From a80e53472209b1c749e02e91ac62c053ac457099 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Tue, 3 Dec 2024 16:11:21 -0500 Subject: [PATCH 01/16] tools/power turbostat: version 2025.01.14 Fix checkpatch whitespace issues since 2024.11.30 Summary of Changes since 2024.11.30: Enable SysWatt by default. Add initial PTL, CWF platform support. Refuse to run on unsupported platforms without --force to avoid not-so-useful measurements mistakenly made using obsolete versions. Harden initial PMT code in response to early use. Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 36 ++++++++++++--------------- 1 file changed, 16 insertions(+), 20 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 0f2475fa9fa4..76d2632e60ac 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -3,7 +3,7 @@ * turbostat -- show CPU frequency and C-state residency * on modern Intel and AMD processors. * - * Copyright (c) 2024 Intel Corporation. + * Copyright (c) 2025 Intel Corporation. * Len Brown */ @@ -271,11 +271,11 @@ struct msr_counter bic[] = { #define BIC_Sys_J (1ULL << 60) #define BIC_NMI (1ULL << 61) -#define BIC_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die ) -#define BIC_THERMAL_PWR ( BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__ | BIC_SysWatt) +#define BIC_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die) +#define BIC_THERMAL_PWR (BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__ | BIC_SysWatt) #define BIC_FREQUENCY (BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_SAMMHz | BIC_SAMACTMHz | BIC_UNCORE_MHZ) #define BIC_IDLE (BIC_Busy | BIC_sysfs | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_SAM_mc6 | BIC_Diec6) -#define BIC_OTHER ( BIC_IRQ | BIC_NMI | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) +#define BIC_OTHER (BIC_IRQ | BIC_NMI | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) #define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC) @@ -1593,8 +1593,7 @@ struct pmt_counter { * PMT telemetry directory iterator. * Used to iterate telemetry files in sysfs in correct order. */ -struct pmt_diriter_t -{ +struct pmt_diriter_t { DIR *dir; struct dirent **namelist; unsigned int num_names; @@ -1604,6 +1603,7 @@ struct pmt_diriter_t int pmt_telemdir_filter(const struct dirent *e) { unsigned int dummy; + return sscanf(e->d_name, "telem%u", &dummy); } @@ -1617,7 +1617,7 @@ int pmt_telemdir_sort(const struct dirent **a, const struct dirent **b) return aidx >= bidx; } -const struct dirent* pmt_diriter_next(struct pmt_diriter_t *iter) +const struct dirent *pmt_diriter_next(struct pmt_diriter_t *iter) { const struct dirent *ret = NULL; @@ -1633,7 +1633,7 @@ const struct dirent* pmt_diriter_next(struct pmt_diriter_t *iter) return ret; } -const struct dirent* pmt_diriter_begin(struct pmt_diriter_t *iter, const char *pmt_root_path) +const struct dirent *pmt_diriter_begin(struct pmt_diriter_t *iter, const char *pmt_root_path) { int num_names = iter->num_names; @@ -2302,7 +2302,7 @@ void help(void) " -h, --help\n" " print this help message\n" " -v, --version\n" - " print version information\n" "\n" "For more help, run \"man turbostat\"\n"); + " print version information\n\nFor more help, run \"man turbostat\"\n"); } /* @@ -9053,18 +9053,16 @@ struct pmt_mmio *pmt_mmio_open(unsigned int target_guid) return NULL; } - for (;entry != NULL; entry = pmt_diriter_next(&pmt_iter)) { - if (fstatat(dirfd(pmt_iter.dir), entry->d_name, &st, 0) == -1) { + for ( ; entry != NULL; entry = pmt_diriter_next(&pmt_iter)) { + if (fstatat(dirfd(pmt_iter.dir), entry->d_name, &st, 0) == -1) break; - } if (!S_ISDIR(st.st_mode)) continue; fd_telem_dir = openat(dirfd(pmt_iter.dir), entry->d_name, O_RDONLY); - if (fd_telem_dir == -1) { + if (fd_telem_dir == -1) break; - } if (parse_telem_info_file(fd_telem_dir, "guid", "%lx", &guid)) { close(fd_telem_dir); @@ -9425,7 +9423,7 @@ int get_and_dump_counters(void) void print_version() { - fprintf(outf, "turbostat version 2024.11.30 - Len Brown \n"); + fprintf(outf, "turbostat version 2025.01.14 - Len Brown \n"); } #define COMMAND_LINE_SIZE 2048 @@ -9750,7 +9748,7 @@ next: } if ((msr_num == 0) && (path == NULL) && (perf_device[0] == '\0' || perf_event[0] == '\0')) { - fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter | perf/device/event ) required\n"); + fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter | perf/device/event) required\n"); fail++; } @@ -9822,9 +9820,8 @@ int pmt_parse_from_path(const char *target_path, unsigned int *out_guid, unsigne dirname = pmt_diriter_next(&pmt_iter)) { fd_telem_dir = openat(dirfd(pmt_iter.dir), dirname->d_name, O_RDONLY | O_DIRECTORY); - if (fd_telem_dir == -1) { + if (fd_telem_dir == -1) continue; - } if (parse_telem_info_file(fd_telem_dir, "guid", "%lx", &guid)) { fprintf(stderr, "%s: Failed to parse the guid file: %s", __func__, strerror(errno)); @@ -9962,9 +9959,8 @@ void parse_add_command_pmt(char *add_command) goto next; } - if (sscanf(add_command, "seq=%x", &seq) == 1) { + if (sscanf(add_command, "seq=%x", &seq) == 1) goto next; - } if (strncmp(add_command, direct_path_prefix, strlen(direct_path_prefix)) == 0) { direct_path = add_command + strlen(direct_path_prefix); -- 2.51.0 From 1a202afeaa370970413846c2cb09b383875e753c Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Fri, 17 Jan 2025 13:36:59 +0100 Subject: [PATCH 02/16] tools/power turbostat: Add tcore clock PMT type Some PMT counters, for example module c1e residency on Intel Clearwater Forest, are reported using tcore clock type. Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 32 ++++++++++++++++++++++++--- 1 file changed, 29 insertions(+), 3 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 76d2632e60ac..ecaa4e0fb2c0 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -1538,6 +1538,8 @@ static struct msr_counter_arch_info msr_counter_arch_infos[] = { #define PMT_MTL_DC6_GUID 0x1a067102 #define PMT_MTL_DC6_SEQ 0 +unsigned long long tcore_clock_freq_hz = 800000000; + #define PMT_COUNTER_NAME_SIZE_BYTES 16 #define PMT_COUNTER_TYPE_NAME_SIZE_BYTES 32 @@ -1560,6 +1562,7 @@ struct pmt_mmio { enum pmt_datatype { PMT_TYPE_RAW, PMT_TYPE_XTAL_TIME, + PMT_TYPE_TCORE_CLOCK, }; struct pmt_domain_info { @@ -2474,6 +2477,7 @@ void print_header(char *delim) break; case PMT_TYPE_XTAL_TIME: + case PMT_TYPE_TCORE_CLOCK: outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), ppmt->name); break; } @@ -2548,6 +2552,7 @@ void print_header(char *delim) break; case PMT_TYPE_XTAL_TIME: + case PMT_TYPE_TCORE_CLOCK: outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), ppmt->name); break; } @@ -2679,6 +2684,7 @@ void print_header(char *delim) break; case PMT_TYPE_XTAL_TIME: + case PMT_TYPE_TCORE_CLOCK: outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), ppmt->name); break; } @@ -2997,7 +3003,7 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data for (i = 0, ppmt = sys.pmt_tp; ppmt; i++, ppmt = ppmt->next) { const unsigned long value_raw = t->pmt_counter[i]; - const double value_converted = 100.0 * value_raw / crystal_hz / interval_float; + double value_converted; switch (ppmt->type) { case PMT_TYPE_RAW: if (pmt_counter_get_width(ppmt) <= 32) @@ -3009,8 +3015,13 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data break; case PMT_TYPE_XTAL_TIME: + value_converted = 100.0 * value_raw / crystal_hz / interval_float; outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); break; + + case PMT_TYPE_TCORE_CLOCK: + value_converted = 100.0 * value_raw / tcore_clock_freq_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); } } @@ -3077,7 +3088,7 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data for (i = 0, ppmt = sys.pmt_cp; ppmt; i++, ppmt = ppmt->next) { const unsigned long value_raw = c->pmt_counter[i]; - const double value_converted = 100.0 * value_raw / crystal_hz / interval_float; + double value_converted; switch (ppmt->type) { case PMT_TYPE_RAW: if (pmt_counter_get_width(ppmt) <= 32) @@ -3089,8 +3100,13 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data break; case PMT_TYPE_XTAL_TIME: + value_converted = 100.0 * value_raw / crystal_hz / interval_float; outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); break; + + case PMT_TYPE_TCORE_CLOCK: + value_converted = 100.0 * value_raw / tcore_clock_freq_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); } } @@ -3275,7 +3291,7 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data for (i = 0, ppmt = sys.pmt_pp; ppmt; i++, ppmt = ppmt->next) { const unsigned long value_raw = p->pmt_counter[i]; - const double value_converted = 100.0 * value_raw / crystal_hz / interval_float; + double value_converted; switch (ppmt->type) { case PMT_TYPE_RAW: if (pmt_counter_get_width(ppmt) <= 32) @@ -3287,8 +3303,13 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data break; case PMT_TYPE_XTAL_TIME: + value_converted = 100.0 * value_raw / crystal_hz / interval_float; outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); break; + + case PMT_TYPE_TCORE_CLOCK: + value_converted = 100.0 * value_raw / tcore_clock_freq_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); } } @@ -10016,6 +10037,11 @@ next: has_type = true; } + if (strcmp("tcore_clock", type_name) == 0) { + type = PMT_TYPE_TCORE_CLOCK; + has_type = true; + } + if (!has_type) { printf("%s: invalid %s: %s\n", __func__, "type", type_name); exit(1); -- 2.51.0 From b32c36975da48afc9089f8b61f7b2dcc40e479d2 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Mon, 27 Jan 2025 16:42:19 -0600 Subject: [PATCH 03/16] tools/power turbostat: Fix forked child affinity regression In "one-shot" mode, turbostat 1. takes a counter snapshot 2. forks and waits for a child 3. takes the end counter snapshot and prints the result. But turbostat counter snapshots currently use affinity to travel around the system so that counter reads are "local", and this affinity must be cleared between #1 and #2 above. The offending commit removed that reset that allowed the child to run on cpu_present_set. Fix that issue, and improve upon the original by using cpu_possible_set for the child. This allows the child to also run on CPUs that hotplug online during its runtime. Reported-by: Zhang Rui Fixes: 7bb3fe27ad4f ("tools/power/turbostat: Obey allowed CPUs during startup") Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 54 ++++++++++++++++++++++++++- 1 file changed, 52 insertions(+), 2 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index ecaa4e0fb2c0..1f188a0908da 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -1120,8 +1120,8 @@ int backwards_count; char *progname; #define CPU_SUBSET_MAXCPUS 1024 /* need to use before probe... */ -cpu_set_t *cpu_present_set, *cpu_effective_set, *cpu_allowed_set, *cpu_affinity_set, *cpu_subset; -size_t cpu_present_setsize, cpu_effective_setsize, cpu_allowed_setsize, cpu_affinity_setsize, cpu_subset_size; +cpu_set_t *cpu_present_set, *cpu_possible_set, *cpu_effective_set, *cpu_allowed_set, *cpu_affinity_set, *cpu_subset; +size_t cpu_present_setsize, cpu_possible_setsize, cpu_effective_setsize, cpu_allowed_setsize, cpu_affinity_setsize, cpu_subset_size; #define MAX_ADDED_THREAD_COUNTERS 24 #define MAX_ADDED_CORE_COUNTERS 8 #define MAX_ADDED_PACKAGE_COUNTERS 16 @@ -8488,6 +8488,33 @@ int dir_filter(const struct dirent *dirp) return 0; } +char *possible_file = "/sys/devices/system/cpu/possible"; +char possible_buf[1024]; + +int initialize_cpu_possible_set(void) +{ + FILE *fp; + + fp = fopen(possible_file, "r"); + if (!fp) { + warn("open %s", possible_file); + return -1; + } + if (fread(possible_buf, sizeof(char), 1024, fp) == 0) { + warn("read %s", possible_file); + goto err; + } + if (parse_cpu_str(possible_buf, cpu_possible_set, cpu_possible_setsize)) { + warnx("%s: cpu str malformat %s\n", possible_file, cpu_effective_str); + goto err; + } + return 0; + +err: + fclose(fp); + return -1; +} + void topology_probe(bool startup) { int i; @@ -8519,6 +8546,16 @@ void topology_probe(bool startup) CPU_ZERO_S(cpu_present_setsize, cpu_present_set); for_all_proc_cpus(mark_cpu_present); + /* + * Allocate and initialize cpu_possible_set + */ + cpu_possible_set = CPU_ALLOC((topo.max_cpu_num + 1)); + if (cpu_possible_set == NULL) + err(3, "CPU_ALLOC"); + cpu_possible_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); + CPU_ZERO_S(cpu_possible_setsize, cpu_possible_set); + initialize_cpu_possible_set(); + /* * Allocate and initialize cpu_effective_set */ @@ -9371,6 +9408,18 @@ void turbostat_init() } } +void affinitize_child(void) +{ + /* Prefer cpu_possible_set, if available */ + if (sched_setaffinity(0, cpu_possible_setsize, cpu_possible_set)) { + warn("sched_setaffinity cpu_possible_set"); + + /* Otherwise, allow child to run on same cpu set as turbostat */ + if (sched_setaffinity(0, cpu_allowed_setsize, cpu_allowed_set)) + warn("sched_setaffinity cpu_allowed_set"); + } +} + int fork_it(char **argv) { pid_t child_pid; @@ -9386,6 +9435,7 @@ int fork_it(char **argv) child_pid = fork(); if (!child_pid) { /* child */ + affinitize_child(); execvp(argv[0], argv); err(errno, "exec %s", argv[0]); } else { -- 2.51.0 From 5499b5ac0b2c661cc37190a23a4aee9308b3d3ee Mon Sep 17 00:00:00 2001 From: Len Brown Date: Mon, 27 Jan 2025 20:58:42 -0600 Subject: [PATCH 04/16] tools/power turbostat: Harden one-shot mode against cpu offline when turbostat interval mode can't migrate to a CPU, it complains, prints no data, re-initializes with the new CPU configuration and starts a new interval. But this strategy in the face of a CPU hotplug offline during an interval doesn't help in one-shot mode. When the missing CPU is discovered at the end of the interval, the forked program has already returned and there is nothing left for a new interval to measure. So instead of aborting get_coutners() and delta_cpu() if a missing CPU is detected, complain, but carry on and output what statistics are actually present. Use the same strategy for delta_cpu when aperf:mperf are observed to have been reset -- complain, but carry on and print data for the CPUs that are still present. Interval mode error handling is unchanged. One-shot mode can now do this: $ sudo chcpu -e 1 ; sudo ./turbostat --quiet --show PkgWatt,Busy%,CPU chcpu -d 1 CPU 1 enabled CPU 1 disabled get_counters: Could not migrate to CPU 1 ./turbostat: Counter reset detected 0.036920 sec CPU Busy% PkgWatt - 0.00 10.00 0 99.73 10.00 1 0.00 2 91.53 3 16.83 Suggested-by: Zhang Rui Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 27 ++++++++++++--------------- 1 file changed, 12 insertions(+), 15 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 1f188a0908da..8df08819e7b4 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -2063,6 +2063,8 @@ int for_all_cpus(int (func) (struct thread_data *, struct core_data *, struct pk { int retval, pkg_no, core_no, thread_no, node_no; + retval = 0; + for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) { for (core_no = 0; core_no < topo.cores_per_node; ++core_no) { @@ -2078,14 +2080,12 @@ int for_all_cpus(int (func) (struct thread_data *, struct core_data *, struct pk c = GET_CORE(core_base, core_no, node_no, pkg_no); p = GET_PKG(pkg_base, pkg_no); - retval = func(t, c, p); - if (retval) - return retval; + retval |= func(t, c, p); } } } } - return 0; + return retval; } int is_cpu_first_thread_in_core(struct thread_data *t, struct core_data *c, struct pkg_data *p) @@ -3620,12 +3620,10 @@ int delta_cpu(struct thread_data *t, struct core_data *c, /* always calculate thread delta */ retval = delta_thread(t, t2, c2); /* c2 is core delta */ - if (retval) - return retval; /* calculate package delta only for 1st core in package */ if (is_cpu_first_core_in_package(t, c, p)) - retval = delta_package(p, p2); + retval |= delta_package(p, p2); return retval; } @@ -5748,6 +5746,8 @@ int for_all_cpus_2(int (func) (struct thread_data *, struct core_data *, { int retval, pkg_no, node_no, core_no, thread_no; + retval = 0; + for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) { for (core_no = 0; core_no < topo.cores_per_node; ++core_no) { @@ -5769,14 +5769,12 @@ int for_all_cpus_2(int (func) (struct thread_data *, struct core_data *, p = GET_PKG(pkg_base, pkg_no); p2 = GET_PKG(pkg_base2, pkg_no); - retval = func(t, c, p, t2, c2, p2); - if (retval) - return retval; + retval |= func(t, c, p, t2, c2, p2); } } } } - return 0; + return retval; } /* @@ -9462,10 +9460,9 @@ int fork_it(char **argv) timersub(&tv_odd, &tv_even, &tv_delta); if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) fprintf(outf, "%s: Counter reset detected\n", progname); - else { - compute_average(EVEN_COUNTERS); - format_all_counters(EVEN_COUNTERS); - } + + compute_average(EVEN_COUNTERS); + format_all_counters(EVEN_COUNTERS); fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec / 1000000.0); -- 2.51.0 From 5ce1e9bbb2a1d43cf9e613cb03e65ecdfd309fe9 Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Fri, 17 Jan 2025 13:50:29 +0100 Subject: [PATCH 05/16] tools/power turbostat: Add CPU%c1e BIC for CWF Intel Clearwater Forest report PMT telemetry with GUID 0x14421519, which can be used to obtain module c1e residency counter of type tcore clock. Add early support for the counter by using heuristic that should work for the Clearwater Forest platforms. Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 68 +++++++++++++++++++++++++++ 1 file changed, 68 insertions(+) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 8df08819e7b4..364a44a7d7ae 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -205,6 +205,7 @@ struct msr_counter bic[] = { { 0x0, "SysWatt", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "Sys_J", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "NMI", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU%c1e", NULL, 0, 0, 0, NULL, 0 }, }; #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter)) @@ -270,6 +271,7 @@ struct msr_counter bic[] = { #define BIC_SysWatt (1ULL << 59) #define BIC_Sys_J (1ULL << 60) #define BIC_NMI (1ULL << 61) +#define BIC_CPU_c1e (1ULL << 62) #define BIC_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die) #define BIC_THERMAL_PWR (BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__ | BIC_SysWatt) @@ -1538,6 +1540,14 @@ static struct msr_counter_arch_info msr_counter_arch_infos[] = { #define PMT_MTL_DC6_GUID 0x1a067102 #define PMT_MTL_DC6_SEQ 0 +#define PMT_COUNTER_CWF_MC1E_OFFSET_BASE 20936 +#define PMT_COUNTER_CWF_MC1E_OFFSET_INCREMENT 24 +#define PMT_COUNTER_CWF_MC1E_NUM_MODULES_PER_FILE 12 +#define PMT_COUNTER_CWF_CPUS_PER_MODULE 4 +#define PMT_COUNTER_CWF_MC1E_LSB 0 +#define PMT_COUNTER_CWF_MC1E_MSB 63 +#define PMT_CWF_MC1E_GUID 0x14421519 + unsigned long long tcore_clock_freq_hz = 800000000; #define PMT_COUNTER_NAME_SIZE_BYTES 16 @@ -9367,11 +9377,69 @@ int pmt_add_counter(unsigned int guid, unsigned int seq, const char *name, enum void pmt_init(void) { + int cpu_num; + unsigned long seq, offset, mod_num; + if (BIC_IS_ENABLED(BIC_Diec6)) { pmt_add_counter(PMT_MTL_DC6_GUID, PMT_MTL_DC6_SEQ, "Die%c6", PMT_TYPE_XTAL_TIME, PMT_COUNTER_MTL_DC6_LSB, PMT_COUNTER_MTL_DC6_MSB, PMT_COUNTER_MTL_DC6_OFFSET, SCOPE_PACKAGE, FORMAT_DELTA, 0, PMT_OPEN_TRY); } + + if (BIC_IS_ENABLED(BIC_CPU_c1e)) { + seq = 0; + offset = PMT_COUNTER_CWF_MC1E_OFFSET_BASE; + mod_num = 0; /* Relative module number for current PMT file. */ + + /* Open the counter for each CPU. */ + for (cpu_num = 0; cpu_num < topo.max_cpu_num;) { + + if (cpu_is_not_allowed(cpu_num)) + goto next_loop_iter; + + /* + * Set the scope to CPU, even though CWF report the counter per module. + * CPUs inside the same module will read from the same location, instead of reporting zeros. + * + * CWF with newer firmware might require a PMT_TYPE_XTAL_TIME intead of PMT_TYPE_TCORE_CLOCK. + */ + pmt_add_counter(PMT_CWF_MC1E_GUID, seq, "CPU%c1e", PMT_TYPE_TCORE_CLOCK, + PMT_COUNTER_CWF_MC1E_LSB, PMT_COUNTER_CWF_MC1E_MSB, offset, SCOPE_CPU, + FORMAT_DELTA, cpu_num, PMT_OPEN_TRY); + + /* + * Rather complex logic for each time we go to the next loop iteration, + * so keep it as a label. + */ +next_loop_iter: + /* + * Advance the cpu number and check if we should also advance offset to + * the next counter inside the PMT file. + * + * On Clearwater Forest platform, the counter is reported per module, + * so open the same counter for all of the CPUs inside the module. + * That way, reported table show the correct value for all of the CPUs inside the module, + * instead of zeros. + */ + ++cpu_num; + if (cpu_num % PMT_COUNTER_CWF_CPUS_PER_MODULE == 0) { + offset += PMT_COUNTER_CWF_MC1E_OFFSET_INCREMENT; + ++mod_num; + } + + /* + * There are PMT_COUNTER_CWF_MC1E_NUM_MODULES_PER_FILE in each PMT file. + * + * If that number is reached, seq must be incremented to advance to the next file in a sequence. + * Offset inside that file and a module counter has to be reset. + */ + if (mod_num == PMT_COUNTER_CWF_MC1E_NUM_MODULES_PER_FILE) { + ++seq; + offset = PMT_COUNTER_CWF_MC1E_OFFSET_BASE; + mod_num = 0; + } + } + } } void turbostat_init() -- 2.51.0 From 2c4627c8ced77855b106c7104ecab70837d53799 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Sun, 2 Feb 2025 10:43:02 -0600 Subject: [PATCH 06/16] tools/power turbostat: version 2025.02.02 Summary of Changes since 2024.11.30: Fix regression in 2023.11.07 that affinitized forked child in one-shot mode. Harden one-shot mode against hotplug online/offline Enable RAPL SysWatt column by default. Add initial PTL, CWF platform support. Harden initial PMT code in response to early use. Enable first built-in PMT counter: CWF c1e residency Refuse to run on unsupported platforms without --force, to encourage updating to a version that supports the system, and to avoid no-so-useful measurement results. Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 364a44a7d7ae..8d5011a0bf60 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -9559,7 +9559,7 @@ int get_and_dump_counters(void) void print_version() { - fprintf(outf, "turbostat version 2025.01.14 - Len Brown \n"); + fprintf(outf, "turbostat version 2025.02.02 - Len Brown \n"); } #define COMMAND_LINE_SIZE 2048 -- 2.51.0 From 1c7c7388e6c31f46b26a884d80b45efbad8237b2 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Thu, 6 Feb 2025 21:46:24 -0600 Subject: [PATCH 07/16] tools/power turbostat: Clustered Uncore MHz counters should honor show/hide options MIME-Version: 1.0 Content-Type: text/plain; charset=utf8 Content-Transfer-Encoding: 8bit The clustered uncore frequency counters, UMHz*.* should honor the --show and --hide options. All non-specified counters should be implicityly hidden. But when --show was used, UMHz*.* showed up anyway: $ sudo turbostat -q -S --show Busy% Busy%  UMHz0.0  UMHz1.0  UMHz2.0  UMHz3.0  UMHz4.0 Indeed, there was no string that can be used to explicitly show or hide clustered uncore counters. Even through they are dynamically probed and added, group the clustered UMHz*.* counters with the legacy built-in-counter "UncMHz" for show/hide. turbostat --show Busy% does not show UMHz*.*. turbostat --show UncMHz shows either UncMHz or UMHz*.*, if present turbostat --hide UncMHz hides either UncMHz or UMHz*.*, if present Reported-by: Artem Bityutskiy Signed-off-by: Len Brown Tested-by: Artem Bityutskiy --- tools/power/x86/turbostat/turbostat.8 | 1 + tools/power/x86/turbostat/turbostat.c | 13 ++++++++++++- 2 files changed, 13 insertions(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 99bf905ade81..ed258f248152 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -199,6 +199,7 @@ The system configuration dump (if --quiet is not used) is followed by statistics \fBUncMHz\fP per-package uncore MHz, instantaneous sample. .PP \fBUMHz1.0\fP per-package uncore MHz for domain=1 and fabric_cluster=0, instantaneous sample. System summary is the average of all packages. +For the "--show" and "--hide" options, use "UncMHz" to operate on all UMHz*.* as a group. .SH TOO MUCH INFORMATION EXAMPLE By default, turbostat dumps all possible information -- a system configuration header, followed by columns for all counters. This is ideal for remote debugging, use the "--out" option to save everything to a text file, and get that file to the expert helping you debug. diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 8d5011a0bf60..5eef95956c2f 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -6703,7 +6703,18 @@ static void probe_intel_uncore_frequency_cluster(void) sprintf(path, "%s/current_freq_khz", path_base); sprintf(name_buf, "UMHz%d.%d", domain_id, cluster_id); - add_counter(0, path, name_buf, 0, SCOPE_PACKAGE, COUNTER_K2M, FORMAT_AVERAGE, 0, package_id); + /* + * Once add_couter() is called, that counter is always read + * and reported -- So it is effectively (enabled & present). + * Only call add_counter() here if legacy BIC_UNCORE_MHZ (UncMHz) + * is (enabled). Since we are in this routine, we + * know we will not probe and set (present) the legacy counter. + * + * This allows "--show/--hide UncMHz" to be effective for + * the clustered MHz counters, as a group. + */ + if BIC_IS_ENABLED(BIC_UNCORE_MHZ) + add_counter(0, path, name_buf, 0, SCOPE_PACKAGE, COUNTER_K2M, FORMAT_AVERAGE, 0, package_id); if (quiet) continue; -- 2.51.0 From b312d880fb462d4759396950865ec914de9d253c Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Sat, 8 Feb 2025 10:44:26 +0800 Subject: [PATCH 08/16] tools/power turbostat: Allow Zero return value for some RAPL registers turbostat aborted with below messages on a dual-package system, turbostat: turbostat.c:3744: rapl_counter_accumulate: Assertion `dst->unit == src->unit' failed. Aborted This is because 1. the MSR_DRAM_PERF_STATUS returns Zero for one package, and non-Zero for another package 2. probe_msr() treats Zero return value as a failure so this feature is enabled on one package, and disabled for another package. 3. turbostat aborts because the feature is invalid on some package Unlike the RAPL energy counter registers, MSR_DRAM_PERF_STATUS can return Zero value, and this should not be treated as a failure. Fix the problem by allowing Zero return value for RAPL registers other than the energy counters. Fixes: 7c6fee25bdf5 ("tools/power turbostat: Check for non-zero value when MSR probing") Reported-by: Artem Bityutskiy Signed-off-by: Zhang Rui Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 27 ++++++++++++++++++--------- 1 file changed, 18 insertions(+), 9 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 5eef95956c2f..d26008f37a2c 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -2211,7 +2211,7 @@ int get_msr(int cpu, off_t offset, unsigned long long *msr) return 0; } -int probe_msr(int cpu, off_t offset) +int probe_rapl_msr(int cpu, off_t offset, int index) { ssize_t retval; unsigned long long value; @@ -2220,13 +2220,22 @@ int probe_msr(int cpu, off_t offset) retval = pread(get_msr_fd(cpu), &value, sizeof(value), offset); - /* - * Expect MSRs to accumulate some non-zero value since the system was powered on. - * Treat zero as a read failure. - */ - if (retval != sizeof(value) || value == 0) + /* if the read failed, the probe fails */ + if (retval != sizeof(value)) return 1; + /* If an Energy Status Counter MSR returns 0, the probe fails */ + switch (index) { + case RAPL_RCI_INDEX_ENERGY_PKG: + case RAPL_RCI_INDEX_ENERGY_CORES: + case RAPL_RCI_INDEX_DRAM: + case RAPL_RCI_INDEX_GFX: + case RAPL_RCI_INDEX_ENERGY_PLATFORM: + if (value == 0) + return 1; + } + + /* PKG,DRAM_PERF_STATUS MSRs, can return any value */ return 0; } @@ -7907,7 +7916,7 @@ void rapl_perf_init(void) rci->flags[cai->rci_index] = cai->flags; /* Use MSR for this counter */ - } else if (!no_msr && cai->msr && probe_msr(cpu, cai->msr) == 0) { + } else if (!no_msr && cai->msr && probe_rapl_msr(cpu, cai->msr, cai->rci_index) == 0) { rci->source[cai->rci_index] = COUNTER_SOURCE_MSR; rci->msr[cai->rci_index] = cai->msr; rci->msr_mask[cai->rci_index] = cai->msr_mask; @@ -8045,7 +8054,7 @@ void msr_perf_init_(void) cai->present = true; /* User MSR for this counter */ - } else if (!no_msr && cai->msr && probe_msr(cpu, cai->msr) == 0) { + } else if (!no_msr && cai->msr && probe_rapl_msr(cpu, cai->msr, cai->rci_index) == 0) { cci->source[cai->rci_index] = COUNTER_SOURCE_MSR; cci->msr[cai->rci_index] = cai->msr; cci->msr_mask[cai->rci_index] = cai->msr_mask; @@ -8159,7 +8168,7 @@ void cstate_perf_init_(bool soft_c1) /* User MSR for this counter */ } else if (!no_msr && cai->msr && pkg_cstate_limit >= cai->pkg_cstate_limit - && probe_msr(cpu, cai->msr) == 0) { + && probe_rapl_msr(cpu, cai->msr, cai->rci_index) == 0) { cci->source[cai->rci_index] = COUNTER_SOURCE_MSR; cci->msr[cai->rci_index] = cai->msr; } -- 2.51.0 From 5132681dcd96b2a8c357b6e5d93e9876924bb80b Mon Sep 17 00:00:00 2001 From: Artem Bityutskiy Date: Sat, 8 Feb 2025 12:55:42 +0200 Subject: [PATCH 09/16] tools/power turbostat: Fix names matching Fix the 'find_msrp_by_name()' function which returns incorrect matches for cases like this: s1 = "C1-"; find_msrp_by_name(head, s1); Inside 'find_msrp_by_name()': ... s2 = "C1" if !(strcnmp(s1, s2, len(s2))) // Incorrect match! return mp; Full strings should be match istead. Switch to 'strcmp()' to fix the problem. Signed-off-by: Artem Bityutskiy Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index d26008f37a2c..d3af2bf307e1 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -9612,7 +9612,7 @@ struct msr_counter *find_msrp_by_name(struct msr_counter *head, char *name) for (mp = head; mp; mp = mp->next) { if (debug) fprintf(stderr, "%s: %s %s\n", __func__, name, mp->name); - if (!strncmp(name, mp->name, strlen(mp->name))) + if (!strcmp(name, mp->name)) return mp; } return NULL; -- 2.51.0 From ed625c61b85c2333324dca43146c4ebabf8b236f Mon Sep 17 00:00:00 2001 From: Artem Bityutskiy Date: Sat, 8 Feb 2025 13:53:19 +0200 Subject: [PATCH 10/16] tools/power turbostat: Add idle governor statistics reporting The idle governor provides the following per-idle state sysfs files: * above - Indicates overshoots, where a more shallow state should have been requested (if avaliale and enabled). * below - Indicates undershoots, where a deeper state should have been requested (if available and enabled). These files offer valuable insights into how effectively the Linux kernel idle governor selects idle states for a given workload. This commit adds support for these files in turbostat. Expose the contents of these files with the following naming convention: * C1: The number of times the C1 state was requested (existing counter). * C1+: The number of times the idle governor selected C1, but a deeper idle state should have been selected instead. * C1-: The number of times the idle governor selected C1, but a shallower idle state should have been selected instead. Signed-off-by: Artem Bityutskiy Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.8 | 4 +++ tools/power/x86/turbostat/turbostat.c | 48 ++++++++++++++++++++++----- 2 files changed, 44 insertions(+), 8 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index ed258f248152..52d727e29ea7 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -160,6 +160,10 @@ The system configuration dump (if --quiet is not used) is followed by statistics .PP \fBC1, C2, C3...\fP The number times Linux requested the C1, C2, C3 idle state during the measurement interval. The system summary line shows the sum for all CPUs. These are C-state names as exported in /sys/devices/system/cpu/cpu*/cpuidle/state*/name. While their names are generic, their attributes are processor specific. They the system description section of output shows what MWAIT sub-states they are mapped to on each system. .PP +\fBC1+, C2+, C3+...\fP The idle governor idle state misprediction statistics. Inidcates the number times Linux requested the C1, C2, C3 idle state during the measurement interval, but should have requested a deeper idle state (if it exists and enabled). These statistics come from the /sys/devices/system/cpu/cpu*/cpuidle/state*/below file. +.PP +\fBC1-, C2-, C3-...\fP The idle governor idle state misprediction statistics. Inidcates the number times Linux requested the C1, C2, C3 idle state during the measurement interval, but should have requested a shallower idle state (if it exists and enabled). These statistics come from the /sys/devices/system/cpu/cpu*/cpuidle/state*/above file. +.PP \fBC1%, C2%, C3%\fP The residency percentage that Linux requested C1, C2, C3.... The system summary is the average of all CPUs in the system. Note that these are software, reflecting what was requested. The hardware counters reflect what was actually achieved. .PP \fBCPU%c1, CPU%c3, CPU%c6, CPU%c7\fP show the percentage residency in hardware core idle states. These numbers are from hardware residency counters. diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index d3af2bf307e1..f29e47fe4249 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -10265,6 +10265,7 @@ void probe_sysfs(void) char name_buf[16]; FILE *input; int state; + int min_state = 1024, max_state = 0; char *sp; for (state = 10; state >= 0; --state) { @@ -10296,6 +10297,11 @@ void probe_sysfs(void) continue; add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC, FORMAT_PERCENT, SYSFS_PERCPU, 0); + + if (state > max_state) + max_state = state; + if (state < min_state) + min_state = state; } for (state = 10; state >= 0; --state) { @@ -10306,26 +10312,52 @@ void probe_sysfs(void) continue; if (!fgets(name_buf, sizeof(name_buf), input)) err(1, "%s: failed to read file", path); - /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */ - sp = strchr(name_buf, '-'); - if (!sp) - sp = strchrnul(name_buf, '\n'); - *sp = '\0'; fclose(input); remove_underbar(name_buf); - sprintf(path, "cpuidle/state%d/usage", state); - if (!DO_BIC(BIC_sysfs) && !is_deferred_add(name_buf)) continue; if (is_deferred_skip(name_buf)) continue; + /* truncate "C1-HSW\n" to "C1", or truncate "C1\n" to "C1" */ + sp = strchr(name_buf, '-'); + if (!sp) + sp = strchrnul(name_buf, '\n'); + + /* + * The 'below' sysfs file always contains 0 for the deepest state (largest index), + * do not add it. + */ + if (state != max_state) { + /* + * Add 'C1+' for C1, and so on. The 'below' sysfs file always contains 0 for + * the last state, so do not add it. + */ + + *sp = '+'; + *(sp + 1) = '\0'; + sprintf(path, "cpuidle/state%d/below", state); + add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS, FORMAT_DELTA, SYSFS_PERCPU, 0); + } + + *sp = '\0'; + sprintf(path, "cpuidle/state%d/usage", state); add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS, FORMAT_DELTA, SYSFS_PERCPU, 0); - } + /* + * The 'above' sysfs file always contains 0 for the shallowest state (smallest + * index), do not add it. + */ + if (state != min_state) { + *sp = '-'; + *(sp + 1) = '\0'; + sprintf(path, "cpuidle/state%d/above", state); + add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS, FORMAT_DELTA, SYSFS_PERCPU, 0); + } + } } /* -- 2.51.0 From eb187540d13ae260b91dbca7257bc44bd83ca8c6 Mon Sep 17 00:00:00 2001 From: Justin Ernst Date: Wed, 19 Mar 2025 15:27:31 -0500 Subject: [PATCH 11/16] tools/power turbostat: Increase CPU_SUBSET_MAXCPUS to 8192 On systems with >= 1024 cpus (in my case 1152), turbostat fails with the error output: "turbostat: /sys/fs/cgroup/cpuset.cpus.effective: cpu str malformat 0-1151" A similar error appears with the use of turbostat --cpu when the inputted cpu range contains a cpu number >= 1024: # turbostat -c 1100-1151 "--cpu 1100-1151" malformed ... Both errors are caused by parse_cpu_str() reaching its limit of CPU_SUBSET_MAXCPUS. It's a good idea to limit the maximum cpu number being parsed, but 1024 is too low. For a small increase in compute and allocated memory, increasing CPU_SUBSET_MAXCPUS brings support for parsing cpu numbers >= 1024. Increase CPU_SUBSET_MAXCPUS to 8192, a common setting for CONFIG_NR_CPUS on x86_64. Signed-off-by: Justin Ernst Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index f29e47fe4249..218aca958923 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -1121,7 +1121,7 @@ end: int backwards_count; char *progname; -#define CPU_SUBSET_MAXCPUS 1024 /* need to use before probe... */ +#define CPU_SUBSET_MAXCPUS 8192 /* need to use before probe... */ cpu_set_t *cpu_present_set, *cpu_possible_set, *cpu_effective_set, *cpu_allowed_set, *cpu_affinity_set, *cpu_subset; size_t cpu_present_setsize, cpu_possible_setsize, cpu_effective_setsize, cpu_allowed_setsize, cpu_affinity_setsize, cpu_subset_size; #define MAX_ADDED_THREAD_COUNTERS 24 -- 2.51.0 From f729775f79a9c942c6c82ed6b44bd030afe10423 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Sun, 6 Apr 2025 11:18:39 -0400 Subject: [PATCH 12/16] tools/power turbostat: report CoreThr per measurement interval The CoreThr column displays total thermal throttling events since boot time. Change it to report events during the measurement interval. This is more useful for showing a user the current conditions. Total events since boot time are still available to the user via /sys/devices/system/cpu/cpu*/thermal_throttle/* Document CoreThr on turbostat.8 Fixes: eae97e053fe30 ("turbostat: Support thermal throttle count print") Reported-by: Arjan van de Ven Signed-off-by: Len Brown Cc: Chen Yu --- tools/power/x86/turbostat/turbostat.8 | 2 ++ tools/power/x86/turbostat/turbostat.c | 2 +- 2 files changed, 3 insertions(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 52d727e29ea7..144565151e1e 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -172,6 +172,8 @@ The system configuration dump (if --quiet is not used) is followed by statistics .PP \fBPkgTmp\fP Degrees Celsius reported by the per-package Package Thermal Monitor. .PP +\fBCoreThr\fP Core Thermal Throttling events during the measurement interval. Note that events since boot can be find in /sys/devices/system/cpu/cpu*/thermal_throttle/* +.PP \fBGFX%rc6\fP The percentage of time the GPU is in the "render C6" state, rc6, during the measurement interval. From /sys/class/drm/card0/power/rc6_residency_ms or /sys/class/drm/card0/gt/gt0/rc6_residency_ms or /sys/class/drm/card0/device/tile0/gtN/gtidle/idle_residency_ms depending on the graphics driver being used. .PP \fBGFXMHz\fP Instantaneous snapshot of what sysfs presents at the end of the measurement interval. From /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz or /sys/class/drm/card0/gt_cur_freq_mhz or /sys/class/drm/card0/gt/gt0/rps_cur_freq_mhz or /sys/class/drm/card0/device/tile0/gtN/freq0/cur_freq depending on the graphics driver being used. diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 218aca958923..70e17d4ad9b6 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -3485,7 +3485,7 @@ void delta_core(struct core_data *new, struct core_data *old) old->c6 = new->c6 - old->c6; old->c7 = new->c7 - old->c7; old->core_temp_c = new->core_temp_c; - old->core_throt_cnt = new->core_throt_cnt; + old->core_throt_cnt = new->core_throt_cnt - old->core_throt_cnt; old->mc6_us = new->mc6_us - old->mc6_us; DELTA_WRAP32(new->core_energy.raw_value, old->core_energy.raw_value); -- 2.51.0 From 3ae8508663372b93c5556a887e96ed0ca5df0711 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Sun, 6 Apr 2025 12:23:22 -0400 Subject: [PATCH 13/16] tools/power turbostat: Document GNR UncMHz domain convention Document that on Intel Granite Rapids Systems, Uncore domains 0-2 are CPU domains, and uncore domains 3-4 are IO domains. Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.8 | 1 + 1 file changed, 1 insertion(+) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 144565151e1e..e86493880c16 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -205,6 +205,7 @@ The system configuration dump (if --quiet is not used) is followed by statistics \fBUncMHz\fP per-package uncore MHz, instantaneous sample. .PP \fBUMHz1.0\fP per-package uncore MHz for domain=1 and fabric_cluster=0, instantaneous sample. System summary is the average of all packages. +Intel Granite Rapids systems use domains 0-2 for CPUs, and 3-4 for IO, with cluster always 0. For the "--show" and "--hide" options, use "UncMHz" to operate on all UMHz*.* as a group. .SH TOO MUCH INFORMATION EXAMPLE By default, turbostat dumps all possible information -- a system configuration header, followed by columns for all counters. -- 2.51.0 From f8b136ef2605c1bf62020462d10e35228760aa19 Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Wed, 19 Mar 2025 08:53:07 +0800 Subject: [PATCH 14/16] tools/power turbostat: Restore GFX sysfs fflush() call Do fflush() to discard the buffered data, before each read of the graphics sysfs knobs. Fixes: ba99a4fc8c24 ("tools/power turbostat: Remove unnecessary fflush() call") Signed-off-by: Zhang Rui Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 1 + 1 file changed, 1 insertion(+) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 70e17d4ad9b6..c9a34c16c7a8 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -6039,6 +6039,7 @@ int snapshot_graphics(int idx) int retval; rewind(gfx_info[idx].fp); + fflush(gfx_info[idx].fp); switch (idx) { case GFX_rc6: -- 2.51.0 From 994633894f208a0151baaee1688ab3c431912553 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Sun, 6 Apr 2025 12:53:18 -0400 Subject: [PATCH 15/16] tools/power turbostat: re-factor sysfs code Probe cpuidle "sysfs" residency and counts separately, since soon we will make one disabled on, and the other disabled off. Clarify that some BIC (build-in-counters) are actually "groups". since we're about to re-name some of those groups. no functional change. Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 31 ++++++++++++++++++--------- 1 file changed, 21 insertions(+), 10 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index c9a34c16c7a8..df0391bedcde 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -273,10 +273,10 @@ struct msr_counter bic[] = { #define BIC_NMI (1ULL << 61) #define BIC_CPU_c1e (1ULL << 62) -#define BIC_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die) -#define BIC_THERMAL_PWR (BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__ | BIC_SysWatt) -#define BIC_FREQUENCY (BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_SAMMHz | BIC_SAMACTMHz | BIC_UNCORE_MHZ) -#define BIC_IDLE (BIC_Busy | BIC_sysfs | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_SAM_mc6 | BIC_Diec6) +#define BIC_GROUP_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die) +#define BIC_GROUP_THERMAL_PWR (BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__ | BIC_SysWatt) +#define BIC_GROUP_FREQUENCY (BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_SAMMHz | BIC_SAMACTMHz | BIC_UNCORE_MHZ) +#define BIC_GROUP_IDLE (BIC_Busy | BIC_sysfs | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_SAM_mc6 | BIC_Diec6) #define BIC_OTHER (BIC_IRQ | BIC_NMI | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) #define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC) @@ -2354,16 +2354,16 @@ unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode) retval |= ~0; break; } else if (!strcmp(name_list, "topology")) { - retval |= BIC_TOPOLOGY; + retval |= BIC_GROUP_TOPOLOGY; break; } else if (!strcmp(name_list, "power")) { - retval |= BIC_THERMAL_PWR; + retval |= BIC_GROUP_THERMAL_PWR; break; } else if (!strcmp(name_list, "idle")) { - retval |= BIC_IDLE; + retval |= BIC_GROUP_IDLE; break; } else if (!strcmp(name_list, "frequency")) { - retval |= BIC_FREQUENCY; + retval |= BIC_GROUP_FREQUENCY; break; } else if (!strcmp(name_list, "other")) { retval |= BIC_OTHER; @@ -10260,7 +10260,7 @@ int is_deferred_skip(char *name) return 0; } -void probe_sysfs(void) +void probe_cpuidle_residency(void) { char path[64]; char name_buf[16]; @@ -10304,6 +10304,16 @@ void probe_sysfs(void) if (state < min_state) min_state = state; } +} + +void probe_cpuidle_counts(void) +{ + char path[64]; + char name_buf[16]; + FILE *input; + int state; + int min_state = 1024, max_state = 0; + char *sp; for (state = 10; state >= 0; --state) { @@ -10602,7 +10612,8 @@ skip_cgroup_setting: print_bootcmd(); } - probe_sysfs(); + probe_cpuidle_residency(); + probe_cpuidle_counts(); if (!getuid()) set_rlimit(); -- 2.51.0 From ec4acd3166d8a7a03b059d01b9c6f11a658e833f Mon Sep 17 00:00:00 2001 From: Len Brown Date: Sun, 6 Apr 2025 14:29:57 -0400 Subject: [PATCH 16/16] tools/power turbostat: disable "cpuidle" invocation counters, by default Create "pct_idle" counter group, the sofware notion of residency so it can now be singled out, independent of other counter groups. Create "cpuidle" group, the cpuidle invocation counts. Disable "cpuidle", by default. Create "swidle" = "cpuidle" + "pct_idle". Undocument "sysfs", the old name for "swidle", but keep it working for backwards compatibilty. Create "hwidle", all the HW idle counters Modify "idle", enabled by default "idle" = "hwidle" + "pct_idle" (and now excludes "cpuidle") Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.8 | 12 +++++----- tools/power/x86/turbostat/turbostat.c | 34 +++++++++++++++++++++------ 2 files changed, 33 insertions(+), 13 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index e86493880c16..b74ed916057e 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -100,7 +100,7 @@ The column name "all" can be used to enable all disabled-by-default built-in cou .PP \fB--show column\fP show only the specified built-in columns. May be invoked multiple times, or with a comma-separated list of column names. .PP -\fB--show CATEGORY --hide CATEGORY\fP Show and hide also accept a single CATEGORY of columns: "all", "topology", "idle", "frequency", "power", "sysfs", "other". +\fB--show CATEGORY --hide CATEGORY\fP Show and hide also accept a single CATEGORY of columns: "all", "topology", "idle", "frequency", "power", "cpuidle", "hwidle", "swidle", "other". "idle" (enabled by default), includes "hwidle" and "idle_pct". "cpuidle" (default disabled) includes cpuidle software invocation counters. "swidle" includes "cpuidle" plus "idle_pct". "hwidle" includes only hardware based idle residency counters. Older versions of turbostat used the term "sysfs" for what is now "swidle". .PP \fB--Dump\fP displays the raw counter values. .PP @@ -158,15 +158,15 @@ The system configuration dump (if --quiet is not used) is followed by statistics .PP \fBSMI\fP The number of System Management Interrupts serviced CPU during the measurement interval. While this counter is actually per-CPU, SMI are triggered on all processors, so the number should be the same for all CPUs. .PP -\fBC1, C2, C3...\fP The number times Linux requested the C1, C2, C3 idle state during the measurement interval. The system summary line shows the sum for all CPUs. These are C-state names as exported in /sys/devices/system/cpu/cpu*/cpuidle/state*/name. While their names are generic, their attributes are processor specific. They the system description section of output shows what MWAIT sub-states they are mapped to on each system. +\fBC1, C2, C3...\fP The number times Linux requested the C1, C2, C3 idle state during the measurement interval. The system summary line shows the sum for all CPUs. These are C-state names as exported in /sys/devices/system/cpu/cpu*/cpuidle/state*/name. While their names are generic, their attributes are processor specific. They the system description section of output shows what MWAIT sub-states they are mapped to on each system. These counters are in the "cpuidle" group, which is disabled, by default. .PP -\fBC1+, C2+, C3+...\fP The idle governor idle state misprediction statistics. Inidcates the number times Linux requested the C1, C2, C3 idle state during the measurement interval, but should have requested a deeper idle state (if it exists and enabled). These statistics come from the /sys/devices/system/cpu/cpu*/cpuidle/state*/below file. +\fBC1+, C2+, C3+...\fP The idle governor idle state misprediction statistics. Inidcates the number times Linux requested the C1, C2, C3 idle state during the measurement interval, but should have requested a deeper idle state (if it exists and enabled). These statistics come from the /sys/devices/system/cpu/cpu*/cpuidle/state*/below file. These counters are in the "cpuidle" group, which is disabled, by default. .PP -\fBC1-, C2-, C3-...\fP The idle governor idle state misprediction statistics. Inidcates the number times Linux requested the C1, C2, C3 idle state during the measurement interval, but should have requested a shallower idle state (if it exists and enabled). These statistics come from the /sys/devices/system/cpu/cpu*/cpuidle/state*/above file. +\fBC1-, C2-, C3-...\fP The idle governor idle state misprediction statistics. Inidcates the number times Linux requested the C1, C2, C3 idle state during the measurement interval, but should have requested a shallower idle state (if it exists and enabled). These statistics come from the /sys/devices/system/cpu/cpu*/cpuidle/state*/above file. These counters are in the "cpuidle" group, which is disabled, by default. .PP -\fBC1%, C2%, C3%\fP The residency percentage that Linux requested C1, C2, C3.... The system summary is the average of all CPUs in the system. Note that these are software, reflecting what was requested. The hardware counters reflect what was actually achieved. +\fBC1%, C2%, C3%\fP The residency percentage that Linux requested C1, C2, C3.... The system summary is the average of all CPUs in the system. Note that these are software, reflecting what was requested. The hardware counters reflect what was actually achieved. These counters are in the "pct_idle" group, which is enabled by default. .PP -\fBCPU%c1, CPU%c3, CPU%c6, CPU%c7\fP show the percentage residency in hardware core idle states. These numbers are from hardware residency counters. +\fBCPU%c1, CPU%c3, CPU%c6, CPU%c7\fP show the percentage residency in hardware core idle states. These numbers are from hardware residency counters and are in the "hwidle" group, which is enabled, by default. .PP \fBCoreTmp\fP Degrees Celsius reported by the per-core Digital Thermal Sensor. .PP diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index df0391bedcde..ab184f95cdaf 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -153,7 +153,7 @@ struct msr_counter bic[] = { { 0x0, "TSC_MHz", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "IRQ", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "SMI", NULL, 32, 0, FORMAT_DELTA, NULL, 0 }, - { 0x0, "sysfs", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "cpuidle", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "CPU%c1", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "CPU%c3", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "CPU%c6", NULL, 0, 0, 0, NULL, 0 }, @@ -206,6 +206,7 @@ struct msr_counter bic[] = { { 0x0, "Sys_J", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "NMI", NULL, 0, 0, 0, NULL, 0 }, { 0x0, "CPU%c1e", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "pct_idle", NULL, 0, 0, 0, NULL, 0 }, }; #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter)) @@ -219,7 +220,7 @@ struct msr_counter bic[] = { #define BIC_TSC_MHz (1ULL << 7) #define BIC_IRQ (1ULL << 8) #define BIC_SMI (1ULL << 9) -#define BIC_sysfs (1ULL << 10) +#define BIC_cpuidle (1ULL << 10) #define BIC_CPU_c1 (1ULL << 11) #define BIC_CPU_c3 (1ULL << 12) #define BIC_CPU_c6 (1ULL << 13) @@ -272,17 +273,20 @@ struct msr_counter bic[] = { #define BIC_Sys_J (1ULL << 60) #define BIC_NMI (1ULL << 61) #define BIC_CPU_c1e (1ULL << 62) +#define BIC_pct_idle (1ULL << 63) #define BIC_GROUP_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die) #define BIC_GROUP_THERMAL_PWR (BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__ | BIC_SysWatt) #define BIC_GROUP_FREQUENCY (BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_SAMMHz | BIC_SAMACTMHz | BIC_UNCORE_MHZ) -#define BIC_GROUP_IDLE (BIC_Busy | BIC_sysfs | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_SAM_mc6 | BIC_Diec6) +#define BIC_GROUP_HW_IDLE (BIC_Busy | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_SAM_mc6 | BIC_Diec6) +#define BIC_GROUP_SW_IDLE (BIC_Busy | BIC_cpuidle | BIC_pct_idle ) +#define BIC_GROUP_IDLE (BIC_GROUP_HW_IDLE | BIC_pct_idle) #define BIC_OTHER (BIC_IRQ | BIC_NMI | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) -#define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC) +#define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC | BIC_cpuidle) unsigned long long bic_enabled = (0xFFFFFFFFFFFFFFFFULL & ~BIC_DISABLED_BY_DEFAULT); -unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC_X2APIC; +unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_cpuidle | BIC_pct_idle | BIC_APIC | BIC_X2APIC; #define DO_BIC(COUNTER_NAME) (bic_enabled & bic_present & COUNTER_NAME) #define DO_BIC_READ(COUNTER_NAME) (bic_present & COUNTER_NAME) @@ -2362,6 +2366,15 @@ unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode) } else if (!strcmp(name_list, "idle")) { retval |= BIC_GROUP_IDLE; break; + } else if (!strcmp(name_list, "swidle")) { + retval |= BIC_GROUP_SW_IDLE; + break; + } else if (!strcmp(name_list, "sysfs")) { /* legacy compatibility */ + retval |= BIC_GROUP_SW_IDLE; + break; + } else if (!strcmp(name_list, "hwidle")) { + retval |= BIC_GROUP_HW_IDLE; + break; } else if (!strcmp(name_list, "frequency")) { retval |= BIC_GROUP_FREQUENCY; break; @@ -2372,6 +2385,7 @@ unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode) } if (i == MAX_BIC) { + fprintf(stderr, "deferred %s\n", name_list); if (mode == SHOW_LIST) { deferred_add_names[deferred_add_index++] = name_list; if (deferred_add_index >= MAX_DEFERRED) { @@ -10269,6 +10283,9 @@ void probe_cpuidle_residency(void) int min_state = 1024, max_state = 0; char *sp; + if (!DO_BIC(BIC_pct_idle)) + return; + for (state = 10; state >= 0; --state) { sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state); @@ -10291,7 +10308,7 @@ void probe_cpuidle_residency(void) sprintf(path, "cpuidle/state%d/time", state); - if (!DO_BIC(BIC_sysfs) && !is_deferred_add(name_buf)) + if (!DO_BIC(BIC_pct_idle) && !is_deferred_add(name_buf)) continue; if (is_deferred_skip(name_buf)) @@ -10315,6 +10332,9 @@ void probe_cpuidle_counts(void) int min_state = 1024, max_state = 0; char *sp; + if (!DO_BIC(BIC_cpuidle)) + return; + for (state = 10; state >= 0; --state) { sprintf(path, "/sys/devices/system/cpu/cpu%d/cpuidle/state%d/name", base_cpu, state); @@ -10327,7 +10347,7 @@ void probe_cpuidle_counts(void) remove_underbar(name_buf); - if (!DO_BIC(BIC_sysfs) && !is_deferred_add(name_buf)) + if (!DO_BIC(BIC_cpuidle) && !is_deferred_add(name_buf)) continue; if (is_deferred_skip(name_buf)) -- 2.51.0