[PATCH 0/3] perf intel-pt: Fix PEBS-via-PT data_src etc

[PATCH 0/3] perf intel-pt: Fix PEBS-via-PT data_src etc

[Adrian Hunter]

Hi

Here are some Intel PT related fixes.


Adrian Hunter (3):
      perf intel-pt: Fix PEBS-via-PT data_src
      perf intel-pt: Do not default to recording all switch events
      perf scripts python: exported-sql-viewer.py: Fix pattern matching with Python 3

 tools/perf/Documentation/perf-config.txt         |   4 +
 tools/perf/arch/x86/util/intel-pt.c              |  16 +-
 tools/perf/scripts/python/exported-sql-viewer.py |   5 +-
 tools/perf/util/intel-pt.c                       | 205 ++++++++++++++++++++++-
 4 files changed, 225 insertions(+), 5 deletions(-)


Regards
Adrian

[PATCH 1/3] perf intel-pt: Fix PEBS-via-PT data_src

[Adrian Hunter]

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).

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@intel.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;
 		}
 	}
-- 
2.45.2

Re: [PATCH 1/3] perf intel-pt: Fix PEBS-via-PT data_src

[Liang, Kan]

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@linux.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@intel.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;
>  		}
>  	}

[PATCH 2/3] perf intel-pt: Do not default to recording all switch events

[Adrian Hunter]

On systems with many CPUs, recording extra context switch events can be
excessive and unnecessary. Add perf config intel-pt.all-switch-events=false
to control the behaviour.

Example:

 # perf config intel-pt.all-switch-events=false
 # perf record -eintel_pt//u uname
 Linux
 [ perf record: Woken up 1 times to write data ]
 [ perf record: Captured and wrote 0.082 MB perf.data ]
 # perf script -D | grep PERF_RECORD_SWITCH | awk '{print $5}' | uniq -c
       5 PERF_RECORD_SWITCH
 # perf config intel-pt.all-switch-events=true
 # perf record -eintel_pt//u uname
 Linux
 [ perf record: Woken up 1 times to write data ]
 [ perf record: Captured and wrote 0.102 MB perf.data ]
 # perf script -D | grep PERF_RECORD_SWITCH | awk '{print $5}' | uniq -c
     180 PERF_RECORD_SWITCH_CPU_WIDE

Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
---
 tools/perf/Documentation/perf-config.txt |  4 ++++
 tools/perf/arch/x86/util/intel-pt.c      | 16 +++++++++++++++-
 2 files changed, 19 insertions(+), 1 deletion(-)

diff --git a/tools/perf/Documentation/perf-config.txt b/tools/perf/Documentation/perf-config.txt
index 36ebebc875ea..c6f335659667 100644
--- a/tools/perf/Documentation/perf-config.txt
+++ b/tools/perf/Documentation/perf-config.txt
@@ -708,6 +708,10 @@ intel-pt.*::
 		the maximum is exceeded there will be a "Never-ending loop"
 		error. The default is 100000.
 
+	intel-pt.all-switch-events::
+		If the user has permission to do so, always record all context
+		switch events on all CPUs.
+
 auxtrace.*::
 
 	auxtrace.dumpdir::
diff --git a/tools/perf/arch/x86/util/intel-pt.c b/tools/perf/arch/x86/util/intel-pt.c
index 8f235d8b67b6..add33cb5d1da 100644
--- a/tools/perf/arch/x86/util/intel-pt.c
+++ b/tools/perf/arch/x86/util/intel-pt.c
@@ -19,6 +19,7 @@
 #include "../../../util/evlist.h"
 #include "../../../util/evsel.h"
 #include "../../../util/evsel_config.h"
+#include "../../../util/config.h"
 #include "../../../util/cpumap.h"
 #include "../../../util/mmap.h"
 #include <subcmd/parse-options.h>
@@ -52,6 +53,7 @@ struct intel_pt_recording {
 	struct perf_pmu			*intel_pt_pmu;
 	int				have_sched_switch;
 	struct evlist		*evlist;
+	bool				all_switch_events;
 	bool				snapshot_mode;
 	bool				snapshot_init_done;
 	size_t				snapshot_size;
@@ -794,7 +796,7 @@ static int intel_pt_recording_options(struct auxtrace_record *itr,
 			bool cpu_wide = !target__none(&opts->target) &&
 					!target__has_task(&opts->target);
 
-			if (!cpu_wide && perf_can_record_cpu_wide()) {
+			if (ptr->all_switch_events && !cpu_wide && perf_can_record_cpu_wide()) {
 				struct evsel *switch_evsel;
 
 				switch_evsel = evlist__add_dummy_on_all_cpus(evlist);
@@ -1178,6 +1180,16 @@ static u64 intel_pt_reference(struct auxtrace_record *itr __maybe_unused)
 	return rdtsc();
 }
 
+static int intel_pt_perf_config(const char *var, const char *value, void *data)
+{
+	struct intel_pt_recording *ptr = data;
+
+	if (!strcmp(var, "intel-pt.all-switch-events"))
+		ptr->all_switch_events = perf_config_bool(var, value);
+
+	return 0;
+}
+
 struct auxtrace_record *intel_pt_recording_init(int *err)
 {
 	struct perf_pmu *intel_pt_pmu = perf_pmus__find(INTEL_PT_PMU_NAME);
@@ -1197,6 +1209,8 @@ struct auxtrace_record *intel_pt_recording_init(int *err)
 		return NULL;
 	}
 
+	perf_config(intel_pt_perf_config, ptr);
+
 	ptr->intel_pt_pmu = intel_pt_pmu;
 	ptr->itr.recording_options = intel_pt_recording_options;
 	ptr->itr.info_priv_size = intel_pt_info_priv_size;
-- 
2.45.2

Re: [PATCH 2/3] perf intel-pt: Do not default to recording all switch events

[Arnaldo Carvalho de Melo]

On Mon, May 12, 2025 at 12:39:31PM +0300, Adrian Hunter wrote:
> On systems with many CPUs, recording extra context switch events can be
> excessive and unnecessary. Add perf config intel-pt.all-switch-events=false
> to control the behaviour.
> 
> Example:
> 
>  # perf config intel-pt.all-switch-events=false
>  # perf record -eintel_pt//u uname
>  Linux
>  [ perf record: Woken up 1 times to write data ]
>  [ perf record: Captured and wrote 0.082 MB perf.data ]
>  # perf script -D | grep PERF_RECORD_SWITCH | awk '{print $5}' | uniq -c
>        5 PERF_RECORD_SWITCH
>  # perf config intel-pt.all-switch-events=true
>  # perf record -eintel_pt//u uname
>  Linux
>  [ perf record: Woken up 1 times to write data ]
>  [ perf record: Captured and wrote 0.102 MB perf.data ]
>  # perf script -D | grep PERF_RECORD_SWITCH | awk '{print $5}' | uniq -c
>      180 PERF_RECORD_SWITCH_CPU_WIDE

Thanks, tested and applied the series.

BTW, I used 'perf report --stats' that shows the number of PERF_RECORD_
events:

Committer testing:

While doing a make -j28 allmodconfig:

  root@five:~# grep "model name" -m1 /proc/cpuinfo
  model name    : Intel(R) Core(TM) i7-14700K
  root@five:~#
  root@five:~# perf config intel-pt.all-switch-events=false
  root@five:~# perf record -e intel_pt//u uname
  Linux
  [ perf record: Woken up 2 times to write data ]
  [ perf record: Captured and wrote 0.019 MB perf.data ]
  root@five:~# perf report --stats | grep SWITCH_CPU_WIDE
  root@five:~# 
  root@five:~# perf config intel-pt.all-switch-events=true
  root@five:~# perf record -e intel_pt//u uname
  Linux
  [ perf record: Woken up 1 times to write data ]
  [ perf record: Captured and wrote 0.047 MB perf.data ]
  root@five:~# perf report --stats | grep SWITCH_CPU_WIDE
       SWITCH_CPU_WIDE events:        542  (96.4%)
  root@five:~#

- Arnaldo

[PATCH 3/3] perf scripts python: exported-sql-viewer.py: Fix pattern matching with Python 3

[Adrian Hunter]

The script allows the user to enter patterns to find symbols.  The pattern
matching characters are converted for use in SQL.  For PostgreSQL the
conversion involves using the Python maketrans() method which is slightly
different in Python 3 compared with Python 2.  Fix to work in Python 3.

Fixes: beda0e725e5f ("perf script python: Add Python3 support to exported-sql-viewer.py")
Signed-off-by: Adrian Hunter <adrian.hunter@intel.com>
---
 tools/perf/scripts/python/exported-sql-viewer.py | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)

diff --git a/tools/perf/scripts/python/exported-sql-viewer.py b/tools/perf/scripts/python/exported-sql-viewer.py
index 121cf61ba1b3..e0b2e7268ef6 100755
--- a/tools/perf/scripts/python/exported-sql-viewer.py
+++ b/tools/perf/scripts/python/exported-sql-viewer.py
@@ -680,7 +680,10 @@ class CallGraphModelBase(TreeModel):
 				s = value.replace("%", "\\%")
 				s = s.replace("_", "\\_")
 				# Translate * and ? into SQL LIKE pattern characters % and _
-				trans = string.maketrans("*?", "%_")
+				if sys.version_info[0] == 3:
+					trans = str.maketrans("*?", "%_")
+				else:
+					trans = string.maketrans("*?", "%_")
 				match = " LIKE '" + str(s).translate(trans) + "'"
 			else:
 				match = " GLOB '" + str(value) + "'"
-- 
2.45.2