> by Type" as-is, instead of assuming they are L1 load/store misses.
>> What do you think?
> 
> I did some checking accross PPRs, and this counter seems to have changed
> names multiple times throughout the PPR revisions. 
> 
> Zen1 PPR (54945 Rev 1.14 - April 15, 2017) lists counter called "LsMabAllocPipe"
> with 5 subcounters that have different names compared to ones we see in
> the mainline right now. PPRs for stepping B2
> onwards change this to the 3 sub-counter and primary counter name
> we see right now. This public description still changes accross various
> PPR revisions, which is why I had this set to what it was. The lastest
> PPR I can find is indeed lists it as "LS MAB Allocates by Type";
> I will change it to that with the fuffix of tehe sub-counter name. Since
> the same counter is in Zen2 as well, I will make the same changes there
> too.

Thanks, yes, and if you look at the Software Optimization Guide that I
just added to the bugzilla [1], Figure 7 "Load-Store Unit" on page 40
shows a MAB block separate from the Data Cache block.

Kim

[1] https://bugzilla.kernel.org/show_bug.cgi?id=206537