Re: [PATCH 6/9] perf report: Support instruction latency

From: Athira Rajeev
Date: Fri Feb 05 2021 - 19:15:07 EST




> On 04-Feb-2021, at 8:49 PM, Liang, Kan <kan.liang@xxxxxxxxxxxxxxx> wrote:
>
>
>
> On 2/4/2021 8:11 AM, Athira Rajeev wrote:
>>> On 03-Feb-2021, at 1:39 AM, kan.liang@xxxxxxxxxxxxxxx wrote:
>>>
>>> From: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
>>>
>>> The instruction latency information can be recorded on some platforms,
>>> e.g., the Intel Sapphire Rapids server. With both memory latency
>>> (weight) and the new instruction latency information, users can easily
>>> locate the expensive load instructions, and also understand the time
>>> spent in different stages. The users can optimize their applications
>>> in different pipeline stages.
>>>
>>> The 'weight' field is shared among different architectures. Reusing the
>>> 'weight' field may impacts other architectures. Add a new field to store
>>> the instruction latency.
>>>
>>> Like the 'weight' support, introduce a 'ins_lat' for the global
>>> instruction latency, and a 'local_ins_lat' for the local instruction
>>> latency version.
>>>
>>> Add new sort functions, INSTR Latency and Local INSTR Latency,
>>> accordingly.
>>>
>>> Add local_ins_lat to the default_mem_sort_order[].
>>>
>>> Signed-off-by: Kan Liang <kan.liang@xxxxxxxxxxxxxxx>
>>> ---
>>> tools/perf/Documentation/perf-report.txt | 6 +++-
>>> tools/perf/util/event.h | 1 +
>>> tools/perf/util/evsel.c | 4 ++-
>>> tools/perf/util/hist.c | 12 ++++++--
>>> tools/perf/util/hist.h | 2 ++
>>> tools/perf/util/intel-pt.c | 5 ++--
>>> tools/perf/util/session.c | 8 ++++--
>>> tools/perf/util/sort.c | 47 +++++++++++++++++++++++++++++++-
>>> tools/perf/util/sort.h | 3 ++
>>> tools/perf/util/synthetic-events.c | 4 ++-
>>> 10 files changed, 81 insertions(+), 11 deletions(-)
>>>
>>> diff --git a/tools/perf/Documentation/perf-report.txt b/tools/perf/Documentation/perf-report.txt
>>> index 826b5a9..0565b7c 100644
>>> --- a/tools/perf/Documentation/perf-report.txt
>>> +++ b/tools/perf/Documentation/perf-report.txt
>>> @@ -108,6 +108,9 @@ OPTIONS
>>> - period: Raw number of event count of sample
>>> - time: Separate the samples by time stamp with the resolution specified by
>>> --time-quantum (default 100ms). Specify with overhead and before it.
>>> + - ins_lat: Instruction latency in core cycles. This is the global
>>> + instruction latency
>>> + - local_ins_lat: Local instruction latency version
>>>
>>> By default, comm, dso and symbol keys are used.
>>> (i.e. --sort comm,dso,symbol)
>>> @@ -154,7 +157,8 @@ OPTIONS
>>> - blocked: reason of blocked load access for the data at the time of the sample
>>>
>>> And the default sort keys are changed to local_weight, mem, sym, dso,
>>> - symbol_daddr, dso_daddr, snoop, tlb, locked, blocked, see '--mem-mode'.
>>> + symbol_daddr, dso_daddr, snoop, tlb, locked, blocked, local_ins_lat,
>>> + see '--mem-mode'.
>>>
>>> If the data file has tracepoint event(s), following (dynamic) sort keys
>>> are also available:
>>> diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
>>> index ff403ea..5d50a49 100644
>>> --- a/tools/perf/util/event.h
>>> +++ b/tools/perf/util/event.h
>>> @@ -141,6 +141,7 @@ struct perf_sample {
>>> u16 insn_len;
>>> u8 cpumode;
>>> u16 misc;
>>> + u16 ins_lat;
>>> bool no_hw_idx; /* No hw_idx collected in branch_stack */
>>> char insn[MAX_INSN];
>>> void *raw_data;
>>> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
>>> index 0a2a307..24c0b59 100644
>>> --- a/tools/perf/util/evsel.c
>>> +++ b/tools/perf/util/evsel.c
>>> @@ -2337,8 +2337,10 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
>>> weight.full = *array;
>>> if (type & PERF_SAMPLE_WEIGHT)
>>> data->weight = weight.full;
>>> - else
>>> + else {
>>> data->weight = weight.var1_dw;
>>> + data->ins_lat = weight.var2_w;
>>> + }
>>> array++;
>>> }
>>>
>>> diff --git a/tools/perf/util/hist.c b/tools/perf/util/hist.c
>>> index 6866ab0..8a432f8 100644
>>> --- a/tools/perf/util/hist.c
>>> +++ b/tools/perf/util/hist.c
>>> @@ -209,6 +209,8 @@ void hists__calc_col_len(struct hists *hists, struct hist_entry *h)
>>> hists__new_col_len(hists, HISTC_LOCAL_WEIGHT, 12);
>>> hists__new_col_len(hists, HISTC_GLOBAL_WEIGHT, 12);
>>> hists__new_col_len(hists, HISTC_MEM_BLOCKED, 10);
>>> + hists__new_col_len(hists, HISTC_LOCAL_INS_LAT, 13);
>>> + hists__new_col_len(hists, HISTC_GLOBAL_INS_LAT, 13);
>>> if (symbol_conf.nanosecs)
>>> hists__new_col_len(hists, HISTC_TIME, 16);
>>> else
>>> @@ -286,12 +288,13 @@ static long hist_time(unsigned long htime)
>>> }
>>>
>>> static void he_stat__add_period(struct he_stat *he_stat, u64 period,
>>> - u64 weight)
>>> + u64 weight, u64 ins_lat)
>>> {
>>>
>>> he_stat->period += period;
>>> he_stat->weight += weight;
>>> he_stat->nr_events += 1;
>>> + he_stat->ins_lat += ins_lat;
>>> }
>>>
>>> static void he_stat__add_stat(struct he_stat *dest, struct he_stat *src)
>>> @@ -303,6 +306,7 @@ static void he_stat__add_stat(struct he_stat *dest, struct he_stat *src)
>>> dest->period_guest_us += src->period_guest_us;
>>> dest->nr_events += src->nr_events;
>>> dest->weight += src->weight;
>>> + dest->ins_lat += src->ins_lat;
>>> }
>>>
>>> static void he_stat__decay(struct he_stat *he_stat)
>>> @@ -591,6 +595,7 @@ static struct hist_entry *hists__findnew_entry(struct hists *hists,
>>> int64_t cmp;
>>> u64 period = entry->stat.period;
>>> u64 weight = entry->stat.weight;
>>> + u64 ins_lat = entry->stat.ins_lat;
>>> bool leftmost = true;
>>>
>>> p = &hists->entries_in->rb_root.rb_node;
>>> @@ -609,11 +614,11 @@ static struct hist_entry *hists__findnew_entry(struct hists *hists,
>>>
>>> if (!cmp) {
>>> if (sample_self) {
>>> - he_stat__add_period(&he->stat, period, weight);
>>> + he_stat__add_period(&he->stat, period, weight, ins_lat);
>>> hist_entry__add_callchain_period(he, period);
>>> }
>>> if (symbol_conf.cumulate_callchain)
>>> - he_stat__add_period(he->stat_acc, period, weight);
>>> + he_stat__add_period(he->stat_acc, period, weight, ins_lat);
>>>
>>> /*
>>> * This mem info was allocated from sample__resolve_mem
>>> @@ -723,6 +728,7 @@ __hists__add_entry(struct hists *hists,
>>> .nr_events = 1,
>>> .period = sample->period,
>>> .weight = sample->weight,
>>> + .ins_lat = sample->ins_lat,
>>> },
>>> .parent = sym_parent,
>>> .filtered = symbol__parent_filter(sym_parent) | al->filtered,
>>> diff --git a/tools/perf/util/hist.h b/tools/perf/util/hist.h
>>> index 522486b..36bca33 100644
>>> --- a/tools/perf/util/hist.h
>>> +++ b/tools/perf/util/hist.h
>>> @@ -72,6 +72,8 @@ enum hist_column {
>>> HISTC_DSO_SIZE,
>>> HISTC_SYMBOL_IPC,
>>> HISTC_MEM_BLOCKED,
>>> + HISTC_LOCAL_INS_LAT,
>>> + HISTC_GLOBAL_INS_LAT,
>>> HISTC_NR_COLS, /* Last entry */
>>> };
>>>
>>> diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
>>> index a929f6d..c9477d0 100644
>>> --- a/tools/perf/util/intel-pt.c
>>> +++ b/tools/perf/util/intel-pt.c
>>> @@ -1871,9 +1871,10 @@ static int intel_pt_synth_pebs_sample(struct intel_pt_queue *ptq)
>>> * cycles. Use latency >> 32 to distinguish the
>>> * different format of the mem access latency field.
>>> */
>>> - if (weight > 0)
>>> + if (weight > 0) {
>>> sample.weight = weight & 0xffff;
>>> - else
>>> + sample.ins_lat = items->mem_access_latency & 0xffff;
>>> + } else
>>> sample.weight = items->mem_access_latency;
>>> }
>>> if (!sample.weight && items->has_tsx_aux_info) {
>>> diff --git a/tools/perf/util/session.c b/tools/perf/util/session.c
>>> index a35d8c2..330e591 100644
>>> --- a/tools/perf/util/session.c
>>> +++ b/tools/perf/util/session.c
>>> @@ -1297,8 +1297,12 @@ static void dump_sample(struct evsel *evsel, union perf_event *event,
>>> if (sample_type & PERF_SAMPLE_STACK_USER)
>>> stack_user__printf(&sample->user_stack);
>>>
>>> - if (sample_type & PERF_SAMPLE_WEIGHT_TYPE)
>>> - printf("... weight: %" PRIu64 "\n", sample->weight);
>>> + if (sample_type & PERF_SAMPLE_WEIGHT_TYPE) {
>>> + printf("... weight: %" PRIu64 "", sample->weight);
>>> + if (sample_type & PERF_SAMPLE_WEIGHT_STRUCT)
>>> + printf(",0x%"PRIx16"", sample->ins_lat);
>>> + printf("\n");
>>> + }
>>>
>>> if (sample_type & PERF_SAMPLE_DATA_SRC)
>>> printf(" . data_src: 0x%"PRIx64"\n", sample->data_src);
>>> diff --git a/tools/perf/util/sort.c b/tools/perf/util/sort.c
>>> index 249a03c..e0529f2 100644
>>> --- a/tools/perf/util/sort.c
>>> +++ b/tools/perf/util/sort.c
>>> @@ -36,7 +36,7 @@ const char default_parent_pattern[] = "^sys_|^do_page_fault";
>>> const char *parent_pattern = default_parent_pattern;
>>> const char *default_sort_order = "comm,dso,symbol";
>>> const char default_branch_sort_order[] = "comm,dso_from,symbol_from,symbol_to,cycles";
>>> -const char default_mem_sort_order[] = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked,blocked";
>>> +const char default_mem_sort_order[] = "local_weight,mem,sym,dso,symbol_daddr,dso_daddr,snoop,tlb,locked,blocked,local_ins_lat";
>>> const char default_top_sort_order[] = "dso,symbol";
>>> const char default_diff_sort_order[] = "dso,symbol";
>>> const char default_tracepoint_sort_order[] = "trace";
>>> @@ -1365,6 +1365,49 @@ struct sort_entry sort_global_weight = {
>>> .se_width_idx = HISTC_GLOBAL_WEIGHT,
>>> };
>>>
>>> +static u64 he_ins_lat(struct hist_entry *he)
>>> +{
>>> + return he->stat.nr_events ? he->stat.ins_lat / he->stat.nr_events : 0;
>>> +}
>>> +
>>> +static int64_t
>>> +sort__local_ins_lat_cmp(struct hist_entry *left, struct hist_entry *right)
>>> +{
>>> + return he_ins_lat(left) - he_ins_lat(right);
>>> +}
>>> +
>>> +static int hist_entry__local_ins_lat_snprintf(struct hist_entry *he, char *bf,
>>> + size_t size, unsigned int width)
>>> +{
>>> + return repsep_snprintf(bf, size, "%-*u", width, he_ins_lat(he));
>>> +}
>>> +
>>> +struct sort_entry sort_local_ins_lat = {
>>> + .se_header = "Local INSTR Latency",
>>> + .se_cmp = sort__local_ins_lat_cmp,
>>> + .se_snprintf = hist_entry__local_ins_lat_snprintf,
>>> + .se_width_idx = HISTC_LOCAL_INS_LAT,
>>> +};
>> Hi Kan Liang,
>> Currently with this changes, perf will display "Local INSTR Latency” for the new column in ‘perf mem report’
>> Can we make this header string to be Architecture Specific ?
>
> I don't think current perf supports an arch specific header string. It sounds like a new feature. I think it's a good feature.
>
> Besides the instruction latency, we also introduce a 'block' field for perf mem. I'm not sure if the field applies for PowerPC. You may want to check it as well.

Hi Kan Liang,

Sure, we will check this.
>
> Could you please propose a patch to implement the feature? I think we can have a further discussion based on it.


Ok, We will look into this and come back with an RFC patch.
>
>> Because in other archs, the var2_w of ‘perf_sample_weight’ could be used to capture something else than the Local INSTR Latency.
>> Can we have some weak function to populate the header string ?
>
> I agree that the var2_w has different meanings among architectures. We should not force it to data->ins_lat.
>
> The patch as below should fix it. Does it work for you?

My point about weak function was actually for the arch specific header string. But I guess we should not force it to data->ins_lat
as you mentioned. I checked the below patch defining an ‘arch_perf_parse_sample_weight' for powerpc and it works.

But one observation is that, for cases with kernel having support for PERF_SAMPLE_WEIGHT_STRUCT but missing arch specific support for ‘arch_perf_parse_sample_weight', it will report ‘Local Weight’ wrongly since weak function takes it as 64 bit. Not sure if that is a valid case to consider though.

Thanks
Athira
>
> diff --git a/tools/perf/arch/x86/util/event.c b/tools/perf/arch/x86/util/event.c
> index 047dc00..9b31734 100644
> --- a/tools/perf/arch/x86/util/event.c
> +++ b/tools/perf/arch/x86/util/event.c
> @@ -75,3 +75,28 @@ int perf_event__synthesize_extra_kmaps(struct perf_tool *tool,
> }
>
> #endif
> +
> +void arch_perf_parse_sample_weight(struct perf_sample *data,
> + const __u64 *array, u64 type)
> +{
> + union perf_sample_weight weight;
> +
> + weight.full = *array;
> + if (type & PERF_SAMPLE_WEIGHT)
> + data->weight = weight.full;
> + else {
> + data->weight = weight.var1_dw;
> + data->ins_lat = weight.var2_w;
> + }
> +}
> +
> +void arch_perf_synthesize_sample_weight(const struct perf_sample *data,
> + __u64 *array, u64 type)
> +{
> + *array = data->weight;
> +
> + if (type & PERF_SAMPLE_WEIGHT_STRUCT) {
> + *array &= 0xffffffff;
> + *array |= ((u64)data->ins_lat << 32);
> + }
> +}
> diff --git a/tools/perf/util/event.h b/tools/perf/util/event.h
> index 60752e4..9f123aa 100644
> --- a/tools/perf/util/event.h
> +++ b/tools/perf/util/event.h
> @@ -414,4 +414,7 @@ extern unsigned int proc_map_timeout;
> #define PAGE_SIZE_NAME_LEN 32
> char *get_page_size_name(u64 size, char *str);
>
> +void arch_perf_parse_sample_weight(struct perf_sample *data, const __u64 *array, u64 type);
> +void arch_perf_synthesize_sample_weight(const struct perf_sample *data, __u64 *array, u64 type);
> +
> #endif /* __PERF_RECORD_H */
> diff --git a/tools/perf/util/evsel.c b/tools/perf/util/evsel.c
> index 30b5452..1da8903 100644
> --- a/tools/perf/util/evsel.c
> +++ b/tools/perf/util/evsel.c
> @@ -2106,6 +2106,13 @@ perf_event__check_size(union perf_event *event, unsigned int sample_size)
> return 0;
> }
>
> +void __weak arch_perf_parse_sample_weight(struct perf_sample *data,
> + const __u64 *array,
> + u64 type __maybe_unused)
> +{
> + data->weight = *array;
> +}
> +
> int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
> struct perf_sample *data)
> {
> @@ -2347,16 +2354,8 @@ int evsel__parse_sample(struct evsel *evsel, union perf_event *event,
> }
>
> if (type & PERF_SAMPLE_WEIGHT_TYPE) {
> - union perf_sample_weight weight;
> -
> OVERFLOW_CHECK_u64(array);
> - weight.full = *array;
> - if (type & PERF_SAMPLE_WEIGHT)
> - data->weight = weight.full;
> - else {
> - data->weight = weight.var1_dw;
> - data->ins_lat = weight.var2_w;
> - }
> + arch_perf_parse_sample_weight(data, array, type);
> array++;
> }
>
> diff --git a/tools/perf/util/synthetic-events.c b/tools/perf/util/synthetic-events.c
> index c6f9db3..412f4c3 100644
> --- a/tools/perf/util/synthetic-events.c
> +++ b/tools/perf/util/synthetic-events.c
> @@ -1507,6 +1507,12 @@ size_t perf_event__sample_event_size(const struct perf_sample *sample, u64 type,
> return result;
> }
>
> +void __weak arch_perf_synthesize_sample_weight(const struct perf_sample *data,
> + __u64 *array, u64 type __maybe_unused)
> +{
> + *array = data->weight;
> +}
> +
> int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_format,
> const struct perf_sample *sample)
> {
> @@ -1643,11 +1649,7 @@ int perf_event__synthesize_sample(union perf_event *event, u64 type, u64 read_fo
> }
>
> if (type & PERF_SAMPLE_WEIGHT_TYPE) {
> - *array = sample->weight;
> - if (type & PERF_SAMPLE_WEIGHT_STRUCT) {
> - *array &= 0xffffffff;
> - *array |= ((u64)sample->ins_lat << 32);
> - }
> + arch_perf_synthesize_sample_weight(sample, array, type);
> array++;
> }
>
> Thanks,
> Kan