From: "Liang, Kan" <kan.liang@linux.intel.com> To: Ian Rogers <irogers@google.com> Cc: Carel Si <beibei.si@intel.com>, acme@redhat.com, alexander.shishkin@linux.intel.com, alexandre.torgue@foss.st.com, ak@linux.intel.com, mingo@redhat.com, james.clark@arm.com, jolsa@kernel.org, john.garry@huawei.com, mark.rutland@arm.com, mcoquelin.stm32@gmail.com, namhyung@kernel.org, peterz@infradead.org, eranian@google.com, linux-kernel@vger.kernel.org, linux-mm@kvack.org, lkp@lists.01.org, lkp@intel.com Subject: Re: [LKP] Re: [perf vendor events] 3f5f0df7bf: perf-sanity-tests.perf_all_metrics_test.fail Date: Wed, 13 Apr 2022 14:17:16 -0400 [thread overview] Message-ID: <74d36e66-5b16-ee10-bcae-171e2b681845@linux.intel.com> (raw) In-Reply-To: <CAP-5=fXj+WT3ExNo-fL4d9b_Wm5swnKaCWd67rU85Q=QFg5KUw@mail.gmail.com> On 4/13/2022 1:09 PM, Ian Rogers wrote: > On Wed, Apr 13, 2022 at 9:37 AM Liang, Kan <kan.liang@linux.intel.com> wrote: >> >> >> >> On 4/13/2022 12:03 PM, Ian Rogers wrote: >>> 3) Weak group doesn't fall back to no group: >> >> That's because the group validation code doesn't take pinned events, >> such as the NMI watchdog, into account. >> >> I proposed a kernel patch to fix it, but it's rejected. It should be >> hard to find a generic way to fix it from the kernel side. >> https://lore.kernel.org/lkml/1565977750-76693-1-git-send-email-kan.liang@linux.intel.com/ >> >> Maybe we can workaround it from the perf tool side? >> For example, for each weak group with cycles event and NMI watchdog is >> enabled, add an extra cycles event when opening the group. If the open >> fails with the extra cycles event, fall back to no group. After the >> extra cycles event check, remove the extra cycles. >> >> What do you think? > > Thanks Kan, it is a shame the kernel support is lacking here. I'm not > sure what you are proposing for the perf tool to do. So: > >> for each weak group with cycles event and NMI watchdog > > Okay, let's try Branching_Overhead as mentioned in this report - but > the event is CPU_CLK_UNHALTED.THREAD here :-/ > >> add an extra cycles event when opening the group > > So the perf_event_open doesn't fail here for me: > $ perf stat -e '{BR_INST_RETIRED.NEAR_CALL,BR_INST_RETIRED.NEAR_TAKEN,BR_INST_RETIRED.NOT_TAKEN,BR_INST_RETIRED.CONDITIONAL,CPU_CLK_UNHALTED.THREAD},cycles' > -a sleep 1 > No, I mean modifying the perf tool code and add an extra cycles in the weak group. Here is a very initial POC patch, which should prove the idea. diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index b7fe88beb584..782c3d7f1b32 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -71,7 +71,9 @@ #include "util/bpf_counter.h" #include "util/iostat.h" #include "util/pmu-hybrid.h" +#include "util/util.h" #include "asm/bug.h" +#include "perf-sys.h" #include <linux/time64.h> #include <linux/zalloc.h> @@ -777,6 +779,8 @@ static enum counter_recovery stat_handle_error(struct evsel *counter) return COUNTER_FATAL; } +#define FD(e, x, y) (*(int *)xyarray__entry(e->core.fd, x, y)) + static int __run_perf_stat(int argc, const char **argv, int run_idx) { int interval = stat_config.interval; @@ -793,6 +797,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) struct affinity saved_affinity, *affinity = NULL; int err; bool second_pass = false; + bool has_cycles = false; if (forks) { if (evlist__prepare_workload(evsel_list, &target, argv, is_pipe, workload_exec_failed_signal) < 0) { @@ -821,6 +826,8 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) evlist__for_each_cpu(evlist_cpu_itr, evsel_list, affinity) { counter = evlist_cpu_itr.evsel; + if (counter->core.attr.config == 0x3c) + has_cycles = true; /* * bperf calls evsel__open_per_cpu() in bperf__load(), so * no need to call it again here. @@ -867,6 +874,24 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) counter->supported = true; } + //make it model specific. need to move to a better place + if (counter->supported && !second_pass && has_cycles && counter->weak_group && sysctl__nmi_watchdog_enabled()) { + struct evsel *leader = evsel__leader(counter); + int group_fd = FD(leader, 0, 0); + struct evsel *evsel; + int fd; + + evsel = evsel__new_cycles(0, PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES); + fd = sys_perf_event_open(&evsel->core.attr, -1, 0, group_fd, 0x8); + + if (fd < 0) { + evlist__reset_weak_group(evsel_list, counter, false); + second_pass = true; + } else { + evsel__close(evsel); + } + } + if (second_pass) { /* * Now redo all the weak group after closing them, With the above patch, $ sudo ./perf stat -e '{BR_INST_RETIRED.NEAR_CALL,BR_INST_RETIRED.NEAR_TAKEN,BR_INST_RETIRED.NOT_TAKEN,BR_INST_RETIRED.CONDITIONAL,CPU_CLK_UNHALTED.THREAD}:W' -C0 sleep 1 Performance counter stats for 'CPU(s) 0': 913,369 BR_INST_RETIRED.NEAR_CALL (79.95%) 3,648,433 BR_INST_RETIRED.NEAR_TAKEN (80.00%) 2,481,976 BR_INST_RETIRED.NOT_TAKEN (80.05%) 3,696,298 BR_INST_RETIRED.CONDITIONAL (80.04%) 27,792,053 CPU_CLK_UNHALTED.THREAD (79.96%) 1.002224709 seconds time elapsed Thanks, Kan
WARNING: multiple messages have this Message-ID (diff)
From: Liang, Kan <kan.liang@linux.intel.com> To: lkp@lists.01.org Subject: Re: [perf vendor events] 3f5f0df7bf: perf-sanity-tests.perf_all_metrics_test.fail Date: Wed, 13 Apr 2022 14:17:16 -0400 [thread overview] Message-ID: <74d36e66-5b16-ee10-bcae-171e2b681845@linux.intel.com> (raw) In-Reply-To: <CAP-5=fXj+WT3ExNo-fL4d9b_Wm5swnKaCWd67rU85Q=QFg5KUw@mail.gmail.com> [-- Attachment #1: Type: text/plain, Size: 4755 bytes --] On 4/13/2022 1:09 PM, Ian Rogers wrote: > On Wed, Apr 13, 2022 at 9:37 AM Liang, Kan <kan.liang@linux.intel.com> wrote: >> >> >> >> On 4/13/2022 12:03 PM, Ian Rogers wrote: >>> 3) Weak group doesn't fall back to no group: >> >> That's because the group validation code doesn't take pinned events, >> such as the NMI watchdog, into account. >> >> I proposed a kernel patch to fix it, but it's rejected. It should be >> hard to find a generic way to fix it from the kernel side. >> https://lore.kernel.org/lkml/1565977750-76693-1-git-send-email-kan.liang(a)linux.intel.com/ >> >> Maybe we can workaround it from the perf tool side? >> For example, for each weak group with cycles event and NMI watchdog is >> enabled, add an extra cycles event when opening the group. If the open >> fails with the extra cycles event, fall back to no group. After the >> extra cycles event check, remove the extra cycles. >> >> What do you think? > > Thanks Kan, it is a shame the kernel support is lacking here. I'm not > sure what you are proposing for the perf tool to do. So: > >> for each weak group with cycles event and NMI watchdog > > Okay, let's try Branching_Overhead as mentioned in this report - but > the event is CPU_CLK_UNHALTED.THREAD here :-/ > >> add an extra cycles event when opening the group > > So the perf_event_open doesn't fail here for me: > $ perf stat -e '{BR_INST_RETIRED.NEAR_CALL,BR_INST_RETIRED.NEAR_TAKEN,BR_INST_RETIRED.NOT_TAKEN,BR_INST_RETIRED.CONDITIONAL,CPU_CLK_UNHALTED.THREAD},cycles' > -a sleep 1 > No, I mean modifying the perf tool code and add an extra cycles in the weak group. Here is a very initial POC patch, which should prove the idea. diff --git a/tools/perf/builtin-stat.c b/tools/perf/builtin-stat.c index b7fe88beb584..782c3d7f1b32 100644 --- a/tools/perf/builtin-stat.c +++ b/tools/perf/builtin-stat.c @@ -71,7 +71,9 @@ #include "util/bpf_counter.h" #include "util/iostat.h" #include "util/pmu-hybrid.h" +#include "util/util.h" #include "asm/bug.h" +#include "perf-sys.h" #include <linux/time64.h> #include <linux/zalloc.h> @@ -777,6 +779,8 @@ static enum counter_recovery stat_handle_error(struct evsel *counter) return COUNTER_FATAL; } +#define FD(e, x, y) (*(int *)xyarray__entry(e->core.fd, x, y)) + static int __run_perf_stat(int argc, const char **argv, int run_idx) { int interval = stat_config.interval; @@ -793,6 +797,7 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) struct affinity saved_affinity, *affinity = NULL; int err; bool second_pass = false; + bool has_cycles = false; if (forks) { if (evlist__prepare_workload(evsel_list, &target, argv, is_pipe, workload_exec_failed_signal) < 0) { @@ -821,6 +826,8 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) evlist__for_each_cpu(evlist_cpu_itr, evsel_list, affinity) { counter = evlist_cpu_itr.evsel; + if (counter->core.attr.config == 0x3c) + has_cycles = true; /* * bperf calls evsel__open_per_cpu() in bperf__load(), so * no need to call it again here. @@ -867,6 +874,24 @@ static int __run_perf_stat(int argc, const char **argv, int run_idx) counter->supported = true; } + //make it model specific. need to move to a better place + if (counter->supported && !second_pass && has_cycles && counter->weak_group && sysctl__nmi_watchdog_enabled()) { + struct evsel *leader = evsel__leader(counter); + int group_fd = FD(leader, 0, 0); + struct evsel *evsel; + int fd; + + evsel = evsel__new_cycles(0, PERF_TYPE_HARDWARE, PERF_COUNT_HW_CPU_CYCLES); + fd = sys_perf_event_open(&evsel->core.attr, -1, 0, group_fd, 0x8); + + if (fd < 0) { + evlist__reset_weak_group(evsel_list, counter, false); + second_pass = true; + } else { + evsel__close(evsel); + } + } + if (second_pass) { /* * Now redo all the weak group after closing them, With the above patch, $ sudo ./perf stat -e '{BR_INST_RETIRED.NEAR_CALL,BR_INST_RETIRED.NEAR_TAKEN,BR_INST_RETIRED.NOT_TAKEN,BR_INST_RETIRED.CONDITIONAL,CPU_CLK_UNHALTED.THREAD}:W' -C0 sleep 1 Performance counter stats for 'CPU(s) 0': 913,369 BR_INST_RETIRED.NEAR_CALL (79.95%) 3,648,433 BR_INST_RETIRED.NEAR_TAKEN (80.00%) 2,481,976 BR_INST_RETIRED.NOT_TAKEN (80.05%) 3,696,298 BR_INST_RETIRED.CONDITIONAL (80.04%) 27,792,053 CPU_CLK_UNHALTED.THREAD (79.96%) 1.002224709 seconds time elapsed Thanks, Kan
next prev parent reply other threads:[~2022-04-13 18:17 UTC|newest] Thread overview: 22+ messages / expand[flat|nested] mbox.gz Atom feed top 2022-03-04 8:33 [perf vendor events] 3f5f0df7bf: perf-sanity-tests.perf_all_metrics_test.fail kernel test robot 2022-03-04 8:33 ` kernel test robot 2022-03-04 18:10 ` Ian Rogers 2022-03-04 18:10 ` Ian Rogers 2022-04-13 7:05 ` [LKP] " Carel Si 2022-04-13 7:05 ` Carel Si 2022-04-13 16:03 ` [LKP] " Ian Rogers 2022-04-13 16:03 ` Ian Rogers 2022-04-13 16:37 ` [LKP] " Liang, Kan 2022-04-13 16:37 ` Liang, Kan 2022-04-13 17:09 ` [LKP] " Ian Rogers 2022-04-13 17:09 ` Ian Rogers 2022-04-13 18:17 ` Liang, Kan [this message] 2022-04-13 18:17 ` Liang, Kan 2022-04-14 16:09 ` [LKP] " Ian Rogers 2022-04-14 16:09 ` Ian Rogers 2022-04-14 19:06 ` [LKP] " Liang, Kan 2022-04-14 19:06 ` Liang, Kan 2022-04-14 22:58 ` [LKP] " Namhyung Kim 2022-04-14 22:58 ` Namhyung Kim 2022-04-18 12:42 ` [LKP] " Liang, Kan 2022-04-18 12:42 ` Liang, Kan
Reply instructions: You may reply publicly to this message via plain-text email using any one of the following methods: * Save the following mbox file, import it into your mail client, and reply-to-all from there: mbox Avoid top-posting and favor interleaved quoting: https://en.wikipedia.org/wiki/Posting_style#Interleaved_style * Reply using the --to, --cc, and --in-reply-to switches of git-send-email(1): git send-email \ --in-reply-to=74d36e66-5b16-ee10-bcae-171e2b681845@linux.intel.com \ --to=kan.liang@linux.intel.com \ --cc=acme@redhat.com \ --cc=ak@linux.intel.com \ --cc=alexander.shishkin@linux.intel.com \ --cc=alexandre.torgue@foss.st.com \ --cc=beibei.si@intel.com \ --cc=eranian@google.com \ --cc=irogers@google.com \ --cc=james.clark@arm.com \ --cc=john.garry@huawei.com \ --cc=jolsa@kernel.org \ --cc=linux-kernel@vger.kernel.org \ --cc=linux-mm@kvack.org \ --cc=lkp@intel.com \ --cc=lkp@lists.01.org \ --cc=mark.rutland@arm.com \ --cc=mcoquelin.stm32@gmail.com \ --cc=mingo@redhat.com \ --cc=namhyung@kernel.org \ --cc=peterz@infradead.org \ /path/to/YOUR_REPLY https://kernel.org/pub/software/scm/git/docs/git-send-email.html * If your mail client supports setting the In-Reply-To header via mailto: links, try the mailto: linkBe sure your reply has a Subject: header at the top and a blank line before the message body.
This is an external index of several public inboxes, see mirroring instructions on how to clone and mirror all data and code used by this external index.