| lists.openwall.net | lists / announce owl-users owl-dev john-users john-dev passwdqc-users yescrypt popa3d-users / oss-security kernel-hardening musl sabotage tlsify passwords / crypt-dev xvendor / Bugtraq Full-Disclosure linux-kernel linux-netdev linux-ext4 linux-hardening linux-cve-announce PHC | |
|
Open Source and information security mailing list archives
| ||
|
Date: Wed, 14 Oct 2020 16:03:26 +0200
From: Jiri Olsa <jolsa@...hat.com>
To: Leo Yan <leo.yan@...aro.org>
Cc: Arnaldo Carvalho de Melo <acme@...nel.org>,
Peter Zijlstra <peterz@...radead.org>,
Ingo Molnar <mingo@...hat.com>,
Mark Rutland <mark.rutland@....com>,
Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Namhyung Kim <namhyung@...nel.org>,
Andi Kleen <ak@...ux.intel.com>,
David Ahern <dsahern@...il.com>,
Don Zickus <dzickus@...hat.com>, Joe Mario <jmario@...hat.com>,
Al Grant <Al.Grant@....com>, James Clark <james.clark@....com>,
linux-kernel@...r.kernel.org
Subject: Re: [PATCH v1 0/8] perf c2c: Refine the organization of metrics
On Wed, Oct 14, 2020 at 06:09:13AM +0100, Leo Yan wrote:
> This patch set is to refine metrics output organization.
>
> If we reivew the current memory metrics in Perf c2c tool, it doesn't
> orgnize the metrics with directive approach; thus user needs to take
> time to dig into every statistics item. On the other hand, if use the
> "summary and breakdown" approach, the output result will be easier for
> reviewing by users, e.g. the output result can firstly give out the
> summary values, and then the later items will breakdown into more
> detailed statistics.
>
> For this reason, this patch is to reorgnize the metrics and it only
> changes for the "Shared Data Cache Line Table": it firstly displays the
> summary values for total records, total loads, total stores; then it
> breaks these summary values into small values, with the order from the
> most near memory node ("CPU Load Hit") to more far nodes
> ("LLC Load Hit", "RMT Load Hit", "Load Dram").
>
> "LLC Load Hit" = "LclHit" + "LclHitm"
>
> "RMT Load Hit" = "RmtHit" + "RmtHitm" \
> -> LLC Load Miss
> "Load Dram" = "Lcl" + "Rmt" /
>
> Another main reason for this patch set is wanting to extend "perf c2c"
> to support Arm SPE memory event, but Arm SPE doesn't contain 'HTIM' tag
> in its default trace data, for this case if want to analyze cache false
> sharing issue, we need to rely on LLC metrics + multi-threading info.
> So this patch set can be friendly to show LLC related metrics in the
> "Shared Data Cache Line Table"; for sorting cache lines with LLC metrics
> which will be sent out with another separate patch set.
>
> Before:
>
> =================================================
> Shared Data Cache Line Table
> =================================================
> #
> # ----------- Cacheline ---------- Total Tot ----- LLC Load Hitm ----- ---- Store Reference ---- --- Load Dram ---- LLC Total ----- Core Load Hit ----- -- LLC Load Hit --
> # Index Address Node PA cnt records Hitm Total Lcl Rmt Total L1Hit L1Miss Lcl Rmt Ld Miss Loads FB L1 L2 Llc Rmt
> # ..... .................. .... ...... ....... ....... ....... ....... ....... ....... ....... ....... ........ ........ ....... ....... ....... ....... ....... ........ ........
> #
> 0 0x55acdcc92100 0 8197 40716 52.18% 3170 3170 0 24466 24437 29 0 0 0 16250 3349 5909 0 3822 0
> 1 0x55acdcc920c0 0 1 4621 31.01% 1884 1884 0 0 0 0 0 0 0 4621 739 0 0 1998 0
> 2 0x55acdcc92080 0 1 4475 16.69% 1014 1014 0 0 0 0 0 0 0 4475 2405 0 0 1056 0
>
>
> After:
>
> =================================================
> Shared Data Cache Line Table
> =================================================
> #
> # ----------- Cacheline ---------- Tot ------- Load Hitm ------- Total Total Total ---- Stores ---- ----- Core Load Hit ----- - LLC Load Hit -- - RMT Load Hit -- --- Load Dram ----
> # Index Address Node PA cnt Hitm Total LclHitm RmtHitm records Loads Stores L1Hit L1Miss FB L1 L2 LclHit LclHitm RmtHit RmtHitm Lcl Rmt
> # ..... .................. .... ...... ....... ....... ....... ....... ....... ....... ....... ....... ....... ....... ....... ....... ........ ....... ........ ....... ........ ........
> #
> 0 0x55acdcc92100 0 8197 52.18% 3170 3170 0 40716 16250 24466 24437 29 3349 5909 0 3822 3170 0 0 0 0
> 1 0x55acdcc920c0 0 1 31.01% 1884 1884 0 4621 4621 0 0 0 739 0 0 1998 1884 0 0 0 0
> 2 0x55acdcc92080 0 1 16.69% 1014 1014 0 4475 4475 0 0 0 2405 0 0 1056 1014 0 0 0 0
I haven't used the tool for some time, so it's fine with me,
but there might be some people already used to see certain
columns in place and I don't want to make them angry unless
there's really good reason for that ;-)
Joe, could you please check on these changes?
thanks,
jirka
>
>
> Leo Yan (8):
> perf c2c: Display the total numbers continuously
> perf c2c: Display "Total Stores" as a standalone metrics
> perf c2c: Organize metrics based on memory hierarchy
> perf c2c: Change header from "LLC Load Hitm" to "Load Hitm"
> perf c2c: Use more explicit headers for HITM
> perf c2c: Change header for LLC local hit
> perf c2c: Correct LLC load hit metrics
> perf c2c: Add metrics "RMT Load Hit"
>
> tools/perf/builtin-c2c.c | 83 +++++++++-------------------------------
> 1 file changed, 18 insertions(+), 65 deletions(-)
>
> --
> 2.17.1
>
Powered by blists - more mailing lists