[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <a3282de2-bc5a-444b-b58b-2dc99c420266@linux.intel.com>
Date: Mon, 12 May 2025 11:33:10 -0400
From: "Liang, Kan" <kan.liang@...ux.intel.com>
To: Adrian Hunter <adrian.hunter@...el.com>,
Arnaldo Carvalho de Melo <acme@...nel.org>, Jiri Olsa <jolsa@...nel.org>,
Namhyung Kim <namhyung@...nel.org>
Cc: Alexander Shishkin <alexander.shishkin@...ux.intel.com>,
Ian Rogers <irogers@...gle.com>, linux-kernel@...r.kernel.org,
linux-perf-users@...r.kernel.org
Subject: Re: [PATCH 1/3] perf intel-pt: Fix PEBS-via-PT data_src
On 2025-05-12 5:39 a.m., Adrian Hunter wrote:
> The Fixes commit did not add support for decoding PEBS-via-PT data_src.
> Fix by adding support.
>
> PEBS-via-PT is a feature of some E-core processors, starting with
> processors based on Tremont microarchitecture. Because the kernel only
> supports Intel PT features that are on all processors, there is no support
> for PEBS-via-PT on hybrids.
>
> Currently that leaves processors based on Tremont, Gracemont and Crestmont,
> however there are no events on Tremont that produce data_src information,
> and for Gracemont and Crestmont there are only:
>
> mem-loads event=0xd0,umask=0x5,ldlat=3
> mem-stores event=0xd0,umask=0x6
>
> Affected processors include Alder Lake N (Gracemont), Sierra Forest
> (Crestmont) and Grand Ridge (Crestmont).
>
I think they are the only processors which need to be patched. For the
future platforms, the PEBS-via-PT should be gone. The kernel data_source
functions have been there for a while. I think it's OK to copy & paste
and without worrying about the sync for future platforms.
The patch looks good to me.
Reviewed-by: Kan Liang <kan.liang@...ux.intel.com>
Thanks,
Kan
> Example:
>
> # perf record -d -e intel_pt/branch=0/ -e mem-loads/aux-output/pp uname
>
> Before:
>
> # perf.before script --itrace=o -Fdata_src
> 0 |OP No|LVL N/A|SNP N/A|TLB N/A|LCK No|BLK N/A
> 0 |OP No|LVL N/A|SNP N/A|TLB N/A|LCK No|BLK N/A
>
> After:
>
> # perf script --itrace=o -Fdata_src
> 10268100142 |OP LOAD|LVL L1 hit|SNP None|TLB L1 or L2 hit|LCK No|BLK N/A
> 10450100442 |OP LOAD|LVL L2 hit|SNP None|TLB L2 miss|LCK No|BLK N/A
>
> Fixes: 975846eddf907 ("perf intel-pt: Add memory information to synthesized PEBS sample")
> Signed-off-by: Adrian Hunter <adrian.hunter@...el.com>
> ---
> tools/perf/util/intel-pt.c | 205 ++++++++++++++++++++++++++++++++++++-
> 1 file changed, 202 insertions(+), 3 deletions(-)
>
> diff --git a/tools/perf/util/intel-pt.c b/tools/perf/util/intel-pt.c
> index 4e8a9b172fbc..9b1011fe4826 100644
> --- a/tools/perf/util/intel-pt.c
> +++ b/tools/perf/util/intel-pt.c
> @@ -127,6 +127,7 @@ struct intel_pt {
>
> bool single_pebs;
> bool sample_pebs;
> + int pebs_data_src_fmt;
> struct evsel *pebs_evsel;
>
> u64 evt_sample_type;
> @@ -175,6 +176,7 @@ enum switch_state {
> struct intel_pt_pebs_event {
> struct evsel *evsel;
> u64 id;
> + int data_src_fmt;
> };
>
> struct intel_pt_queue {
> @@ -2272,7 +2274,146 @@ static void intel_pt_add_lbrs(struct branch_stack *br_stack,
> }
> }
>
> -static int intel_pt_do_synth_pebs_sample(struct intel_pt_queue *ptq, struct evsel *evsel, u64 id)
> +#define P(a, b) PERF_MEM_S(a, b)
> +#define OP_LH (P(OP, LOAD) | P(LVL, HIT))
> +#define LEVEL(x) P(LVLNUM, x)
> +#define REM P(REMOTE, REMOTE)
> +#define SNOOP_NONE_MISS (P(SNOOP, NONE) | P(SNOOP, MISS))
> +
> +#define PERF_PEBS_DATA_SOURCE_GRT_MAX 0x10
> +#define PERF_PEBS_DATA_SOURCE_GRT_MASK (PERF_PEBS_DATA_SOURCE_GRT_MAX - 1)
> +
> +/* Based on kernel __intel_pmu_pebs_data_source_grt() and pebs_data_source */
> +static const u64 pebs_data_source_grt[PERF_PEBS_DATA_SOURCE_GRT_MAX] = {
> + P(OP, LOAD) | P(LVL, MISS) | LEVEL(L3) | P(SNOOP, NA), /* L3 miss|SNP N/A */
> + OP_LH | P(LVL, L1) | LEVEL(L1) | P(SNOOP, NONE), /* L1 hit|SNP None */
> + OP_LH | P(LVL, LFB) | LEVEL(LFB) | P(SNOOP, NONE), /* LFB/MAB hit|SNP None */
> + OP_LH | P(LVL, L2) | LEVEL(L2) | P(SNOOP, NONE), /* L2 hit|SNP None */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, NONE), /* L3 hit|SNP None */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, HIT), /* L3 hit|SNP Hit */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, HITM), /* L3 hit|SNP HitM */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, HITM), /* L3 hit|SNP HitM */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOPX, FWD), /* L3 hit|SNP Fwd */
> + OP_LH | P(LVL, REM_CCE1) | REM | LEVEL(L3) | P(SNOOP, HITM), /* Remote L3 hit|SNP HitM */
> + OP_LH | P(LVL, LOC_RAM) | LEVEL(RAM) | P(SNOOP, HIT), /* RAM hit|SNP Hit */
> + OP_LH | P(LVL, REM_RAM1) | REM | LEVEL(L3) | P(SNOOP, HIT), /* Remote L3 hit|SNP Hit */
> + OP_LH | P(LVL, LOC_RAM) | LEVEL(RAM) | SNOOP_NONE_MISS, /* RAM hit|SNP None or Miss */
> + OP_LH | P(LVL, REM_RAM1) | LEVEL(RAM) | REM | SNOOP_NONE_MISS, /* Remote RAM hit|SNP None or Miss */
> + OP_LH | P(LVL, IO) | LEVEL(NA) | P(SNOOP, NONE), /* I/O hit|SNP None */
> + OP_LH | P(LVL, UNC) | LEVEL(NA) | P(SNOOP, NONE), /* Uncached hit|SNP None */
> +};
> +
> +/* Based on kernel __intel_pmu_pebs_data_source_cmt() and pebs_data_source */
> +static const u64 pebs_data_source_cmt[PERF_PEBS_DATA_SOURCE_GRT_MAX] = {
> + P(OP, LOAD) | P(LVL, MISS) | LEVEL(L3) | P(SNOOP, NA), /* L3 miss|SNP N/A */
> + OP_LH | P(LVL, L1) | LEVEL(L1) | P(SNOOP, NONE), /* L1 hit|SNP None */
> + OP_LH | P(LVL, LFB) | LEVEL(LFB) | P(SNOOP, NONE), /* LFB/MAB hit|SNP None */
> + OP_LH | P(LVL, L2) | LEVEL(L2) | P(SNOOP, NONE), /* L2 hit|SNP None */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, NONE), /* L3 hit|SNP None */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, MISS), /* L3 hit|SNP Hit */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, HIT), /* L3 hit|SNP HitM */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOPX, FWD), /* L3 hit|SNP HitM */
> + OP_LH | P(LVL, L3) | LEVEL(L3) | P(SNOOP, HITM), /* L3 hit|SNP Fwd */
> + OP_LH | P(LVL, REM_CCE1) | REM | LEVEL(L3) | P(SNOOP, HITM), /* Remote L3 hit|SNP HitM */
> + OP_LH | P(LVL, LOC_RAM) | LEVEL(RAM) | P(SNOOP, NONE), /* RAM hit|SNP Hit */
> + OP_LH | LEVEL(RAM) | REM | P(SNOOP, NONE), /* Remote L3 hit|SNP Hit */
> + OP_LH | LEVEL(RAM) | REM | P(SNOOPX, FWD), /* RAM hit|SNP None or Miss */
> + OP_LH | LEVEL(RAM) | REM | P(SNOOP, HITM), /* Remote RAM hit|SNP None or Miss */
> + OP_LH | P(LVL, IO) | LEVEL(NA) | P(SNOOP, NONE), /* I/O hit|SNP None */
> + OP_LH | P(LVL, UNC) | LEVEL(NA) | P(SNOOP, NONE), /* Uncached hit|SNP None */
> +};
> +
> +/* Based on kernel pebs_set_tlb_lock() */
> +static inline void pebs_set_tlb_lock(u64 *val, bool tlb, bool lock)
> +{
> + /*
> + * TLB access
> + * 0 = did not miss 2nd level TLB
> + * 1 = missed 2nd level TLB
> + */
> + if (tlb)
> + *val |= P(TLB, MISS) | P(TLB, L2);
> + else
> + *val |= P(TLB, HIT) | P(TLB, L1) | P(TLB, L2);
> +
> + /* locked prefix */
> + if (lock)
> + *val |= P(LOCK, LOCKED);
> +}
> +
> +/* Based on kernel __grt_latency_data() */
> +static u64 intel_pt_grt_latency_data(u8 dse, bool tlb, bool lock, bool blk,
> + const u64 *pebs_data_source)
> +{
> + u64 val;
> +
> + dse &= PERF_PEBS_DATA_SOURCE_GRT_MASK;
> + val = pebs_data_source[dse];
> +
> + pebs_set_tlb_lock(&val, tlb, lock);
> +
> + if (blk)
> + val |= P(BLK, DATA);
> + else
> + val |= P(BLK, NA);
> +
> + return val;
> +}
> +
> +/* Default value for data source */
> +#define PERF_MEM_NA (PERF_MEM_S(OP, NA) |\
> + PERF_MEM_S(LVL, NA) |\
> + PERF_MEM_S(SNOOP, NA) |\
> + PERF_MEM_S(LOCK, NA) |\
> + PERF_MEM_S(TLB, NA) |\
> + PERF_MEM_S(LVLNUM, NA))
> +
> +enum DATA_SRC_FORMAT {
> + DATA_SRC_FORMAT_ERR = -1,
> + DATA_SRC_FORMAT_NA = 0,
> + DATA_SRC_FORMAT_GRT = 1,
> + DATA_SRC_FORMAT_CMT = 2,
> +};
> +
> +/* Based on kernel grt_latency_data() and cmt_latency_data */
> +static u64 intel_pt_get_data_src(u64 mem_aux_info, int data_src_fmt)
> +{
> + switch (data_src_fmt) {
> + case DATA_SRC_FORMAT_GRT: {
> + union {
> + u64 val;
> + struct {
> + unsigned int dse:4;
> + unsigned int locked:1;
> + unsigned int stlb_miss:1;
> + unsigned int fwd_blk:1;
> + unsigned int reserved:25;
> + };
> + } x = {.val = mem_aux_info};
> + return intel_pt_grt_latency_data(x.dse, x.stlb_miss, x.locked, x.fwd_blk,
> + pebs_data_source_grt);
> + }
> + case DATA_SRC_FORMAT_CMT: {
> + union {
> + u64 val;
> + struct {
> + unsigned int dse:5;
> + unsigned int locked:1;
> + unsigned int stlb_miss:1;
> + unsigned int fwd_blk:1;
> + unsigned int reserved:24;
> + };
> + } x = {.val = mem_aux_info};
> + return intel_pt_grt_latency_data(x.dse, x.stlb_miss, x.locked, x.fwd_blk,
> + pebs_data_source_cmt);
> + }
> + default:
> + return PERF_MEM_NA;
> + }
> +}
> +
> +static int intel_pt_do_synth_pebs_sample(struct intel_pt_queue *ptq, struct evsel *evsel,
> + u64 id, int data_src_fmt)
> {
> const struct intel_pt_blk_items *items = &ptq->state->items;
> struct perf_sample sample;
> @@ -2393,6 +2534,18 @@ static int intel_pt_do_synth_pebs_sample(struct intel_pt_queue *ptq, struct evse
> }
> }
>
> + if (sample_type & PERF_SAMPLE_DATA_SRC) {
> + if (items->has_mem_aux_info && data_src_fmt) {
> + if (data_src_fmt < 0) {
> + pr_err("Intel PT missing data_src info\n");
> + return -1;
> + }
> + sample.data_src = intel_pt_get_data_src(items->mem_aux_info, data_src_fmt);
> + } else {
> + sample.data_src = PERF_MEM_NA;
> + }
> + }
> +
> if (sample_type & PERF_SAMPLE_TRANSACTION && items->has_tsx_aux_info) {
> u64 ax = items->has_rax ? items->rax : 0;
> /* Refer kernel's intel_hsw_transaction() */
> @@ -2413,9 +2566,10 @@ static int intel_pt_synth_single_pebs_sample(struct intel_pt_queue *ptq)
> {
> struct intel_pt *pt = ptq->pt;
> struct evsel *evsel = pt->pebs_evsel;
> + int data_src_fmt = pt->pebs_data_src_fmt;
> u64 id = evsel->core.id[0];
>
> - return intel_pt_do_synth_pebs_sample(ptq, evsel, id);
> + return intel_pt_do_synth_pebs_sample(ptq, evsel, id, data_src_fmt);
> }
>
> static int intel_pt_synth_pebs_sample(struct intel_pt_queue *ptq)
> @@ -2440,7 +2594,7 @@ static int intel_pt_synth_pebs_sample(struct intel_pt_queue *ptq)
> hw_id);
> return intel_pt_synth_single_pebs_sample(ptq);
> }
> - err = intel_pt_do_synth_pebs_sample(ptq, pe->evsel, pe->id);
> + err = intel_pt_do_synth_pebs_sample(ptq, pe->evsel, pe->id, pe->data_src_fmt);
> if (err)
> return err;
> }
> @@ -3407,6 +3561,49 @@ static int intel_pt_process_itrace_start(struct intel_pt *pt,
> event->itrace_start.tid);
> }
>
> +/*
> + * Events with data_src are identified by L1_Hit_Indication
> + * refer https://github.com/intel/perfmon
> + */
> +static int intel_pt_data_src_fmt(struct intel_pt *pt, struct evsel *evsel)
> +{
> + struct perf_env *env = pt->machine->env;
> + int fmt = DATA_SRC_FORMAT_NA;
> +
> + if (!env->cpuid)
> + return DATA_SRC_FORMAT_ERR;
> +
> + /*
> + * PEBS-via-PT is only supported on E-core non-hybrid. Of those only
> + * Gracemont and Crestmont have data_src. Check for:
> + * Alderlake N (Gracemont)
> + * Sierra Forest (Crestmont)
> + * Grand Ridge (Crestmont)
> + */
> +
> + if (!strncmp(env->cpuid, "GenuineIntel,6,190,", 19))
> + fmt = DATA_SRC_FORMAT_GRT;
> +
> + if (!strncmp(env->cpuid, "GenuineIntel,6,175,", 19) ||
> + !strncmp(env->cpuid, "GenuineIntel,6,182,", 19))
> + fmt = DATA_SRC_FORMAT_CMT;
> +
> + if (fmt == DATA_SRC_FORMAT_NA)
> + return fmt;
> +
> + /*
> + * Only data_src events are:
> + * mem-loads event=0xd0,umask=0x5
> + * mem-stores event=0xd0,umask=0x6
> + */
> + if (evsel->core.attr.type == PERF_TYPE_RAW &&
> + ((evsel->core.attr.config & 0xffff) == 0x5d0 ||
> + (evsel->core.attr.config & 0xffff) == 0x6d0))
> + return fmt;
> +
> + return DATA_SRC_FORMAT_NA;
> +}
> +
> static int intel_pt_process_aux_output_hw_id(struct intel_pt *pt,
> union perf_event *event,
> struct perf_sample *sample)
> @@ -3427,6 +3624,7 @@ static int intel_pt_process_aux_output_hw_id(struct intel_pt *pt,
>
> ptq->pebs[hw_id].evsel = evsel;
> ptq->pebs[hw_id].id = sample->id;
> + ptq->pebs[hw_id].data_src_fmt = intel_pt_data_src_fmt(pt, evsel);
>
> return 0;
> }
> @@ -3976,6 +4174,7 @@ static void intel_pt_setup_pebs_events(struct intel_pt *pt)
> }
> pt->single_pebs = true;
> pt->sample_pebs = true;
> + pt->pebs_data_src_fmt = intel_pt_data_src_fmt(pt, evsel);
> pt->pebs_evsel = evsel;
> }
> }
Powered by blists - more mailing lists