perf pmu intel: Generalize SNC cpumask adjustment for multiple platforms
PATCH v5 1/2 perf pmu intel: Generalize SNC cpumask adjustment for multiple platforms
Prepare for supporting more Intel platforms with sub-NUMA clustering by generalizing the GNR specific logic.
---
v5:
Split patch.
v4: lore.kernel.org/[email protected]
Rebase.
v3: lore.kernel.org/[email protected]
Fix a typo.
v2: lore.kernel.org/[email protected]
Split EMR and GNR in the SNC2 IMC cpu map.
v1: lore.kernel.org/[email protected]
tools/perf/arch/x86/util/pmu.c | 44 +++++++++++++++++++++------------- 1 file changed, 27 insertions(+), 17 deletions(-)
tools/perf/arch/x86/util/pmu.c +27 -17
@@ -23,20 +23,28 @@#include "util/env.h" #include "util/header.h" -static bool x86__is_intel_graniterapids (void ) +static bool x86__is_snc_supported (void ) { -static bool checked_if_graniterapids; -static bool is_graniterapids; +static bool checked_if_snc_supported; +static bool is_supported; -if (!checked_if_graniterapids) { -const char *graniterapids_cpuid ="GenuineIntel-6-A[DE]" ; +if (!checked_if_snc_supported) { + +/* Graniterapids supports SNC configuration. */ +static const char *const supported_cpuids[] = { +"GenuineIntel-6-A[DE]" ,/* Graniterapids */ + };char *cpuid =get_cpuid_str ((struct perf_cpu ){0 }); - is_graniterapids = cpuid &&strcmp_cpuid_str (graniterapids_cpuid, cpuid) ==0 ; +for (size_t i =0 ; i <ARRAY_SIZE (supported_cpuids); i++) { + is_supported = cpuid &&strcmp_cpuid_str (supported_cpuids[i], cpuid) ==0 ; +if (is_supported) +break ; + }free (cpuid); - checked_if_graniterapids = true; + checked_if_snc_supported = true; } -return is_graniterapids; +return is_supported; }static struct perf_cpu_map *read_sysfs_cpu_map (const char *sysfs_path)
@@ -133,8 +141,8 @@ static int uncore_imc_snc(struct perf_pmu *pmu)// Compute the IMC SNC using lookup tables. unsigned int imc_num;int snc_nodes =snc_nodes_per_l3_cache (); -const u8 snc2_map[] = {1 ,1 ,0 ,0 ,1 ,1 ,0 ,0 }; -const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 ,1 ,1 ,0 ,0 ,2 ,2 }; +const u8 snc2_map[] = {1 ,1 ,0 ,0 }; +const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 };const u8 *snc_map;size_t snc_map_len;
@@ -157,11 +165,12 @@ static int uncore_imc_snc(struct perf_pmu *pmu)pr_warning ("Unexpected: unable to compute IMC number '%s'\n" , pmu->name);return 0 ; } -if (imc_num >= snc_map_len) { +if (imc_num >= snc_map_len *perf_cpu_map__nr (pmu->cpus)) {pr_warning ("Unexpected IMC %d for SNC%d mapping\n" , imc_num, snc_nodes);return 0 ; } -return snc_map[imc_num]; + +return snc_map[imc_num % snc_map_len]; }static int uncore_cha_imc_compute_cpu_adjust (int pmu_snc)
@@ -201,7 +210,7 @@ static int uncore_cha_imc_compute_cpu_adjust(int pmu_snc)return cpu_adjust[pmu_snc]; } -static void gnr_uncore_cha_imc_adjust_cpumask_for_snc (struct perf_pmu *pmu,bool cha) +static void uncore_cha_imc_adjust_cpumask_for_snc (struct perf_pmu *pmu,bool cha) {// With sub-NUMA clustering (SNC) there is a NUMA node per SNC in the // topology. For example, a two socket graniterapids machine may be set
@@ -301,11 +310,12 @@ void perf_pmu__arch_init(struct perf_pmu *pmu) pmu->mem_events = perf_mem_events_intel_aux; else pmu->mem_events = perf_mem_events_intel; - }else if (x86__is_intel_graniterapids ()) { + }else if (x86__is_snc_supported ()) {if (strstarts (pmu->name,"uncore_cha_" )) -gnr_uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ true); -else if (strstarts (pmu->name,"uncore_imc_" )) -gnr_uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ false); +uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ true); +else if (strstarts (pmu->name,"uncore_imc_" ) && + !strstarts (pmu->name,"uncore_imc_free_running" )) +uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ false); } } }
--
2.53.0.1213.gd9a14994de-goog
PATCH v5 2/2 perf pmu intel: Adjust cpumaks for sub-NUMA clusters on Emeraldrapids
Similar to GNR [1], Emeraldrapids supports sub-NUMA clusters as well. Adjust cpumasks as the logic for GNR in [1].
Tested on Emeraldrapids with SNC2 enabled:
$ perf stat --per-node -e 'UNC_CHA_CLOCKTICKS,UNC_M_CLOCKTICKS' -a -- sleep 1
Performance counter stats for 'system wide':
N0 30 72125876670 UNC_CHA_CLOCKTICKS N0 4 8815163648 UNC_M_CLOCKTICKS N1 30 72124958844 UNC_CHA_CLOCKTICKS N1 4 8815014974 UNC_M_CLOCKTICKS N2 30 72121049022 UNC_CHA_CLOCKTICKS N2 4 8814592626 UNC_M_CLOCKTICKS N3 30 72117133854 UNC_CHA_CLOCKTICKS N3 4 8814012840 UNC_M_CLOCKTICKS
1.001574118 seconds time elapsed
[1] lore.kernel.org/[email protected]
--- tools/perf/arch/x86/util/pmu.c | 56 +++++++++++++++++++++++----------- 1 file changed, 38 insertions(+), 18 deletions(-)
tools/perf/arch/x86/util/pmu.c +38 -18
@@ -30,8 +30,9 @@ static bool x86__is_snc_supported(void)if (!checked_if_snc_supported) { -/* Graniterapids supports SNC configuration. */ +/* Emeraldrapids Graniterapids support SNC configuration. */ static const char *const supported_cpuids[] = { +"GenuineIntel-6-CF" ,/* Emeraldrapids */ "GenuineIntel-6-A[DE]" ,/* Graniterapids */ };char *cpuid =get_cpuid_str ((struct perf_cpu ){0 });
@@ -141,23 +142,42 @@ static int uncore_imc_snc(struct perf_pmu *pmu)// Compute the IMC SNC using lookup tables. unsigned int imc_num;int snc_nodes =snc_nodes_per_l3_cache (); -const u8 snc2_map[] = {1 ,1 ,0 ,0 }; -const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 }; -const u8 *snc_map; -size_t snc_map_len; - -switch (snc_nodes) { -case 2 : - snc_map = snc2_map; - snc_map_len =ARRAY_SIZE (snc2_map); -break ; -case 3 : - snc_map = snc3_map; - snc_map_len =ARRAY_SIZE (snc3_map); -break ; -default : -/* Error or no lookup support for SNC with >3 nodes. */ -return 0 ; +char *cpuid; +static const u8 emr_snc2_map[] = {0 ,0 ,1 ,1 }; +static const u8 gnr_snc2_map[] = {1 ,1 ,0 ,0 }; +static const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 }; +static const u8 *snc_map; +static size_t snc_map_len; + +/* snc_map is not inited yet. We only look up once to avoid expensive operations. */ +if (!snc_map) { +switch (snc_nodes) { +case 2 : + cpuid =get_cpuid_str ((struct perf_cpu ){0 }); +if (cpuid) { +if (strcmp_cpuid_str ("GenuineIntel-6-CF" , cpuid) ==0 ) { + snc_map = emr_snc2_map; + snc_map_len =ARRAY_SIZE (emr_snc2_map); + }else if (strcmp_cpuid_str ("GenuineIntel-6-A[DE]" , cpuid) ==0 ) { + snc_map = gnr_snc2_map; + snc_map_len =ARRAY_SIZE (gnr_snc2_map); + } +free (cpuid); + } +break ; +case 3 : + snc_map = snc3_map; + snc_map_len =ARRAY_SIZE (snc3_map); +break ; +default : +/* Error or no lookup support for SNC with >3 nodes. */ +return 0 ; + } + +if (!snc_map) { +pr_warning ("Unexpected: can not find snc map config" ); +return 0 ; + } }/* Compute SNC for PMU. */
Signature · 1 line
2.53.0.1213.gd9a14994de-goog
Namhyung Kim
maintainerRe: [PATCH v5 2/2] perf pmu intel: Adjust cpumaks for sub-NUMA clusters on Emeraldrapids
Quoted 33 lines · Chun-Tse Shao
Similar to GNR [1], Emeraldrapids supports sub-NUMA clusters as well. Adjust cpumasks as the logic for GNR in [1].
Tested on Emeraldrapids with SNC2 enabled:
$ perf stat --per-node -e 'UNC_CHA_CLOCKTICKS,UNC_M_CLOCKTICKS' -a -- sleep 1Performance counter stats for 'system wide':
N0 30 72125876670 UNC_CHA_CLOCKTICKS N0 4 8815163648 UNC_M_CLOCKTICKS N1 30 72124958844 UNC_CHA_CLOCKTICKS N1 4 8815014974 UNC_M_CLOCKTICKS N2 30 72121049022 UNC_CHA_CLOCKTICKS N2 4 8814592626 UNC_M_CLOCKTICKS N3 30 72117133854 UNC_CHA_CLOCKTICKS N3 4 8814012840 UNC_M_CLOCKTICKS1.001574118 seconds time elapsed[1] lore.kernel.org/[email protected]
Reviewed-by Zide Chen <[email protected]>Reviewed-by Ian Rogers <[email protected]>Signed-off-by Chun-Tse Shao <[email protected]>--- tools/perf/arch/x86/util/pmu.c | 56 +++++++++++++++++++++++----------- 1 file changed, 38 insertions(+), 18 deletions(-)tools/perf/arch/x86/util/pmu.c +0
-0
tools/perf/arch/x86/util/pmu.c static bool x86__is_snc_supported(void) · line 36
Quoted 8 lines
tools/perf/arch/x86/util/pmu.c +2
-1@@ -30,8 +30,9 @@ static bool x86__is_snc_supported(void)if (!checked_if_snc_supported) { -/* Graniterapids supports SNC configuration. */ +/* Emeraldrapids Graniterapids support SNC configuration. */ static const char *const supported_cpuids[] = { +"GenuineIntel-6-CF" ,/* Emeraldrapids */ "GenuineIntel-6-A[DE]" ,/* Graniterapids */
It'd be great if we can share these string literals..
Quoted 32 lines
tools/perf/arch/x86/util/pmu.c +9
-17};char *cpuid =get_cpuid_str ((struct perf_cpu ){0 });@@ -141,23 +142,42 @@ static int uncore_imc_snc(struct perf_pmu *pmu)// Compute the IMC SNC using lookup tables. unsigned int imc_num;int snc_nodes =snc_nodes_per_l3_cache (); -const u8 snc2_map[] = {1 ,1 ,0 ,0 }; -const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 }; -const u8 *snc_map; -size_t snc_map_len; - -switch (snc_nodes) { -case 2 : - snc_map = snc2_map; - snc_map_len =ARRAY_SIZE (snc2_map); -break ; -case 3 : - snc_map = snc3_map; - snc_map_len =ARRAY_SIZE (snc3_map); -break ; -default : -/* Error or no lookup support for SNC with >3 nodes. */ -return 0 ; +char *cpuid; +static const u8 emr_snc2_map[] = {0 ,0 ,1 ,1 }; +static const u8 gnr_snc2_map[] = {1 ,1 ,0 ,0 }; +static const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 }; +static const u8 *snc_map; +static size_t snc_map_len; + +/* snc_map is not inited yet. We only look up once to avoid expensive operations. */ +if (!snc_map) {
tools/perf/arch/x86/util/pmu.c static int uncore_imc_snc(struct perf_pmu *pmu) · line 163
Quoted 10 lines
tools/perf/arch/x86/util/pmu.c +10
-0+switch (snc_nodes) { +case 2 : + cpuid =get_cpuid_str ((struct perf_cpu ){0 }); +if (cpuid) { +if (strcmp_cpuid_str ("GenuineIntel-6-CF" , cpuid) ==0 ) { + snc_map = emr_snc2_map; + snc_map_len =ARRAY_SIZE (emr_snc2_map); + }else if (strcmp_cpuid_str ("GenuineIntel-6-A[DE]" , cpuid) ==0 ) { + snc_map = gnr_snc2_map; + snc_map_len =ARRAY_SIZE (gnr_snc2_map);
... in here as well.
Thanks,
Namhyung
Quoted 23 lines
tools/perf/arch/x86/util/pmu.c +17
-0+ } +free (cpuid); + } +break ; +case 3 : + snc_map = snc3_map; + snc_map_len =ARRAY_SIZE (snc3_map); +break ; +default : +/* Error or no lookup support for SNC with >3 nodes. */ +return 0 ; + } + +if (!snc_map) { +pr_warning ("Unexpected: can not find snc map config" ); +return 0 ; + } }/* Compute SNC for PMU. */ Signature · 1 line
2.53.0.1213.gd9a14994de-goog
Re: [PATCH v5 2/2] perf pmu intel: Adjust cpumaks for sub-NUMA clusters on Emeraldrapids
Thanks, submitted v6 patch:
lore.kernel.org/[email protected] I also added SPR into SNC2 in that patch.
-CT
Quoted 119 lines · Namhyung Kim
Quoted 33 lines · Chun-Tse Shao
Similar to GNR [1], Emeraldrapids supports sub-NUMA clusters as well. Adjust cpumasks as the logic for GNR in [1].
Tested on Emeraldrapids with SNC2 enabled:
$ perf stat --per-node -e 'UNC_CHA_CLOCKTICKS,UNC_M_CLOCKTICKS' -a -- sleep 1Performance counter stats for 'system wide':
N0 30 72125876670 UNC_CHA_CLOCKTICKS N0 4 8815163648 UNC_M_CLOCKTICKS N1 30 72124958844 UNC_CHA_CLOCKTICKS N1 4 8815014974 UNC_M_CLOCKTICKS N2 30 72121049022 UNC_CHA_CLOCKTICKS N2 4 8814592626 UNC_M_CLOCKTICKS N3 30 72117133854 UNC_CHA_CLOCKTICKS N3 4 8814012840 UNC_M_CLOCKTICKS1.001574118 seconds time elapsed[1] lore.kernel.org/[email protected]
Reviewed-by Zide Chen <[email protected]>Reviewed-by Ian Rogers <[email protected]>Signed-off-by Chun-Tse Shao <[email protected]>--- tools/perf/arch/x86/util/pmu.c | 56 +++++++++++++++++++++++----------- 1 file changed, 38 insertions(+), 18 deletions(-)tools/perf/arch/x86/util/pmu.c +0
-0tools/perf/arch/x86/util/pmu.c static bool x86__is_snc_supported(void) · line 36
Quoted 8 lines
tools/perf/arch/x86/util/pmu.c +2
-1@@ -30,8 +30,9 @@ static bool x86__is_snc_supported(void)if (!checked_if_snc_supported) { -/* Graniterapids supports SNC configuration. */ +/* Emeraldrapids Graniterapids support SNC configuration. */ static const char *const supported_cpuids[] = { +"GenuineIntel-6-CF" ,/* Emeraldrapids */ "GenuineIntel-6-A[DE]" ,/* Graniterapids */ It'd be great if we can share these string literals..
Quoted 32 lines
tools/perf/arch/x86/util/pmu.c +9
-17};char *cpuid =get_cpuid_str ((struct perf_cpu ){0 });@@ -141,23 +142,42 @@ static int uncore_imc_snc(struct perf_pmu *pmu)// Compute the IMC SNC using lookup tables. unsigned int imc_num;int snc_nodes =snc_nodes_per_l3_cache (); -const u8 snc2_map[] = {1 ,1 ,0 ,0 }; -const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 }; -const u8 *snc_map; -size_t snc_map_len; - -switch (snc_nodes) { -case 2 : - snc_map = snc2_map; - snc_map_len =ARRAY_SIZE (snc2_map); -break ; -case 3 : - snc_map = snc3_map; - snc_map_len =ARRAY_SIZE (snc3_map); -break ; -default : -/* Error or no lookup support for SNC with >3 nodes. */ -return 0 ; +char *cpuid; +static const u8 emr_snc2_map[] = {0 ,0 ,1 ,1 }; +static const u8 gnr_snc2_map[] = {1 ,1 ,0 ,0 }; +static const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 }; +static const u8 *snc_map; +static size_t snc_map_len; + +/* snc_map is not inited yet. We only look up once to avoid expensive operations. */ +if (!snc_map) {tools/perf/arch/x86/util/pmu.c static int uncore_imc_snc(struct perf_pmu *pmu) · line 163
Quoted 10 lines
tools/perf/arch/x86/util/pmu.c +10
-0+switch (snc_nodes) { +case 2 : + cpuid =get_cpuid_str ((struct perf_cpu ){0 }); +if (cpuid) { +if (strcmp_cpuid_str ("GenuineIntel-6-CF" , cpuid) ==0 ) { + snc_map = emr_snc2_map; + snc_map_len =ARRAY_SIZE (emr_snc2_map); + }else if (strcmp_cpuid_str ("GenuineIntel-6-A[DE]" , cpuid) ==0 ) { + snc_map = gnr_snc2_map; + snc_map_len =ARRAY_SIZE (gnr_snc2_map);... in here as well.
Thanks,
NamhyungQuoted 23 lines
tools/perf/arch/x86/util/pmu.c +17
-1+ } +free (cpuid); + } +break ; +case 3 : + snc_map = snc3_map; + snc_map_len =ARRAY_SIZE (snc3_map); +break ; +default : +/* Error or no lookup support for SNC with >3 nodes. */ +return 0 ; + } + +if (!snc_map) { +pr_warning ("Unexpected: can not find snc map config" ); +return 0 ; + } }/* Compute SNC for PMU. */ --2.53.0.1213.gd9a14994de-goog
Namhyung Kim
maintainerHello,
Quoted 75 lines · Chun-Tse Shao
Prepare for supporting more Intel platforms with sub-NUMA clustering by generalizing the GNR specific logic.
Signed-off-by Chun-Tse Shao <[email protected]>---
v5:
Split patch.v4: lore.kernel.org/[email protected]
Rebase.v3: lore.kernel.org/[email protected]
Fix a typo.v2: lore.kernel.org/[email protected]
Split EMR and GNR in the SNC2 IMC cpu map.v1: lore.kernel.org/[email protected]
tools/perf/arch/x86/util/pmu.c | 44 +++++++++++++++++++++------------- 1 file changed, 27 insertions(+), 17 deletions(-)tools/perf/arch/x86/util/pmu.c +18
-10@@ -23,20 +23,28 @@#include "util/env.h" #include "util/header.h" -static bool x86__is_intel_graniterapids (void ) +static bool x86__is_snc_supported (void ) { -static bool checked_if_graniterapids; -static bool is_graniterapids; +static bool checked_if_snc_supported; +static bool is_supported; -if (!checked_if_graniterapids) { -const char *graniterapids_cpuid ="GenuineIntel-6-A[DE]" ; +if (!checked_if_snc_supported) { + +/* Graniterapids supports SNC configuration. */ +static const char *const supported_cpuids[] = { +"GenuineIntel-6-A[DE]" ,/* Graniterapids */ + };char *cpuid =get_cpuid_str ((struct perf_cpu ){0 }); - is_graniterapids = cpuid &&strcmp_cpuid_str (graniterapids_cpuid, cpuid) ==0 ; +for (size_t i =0 ; i <ARRAY_SIZE (supported_cpuids); i++) { + is_supported = cpuid &&strcmp_cpuid_str (supported_cpuids[i], cpuid) ==0 ; +if (is_supported) +break ; + }free (cpuid); - checked_if_graniterapids = true; + checked_if_snc_supported = true; } -return is_graniterapids; +return is_supported; }static struct perf_cpu_map *read_sysfs_cpu_map (const char *sysfs_path)@@ -133,8 +141,8 @@ static int uncore_imc_snc(struct perf_pmu *pmu)// Compute the IMC SNC using lookup tables. unsigned int imc_num;int snc_nodes =snc_nodes_per_l3_cache (); -const u8 snc2_map[] = {1 ,1 ,0 ,0 ,1 ,1 ,0 ,0 }; -const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 ,1 ,1 ,0 ,0 ,2 ,2 }; +const u8 snc2_map[] = {1 ,1 ,0 ,0 }; +const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 };const u8 *snc_map;size_t snc_map_len;
tools/perf/arch/x86/util/pmu.c static int uncore_imc_snc(struct perf_pmu *pmu) · line 170
Quoted 7 lines
tools/perf/arch/x86/util/pmu.c +1
-1@@ -157,11 +165,12 @@ static int uncore_imc_snc(struct perf_pmu *pmu)pr_warning ("Unexpected: unable to compute IMC number '%s'\n" , pmu->name);return 0 ; } -if (imc_num >= snc_map_len) { +if (imc_num >= snc_map_len *perf_cpu_map__nr (pmu->cpus)) {pr_warning ("Unexpected IMC %d for SNC%d mapping\n" , imc_num, snc_nodes);return 0 ;
Like sashiko said, I'm curious if it'd work well on 1-socket machine which may have the same number of uncore IMC PMUs.
Zide, can you confirm?
Thanks,
Namhyung
Quoted 36 lines
tools/perf/arch/x86/util/pmu.c +8
-7} -return snc_map[imc_num]; + +return snc_map[imc_num % snc_map_len]; }static int uncore_cha_imc_compute_cpu_adjust (int pmu_snc)@@ -201,7 +210,7 @@ static int uncore_cha_imc_compute_cpu_adjust(int pmu_snc)return cpu_adjust[pmu_snc]; } -static void gnr_uncore_cha_imc_adjust_cpumask_for_snc (struct perf_pmu *pmu,bool cha) +static void uncore_cha_imc_adjust_cpumask_for_snc (struct perf_pmu *pmu,bool cha) {// With sub-NUMA clustering (SNC) there is a NUMA node per SNC in the // topology. For example, a two socket graniterapids machine may be set @@ -301,11 +310,12 @@ void perf_pmu__arch_init(struct perf_pmu *pmu) pmu->mem_events = perf_mem_events_intel_aux; else pmu->mem_events = perf_mem_events_intel; - }else if (x86__is_intel_graniterapids ()) { + }else if (x86__is_snc_supported ()) {if (strstarts (pmu->name,"uncore_cha_" )) -gnr_uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ true); -else if (strstarts (pmu->name,"uncore_imc_" )) -gnr_uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ false); +uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ true); +else if (strstarts (pmu->name,"uncore_imc_" ) && + !strstarts (pmu->name,"uncore_imc_free_running" )) +uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ false); } } } --2.53.0.1213.gd9a14994de-goog
Quoted 89 lines · Namhyung Kim
Hello,
Quoted 75 lines · Chun-Tse Shao
Prepare for supporting more Intel platforms with sub-NUMA clustering by generalizing the GNR specific logic.
Signed-off-by Chun-Tse Shao <[email protected]>---
v5:
Split patch.v4: lore.kernel.org/[email protected]
Rebase.v3: lore.kernel.org/[email protected]
Fix a typo.v2: lore.kernel.org/[email protected]
Split EMR and GNR in the SNC2 IMC cpu map.v1: lore.kernel.org/[email protected]
tools/perf/arch/x86/util/pmu.c | 44 +++++++++++++++++++++------------- 1 file changed, 27 insertions(+), 17 deletions(-)tools/perf/arch/x86/util/pmu.c +18
-10@@ -23,20 +23,28 @@#include "util/env.h" #include "util/header.h" -static bool x86__is_intel_graniterapids (void ) +static bool x86__is_snc_supported (void ) { -static bool checked_if_graniterapids; -static bool is_graniterapids; +static bool checked_if_snc_supported; +static bool is_supported; -if (!checked_if_graniterapids) { -const char *graniterapids_cpuid ="GenuineIntel-6-A[DE]" ; +if (!checked_if_snc_supported) { + +/* Graniterapids supports SNC configuration. */ +static const char *const supported_cpuids[] = { +"GenuineIntel-6-A[DE]" ,/* Graniterapids */ + };char *cpuid =get_cpuid_str ((struct perf_cpu ){0 }); - is_graniterapids = cpuid &&strcmp_cpuid_str (graniterapids_cpuid, cpuid) ==0 ; +for (size_t i =0 ; i <ARRAY_SIZE (supported_cpuids); i++) { + is_supported = cpuid &&strcmp_cpuid_str (supported_cpuids[i], cpuid) ==0 ; +if (is_supported) +break ; + }free (cpuid); - checked_if_graniterapids = true; + checked_if_snc_supported = true; } -return is_graniterapids; +return is_supported; }static struct perf_cpu_map *read_sysfs_cpu_map (const char *sysfs_path)@@ -133,8 +141,8 @@ static int uncore_imc_snc(struct perf_pmu *pmu)// Compute the IMC SNC using lookup tables. unsigned int imc_num;int snc_nodes =snc_nodes_per_l3_cache (); -const u8 snc2_map[] = {1 ,1 ,0 ,0 ,1 ,1 ,0 ,0 }; -const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 ,1 ,1 ,0 ,0 ,2 ,2 }; +const u8 snc2_map[] = {1 ,1 ,0 ,0 }; +const u8 snc3_map[] = {1 ,1 ,0 ,0 ,2 ,2 };const u8 *snc_map;size_t snc_map_len;tools/perf/arch/x86/util/pmu.c static int uncore_imc_snc(struct perf_pmu *pmu) · line 170
Quoted 7 lines
tools/perf/arch/x86/util/pmu.c +1
-1@@ -157,11 +165,12 @@ static int uncore_imc_snc(struct perf_pmu *pmu)pr_warning ("Unexpected: unable to compute IMC number '%s'\n" , pmu->name);return 0 ; } -if (imc_num >= snc_map_len) { +if (imc_num >= snc_map_len *perf_cpu_map__nr (pmu->cpus)) {pr_warning ("Unexpected IMC %d for SNC%d mapping\n" , imc_num, snc_nodes);return 0 ;Like sashiko said, I'm curious if it'd work well on 1-socket machine which may have the same number of uncore IMC PMUs.
Quoted 1 line
Zide, can you confirm?
On 1-socket, 1-die systems, the IMC controllers may be the same as on multi-die systems. Theoretically, this could cause issues if NUMA is enabled, but I am not aware of any platforms that support SNC on single-die GNR/EMR SKUs.
I checked on a single-die GNR WS: SNC is not supported, so snc_nodes == 1 and the cpumask adjustment is correctly skipped.
Quoted 40 lines
Thanks,
NamhyungQuoted 36 lines
tools/perf/arch/x86/util/pmu.c +8
-7} -return snc_map[imc_num]; + +return snc_map[imc_num % snc_map_len]; }static int uncore_cha_imc_compute_cpu_adjust (int pmu_snc)@@ -201,7 +210,7 @@ static int uncore_cha_imc_compute_cpu_adjust(int pmu_snc)return cpu_adjust[pmu_snc]; } -static void gnr_uncore_cha_imc_adjust_cpumask_for_snc (struct perf_pmu *pmu,bool cha) +static void uncore_cha_imc_adjust_cpumask_for_snc (struct perf_pmu *pmu,bool cha) {// With sub-NUMA clustering (SNC) there is a NUMA node per SNC in the // topology. For example, a two socket graniterapids machine may be set @@ -301,11 +310,12 @@ void perf_pmu__arch_init(struct perf_pmu *pmu) pmu->mem_events = perf_mem_events_intel_aux; else pmu->mem_events = perf_mem_events_intel; - }else if (x86__is_intel_graniterapids ()) { + }else if (x86__is_snc_supported ()) {if (strstarts (pmu->name,"uncore_cha_" )) -gnr_uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ true); -else if (strstarts (pmu->name,"uncore_imc_" )) -gnr_uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ false); +uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ true); +else if (strstarts (pmu->name,"uncore_imc_" ) && + !strstarts (pmu->name,"uncore_imc_free_running" )) +uncore_cha_imc_adjust_cpumask_for_snc (pmu,/*cha=*/ false); } } } --2.53.0.1213.gd9a14994de-goog