From 86c2bc3da769124e3e856b6e9457be3667c30919 Mon Sep 17 00:00:00 2001 From: Smita Koralahalli Date: Tue, 6 Apr 2021 16:59:41 -0500 Subject: perf vendor events amd: Fix broken L2 Cache Hits from L2 HWPF metric MIME-Version: 1.0 Content-Type: text/plain; charset=UTF-8 Content-Transfer-Encoding: 8bit Commit 08ed77e414ab2342 ("perf vendor events amd: Add recommended events") added the hits event "L2 Cache Hits from L2 HWPF" with the same metric expression as the accesses event "L2 Cache Accesses from L2 HWPF": $ perf list --details ... l2_cache_accesses_from_l2_hwpf [L2 Cache Accesses from L2 HWPF] [l2_pf_hit_l2 + l2_pf_miss_l2_hit_l3 + l2_pf_miss_l2_l3] l2_cache_hits_from_l2_hwpf [L2 Cache Hits from L2 HWPF] [l2_pf_hit_l2 + l2_pf_miss_l2_hit_l3 + l2_pf_miss_l2_l3] ... This was wrong and led to counting hits the same as accesses. Section 2.1.15.2 "Performance Measurement" of "PPR for AMD Family 17h Model 31h B0 - 55803 Rev 0.54 - Sep 12, 2019", documents the hits event with EventCode 0x70 which is the same as l2_pf_hit_l2. Fix this, and massage the description for l2_pf_hit_l2 as the hits event is now the duplicate of l2_pf_hit_l2. AMD recommends using the recommended event over other events if the duplicate exists and maintain both for consistency. Hence, l2_cache_hits_from_l2_hwpf should override l2_pf_hit_l2. Before: # perf stat -M l2_cache_accesses_from_l2_hwpf,l2_cache_hits_from_l2_hwpf sleep 1 Performance counter stats for 'sleep 1': 1,436 l2_pf_miss_l2_l3 # 11114.00 l2_cache_accesses_from_l2_hwpf # 11114.00 l2_cache_hits_from_l2_hwpf 4,482 l2_pf_hit_l2 5,196 l2_pf_miss_l2_hit_l3 1.001765339 seconds time elapsed After: # perf stat -M l2_cache_accesses_from_l2_hwpf sleep 1 Performance counter stats for 'sleep 1': 1,477 l2_pf_miss_l2_l3 # 10442.00 l2_cache_accesses_from_l2_hwpf 3,978 l2_pf_hit_l2 4,987 l2_pf_miss_l2_hit_l3 1.001491186 seconds time elapsed # perf stat -e l2_cache_hits_from_l2_hwpf sleep 1 Performance counter stats for 'sleep 1': 3,983 l2_cache_hits_from_l2_hwpf 1.001329970 seconds time elapsed Note the difference in performance counter values for the accesses versus the hits after the fix, and the hits event now counting the same as l2_pf_hit_l2. Fixes: 08ed77e414ab ("perf vendor events amd: Add recommended events") Bugzilla: https://bugzilla.kernel.org/show_bug.cgi?id=206537 Reviewed-by: Robert Richter Signed-off-by: Smita Koralahalli Tested-by: Arnaldo Carvalho de Melo # On a 3900X Cc: Alexander Shishkin Cc: Ian Rogers Cc: Ingo Molnar Cc: Jiri Olsa Cc: Kim Phillips Cc: Mark Rutland Cc: Martin Liška Cc: Michael Petlan Cc: Namhyung Kim Cc: Peter Zijlstra Cc: Vijay Thakkar Cc: linux-perf-users@vger.kernel.org Link: https://lore.kernel.org/r/20210406215944.113332-2-Smita.KoralahalliChannabasappa@amd.com Signed-off-by: Arnaldo Carvalho de Melo --- tools/perf/pmu-events/arch/x86/amdzen1/cache.json | 2 +- tools/perf/pmu-events/arch/x86/amdzen1/recommended.json | 6 +++--- tools/perf/pmu-events/arch/x86/amdzen2/cache.json | 2 +- tools/perf/pmu-events/arch/x86/amdzen2/recommended.json | 6 +++--- 4 files changed, 8 insertions(+), 8 deletions(-) (limited to 'tools') diff --git a/tools/perf/pmu-events/arch/x86/amdzen1/cache.json b/tools/perf/pmu-events/arch/x86/amdzen1/cache.json index 4ea7ec4f496e..008f1683e540 100644 --- a/tools/perf/pmu-events/arch/x86/amdzen1/cache.json +++ b/tools/perf/pmu-events/arch/x86/amdzen1/cache.json @@ -275,7 +275,7 @@ { "EventName": "l2_pf_hit_l2", "EventCode": "0x70", - "BriefDescription": "L2 prefetch hit in L2.", + "BriefDescription": "L2 prefetch hit in L2. Use l2_cache_hits_from_l2_hwpf instead.", "UMask": "0xff" }, { diff --git a/tools/perf/pmu-events/arch/x86/amdzen1/recommended.json b/tools/perf/pmu-events/arch/x86/amdzen1/recommended.json index 2cfe2d2f3bfd..3c954543d1ae 100644 --- a/tools/perf/pmu-events/arch/x86/amdzen1/recommended.json +++ b/tools/perf/pmu-events/arch/x86/amdzen1/recommended.json @@ -79,10 +79,10 @@ "UMask": "0x70" }, { - "MetricName": "l2_cache_hits_from_l2_hwpf", + "EventName": "l2_cache_hits_from_l2_hwpf", + "EventCode": "0x70", "BriefDescription": "L2 Cache Hits from L2 HWPF", - "MetricExpr": "l2_pf_hit_l2 + l2_pf_miss_l2_hit_l3 + l2_pf_miss_l2_l3", - "MetricGroup": "l2_cache" + "UMask": "0xff" }, { "EventName": "l3_accesses", diff --git a/tools/perf/pmu-events/arch/x86/amdzen2/cache.json b/tools/perf/pmu-events/arch/x86/amdzen2/cache.json index f61b982f83ca..8ba84a48188d 100644 --- a/tools/perf/pmu-events/arch/x86/amdzen2/cache.json +++ b/tools/perf/pmu-events/arch/x86/amdzen2/cache.json @@ -205,7 +205,7 @@ { "EventName": "l2_pf_hit_l2", "EventCode": "0x70", - "BriefDescription": "L2 prefetch hit in L2.", + "BriefDescription": "L2 prefetch hit in L2. Use l2_cache_hits_from_l2_hwpf instead.", "UMask": "0xff" }, { diff --git a/tools/perf/pmu-events/arch/x86/amdzen2/recommended.json b/tools/perf/pmu-events/arch/x86/amdzen2/recommended.json index 2ef91e25e661..1c624cee9ef4 100644 --- a/tools/perf/pmu-events/arch/x86/amdzen2/recommended.json +++ b/tools/perf/pmu-events/arch/x86/amdzen2/recommended.json @@ -79,10 +79,10 @@ "UMask": "0x70" }, { - "MetricName": "l2_cache_hits_from_l2_hwpf", + "EventName": "l2_cache_hits_from_l2_hwpf", + "EventCode": "0x70", "BriefDescription": "L2 Cache Hits from L2 HWPF", - "MetricExpr": "l2_pf_hit_l2 + l2_pf_miss_l2_hit_l3 + l2_pf_miss_l2_l3", - "MetricGroup": "l2_cache" + "UMask": "0xff" }, { "EventName": "l3_accesses", -- cgit v1.2.3