From e5f4e68eed85fa8495d78cd966eecc2b27bb9e53 Mon Sep 17 00:00:00 2001 From: Doug Smythies Date: Mon, 3 Apr 2023 14:11:38 -0700 Subject: [PATCH 01/26] tools/power turbostat: Fix added raw MSR output When using --Summary mode, added MSRs in raw mode always print zeros. Print the actual register contents. Example, with patch: note the added column: --add msr0x64f,u32,package,raw,REASON Where: 0x64F is MSR_CORE_PERF_LIMIT_REASONS Busy% Bzy_MHz PkgTmp PkgWatt CorWatt REASON 0.00 4800 35 1.42 0.76 0x00000000 0.00 4801 34 1.42 0.76 0x00000000 80.08 4531 66 108.17 107.52 0x08000000 98.69 4530 66 133.21 132.54 0x08000000 99.28 4505 66 128.26 127.60 0x0c000400 99.65 4486 68 124.91 124.25 0x0c000400 99.63 4483 68 124.90 124.25 0x0c000400 79.34 4481 41 99.80 99.13 0x0c000000 0.00 4801 41 1.40 0.73 0x0c000000 Where, for the test processor (i5-10600K): PKG Limit #1: 125.000 Watts, 8.000000 sec MSR bit 26 = log; bit 10 = status PKG Limit #2: 136.000 Watts, 0.002441 sec MSR bit 27 = log; bit 11 = status Example, without patch: Busy% Bzy_MHz PkgTmp PkgWatt CorWatt REASON 0.01 4800 35 1.43 0.77 0x00000000 0.00 4801 35 1.39 0.73 0x00000000 83.49 4531 66 112.71 112.06 0x00000000 98.69 4530 68 133.35 132.69 0x00000000 99.31 4500 67 127.96 127.30 0x00000000 99.63 4483 69 124.91 124.25 0x00000000 99.61 4481 69 124.90 124.25 0x00000000 99.61 4481 71 124.92 124.25 0x00000000 59.35 4479 42 75.03 74.37 0x00000000 0.00 4800 42 1.39 0.73 0x00000000 0.00 4801 42 1.42 0.76 0x00000000 c000000 [lenb: simplified patch to apply only to package scope] Signed-off-by: Doug Smythies Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 7 ++++--- 1 file changed, 4 insertions(+), 3 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 7a334377f92b..fca7913f6c84 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -2444,9 +2444,10 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status; for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { - if (mp->format == FORMAT_RAW) - continue; - average.packages.counter[i] += p->counter[i]; + if ((mp->format == FORMAT_RAW) && (topo.num_packages == 0)) + average.packages.counter[i] = p->counter[i]; + else + average.packages.counter[i] += p->counter[i]; } return 0; } From 3ac1d14d0583a2de75d49a5234d767e2590384dd Mon Sep 17 00:00:00 2001 From: Wyes Karny Date: Tue, 3 Oct 2023 05:07:51 +0000 Subject: [PATCH 02/26] tools/power turbostat: Increase the limit for fd opened When running turbostat, a system with 512 cpus reaches the limit for maximum number of file descriptors that can be opened. To solve this problem, the limit is raised to 2^15, which is a large enough number. Below data is collected from AMD server systems while running turbostat: |-----------+-------------------------------| | # of cpus | # of opened fds for turbostat | |-----------+-------------------------------| | 128 | 260 | |-----------+-------------------------------| | 192 | 388 | |-----------+-------------------------------| | 512 | 1028 | |-----------+-------------------------------| So, the new max limit would be sufficient up to 2^14 cpus (but this also depends on how many counters are enabled). Reviewed-by: Doug Smythies Tested-by: Doug Smythies Signed-off-by: Wyes Karny Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 21 +++++++++++++++++++++ 1 file changed, 21 insertions(+) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index fca7913f6c84..2550a0e35914 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -53,6 +53,8 @@ #define NAME_BYTES 20 #define PATH_BYTES 128 +#define MAX_NOFILE 0x8000 + enum counter_scope { SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE }; enum counter_type { COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC }; enum counter_format { FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT }; @@ -6705,6 +6707,22 @@ void cmdline(int argc, char **argv) } } +void set_rlimit(void) +{ + struct rlimit limit; + + if (getrlimit(RLIMIT_NOFILE, &limit) < 0) + err(1, "Failed to get rlimit"); + + if (limit.rlim_max < MAX_NOFILE) + limit.rlim_max = MAX_NOFILE; + if (limit.rlim_cur < MAX_NOFILE) + limit.rlim_cur = MAX_NOFILE; + + if (setrlimit(RLIMIT_NOFILE, &limit) < 0) + err(1, "Failed to set rlimit"); +} + int main(int argc, char **argv) { int fd, ret; @@ -6730,6 +6748,9 @@ skip_cgroup_setting: probe_sysfs(); + if (!getuid()) + set_rlimit(); + turbostat_init(); msr_sum_record(); From 0b13410b52c4636aacb6964a4253a797c0fa0d16 Mon Sep 17 00:00:00 2001 From: Peng Liu Date: Sat, 7 Oct 2023 13:46:22 +0800 Subject: [PATCH 03/26] tools/power turbostat: Fix Bzy_MHz documentation typo The code calculates Bzy_MHz by multiplying TSC_delta * APERF_delta/MPERF_delta The man page erroneously showed that TSC_delta was divided. Signed-off-by: Peng Liu Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.8 | 2 +- 1 file changed, 1 insertion(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 8f08c3fd498d..1ba6340d3b3d 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -370,7 +370,7 @@ below the processor's base frequency. Busy% = MPERF_delta/TSC_delta -Bzy_MHz = TSC_delta/APERF_delta/MPERF_delta/measurement_interval +Bzy_MHz = TSC_delta*APERF_delta/MPERF_delta/measurement_interval Note that these calculations depend on TSC_delta, so they are not reliable during intervals when TSC_MHz is not running at the base frequency. From 227ed18f456a68bbb69807294a9089208663a6d3 Mon Sep 17 00:00:00 2001 From: Chen Yu Date: Sun, 22 Oct 2023 13:52:21 +0800 Subject: [PATCH 04/26] tools/power turbostat: Do not print negative LPI residency turbostat prints the abnormal SYS%LPI across suspend-to-idle: SYS%LPI = 114479815993277.50 This is reproduced by: Run a freeze cycle, e.g. "sleepgraph -m freeze -rtcwake 15". Then do a reboot. After boot up, launch the suspend-idle-idle and check the SYS%LPI field. The slp_so residence counter is in LPIT table, and BIOS does not clears this register across reset. The PMC expects the OS to calculate the LPI residency based on the delta. However, there is an firmware issue that the LPIT gets cleared to 0 during the second suspend to idle after the reboot, which brings negative delta value. [lenb: updated to print "neg" upon this BIOS failure] Reported-by: Todd Brandt Signed-off-by: Chen Yu Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 30 +++++++++++++++++++-------- 1 file changed, 21 insertions(+), 9 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 2550a0e35914..c23703dd54aa 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -991,8 +991,8 @@ struct pkg_data { unsigned long long pc8; unsigned long long pc9; unsigned long long pc10; - unsigned long long cpu_lpi; - unsigned long long sys_lpi; + long long cpu_lpi; + long long sys_lpi; unsigned long long pkg_wtd_core_c0; unsigned long long pkg_any_core_c0; unsigned long long pkg_any_gfxe_c0; @@ -1978,12 +1978,22 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data if (DO_BIC(BIC_Pkgpc10)) outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pc10 / tsc); - if (DO_BIC(BIC_CPU_LPI)) - outp += - sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->cpu_lpi / 1000000.0 / interval_float); - if (DO_BIC(BIC_SYS_LPI)) - outp += - sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->sys_lpi / 1000000.0 / interval_float); + if (DO_BIC(BIC_CPU_LPI)) { + if (p->cpu_lpi >= 0) + outp += + sprintf(outp, "%s%.2f", (printed++ ? delim : ""), + 100.0 * p->cpu_lpi / 1000000.0 / interval_float); + else + outp += sprintf(outp, "%s(neg)", (printed++ ? delim : "")); + } + if (DO_BIC(BIC_SYS_LPI)) { + if (p->sys_lpi >= 0) + outp += + sprintf(outp, "%s%.2f", (printed++ ? delim : ""), + 100.0 * p->sys_lpi / 1000000.0 / interval_float); + else + outp += sprintf(outp, "%s(neg)", (printed++ ? delim : "")); + } if (DO_BIC(BIC_PkgWatt)) outp += @@ -3832,7 +3842,8 @@ void re_initialize(void) { free_all_buffers(); setup_all_buffers(false); - fprintf(outf, "turbostat: re-initialized with num_cpus %d, allowed_cpus %d\n", topo.num_cpus, topo.allowed_cpus); + fprintf(outf, "turbostat: re-initialized with num_cpus %d, allowed_cpus %d\n", topo.num_cpus, + topo.allowed_cpus); } void set_max_cpu_num(void) @@ -6145,6 +6156,7 @@ void topology_update(void) topo.allowed_packages = 0; for_all_cpus(update_topo, ODD_COUNTERS); } + void setup_all_buffers(bool startup) { topology_probe(startup); From bb6181fa6bc942aac3f7f2fa8e3831952a2ef118 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Wed, 20 Dec 2023 13:11:05 -0500 Subject: [PATCH 05/26] tools/power turbostat: Expand probe_intel_uncore_frequency() Print current frequency along with the current (and initial) limits Probe and print uncore config also for machines using the new cluster API Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 88 ++++++++++++++++++++------- 1 file changed, 65 insertions(+), 23 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index c23703dd54aa..bbd2e0edadfa 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -4581,20 +4581,15 @@ static void dump_sysfs_file(char *path) static void probe_intel_uncore_frequency(void) { int i, j; - char path[128]; + char path[256]; if (!genuine_intel) return; - if (access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00", R_OK)) - return; + if (access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00/current_freq_khz", R_OK)) + goto probe_cluster; - /* Cluster level sysfs not supported yet. */ - if (!access("/sys/devices/system/cpu/intel_uncore_frequency/uncore00", R_OK)) - return; - - if (!access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00/current_freq_khz", R_OK)) - BIC_PRESENT(BIC_UNCORE_MHZ); + BIC_PRESENT(BIC_UNCORE_MHZ); if (quiet) return; @@ -4602,26 +4597,73 @@ static void probe_intel_uncore_frequency(void) for (i = 0; i < topo.num_packages; ++i) { for (j = 0; j < topo.num_die; ++j) { int k, l; + char path_base[128]; - sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/min_freq_khz", - i, j); - k = read_sysfs_int(path); - sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/max_freq_khz", - i, j); - l = read_sysfs_int(path); - fprintf(outf, "Uncore Frequency pkg%d die%d: %d - %d MHz ", i, j, k / 1000, l / 1000); + sprintf(path_base, "/sys/devices/system/cpu/intel_uncore_frequency/package_%02d_die_%02d", i, + j); - sprintf(path, - "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/initial_min_freq_khz", - i, j); + sprintf(path, "%s/min_freq_khz", path_base); k = read_sysfs_int(path); - sprintf(path, - "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/initial_max_freq_khz", - i, j); + sprintf(path, "%s/max_freq_khz", path_base); l = read_sysfs_int(path); - fprintf(outf, "(%d - %d MHz)\n", k / 1000, l / 1000); + fprintf(outf, "Uncore Frequency package%d die%d: %d - %d MHz ", i, j, k / 1000, l / 1000); + + sprintf(path, "%s/initial_min_freq_khz", path_base); + k = read_sysfs_int(path); + sprintf(path, "%s/initial_max_freq_khz", path_base); + l = read_sysfs_int(path); + fprintf(outf, "(%d - %d MHz)", k / 1000, l / 1000); + + sprintf(path, "%s/current_freq_khz", path_base); + k = read_sysfs_int(path); + fprintf(outf, " %d MHz\n", k / 1000); } } + return; + +probe_cluster: + if (access("/sys/devices/system/cpu/intel_uncore_frequency/uncore00/current_freq_khz", R_OK)) + return; + + if (quiet) + return; + + for (i = 0;; ++i) { + int k, l; + char path_base[128]; + int package_id, domain_id, cluster_id; + + sprintf(path_base, "/sys/devices/system/cpu/intel_uncore_frequency/uncore%02d", i); + + if (access(path_base, R_OK)) + break; + + sprintf(path, "%s/package_id", path_base); + package_id = read_sysfs_int(path); + + sprintf(path, "%s/domain_id", path_base); + domain_id = read_sysfs_int(path); + + sprintf(path, "%s/fabric_cluster_id", path_base); + cluster_id = read_sysfs_int(path); + + sprintf(path, "%s/min_freq_khz", path_base); + k = read_sysfs_int(path); + sprintf(path, "%s/max_freq_khz", path_base); + l = read_sysfs_int(path); + fprintf(outf, "Uncore Frequency package%d domain%d cluster%d: %d - %d MHz ", package_id, domain_id, + cluster_id, k / 1000, l / 1000); + + sprintf(path, "%s/initial_min_freq_khz", path_base); + k = read_sysfs_int(path); + sprintf(path, "%s/initial_max_freq_khz", path_base); + l = read_sysfs_int(path); + fprintf(outf, "(%d - %d MHz)", k / 1000, l / 1000); + + sprintf(path, "%s/current_freq_khz", path_base); + k = read_sysfs_int(path); + fprintf(outf, " %d MHz\n", k / 1000); + } } static void probe_graphics(void) From fb5ceca046efc84f69fcf9779a013f8a0e63bbff Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Fri, 12 Jan 2024 13:48:14 +0100 Subject: [PATCH 06/26] tools/power turbostat: Print ucode revision only if valid If the MSR read were to fail, turbostat would print "microcode 0x0" Signed-off-by: Patryk Wlazlyn Reviewed-by: Len Brown Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 12 +++++++++--- 1 file changed, 9 insertions(+), 3 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index bbd2e0edadfa..a4a40a6e1b95 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -5679,6 +5679,7 @@ void process_cpuid() unsigned int eax, ebx, ecx, edx; unsigned int fms, family, model, stepping, ecx_flags, edx_flags; unsigned long long ucode_patch = 0; + bool ucode_patch_valid = false; eax = ebx = ecx = edx = 0; @@ -5708,6 +5709,8 @@ void process_cpuid() if (get_msr(sched_getcpu(), MSR_IA32_UCODE_REV, &ucode_patch)) warnx("get_msr(UCODE)"); + else + ucode_patch_valid = true; /* * check max extended function levels of CPUID. @@ -5718,9 +5721,12 @@ void process_cpuid() __cpuid(0x80000000, max_extended_level, ebx, ecx, edx); if (!quiet) { - fprintf(outf, "CPUID(1): family:model:stepping 0x%x:%x:%x (%d:%d:%d) microcode 0x%x\n", - family, model, stepping, family, model, stepping, - (unsigned int)((ucode_patch >> 32) & 0xFFFFFFFF)); + fprintf(outf, "CPUID(1): family:model:stepping 0x%x:%x:%x (%d:%d:%d)", + family, model, stepping, family, model, stepping); + if (ucode_patch_valid) + fprintf(outf, " microcode 0x%x", (unsigned int)((ucode_patch >> 32) & 0xFFFFFFFF)); + fputc('\n', outf); + fprintf(outf, "CPUID(0x80000000): max_extended_levels: 0x%x\n", max_extended_level); fprintf(outf, "CPUID(1): %s %s %s %s %s %s %s %s %s %s\n", ecx_flags & (1 << 0) ? "SSE3" : "-", From 538d505fde20393bce1e6fb95cec82b56cdd22ef Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Mon, 22 Jan 2024 15:22:35 +0100 Subject: [PATCH 07/26] tools/power turbostat: Read base_hz and bclk from CPUID.16H if available If MSRs cannot be read, values can be obtained from cpuid. Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 9 +++++++++ 1 file changed, 9 insertions(+) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index a4a40a6e1b95..c35c48b6a99a 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -5848,6 +5848,15 @@ void process_cpuid() base_mhz = max_mhz = bus_mhz = edx = 0; __cpuid(0x16, base_mhz, max_mhz, bus_mhz, edx); + + bclk = bus_mhz; + + base_hz = base_mhz * 1000000; + has_base_hz = 1; + + if (platform->enable_tsc_tweak) + tsc_tweak = base_hz / tsc_hz; + if (!quiet) fprintf(outf, "CPUID(0x16): base_mhz: %d max_mhz: %d bus_mhz: %d\n", base_mhz, max_mhz, bus_mhz); From b6fe938317eed58e8c687bd5965a956e15fb5828 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Fri, 19 Jan 2024 12:25:42 -0600 Subject: [PATCH 08/26] tools/power turbostat: Fix warning upon failed /dev/cpu_dma_latency read Previously a failed read of /dev/cpu_dma_latency erroneously complained turbostat: capget(CAP_SYS_ADMIN) failed, try "# setcap cap_sys_admin=ep ./turbostat This went unnoticed because this file is typically visible to root, and turbostat was typically run as root. Going forward, when a non-root user can run turbostat... Complain about failed read access to this file only if --debug is used. Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 3 ++- 1 file changed, 2 insertions(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index c35c48b6a99a..531f37e5f92a 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -5545,7 +5545,8 @@ void print_dev_latency(void) fd = open(path, O_RDONLY); if (fd < 0) { - warnx("capget(CAP_SYS_ADMIN) failed, try \"# setcap cap_sys_admin=ep %s\"", progname); + if (debug) + warnx("Read %s failed", path); return; } From 2d2ccd57338779469777d4319152151272994182 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Mon, 5 Feb 2024 15:56:25 -0600 Subject: [PATCH 09/26] tools/power turbostat: enhance -D (debug counter dump) output Eliminate redundant debug output for core and package scope counters. Include name and path for all "ADDED" counters. Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 16 +++++++++++----- 1 file changed, 11 insertions(+), 5 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 531f37e5f92a..60432753fe6a 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -1673,11 +1673,13 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p outp += sprintf(outp, "SMI: %d\n", t->smi_count); for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { - outp += sprintf(outp, "tADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, t->counter[i]); + outp += + sprintf(outp, "tADDED [%d] %8s msr0x%x: %08llX %s\n", i, mp->name, mp->msr_num, + t->counter[i], mp->path); } } - if (c) { + if (c && is_cpu_first_thread_in_core(t, c, p)) { outp += sprintf(outp, "core: %d\n", c->core_id); outp += sprintf(outp, "c3: %016llX\n", c->c3); outp += sprintf(outp, "c6: %016llX\n", c->c6); @@ -1687,12 +1689,14 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p outp += sprintf(outp, "Joules: %0X\n", c->core_energy); for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { - outp += sprintf(outp, "cADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, c->counter[i]); + outp += + sprintf(outp, "cADDED [%d] %8s msr0x%x: %08llX %s\n", i, mp->name, mp->msr_num, + c->counter[i], mp->path); } outp += sprintf(outp, "mc6_us: %016llX\n", c->mc6_us); } - if (p) { + if (p && is_cpu_first_core_in_package(t, c, p)) { outp += sprintf(outp, "package: %d\n", p->package_id); outp += sprintf(outp, "Weighted cores: %016llX\n", p->pkg_wtd_core_c0); @@ -1721,7 +1725,9 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c); for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { - outp += sprintf(outp, "pADDED [%d] msr0x%x: %08llX\n", i, mp->msr_num, p->counter[i]); + outp += + sprintf(outp, "pADDED [%d] %8s msr0x%x: %08llX %s\n", i, mp->name, mp->msr_num, + p->counter[i], mp->path); } } From 3e4048466c396cff52c6d435156dbcd0571e4381 Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Thu, 11 Jan 2024 15:48:09 +0100 Subject: [PATCH 10/26] tools/power turbostat: Add --no-msr option Add --no-msr option to allow users to run turbostat without accessing MSRs via the MSR driver. Signed-off-by: Patryk Wlazlyn Reviewed-by: Len Brown Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.8 | 2 + tools/power/x86/turbostat/turbostat.c | 205 +++++++++++++++++++------- 2 files changed, 151 insertions(+), 56 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 1ba6340d3b3d..28be452fbfe2 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -67,6 +67,8 @@ The column name "all" can be used to enable all disabled-by-default built-in cou .PP \fB--quiet\fP Do not decode and print the system configuration header information. .PP ++\fB--no-msr\fP Disable all the uses of the MSR driver. ++.PP \fB--interval seconds\fP overrides the default 5.0 second measurement interval. .PP \fB--num_iterations num\fP number of the measurement iterations. diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 60432753fe6a..4d5437a9725b 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -36,6 +36,7 @@ #include #include #include +#include #define UNUSED(x) (void)(x) @@ -265,6 +266,7 @@ unsigned int has_hwp_epp; /* IA32_HWP_REQUEST[bits 31:24] */ unsigned int has_hwp_pkg; /* IA32_HWP_REQUEST_PKG */ unsigned int first_counter_read = 1; int ignore_stdin; +bool no_msr; int get_msr(int cpu, off_t offset, unsigned long long *msr); @@ -1282,13 +1284,36 @@ int get_msr_fd(int cpu) sprintf(pathname, "/dev/cpu/%d/msr", cpu); fd = open(pathname, O_RDONLY); if (fd < 0) - err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, or run as root", pathname); + err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, " + "or run with --no-msr, or run as root", pathname); fd_percpu[cpu] = fd; return fd; } +static void bic_disable_msr_access(void) +{ + const unsigned long bic_msrs = + BIC_Avg_MHz | + BIC_Busy | + BIC_Bzy_MHz | + BIC_SMI | + BIC_CPU_c1 | + BIC_CPU_c3 | + BIC_CPU_c6 | + BIC_CPU_c7 | + BIC_Mod_c6 | + BIC_CoreTmp | + BIC_Totl_c0 | + BIC_Any_c0 | + BIC_GFX_c0 | + BIC_CPUGFX | + BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_PkgTmp; + + bic_enabled &= ~bic_msrs; +} + static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags) { return syscall(__NR_perf_event_open, hw_event, pid, cpu, group_fd, flags); @@ -1328,6 +1353,8 @@ int get_msr(int cpu, off_t offset, unsigned long long *msr) { ssize_t retval; + assert(!no_msr); + retval = pread(get_msr_fd(cpu), msr, sizeof(*msr), offset); if (retval != sizeof *msr) @@ -1371,6 +1398,7 @@ void help(void) " Override default 5-second measurement interval\n" " -J, --Joules displays energy in Joules instead of Watts\n" " -l, --list list column headers only\n" + " -M, --no-msr Disable all uses of the MSR driver\n" " -n, --num_iterations num\n" " number of the measurement iterations\n" " -N, --header_iterations num\n" @@ -2597,6 +2625,7 @@ unsigned long long snapshot_sysfs_counter(char *path) int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp) { if (mp->msr_num != 0) { + assert(!no_msr); if (get_msr(cpu, mp->msr_num, counterp)) return -1; } else { @@ -2646,6 +2675,9 @@ int get_epb(int cpu) return epb; msr_fallback: + if (no_msr) + return -1; + get_msr(cpu, MSR_IA32_ENERGY_PERF_BIAS, &msr); return msr & 0xf; @@ -2865,7 +2897,7 @@ retry: if (DO_BIC(BIC_CORE_THROT_CNT)) get_core_throt_cnt(cpu, &c->core_throt_cnt); - if (platform->rapl_msrs & RAPL_AMD_F17H) { + if ((platform->rapl_msrs & RAPL_AMD_F17H) && !no_msr) { if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr)) return -14; c->core_energy = msr & 0xFFFFFFFF; @@ -2930,41 +2962,44 @@ retry: if (DO_BIC(BIC_SYS_LPI)) p->sys_lpi = cpuidle_cur_sys_lpi_us; - if (platform->rapl_msrs & RAPL_PKG) { - if (get_msr_sum(cpu, MSR_PKG_ENERGY_STATUS, &msr)) - return -13; - p->energy_pkg = msr; - } - if (platform->rapl_msrs & RAPL_CORE_ENERGY_STATUS) { - if (get_msr_sum(cpu, MSR_PP0_ENERGY_STATUS, &msr)) - return -14; - p->energy_cores = msr; - } - if (platform->rapl_msrs & RAPL_DRAM) { - if (get_msr_sum(cpu, MSR_DRAM_ENERGY_STATUS, &msr)) - return -15; - p->energy_dram = msr; - } - if (platform->rapl_msrs & RAPL_GFX) { - if (get_msr_sum(cpu, MSR_PP1_ENERGY_STATUS, &msr)) - return -16; - p->energy_gfx = msr; - } - if (platform->rapl_msrs & RAPL_PKG_PERF_STATUS) { - if (get_msr_sum(cpu, MSR_PKG_PERF_STATUS, &msr)) - return -16; - p->rapl_pkg_perf_status = msr; - } - if (platform->rapl_msrs & RAPL_DRAM_PERF_STATUS) { - if (get_msr_sum(cpu, MSR_DRAM_PERF_STATUS, &msr)) - return -16; - p->rapl_dram_perf_status = msr; - } - if (platform->rapl_msrs & RAPL_AMD_F17H) { - if (get_msr_sum(cpu, MSR_PKG_ENERGY_STAT, &msr)) - return -13; - p->energy_pkg = msr; + if (!no_msr) { + if (platform->rapl_msrs & RAPL_PKG) { + if (get_msr_sum(cpu, MSR_PKG_ENERGY_STATUS, &msr)) + return -13; + p->energy_pkg = msr; + } + if (platform->rapl_msrs & RAPL_CORE_ENERGY_STATUS) { + if (get_msr_sum(cpu, MSR_PP0_ENERGY_STATUS, &msr)) + return -14; + p->energy_cores = msr; + } + if (platform->rapl_msrs & RAPL_DRAM) { + if (get_msr_sum(cpu, MSR_DRAM_ENERGY_STATUS, &msr)) + return -15; + p->energy_dram = msr; + } + if (platform->rapl_msrs & RAPL_GFX) { + if (get_msr_sum(cpu, MSR_PP1_ENERGY_STATUS, &msr)) + return -16; + p->energy_gfx = msr; + } + if (platform->rapl_msrs & RAPL_PKG_PERF_STATUS) { + if (get_msr_sum(cpu, MSR_PKG_PERF_STATUS, &msr)) + return -16; + p->rapl_pkg_perf_status = msr; + } + if (platform->rapl_msrs & RAPL_DRAM_PERF_STATUS) { + if (get_msr_sum(cpu, MSR_DRAM_PERF_STATUS, &msr)) + return -16; + p->rapl_dram_perf_status = msr; + } + if (platform->rapl_msrs & RAPL_AMD_F17H) { + if (get_msr_sum(cpu, MSR_PKG_ENERGY_STAT, &msr)) + return -13; + p->energy_pkg = msr; + } } + if (DO_BIC(BIC_PkgTmp)) { if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) return -17; @@ -3072,7 +3107,7 @@ void probe_cst_limit(void) unsigned long long msr; int *pkg_cstate_limits; - if (!platform->has_nhm_msrs) + if (!platform->has_nhm_msrs || no_msr) return; switch (platform->cst_limit) { @@ -3116,7 +3151,7 @@ static void dump_platform_info(void) unsigned long long msr; unsigned int ratio; - if (!platform->has_nhm_msrs) + if (!platform->has_nhm_msrs || no_msr) return; get_msr(base_cpu, MSR_PLATFORM_INFO, &msr); @@ -3134,7 +3169,7 @@ static void dump_power_ctl(void) { unsigned long long msr; - if (!platform->has_nhm_msrs) + if (!platform->has_nhm_msrs || no_msr) return; get_msr(base_cpu, MSR_IA32_POWER_CTL, &msr); @@ -3340,7 +3375,7 @@ static void dump_cst_cfg(void) { unsigned long long msr; - if (!platform->has_nhm_msrs) + if (!platform->has_nhm_msrs || no_msr) return; get_msr(base_cpu, MSR_PKG_CST_CONFIG_CONTROL, &msr); @@ -3412,7 +3447,7 @@ void print_irtl(void) { unsigned long long msr; - if (!platform->has_irtl_msrs) + if (!platform->has_irtl_msrs || no_msr) return; if (platform->supported_cstates & PC3) { @@ -4193,6 +4228,8 @@ int get_msr_sum(int cpu, off_t offset, unsigned long long *msr) int ret, idx; unsigned long long msr_cur, msr_last; + assert(!no_msr); + if (!per_cpu_msr_sum) return 1; @@ -4221,6 +4258,8 @@ static int update_msr_sum(struct thread_data *t, struct core_data *c, struct pkg UNUSED(c); UNUSED(p); + assert(!no_msr); + for (i = IDX_PKG_ENERGY; i < IDX_COUNT; i++) { unsigned long long msr_cur, msr_last; off_t offset; @@ -4465,7 +4504,7 @@ void check_permissions(void) sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); if (euidaccess(pathname, R_OK)) { do_exit++; - warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr"); + warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr, or run with --no-msr"); } /* if all else fails, thell them to be root */ @@ -4482,7 +4521,7 @@ void probe_bclk(void) unsigned long long msr; unsigned int base_ratio; - if (!platform->has_nhm_msrs) + if (!platform->has_nhm_msrs || no_msr) return; if (platform->bclk_freq == BCLK_100MHZ) @@ -4522,7 +4561,7 @@ static void dump_turbo_ratio_info(void) if (!has_turbo) return; - if (!platform->has_nhm_msrs) + if (!platform->has_nhm_msrs || no_msr) return; if (platform->trl_msrs & TRL_LIMIT2) @@ -4845,6 +4884,9 @@ int print_hwp(struct thread_data *t, struct core_data *c, struct pkg_data *p) UNUSED(c); UNUSED(p); + if (no_msr) + return 0; + if (!has_hwp) return 0; @@ -4931,6 +4973,9 @@ int print_perf_limit(struct thread_data *t, struct core_data *c, struct pkg_data UNUSED(c); UNUSED(p); + if (no_msr) + return 0; + cpu = t->cpu_id; /* per-package */ @@ -5264,7 +5309,7 @@ int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p) */ void probe_rapl(void) { - if (!platform->rapl_msrs) + if (!platform->rapl_msrs || no_msr) return; if (genuine_intel) @@ -5320,7 +5365,7 @@ int set_temperature_target(struct thread_data *t, struct core_data *c, struct pk } /* Temperature Target MSR is Nehalem and newer only */ - if (!platform->has_nhm_msrs) + if (!platform->has_nhm_msrs || no_msr) goto guess; if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr)) @@ -5367,6 +5412,9 @@ int print_thermal(struct thread_data *t, struct core_data *c, struct pkg_data *p UNUSED(c); UNUSED(p); + if (no_msr) + return 0; + if (!(do_dts || do_ptm)) return 0; @@ -5464,6 +5512,9 @@ void decode_feature_control_msr(void) { unsigned long long msr; + if (no_msr) + return; + if (!get_msr(base_cpu, MSR_IA32_FEAT_CTL, &msr)) fprintf(outf, "cpu%d: MSR_IA32_FEATURE_CONTROL: 0x%08llx (%sLocked %s)\n", base_cpu, msr, msr & FEAT_CTL_LOCKED ? "" : "UN-", msr & (1 << 18) ? "SGX" : ""); @@ -5473,6 +5524,9 @@ void decode_misc_enable_msr(void) { unsigned long long msr; + if (no_msr) + return; + if (!genuine_intel) return; @@ -5490,6 +5544,9 @@ void decode_misc_feature_control(void) { unsigned long long msr; + if (no_msr) + return; + if (!platform->has_msr_misc_feature_control) return; @@ -5511,6 +5568,9 @@ void decode_misc_pwr_mgmt_msr(void) { unsigned long long msr; + if (no_msr) + return; + if (!platform->has_msr_misc_pwr_mgmt) return; @@ -5530,6 +5590,9 @@ void decode_c6_demotion_policy_msr(void) { unsigned long long msr; + if (no_msr) + return; + if (!platform->has_msr_c6_demotion_policy_config) return; @@ -5626,7 +5689,7 @@ void probe_cstates(void) if (platform->has_msr_module_c6_res_ms) BIC_PRESENT(BIC_Mod_c6); - if (platform->has_ext_cst_msrs) { + if (platform->has_ext_cst_msrs && !no_msr) { BIC_PRESENT(BIC_Totl_c0); BIC_PRESENT(BIC_Any_c0); BIC_PRESENT(BIC_GFX_c0); @@ -5714,10 +5777,12 @@ void process_cpuid() ecx_flags = ecx; edx_flags = edx; - if (get_msr(sched_getcpu(), MSR_IA32_UCODE_REV, &ucode_patch)) - warnx("get_msr(UCODE)"); - else - ucode_patch_valid = true; + if (!no_msr) { + if (get_msr(sched_getcpu(), MSR_IA32_UCODE_REV, &ucode_patch)) + warnx("get_msr(UCODE)"); + else + ucode_patch_valid = true; + } /* * check max extended function levels of CPUID. @@ -5892,7 +5957,7 @@ void probe_pm_features(void) probe_thermal(); - if (platform->has_nhm_msrs) + if (platform->has_nhm_msrs && !no_msr) BIC_PRESENT(BIC_SMI); if (!quiet) @@ -6252,8 +6317,10 @@ void turbostat_init() { setup_all_buffers(true); set_base_cpu(); - check_dev_msr(); - check_permissions(); + if (!no_msr) { + check_dev_msr(); + check_permissions(); + } process_cpuid(); probe_pm_features(); linux_perf_init(); @@ -6370,6 +6437,9 @@ int add_counter(unsigned int msr_num, char *path, char *name, { struct msr_counter *msrp; + if (no_msr && msr_num) + errx(1, "Requested MSR counter 0x%x, but in --no-msr mode", msr_num); + msrp = calloc(1, sizeof(struct msr_counter)); if (msrp == NULL) { perror("calloc"); @@ -6674,6 +6744,7 @@ void cmdline(int argc, char **argv) { "list", no_argument, 0, 'l' }, { "out", required_argument, 0, 'o' }, { "quiet", no_argument, 0, 'q' }, + { "no-msr", no_argument, 0, 'M' }, { "show", required_argument, 0, 's' }, { "Summary", no_argument, 0, 'S' }, { "TCC", required_argument, 0, 'T' }, @@ -6683,7 +6754,22 @@ void cmdline(int argc, char **argv) progname = argv[0]; - while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qST:v", long_options, &option_index)) != -1) { + /* + * Parse some options early, because they may make other options invalid, + * like adding the MSR counter with --add and at the same time using --no-msr. + */ + while ((opt = getopt_long_only(argc, argv, "M", long_options, &option_index)) != -1) { + switch (opt) { + case 'M': + no_msr = 1; + break; + default: + break; + } + } + optind = 0; + + while ((opt = getopt_long_only(argc, argv, "+C:c:Dde:hi:Jn:o:qMST:v", long_options, &option_index)) != -1) { switch (opt) { case 'a': parse_add_command(optarg); @@ -6741,6 +6827,9 @@ void cmdline(int argc, char **argv) case 'q': quiet = 1; break; + case 'M': + /* Parsed earlier */ + break; case 'n': num_iterations = strtod(optarg, NULL); @@ -6817,6 +6906,9 @@ skip_cgroup_setting: outf = stderr; cmdline(argc, argv); + if (no_msr) + bic_disable_msr_access(); + if (!quiet) { print_version(); print_bootcmd(); @@ -6829,7 +6921,8 @@ skip_cgroup_setting: turbostat_init(); - msr_sum_record(); + if (!no_msr) + msr_sum_record(); /* dump counters and exit */ if (dump_only) From a0e86c90b83c118985260e36490583b5a38d4359 Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Thu, 11 Jan 2024 15:58:02 +0100 Subject: [PATCH 11/26] tools/power turbostat: Add --no-perf option Add the --no-perf option to allow users to run turbostat without accessing perf. Signed-off-by: Patryk Wlazlyn Reviewed-by: Len Brown Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.8 | 2 ++ tools/power/x86/turbostat/turbostat.c | 25 ++++++++++++++++++++++--- 2 files changed, 24 insertions(+), 3 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 28be452fbfe2..567327b004e6 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -69,6 +69,8 @@ The column name "all" can be used to enable all disabled-by-default built-in cou .PP +\fB--no-msr\fP Disable all the uses of the MSR driver. +.PP ++\fB--no-perf\fP Disable all the uses of the perf API. ++.PP \fB--interval seconds\fP overrides the default 5.0 second measurement interval. .PP \fB--num_iterations num\fP number of the measurement iterations. diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 4d5437a9725b..bad2fec7f342 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -267,6 +267,7 @@ unsigned int has_hwp_pkg; /* IA32_HWP_REQUEST_PKG */ unsigned int first_counter_read = 1; int ignore_stdin; bool no_msr; +bool no_perf; int get_msr(int cpu, off_t offset, unsigned long long *msr); @@ -1314,8 +1315,17 @@ static void bic_disable_msr_access(void) bic_enabled &= ~bic_msrs; } +static void bic_disable_perf_access(void) +{ + const unsigned long bic_perf = BIC_IPC; + + bic_enabled &= ~bic_perf; +} + static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags) { + assert(!no_perf); + return syscall(__NR_perf_event_open, hw_event, pid, cpu, group_fd, flags); } @@ -1332,8 +1342,8 @@ static int perf_instr_count_open(int cpu_num) /* counter for cpu_num, including user + kernel and all processes */ fd = perf_event_open(&pea, -1, cpu_num, -1, 0); if (fd == -1) { - warnx("capget(CAP_PERFMON) failed, try \"# setcap cap_sys_admin=ep %s\"", progname); - BIC_NOT_PRESENT(BIC_IPC); + warnx("capget(CAP_PERFMON) failed, try \"# setcap cap_sys_admin=ep %s\" or use --no-perf", progname); + bic_disable_perf_access(); } return fd; @@ -1399,6 +1409,7 @@ void help(void) " -J, --Joules displays energy in Joules instead of Watts\n" " -l, --list list column headers only\n" " -M, --no-msr Disable all uses of the MSR driver\n" + " -P, --no-perf Disable all uses of the perf API\n" " -n, --num_iterations num\n" " number of the measurement iterations\n" " -N, --header_iterations num\n" @@ -6745,6 +6756,7 @@ void cmdline(int argc, char **argv) { "out", required_argument, 0, 'o' }, { "quiet", no_argument, 0, 'q' }, { "no-msr", no_argument, 0, 'M' }, + { "no-perf", no_argument, 0, 'P' }, { "show", required_argument, 0, 's' }, { "Summary", no_argument, 0, 'S' }, { "TCC", required_argument, 0, 'T' }, @@ -6758,11 +6770,14 @@ void cmdline(int argc, char **argv) * Parse some options early, because they may make other options invalid, * like adding the MSR counter with --add and at the same time using --no-msr. */ - while ((opt = getopt_long_only(argc, argv, "M", long_options, &option_index)) != -1) { + while ((opt = getopt_long_only(argc, argv, "MP", long_options, &option_index)) != -1) { switch (opt) { case 'M': no_msr = 1; break; + case 'P': + no_perf = 1; + break; default: break; } @@ -6828,6 +6843,7 @@ void cmdline(int argc, char **argv) quiet = 1; break; case 'M': + case 'P': /* Parsed earlier */ break; case 'n': @@ -6909,6 +6925,9 @@ skip_cgroup_setting: if (no_msr) bic_disable_msr_access(); + if (no_perf) + bic_disable_perf_access(); + if (!quiet) { print_version(); print_bootcmd(); From e48934c9f1048ed4640b60321baf1986d1a470e1 Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Thu, 11 Jan 2024 16:42:22 +0100 Subject: [PATCH 12/26] tools/power turbostat: Add reading aperf and mperf via perf API By using the perf API we spend less time in between the reads of the counters, resulting in more accurate calculations of the dependent metrics. Using perf API is also usually faster overall, although cache miss, if we get one, is more costly when using perf vs MSR driver. We would fallback to the msr reads if the sysfs isn't there or when in --no-perf mode. Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 374 +++++++++++++++++++++----- 1 file changed, 301 insertions(+), 73 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index bad2fec7f342..1294c46c2170 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -59,6 +59,7 @@ enum counter_scope { SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE }; enum counter_type { COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC }; enum counter_format { FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT }; +enum amperf_source { AMPERF_SOURCE_PERF, AMPERF_SOURCE_MSR }; struct msr_counter { unsigned int msr_num; @@ -207,10 +208,13 @@ unsigned long long bic_present = BIC_USEC | BIC_TOD | BIC_sysfs | BIC_APIC | BIC #define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT) #define BIC_IS_ENABLED(COUNTER_BIT) (bic_enabled & COUNTER_BIT) +struct amperf_group_fd; + char *proc_stat = "/proc/stat"; FILE *outf; int *fd_percpu; int *fd_instr_count_percpu; +struct amperf_group_fd *fd_amperf_percpu; /* File descriptors for perf group with APERF and MPERF counters. */ struct timeval interval_tv = { 5, 0 }; struct timespec interval_ts = { 5, 0 }; @@ -268,6 +272,7 @@ unsigned int first_counter_read = 1; int ignore_stdin; bool no_msr; bool no_perf; +enum amperf_source amperf_source; int get_msr(int cpu, off_t offset, unsigned long long *msr); @@ -1296,9 +1301,6 @@ int get_msr_fd(int cpu) static void bic_disable_msr_access(void) { const unsigned long bic_msrs = - BIC_Avg_MHz | - BIC_Busy | - BIC_Bzy_MHz | BIC_SMI | BIC_CPU_c1 | BIC_CPU_c3 | @@ -1329,21 +1331,30 @@ static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu return syscall(__NR_perf_event_open, hw_event, pid, cpu, group_fd, flags); } -static int perf_instr_count_open(int cpu_num) +static long open_perf_counter_or_fail(int cpu, unsigned int type, unsigned int config, int group_fd, __u64 read_format) { - struct perf_event_attr pea; - int fd; + struct perf_event_attr attr; + const pid_t pid = -1; + const unsigned long flags = 0; - memset(&pea, 0, sizeof(struct perf_event_attr)); - pea.type = PERF_TYPE_HARDWARE; - pea.size = sizeof(struct perf_event_attr); - pea.config = PERF_COUNT_HW_INSTRUCTIONS; + memset(&attr, 0, sizeof(struct perf_event_attr)); - /* counter for cpu_num, including user + kernel and all processes */ - fd = perf_event_open(&pea, -1, cpu_num, -1, 0); + attr.type = type; + attr.size = sizeof(struct perf_event_attr); + attr.config = config; + attr.disabled = 0; + attr.sample_type = PERF_SAMPLE_IDENTIFIER; + attr.read_format = read_format; + + const int fd = perf_event_open(&attr, pid, cpu, group_fd, flags); if (fd == -1) { - warnx("capget(CAP_PERFMON) failed, try \"# setcap cap_sys_admin=ep %s\" or use --no-perf", progname); - bic_disable_perf_access(); + if (errno == EACCES) { + errx(1, "capget(CAP_PERFMON) failed, try \"# setcap cap_sys_admin=ep %s\"" + " or use --no-perf or run as root", progname); + } else { + perror("perf_event_open"); + errx(1, "use --no-perf or run as root"); + } } return fd; @@ -1354,7 +1365,8 @@ int get_instr_count_fd(int cpu) if (fd_instr_count_percpu[cpu]) return fd_instr_count_percpu[cpu]; - fd_instr_count_percpu[cpu] = perf_instr_count_open(cpu); + fd_instr_count_percpu[cpu] = + open_perf_counter_or_fail(cpu, PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS, -1, 0); return fd_instr_count_percpu[cpu]; } @@ -2762,6 +2774,175 @@ int get_core_throt_cnt(int cpu, unsigned long long *cnt) return 0; } +struct amperf_group_fd { + int aperf; /* Also the group descriptor */ + int mperf; +}; + +static unsigned int read_perf_counter_info(const char *const path, const char *const parse_format) +{ + int fdmt; + char buf[16]; + unsigned int v; + + fdmt = open(path, O_RDONLY, 0); + if (fdmt == -1) + errx(1, "Failed to read perf counter info %s\n", path); + + if (read(fdmt, buf, sizeof(buf)) <= 0) + return 0; + + buf[sizeof(buf) - 1] = '\0'; + + if (sscanf(buf, parse_format, &v) != 1) + errx(1, "Failed to parse perf counter info %s\n", path); + + close(fdmt); + + return v; +} + +static unsigned read_msr_type(void) +{ + const char *const path = "/sys/bus/event_source/devices/msr/type"; + const char *const format = "%u"; + + return read_perf_counter_info(path, format); +} + +static unsigned read_aperf_config(void) +{ + const char *const path = "/sys/bus/event_source/devices/msr/events/aperf"; + const char *const format = "event=%x"; + + return read_perf_counter_info(path, format); +} + +static unsigned read_mperf_config(void) +{ + const char *const path = "/sys/bus/event_source/devices/msr/events/mperf"; + const char *const format = "event=%x"; + + return read_perf_counter_info(path, format); +} + +static struct amperf_group_fd open_amperf_fd(int cpu) +{ + const unsigned int msr_type = read_msr_type(); + const unsigned int aperf_config = read_aperf_config(); + const unsigned int mperf_config = read_mperf_config(); + struct amperf_group_fd fds = {.aperf = -1,.mperf = -1 }; + + fds.aperf = open_perf_counter_or_fail(cpu, msr_type, aperf_config, -1, PERF_FORMAT_GROUP); + fds.mperf = open_perf_counter_or_fail(cpu, msr_type, mperf_config, fds.aperf, PERF_FORMAT_GROUP); + + return fds; +} + +static int get_amperf_fd(int cpu) +{ + assert(fd_amperf_percpu); + + if (fd_amperf_percpu[cpu].aperf) + return fd_amperf_percpu[cpu].aperf; + + fd_amperf_percpu[cpu] = open_amperf_fd(cpu); + + return fd_amperf_percpu[cpu].aperf; +} + +/* Read APERF, MPERF and TSC using the perf API. */ +static int read_aperf_mperf_tsc_perf(struct thread_data *t, int cpu) +{ + union { + struct { + unsigned long nr_entries; + unsigned long aperf; + unsigned long mperf; + }; + + unsigned long as_array[3]; + } cnt; + + const int fd_amperf = get_amperf_fd(cpu); + + /* + * Read the TSC with rdtsc, because we want the absolute value and not + * the offset from the start of the counter. + */ + t->tsc = rdtsc(); + + const int n = read(fd_amperf, &cnt.as_array[0], sizeof(cnt.as_array)); + if (n != sizeof(cnt.as_array)) + return -2; + + t->aperf = cnt.aperf * aperf_mperf_multiplier; + t->mperf = cnt.mperf * aperf_mperf_multiplier; + + return 0; +} + +/* Read APERF, MPERF and TSC using the MSR driver and rdtsc instruction. */ +static int read_aperf_mperf_tsc_msr(struct thread_data *t, int cpu) +{ + unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time; + int aperf_mperf_retry_count = 0; + + /* + * The TSC, APERF and MPERF must be read together for + * APERF/MPERF and MPERF/TSC to give accurate results. + * + * Unfortunately, APERF and MPERF are read by + * individual system call, so delays may occur + * between them. If the time to read them + * varies by a large amount, we re-read them. + */ + + /* + * This initial dummy APERF read has been seen to + * reduce jitter in the subsequent reads. + */ + + if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) + return -3; + +retry: + t->tsc = rdtsc(); /* re-read close to APERF */ + + tsc_before = t->tsc; + + if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) + return -3; + + tsc_between = rdtsc(); + + if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf)) + return -4; + + tsc_after = rdtsc(); + + aperf_time = tsc_between - tsc_before; + mperf_time = tsc_after - tsc_between; + + /* + * If the system call latency to read APERF and MPERF + * differ by more than 2x, then try again. + */ + if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) { + aperf_mperf_retry_count++; + if (aperf_mperf_retry_count < 5) + goto retry; + else + warnx("cpu%d jitter %lld %lld", cpu, aperf_time, mperf_time); + } + aperf_mperf_retry_count = 0; + + t->aperf = t->aperf * aperf_mperf_multiplier; + t->mperf = t->mperf * aperf_mperf_multiplier; + + return 0; +} + /* * get_counters(...) * migrate to cpu @@ -2771,7 +2952,6 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { int cpu = t->cpu_id; unsigned long long msr; - int aperf_mperf_retry_count = 0; struct msr_counter *mp; int i; @@ -2784,63 +2964,26 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) if (first_counter_read) get_apic_id(t); -retry: + t->tsc = rdtsc(); /* we are running on local CPU of interest */ if (DO_BIC(BIC_Avg_MHz) || DO_BIC(BIC_Busy) || DO_BIC(BIC_Bzy_MHz) || DO_BIC(BIC_IPC) || soft_c1_residency_display(BIC_Avg_MHz)) { - unsigned long long tsc_before, tsc_between, tsc_after, aperf_time, mperf_time; + int status = -1; - /* - * The TSC, APERF and MPERF must be read together for - * APERF/MPERF and MPERF/TSC to give accurate results. - * - * Unfortunately, APERF and MPERF are read by - * individual system call, so delays may occur - * between them. If the time to read them - * varies by a large amount, we re-read them. - */ + assert(!no_perf || !no_msr); - /* - * This initial dummy APERF read has been seen to - * reduce jitter in the subsequent reads. - */ - - if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) - return -3; - - t->tsc = rdtsc(); /* re-read close to APERF */ - - tsc_before = t->tsc; - - if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) - return -3; - - tsc_between = rdtsc(); - - if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf)) - return -4; - - tsc_after = rdtsc(); - - aperf_time = tsc_between - tsc_before; - mperf_time = tsc_after - tsc_between; - - /* - * If the system call latency to read APERF and MPERF - * differ by more than 2x, then try again. - */ - if ((aperf_time > (2 * mperf_time)) || (mperf_time > (2 * aperf_time))) { - aperf_mperf_retry_count++; - if (aperf_mperf_retry_count < 5) - goto retry; - else - warnx("cpu%d jitter %lld %lld", cpu, aperf_time, mperf_time); + switch (amperf_source) { + case AMPERF_SOURCE_PERF: + status = read_aperf_mperf_tsc_perf(t, cpu); + break; + case AMPERF_SOURCE_MSR: + status = read_aperf_mperf_tsc_msr(t, cpu); + break; } - aperf_mperf_retry_count = 0; - t->aperf = t->aperf * aperf_mperf_multiplier; - t->mperf = t->mperf * aperf_mperf_multiplier; + if (status != 0) + return status; } if (DO_BIC(BIC_IPC)) @@ -3516,6 +3659,21 @@ void free_fd_percpu(void) free(fd_percpu); } +void free_fd_amperf_percpu(void) +{ + int i; + + for (i = 0; i < topo.max_cpu_num + 1; ++i) { + if (fd_amperf_percpu[i].mperf != 0) + close(fd_amperf_percpu[i].mperf); + + if (fd_amperf_percpu[i].aperf != 0) + close(fd_amperf_percpu[i].aperf); + } + + free(fd_amperf_percpu); +} + void free_all_buffers(void) { int i; @@ -3557,6 +3715,7 @@ void free_all_buffers(void) outp = NULL; free_fd_percpu(); + free_fd_amperf_percpu(); free(irq_column_2_cpu); free(irqs_per_cpu); @@ -5647,17 +5806,62 @@ void print_dev_latency(void) */ void linux_perf_init(void) { - if (!BIC_IS_ENABLED(BIC_IPC)) - return; - if (access("/proc/sys/kernel/perf_event_paranoid", F_OK)) return; - fd_instr_count_percpu = calloc(topo.max_cpu_num + 1, sizeof(int)); - if (fd_instr_count_percpu == NULL) - err(-1, "calloc fd_instr_count_percpu"); + if (BIC_IS_ENABLED(BIC_IPC) && has_aperf) { + fd_instr_count_percpu = calloc(topo.max_cpu_num + 1, sizeof(int)); + if (fd_instr_count_percpu == NULL) + err(-1, "calloc fd_instr_count_percpu"); + } - BIC_PRESENT(BIC_IPC); + const bool aperf_required = BIC_IS_ENABLED(BIC_Avg_MHz) || BIC_IS_ENABLED(BIC_Busy) || + BIC_IS_ENABLED(BIC_Bzy_MHz) || BIC_IS_ENABLED(BIC_IPC); + if (aperf_required && has_aperf && amperf_source == AMPERF_SOURCE_PERF) { + fd_amperf_percpu = calloc(topo.max_cpu_num + 1, sizeof(*fd_amperf_percpu)); + if (fd_amperf_percpu == NULL) + err(-1, "calloc fd_amperf_percpu"); + } +} + +static int has_amperf_access_via_msr(void) +{ + const int cpu = sched_getcpu(); + unsigned long long dummy; + + if (get_msr(cpu, MSR_IA32_APERF, &dummy)) + return 0; + + if (get_msr(cpu, MSR_IA32_MPERF, &dummy)) + return 0; + + return 1; +} + +static int has_amperf_access_via_perf(void) +{ + if (access("/sys/bus/event_source/devices/msr/type", F_OK)) + return 0; + + if (access("/sys/bus/event_source/devices/msr/events/aperf", F_OK)) + return 0; + + if (access("/sys/bus/event_source/devices/msr/events/mperf", F_OK)) + return 0; + + return 1; +} + +/* Check if we can access APERF and MPERF */ +static int has_amperf_access(void) +{ + if (!no_msr && has_amperf_access_via_msr()) + return 1; + + if (!no_perf && has_amperf_access_via_perf()) + return 1; + + return 0; } void probe_cstates(void) @@ -5845,10 +6049,11 @@ void process_cpuid() __cpuid(0x6, eax, ebx, ecx, edx); has_aperf = ecx & (1 << 0); - if (has_aperf) { + if (has_aperf && has_amperf_access()) { BIC_PRESENT(BIC_Avg_MHz); BIC_PRESENT(BIC_Busy); BIC_PRESENT(BIC_Bzy_MHz); + BIC_PRESENT(BIC_IPC); } do_dts = eax & (1 << 0); if (do_dts) @@ -6324,6 +6529,19 @@ void set_base_cpu(void) err(-ENODEV, "No valid cpus found"); } +static void set_amperf_source(void) +{ + amperf_source = AMPERF_SOURCE_PERF; + + if (no_perf || !has_amperf_access_via_perf()) + amperf_source = AMPERF_SOURCE_MSR; + + if (quiet || !debug) + return; + + fprintf(outf, "aperf/mperf source preference: %s\n", amperf_source == AMPERF_SOURCE_MSR ? "msr" : "perf"); +} + void turbostat_init() { setup_all_buffers(true); @@ -6334,6 +6552,7 @@ void turbostat_init() } process_cpuid(); probe_pm_features(); + set_amperf_source(); linux_perf_init(); for_all_cpus(get_cpu_type, ODD_COUNTERS); @@ -6341,6 +6560,15 @@ void turbostat_init() if (DO_BIC(BIC_IPC)) (void)get_instr_count_fd(base_cpu); + + /* + * If TSC tweak is needed, but couldn't get it, + * disable more BICs, since it can't be reported accurately. + */ + if (platform->enable_tsc_tweak && !has_base_hz) { + bic_enabled &= ~BIC_Busy; + bic_enabled &= ~BIC_Bzy_MHz; + } } int fork_it(char **argv) From 5088741ec805cd249e27c7176ed09bdab164960e Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Mon, 15 Jan 2024 19:04:21 +0100 Subject: [PATCH 13/26] tools/power turbostat: detect and disable unavailable BICs at runtime To allow unprivileged user to run turbostat seamlessly. Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 192 +++++++++++++++++--------- 1 file changed, 127 insertions(+), 65 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 1294c46c2170..30db6e92193a 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -1317,13 +1317,6 @@ static void bic_disable_msr_access(void) bic_enabled &= ~bic_msrs; } -static void bic_disable_perf_access(void) -{ - const unsigned long bic_perf = BIC_IPC; - - bic_enabled &= ~bic_perf; -} - static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags) { assert(!no_perf); @@ -1331,12 +1324,14 @@ static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu return syscall(__NR_perf_event_open, hw_event, pid, cpu, group_fd, flags); } -static long open_perf_counter_or_fail(int cpu, unsigned int type, unsigned int config, int group_fd, __u64 read_format) +static long open_perf_counter(int cpu, unsigned int type, unsigned int config, int group_fd, __u64 read_format) { struct perf_event_attr attr; const pid_t pid = -1; const unsigned long flags = 0; + assert(!no_perf); + memset(&attr, 0, sizeof(struct perf_event_attr)); attr.type = type; @@ -1347,15 +1342,6 @@ static long open_perf_counter_or_fail(int cpu, unsigned int type, unsigned int c attr.read_format = read_format; const int fd = perf_event_open(&attr, pid, cpu, group_fd, flags); - if (fd == -1) { - if (errno == EACCES) { - errx(1, "capget(CAP_PERFMON) failed, try \"# setcap cap_sys_admin=ep %s\"" - " or use --no-perf or run as root", progname); - } else { - perror("perf_event_open"); - errx(1, "use --no-perf or run as root"); - } - } return fd; } @@ -1365,8 +1351,7 @@ int get_instr_count_fd(int cpu) if (fd_instr_count_percpu[cpu]) return fd_instr_count_percpu[cpu]; - fd_instr_count_percpu[cpu] = - open_perf_counter_or_fail(cpu, PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS, -1, 0); + fd_instr_count_percpu[cpu] = open_perf_counter(cpu, PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS, -1, 0); return fd_instr_count_percpu[cpu]; } @@ -2833,8 +2818,8 @@ static struct amperf_group_fd open_amperf_fd(int cpu) const unsigned int mperf_config = read_mperf_config(); struct amperf_group_fd fds = {.aperf = -1,.mperf = -1 }; - fds.aperf = open_perf_counter_or_fail(cpu, msr_type, aperf_config, -1, PERF_FORMAT_GROUP); - fds.mperf = open_perf_counter_or_fail(cpu, msr_type, mperf_config, fds.aperf, PERF_FORMAT_GROUP); + fds.aperf = open_perf_counter(cpu, msr_type, aperf_config, -1, PERF_FORMAT_GROUP); + fds.mperf = open_perf_counter(cpu, msr_type, mperf_config, fds.aperf, PERF_FORMAT_GROUP); return fds; } @@ -4509,7 +4494,8 @@ release_msr: /* * set_my_sched_priority(pri) - * return previous + * return previous priority on success + * return value < -20 on failure */ int set_my_sched_priority(int priority) { @@ -4519,16 +4505,16 @@ int set_my_sched_priority(int priority) errno = 0; original_priority = getpriority(PRIO_PROCESS, 0); if (errno && (original_priority == -1)) - err(errno, "getpriority"); + return -21; retval = setpriority(PRIO_PROCESS, 0, priority); if (retval) - errx(retval, "capget(CAP_SYS_NICE) failed,try \"# setcap cap_sys_nice=ep %s\"", progname); + return -21; errno = 0; retval = getpriority(PRIO_PROCESS, 0); if (retval != priority) - err(retval, "getpriority(%d) != setpriority(%d)", retval, priority); + return -21; return original_priority; } @@ -4543,6 +4529,9 @@ void turbostat_loop() /* * elevate own priority for interval mode + * + * ignore on error - we probably don't have permission to set it, but + * it's not a big deal */ set_my_sched_priority(-20); @@ -4628,10 +4617,13 @@ void check_dev_msr() struct stat sb; char pathname[32]; + if (no_msr) + return; + sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); if (stat(pathname, &sb)) if (system("/sbin/modprobe msr > /dev/null 2>&1")) - err(-5, "no /dev/cpu/0/msr, Try \"# modprobe msr\" "); + no_msr = 1; } /* @@ -4643,47 +4635,51 @@ int check_for_cap_sys_rawio(void) { cap_t caps; cap_flag_value_t cap_flag_value; + int ret = 0; caps = cap_get_proc(); if (caps == NULL) - err(-6, "cap_get_proc\n"); - - if (cap_get_flag(caps, CAP_SYS_RAWIO, CAP_EFFECTIVE, &cap_flag_value)) - err(-6, "cap_get\n"); - - if (cap_flag_value != CAP_SET) { - warnx("capget(CAP_SYS_RAWIO) failed," " try \"# setcap cap_sys_rawio=ep %s\"", progname); return 1; + + if (cap_get_flag(caps, CAP_SYS_RAWIO, CAP_EFFECTIVE, &cap_flag_value)) { + ret = 1; + goto free_and_exit; } + if (cap_flag_value != CAP_SET) { + ret = 1; + goto free_and_exit; + } + +free_and_exit: if (cap_free(caps) == -1) err(-6, "cap_free\n"); - return 0; + return ret; } -void check_permissions(void) +void check_msr_permission(void) { - int do_exit = 0; + int failed = 0; char pathname[32]; + if (no_msr) + return; + /* check for CAP_SYS_RAWIO */ - do_exit += check_for_cap_sys_rawio(); + failed += check_for_cap_sys_rawio(); /* test file permissions */ sprintf(pathname, "/dev/cpu/%d/msr", base_cpu); if (euidaccess(pathname, R_OK)) { - do_exit++; - warn("/dev/cpu/0/msr open failed, try chown or chmod +r /dev/cpu/*/msr, or run with --no-msr"); + failed++; } - /* if all else fails, thell them to be root */ - if (do_exit) - if (getuid() != 0) - warnx("... or simply run as root"); - - if (do_exit) - exit(-6); + if (failed) { + warnx("Failed to access %s. Some of the counters may not be available\n" + "\tRun as root to enable them or use %s to disable the access explicitly", pathname, "--no-msr"); + no_msr = 1; + } } void probe_bclk(void) @@ -5800,6 +5796,28 @@ void print_dev_latency(void) close(fd); } +static int has_instr_count_access(void) +{ + int fd; + int has_access; + + if (no_perf) + return 0; + + fd = open_perf_counter(base_cpu, PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS, -1, 0); + has_access = fd != -1; + + if (fd != -1) + close(fd); + + if (!has_access) + warnx("Failed to access %s. Some of the counters may not be available\n" + "\tRun as root to enable them or use %s to disable the access explicitly", + "instructions retired perf counter", "--no-perf"); + + return has_access; +} + /* * Linux-perf manages the HW instructions-retired counter * by enabling when requested, and hiding rollover @@ -5826,13 +5844,15 @@ void linux_perf_init(void) static int has_amperf_access_via_msr(void) { - const int cpu = sched_getcpu(); unsigned long long dummy; - if (get_msr(cpu, MSR_IA32_APERF, &dummy)) + if (no_msr) return 0; - if (get_msr(cpu, MSR_IA32_MPERF, &dummy)) + if (get_msr(base_cpu, MSR_IA32_APERF, &dummy)) + return 0; + + if (get_msr(base_cpu, MSR_IA32_MPERF, &dummy)) return 0; return 1; @@ -5840,16 +5860,44 @@ static int has_amperf_access_via_msr(void) static int has_amperf_access_via_perf(void) { - if (access("/sys/bus/event_source/devices/msr/type", F_OK)) + struct amperf_group_fd fds; + + /* + * Cache the last result, so we don't warn the user multiple times + * + * Negative means cached, no access + * Zero means not cached + * Positive means cached, has access + */ + static int has_access_cached; + + if (no_perf) return 0; - if (access("/sys/bus/event_source/devices/msr/events/aperf", F_OK)) - return 0; + if (has_access_cached != 0) + return has_access_cached > 0; - if (access("/sys/bus/event_source/devices/msr/events/mperf", F_OK)) - return 0; + fds = open_amperf_fd(base_cpu); + has_access_cached = (fds.aperf != -1) && (fds.mperf != -1); - return 1; + if (fds.aperf == -1) + warnx("Failed to access %s. Some of the counters may not be available\n" + "\tRun as root to enable them or use %s to disable the access explicitly", + "APERF perf counter", "--no-perf"); + else + close(fds.aperf); + + if (fds.mperf == -1) + warnx("Failed to access %s. Some of the counters may not be available\n" + "\tRun as root to enable them or use %s to disable the access explicitly", + "MPERF perf counter", "--no-perf"); + else + close(fds.mperf); + + if (has_access_cached == 0) + has_access_cached = -1; + + return has_access_cached > 0; } /* Check if we can access APERF and MPERF */ @@ -6542,14 +6590,34 @@ static void set_amperf_source(void) fprintf(outf, "aperf/mperf source preference: %s\n", amperf_source == AMPERF_SOURCE_MSR ? "msr" : "perf"); } +void check_msr_access(void) +{ + check_dev_msr(); + check_msr_permission(); + + if (no_msr) + bic_disable_msr_access(); +} + +void check_perf_access(void) +{ + if (no_perf || !has_instr_count_access()) + bic_enabled &= ~BIC_IPC; + + if (!has_amperf_access()) { + bic_enabled &= ~BIC_Avg_MHz; + bic_enabled &= ~BIC_Busy; + bic_enabled &= ~BIC_Bzy_MHz; + bic_enabled &= ~BIC_IPC; + } +} + void turbostat_init() { setup_all_buffers(true); set_base_cpu(); - if (!no_msr) { - check_dev_msr(); - check_permissions(); - } + check_msr_access(); + check_perf_access(); process_cpuid(); probe_pm_features(); set_amperf_source(); @@ -7150,12 +7218,6 @@ skip_cgroup_setting: outf = stderr; cmdline(argc, argv); - if (no_msr) - bic_disable_msr_access(); - - if (no_perf) - bic_disable_perf_access(); - if (!quiet) { print_version(); print_bootcmd(); From aed48c48fa65abdd584e14f7d0273711bc10d223 Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Tue, 30 Jan 2024 23:57:07 +0100 Subject: [PATCH 14/26] tools/power turbostat: add early exits for permission checks Checking early if the permissions are even needed gets rid of the warnings about some of them missing. Earlier we issued a warning in case of missing MSR and/or perf permissions, even when user never asked for counters that require those. Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 66 +++++++++++++++++++++++++-- 1 file changed, 61 insertions(+), 5 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 30db6e92193a..e5e01b58992e 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -5818,6 +5818,14 @@ static int has_instr_count_access(void) return has_access; } +bool is_aperf_access_required(void) +{ + return BIC_IS_ENABLED(BIC_Avg_MHz) + || BIC_IS_ENABLED(BIC_Busy) + || BIC_IS_ENABLED(BIC_Bzy_MHz) + || BIC_IS_ENABLED(BIC_IPC); +} + /* * Linux-perf manages the HW instructions-retired counter * by enabling when requested, and hiding rollover @@ -5833,8 +5841,7 @@ void linux_perf_init(void) err(-1, "calloc fd_instr_count_percpu"); } - const bool aperf_required = BIC_IS_ENABLED(BIC_Avg_MHz) || BIC_IS_ENABLED(BIC_Busy) || - BIC_IS_ENABLED(BIC_Bzy_MHz) || BIC_IS_ENABLED(BIC_IPC); + const bool aperf_required = is_aperf_access_required(); if (aperf_required && has_aperf && amperf_source == AMPERF_SOURCE_PERF) { fd_amperf_percpu = calloc(topo.max_cpu_num + 1, sizeof(*fd_amperf_percpu)); if (fd_amperf_percpu == NULL) @@ -5903,6 +5910,9 @@ static int has_amperf_access_via_perf(void) /* Check if we can access APERF and MPERF */ static int has_amperf_access(void) { + if (!is_aperf_access_required()) + return 0; + if (!no_msr && has_amperf_access_via_msr()) return 1; @@ -6581,7 +6591,8 @@ static void set_amperf_source(void) { amperf_source = AMPERF_SOURCE_PERF; - if (no_perf || !has_amperf_access_via_perf()) + const bool aperf_required = is_aperf_access_required(); + if (no_perf || !aperf_required || !has_amperf_access_via_perf()) amperf_source = AMPERF_SOURCE_MSR; if (quiet || !debug) @@ -6590,8 +6601,51 @@ static void set_amperf_source(void) fprintf(outf, "aperf/mperf source preference: %s\n", amperf_source == AMPERF_SOURCE_MSR ? "msr" : "perf"); } +bool is_msr_access_required(void) +{ + /* TODO: add detection for dynamic counters from add_counter() */ + if (no_msr) + return false; + + return BIC_IS_ENABLED(BIC_SMI) + || BIC_IS_ENABLED(BIC_CPU_c1) + || BIC_IS_ENABLED(BIC_CPU_c3) + || BIC_IS_ENABLED(BIC_CPU_c6) + || BIC_IS_ENABLED(BIC_CPU_c7) + || BIC_IS_ENABLED(BIC_Mod_c6) + || BIC_IS_ENABLED(BIC_CoreTmp) + || BIC_IS_ENABLED(BIC_Totl_c0) + || BIC_IS_ENABLED(BIC_Any_c0) + || BIC_IS_ENABLED(BIC_GFX_c0) + || BIC_IS_ENABLED(BIC_CPUGFX) + || BIC_IS_ENABLED(BIC_Pkgpc3) + || BIC_IS_ENABLED(BIC_Pkgpc6) + || BIC_IS_ENABLED(BIC_Pkgpc2) + || BIC_IS_ENABLED(BIC_Pkgpc7) + || BIC_IS_ENABLED(BIC_Pkgpc8) + || BIC_IS_ENABLED(BIC_Pkgpc9) + || BIC_IS_ENABLED(BIC_Pkgpc10) + || BIC_IS_ENABLED(BIC_CorWatt) + || BIC_IS_ENABLED(BIC_Cor_J) + || BIC_IS_ENABLED(BIC_PkgWatt) + || BIC_IS_ENABLED(BIC_CorWatt) + || BIC_IS_ENABLED(BIC_GFXWatt) + || BIC_IS_ENABLED(BIC_RAMWatt) + || BIC_IS_ENABLED(BIC_Pkg_J) + || BIC_IS_ENABLED(BIC_Cor_J) + || BIC_IS_ENABLED(BIC_GFX_J) + || BIC_IS_ENABLED(BIC_RAM_J) + || BIC_IS_ENABLED(BIC_PKG__) + || BIC_IS_ENABLED(BIC_RAM__) + || BIC_IS_ENABLED(BIC_PkgTmp) + || (is_aperf_access_required() && !has_amperf_access_via_perf()); +} + void check_msr_access(void) { + if (!is_msr_access_required()) + no_msr = 1; + check_dev_msr(); check_msr_permission(); @@ -6601,10 +6655,12 @@ void check_msr_access(void) void check_perf_access(void) { - if (no_perf || !has_instr_count_access()) + const bool intrcount_required = BIC_IS_ENABLED(BIC_IPC); + if (no_perf || !intrcount_required || !has_instr_count_access()) bic_enabled &= ~BIC_IPC; - if (!has_amperf_access()) { + const bool aperf_required = is_aperf_access_required(); + if (!aperf_required || !has_amperf_access()) { bic_enabled &= ~BIC_Avg_MHz; bic_enabled &= ~BIC_Busy; bic_enabled &= ~BIC_Bzy_MHz; From 4a1bb4dad5d16669e841410944e7bc84ef7263fc Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Thu, 14 Mar 2024 11:36:55 +0100 Subject: [PATCH 15/26] tools/power turbostat: Clear added counters when in no-msr mode If user request --no-msr or is not able to access the MSRs, turbostat should clear all the counters added with --add. Because MSR access permission checks are done after the cmdline is parsed, the decision has to be defered up until the transition into no-msr mode happen. Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 47 ++++++++++++++++++++++++++- 1 file changed, 46 insertions(+), 1 deletion(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index e5e01b58992e..b4a892bf22bf 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -1160,6 +1160,37 @@ struct sys_counters { struct msr_counter *pp; } sys; +void free_sys_counters(void) +{ + struct msr_counter *p = sys.tp, *pnext = NULL; + while (p) { + pnext = p->next; + free(p); + p = pnext; + } + + p = sys.cp, pnext = NULL; + while (p) { + pnext = p->next; + free(p); + p = pnext; + } + + p = sys.pp, pnext = NULL; + while (p) { + pnext = p->next; + free(p); + p = pnext; + } + + sys.added_thread_counters = 0; + sys.added_core_counters = 0; + sys.added_package_counters = 0; + sys.tp = NULL; + sys.cp = NULL; + sys.pp = NULL; +} + struct system_summary { struct thread_data threads; struct core_data cores; @@ -1315,6 +1346,8 @@ static void bic_disable_msr_access(void) BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_PkgTmp; bic_enabled &= ~bic_msrs; + + free_sys_counters(); } static long perf_event_open(struct perf_event_attr *hw_event, pid_t pid, int cpu, int group_fd, unsigned long flags) @@ -6601,12 +6634,24 @@ static void set_amperf_source(void) fprintf(outf, "aperf/mperf source preference: %s\n", amperf_source == AMPERF_SOURCE_MSR ? "msr" : "perf"); } +bool has_added_counters(void) +{ + /* + * It only makes sense to call this after the command line is parsed, + * otherwise sys structure is not populated. + */ + + return sys.added_core_counters | sys.added_thread_counters | sys.added_package_counters; +} + bool is_msr_access_required(void) { - /* TODO: add detection for dynamic counters from add_counter() */ if (no_msr) return false; + if (has_added_counters()) + return true; + return BIC_IS_ENABLED(BIC_SMI) || BIC_IS_ENABLED(BIC_CPU_c1) || BIC_IS_ENABLED(BIC_CPU_c3) From ebf8449caba1df2eb6ba0b465fe15dc06d3b9135 Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Thu, 15 Feb 2024 12:50:19 +0100 Subject: [PATCH 16/26] tools/power turbostat: Add proper re-initialization for perf file descriptors Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 26 ++++++++++++++++++++++++++ 1 file changed, 26 insertions(+) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index b4a892bf22bf..a380829c5890 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -3669,18 +3669,25 @@ void free_fd_percpu(void) { int i; + if (!fd_percpu) + return; + for (i = 0; i < topo.max_cpu_num + 1; ++i) { if (fd_percpu[i] != 0) close(fd_percpu[i]); } free(fd_percpu); + fd_percpu = NULL; } void free_fd_amperf_percpu(void) { int i; + if (!fd_amperf_percpu) + return; + for (i = 0; i < topo.max_cpu_num + 1; ++i) { if (fd_amperf_percpu[i].mperf != 0) close(fd_amperf_percpu[i].mperf); @@ -3690,6 +3697,21 @@ void free_fd_amperf_percpu(void) } free(fd_amperf_percpu); + fd_amperf_percpu = NULL; +} + +void free_fd_instr_count_percpu(void) +{ + if (!fd_instr_count_percpu) + return; + + for (int i = 0; i < topo.max_cpu_num + 1; ++i) { + if (fd_instr_count_percpu[i] != 0) + close(fd_instr_count_percpu[i]); + } + + free(fd_instr_count_percpu); + fd_instr_count_percpu = NULL; } void free_all_buffers(void) @@ -3733,6 +3755,7 @@ void free_all_buffers(void) outp = NULL; free_fd_percpu(); + free_fd_instr_count_percpu(); free_fd_amperf_percpu(); free(irq_column_2_cpu); @@ -4067,10 +4090,13 @@ static void update_effective_set(bool startup) err(1, "%s: cpu str malformat %s\n", PATH_EFFECTIVE_CPUS, cpu_effective_str); } +void linux_perf_init(void); + void re_initialize(void) { free_all_buffers(); setup_all_buffers(false); + linux_perf_init(); fprintf(outf, "turbostat: re-initialized with num_cpus %d, allowed_cpus %d\n", topo.num_cpus, topo.allowed_cpus); } From 05a2f07db8883b027c0b4a475fcc586278922b8d Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Tue, 5 Mar 2024 12:27:27 +0100 Subject: [PATCH 17/26] tools/power turbostat: read RAPL counters via perf Some of the future Intel platforms will require reading the RAPL counters via perf and not MSR. On current platforms we can still read them using both ways. Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 786 +++++++++++++++++++++----- 1 file changed, 649 insertions(+), 137 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index a380829c5890..283dffb987b5 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -37,6 +37,7 @@ #include #include #include +#include #define UNUSED(x) (void)(x) @@ -60,6 +61,7 @@ enum counter_scope { SCOPE_CPU, SCOPE_CORE, SCOPE_PACKAGE }; enum counter_type { COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC }; enum counter_format { FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT }; enum amperf_source { AMPERF_SOURCE_PERF, AMPERF_SOURCE_MSR }; +enum rapl_source { RAPL_SOURCE_NONE, RAPL_SOURCE_PERF, RAPL_SOURCE_MSR }; struct msr_counter { unsigned int msr_num; @@ -958,6 +960,175 @@ size_t cpu_present_setsize, cpu_effective_setsize, cpu_allowed_setsize, cpu_affi #define MAX_ADDED_THREAD_COUNTERS 24 #define BITMASK_SIZE 32 +/* Indexes used to map data read from perf and MSRs into global variables */ +enum rapl_rci_index { + RAPL_RCI_INDEX_ENERGY_PKG = 0, + RAPL_RCI_INDEX_ENERGY_CORES = 1, + RAPL_RCI_INDEX_DRAM = 2, + RAPL_RCI_INDEX_GFX = 3, + RAPL_RCI_INDEX_PKG_PERF_STATUS = 4, + RAPL_RCI_INDEX_DRAM_PERF_STATUS = 5, + RAPL_RCI_INDEX_CORE_ENERGY = 6, + NUM_RAPL_COUNTERS, +}; + +enum rapl_unit { + RAPL_UNIT_INVALID, + RAPL_UNIT_JOULES, + RAPL_UNIT_WATTS, +}; + +struct rapl_counter_info_t { + unsigned long long data[NUM_RAPL_COUNTERS]; + enum rapl_source source[NUM_RAPL_COUNTERS]; + unsigned long long flags[NUM_RAPL_COUNTERS]; + double scale[NUM_RAPL_COUNTERS]; + enum rapl_unit unit[NUM_RAPL_COUNTERS]; + + union { + /* Active when source == RAPL_SOURCE_MSR */ + struct { + unsigned long long msr[NUM_RAPL_COUNTERS]; + unsigned long long msr_mask[NUM_RAPL_COUNTERS]; + int msr_shift[NUM_RAPL_COUNTERS]; + }; + }; + + int fd_perf; +}; + +/* struct rapl_counter_info_t for each RAPL domain */ +struct rapl_counter_info_t *rapl_counter_info_perdomain; + +#define RAPL_COUNTER_FLAG_USE_MSR_SUM (1u << 1) + +struct rapl_counter_arch_info { + int feature_mask; /* Mask for testing if the counter is supported on host */ + const char *perf_subsys; + const char *perf_name; + unsigned long long msr; + unsigned long long msr_mask; + int msr_shift; /* Positive mean shift right, negative mean shift left */ + double *platform_rapl_msr_scale; /* Scale applied to values read by MSR (platform dependent, filled at runtime) */ + unsigned int rci_index; /* Maps data from perf counters to global variables */ + unsigned long long bic; + double compat_scale; /* Some counters require constant scaling to be in the same range as other, similar ones */ + unsigned long long flags; +}; + +static const struct rapl_counter_arch_info rapl_counter_arch_infos[] = { + { + .feature_mask = RAPL_PKG, + .perf_subsys = "power", + .perf_name = "energy-pkg", + .msr = MSR_PKG_ENERGY_STATUS, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_energy_units, + .rci_index = RAPL_RCI_INDEX_ENERGY_PKG, + .bic = BIC_PkgWatt | BIC_Pkg_J, + .compat_scale = 1.0, + .flags = RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, + { + .feature_mask = RAPL_AMD_F17H, + .perf_subsys = "power", + .perf_name = "energy-pkg", + .msr = MSR_PKG_ENERGY_STAT, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_energy_units, + .rci_index = RAPL_RCI_INDEX_ENERGY_PKG, + .bic = BIC_PkgWatt | BIC_Pkg_J, + .compat_scale = 1.0, + .flags = RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, + { + .feature_mask = RAPL_CORE_ENERGY_STATUS, + .perf_subsys = "power", + .perf_name = "energy-cores", + .msr = MSR_PP0_ENERGY_STATUS, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_energy_units, + .rci_index = RAPL_RCI_INDEX_ENERGY_CORES, + .bic = BIC_CorWatt | BIC_Cor_J, + .compat_scale = 1.0, + .flags = RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, + { + .feature_mask = RAPL_DRAM, + .perf_subsys = "power", + .perf_name = "energy-ram", + .msr = MSR_DRAM_ENERGY_STATUS, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_dram_energy_units, + .rci_index = RAPL_RCI_INDEX_DRAM, + .bic = BIC_RAMWatt | BIC_RAM_J, + .compat_scale = 1.0, + .flags = RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, + { + .feature_mask = RAPL_GFX, + .perf_subsys = "power", + .perf_name = "energy-gpu", + .msr = MSR_PP1_ENERGY_STATUS, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_energy_units, + .rci_index = RAPL_RCI_INDEX_GFX, + .bic = BIC_GFXWatt | BIC_GFX_J, + .compat_scale = 1.0, + .flags = RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, + { + .feature_mask = RAPL_PKG_PERF_STATUS, + .perf_subsys = NULL, + .perf_name = NULL, + .msr = MSR_PKG_PERF_STATUS, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_time_units, + .rci_index = RAPL_RCI_INDEX_PKG_PERF_STATUS, + .bic = BIC_PKG__, + .compat_scale = 100.0, + .flags = RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, + { + .feature_mask = RAPL_DRAM_PERF_STATUS, + .perf_subsys = NULL, + .perf_name = NULL, + .msr = MSR_DRAM_PERF_STATUS, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_time_units, + .rci_index = RAPL_RCI_INDEX_DRAM_PERF_STATUS, + .bic = BIC_RAM__, + .compat_scale = 100.0, + .flags = RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, + { + .feature_mask = RAPL_AMD_F17H, + .perf_subsys = NULL, + .perf_name = NULL, + .msr = MSR_CORE_ENERGY_STAT, + .msr_mask = 0xFFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_energy_units, + .rci_index = RAPL_RCI_INDEX_CORE_ENERGY, + .bic = BIC_CorWatt | BIC_Cor_J, + .compat_scale = 1.0, + .flags = 0, + }, +}; + +struct rapl_counter { + unsigned long long raw_value; + enum rapl_unit unit; + double scale; +}; + struct thread_data { struct timeval tv_begin; struct timeval tv_end; @@ -984,7 +1155,7 @@ struct core_data { unsigned long long c7; unsigned long long mc6_us; /* duplicate as per-core for now, even though per module */ unsigned int core_temp_c; - unsigned int core_energy; /* MSR_CORE_ENERGY_STAT */ + struct rapl_counter core_energy; /* MSR_CORE_ENERGY_STAT */ unsigned int core_id; unsigned long long core_throt_cnt; unsigned long long counter[MAX_ADDED_COUNTERS]; @@ -1009,12 +1180,12 @@ struct pkg_data { unsigned int gfx_mhz; unsigned int gfx_act_mhz; unsigned int package_id; - unsigned long long energy_pkg; /* MSR_PKG_ENERGY_STATUS */ - unsigned long long energy_dram; /* MSR_DRAM_ENERGY_STATUS */ - unsigned long long energy_cores; /* MSR_PP0_ENERGY_STATUS */ - unsigned long long energy_gfx; /* MSR_PP1_ENERGY_STATUS */ - unsigned long long rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */ - unsigned long long rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */ + struct rapl_counter energy_pkg; /* MSR_PKG_ENERGY_STATUS */ + struct rapl_counter energy_dram; /* MSR_DRAM_ENERGY_STATUS */ + struct rapl_counter energy_cores; /* MSR_PP0_ENERGY_STATUS */ + struct rapl_counter energy_gfx; /* MSR_PP1_ENERGY_STATUS */ + struct rapl_counter rapl_pkg_perf_status; /* MSR_PKG_PERF_STATUS */ + struct rapl_counter rapl_dram_perf_status; /* MSR_DRAM_PERF_STATUS */ unsigned int pkg_temp_c; unsigned int uncore_mhz; unsigned long long counter[MAX_ADDED_COUNTERS]; @@ -1403,6 +1574,21 @@ int get_msr(int cpu, off_t offset, unsigned long long *msr) return 0; } +int probe_msr(int cpu, off_t offset) +{ + ssize_t retval; + unsigned long long dummy; + + assert(!no_msr); + + retval = pread(get_msr_fd(cpu), &dummy, sizeof(dummy), offset); + + if (retval != sizeof(dummy)) + return 1; + + return 0; +} + #define MAX_DEFERRED 16 char *deferred_add_names[MAX_DEFERRED]; char *deferred_skip_names[MAX_DEFERRED]; @@ -1755,7 +1941,11 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p outp += sprintf(outp, "c7: %016llX\n", c->c7); outp += sprintf(outp, "DTS: %dC\n", c->core_temp_c); outp += sprintf(outp, "cpu_throt_count: %016llX\n", c->core_throt_cnt); - outp += sprintf(outp, "Joules: %0X\n", c->core_energy); + + const unsigned long long energy_value = c->core_energy.raw_value * c->core_energy.scale; + const double energy_scale = c->core_energy.scale; + if (c->core_energy.unit == RAPL_UNIT_JOULES) + outp += sprintf(outp, "Joules: %0llX (scale: %lf)\n", energy_value, energy_scale); for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { outp += @@ -1785,12 +1975,12 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p outp += sprintf(outp, "pc10: %016llX\n", p->pc10); outp += sprintf(outp, "cpu_lpi: %016llX\n", p->cpu_lpi); outp += sprintf(outp, "sys_lpi: %016llX\n", p->sys_lpi); - outp += sprintf(outp, "Joules PKG: %0llX\n", p->energy_pkg); - outp += sprintf(outp, "Joules COR: %0llX\n", p->energy_cores); - outp += sprintf(outp, "Joules GFX: %0llX\n", p->energy_gfx); - outp += sprintf(outp, "Joules RAM: %0llX\n", p->energy_dram); - outp += sprintf(outp, "Throttle PKG: %0llX\n", p->rapl_pkg_perf_status); - outp += sprintf(outp, "Throttle RAM: %0llX\n", p->rapl_dram_perf_status); + outp += sprintf(outp, "Joules PKG: %0llX\n", p->energy_pkg.raw_value); + outp += sprintf(outp, "Joules COR: %0llX\n", p->energy_cores.raw_value); + outp += sprintf(outp, "Joules GFX: %0llX\n", p->energy_gfx.raw_value); + outp += sprintf(outp, "Joules RAM: %0llX\n", p->energy_dram.raw_value); + outp += sprintf(outp, "Throttle PKG: %0llX\n", p->rapl_pkg_perf_status.raw_value); + outp += sprintf(outp, "Throttle RAM: %0llX\n", p->rapl_dram_perf_status.raw_value); outp += sprintf(outp, "PTM: %dC\n", p->pkg_temp_c); for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { @@ -1805,6 +1995,23 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p return 0; } +double rapl_counter_get_value(const struct rapl_counter *c, enum rapl_unit desired_unit, double interval) +{ + assert(desired_unit != RAPL_UNIT_INVALID); + + /* + * For now we don't expect anything other than joules, + * so just simplify the logic. + */ + assert(c->unit == RAPL_UNIT_JOULES); + + const double scaled = c->raw_value * c->scale; + + if (desired_unit == RAPL_UNIT_WATTS) + return scaled / interval; + return scaled; +} + /* * column formatting convention & formats */ @@ -1998,9 +2205,11 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data if (DO_BIC(BIC_CorWatt) && platform->has_per_core_rapl) outp += - sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units / interval_float); + sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&c->core_energy, RAPL_UNIT_WATTS, interval_float)); if (DO_BIC(BIC_Cor_J) && platform->has_per_core_rapl) - outp += sprintf(outp, fmt8, (printed++ ? delim : ""), c->core_energy * rapl_energy_units); + outp += sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&c->core_energy, RAPL_UNIT_JOULES, interval_float)); /* print per-package data only for 1st core in package */ if (!is_cpu_first_core_in_package(t, c, p)) @@ -2072,34 +2281,40 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data if (DO_BIC(BIC_PkgWatt)) outp += - sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units / interval_float); - + sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&p->energy_pkg, RAPL_UNIT_WATTS, interval_float)); if (DO_BIC(BIC_CorWatt) && !platform->has_per_core_rapl) outp += - sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units / interval_float); + sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&p->energy_cores, RAPL_UNIT_WATTS, interval_float)); if (DO_BIC(BIC_GFXWatt)) outp += - sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units / interval_float); + sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&p->energy_gfx, RAPL_UNIT_WATTS, interval_float)); if (DO_BIC(BIC_RAMWatt)) outp += sprintf(outp, fmt8, (printed++ ? delim : ""), - p->energy_dram * rapl_dram_energy_units / interval_float); + rapl_counter_get_value(&p->energy_dram, RAPL_UNIT_WATTS, interval_float)); if (DO_BIC(BIC_Pkg_J)) - outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_pkg * rapl_energy_units); + outp += sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&p->energy_pkg, RAPL_UNIT_JOULES, interval_float)); if (DO_BIC(BIC_Cor_J) && !platform->has_per_core_rapl) - outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_cores * rapl_energy_units); + outp += sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&p->energy_cores, RAPL_UNIT_JOULES, interval_float)); if (DO_BIC(BIC_GFX_J)) - outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_gfx * rapl_energy_units); + outp += sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&p->energy_gfx, RAPL_UNIT_JOULES, interval_float)); if (DO_BIC(BIC_RAM_J)) - outp += sprintf(outp, fmt8, (printed++ ? delim : ""), p->energy_dram * rapl_dram_energy_units); + outp += sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&p->energy_dram, RAPL_UNIT_JOULES, interval_float)); if (DO_BIC(BIC_PKG__)) outp += sprintf(outp, fmt8, (printed++ ? delim : ""), - 100.0 * p->rapl_pkg_perf_status * rapl_time_units / interval_float); + rapl_counter_get_value(&p->rapl_pkg_perf_status, RAPL_UNIT_WATTS, interval_float)); if (DO_BIC(BIC_RAM__)) outp += sprintf(outp, fmt8, (printed++ ? delim : ""), - 100.0 * p->rapl_dram_perf_status * rapl_time_units / interval_float); + rapl_counter_get_value(&p->rapl_dram_perf_status, RAPL_UNIT_WATTS, interval_float)); /* UncMHz */ if (DO_BIC(BIC_UNCORE_MHZ)) outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->uncore_mhz); @@ -2208,12 +2423,13 @@ int delta_package(struct pkg_data *new, struct pkg_data *old) old->gfx_mhz = new->gfx_mhz; old->gfx_act_mhz = new->gfx_act_mhz; - old->energy_pkg = new->energy_pkg - old->energy_pkg; - old->energy_cores = new->energy_cores - old->energy_cores; - old->energy_gfx = new->energy_gfx - old->energy_gfx; - old->energy_dram = new->energy_dram - old->energy_dram; - old->rapl_pkg_perf_status = new->rapl_pkg_perf_status - old->rapl_pkg_perf_status; - old->rapl_dram_perf_status = new->rapl_dram_perf_status - old->rapl_dram_perf_status; + old->energy_pkg.raw_value = new->energy_pkg.raw_value - old->energy_pkg.raw_value; + old->energy_cores.raw_value = new->energy_cores.raw_value - old->energy_cores.raw_value; + old->energy_gfx.raw_value = new->energy_gfx.raw_value - old->energy_gfx.raw_value; + old->energy_dram.raw_value = new->energy_dram.raw_value - old->energy_dram.raw_value; + old->rapl_pkg_perf_status.raw_value = new->rapl_pkg_perf_status.raw_value - old->rapl_pkg_perf_status.raw_value; + old->rapl_dram_perf_status.raw_value = + new->rapl_dram_perf_status.raw_value - old->rapl_dram_perf_status.raw_value; for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { if (mp->format == FORMAT_RAW) @@ -2237,7 +2453,7 @@ void delta_core(struct core_data *new, struct core_data *old) old->core_throt_cnt = new->core_throt_cnt; old->mc6_us = new->mc6_us - old->mc6_us; - DELTA_WRAP32(new->core_energy, old->core_energy); + DELTA_WRAP32(new->core_energy.raw_value, old->core_energy.raw_value); for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { if (mp->format == FORMAT_RAW) @@ -2364,6 +2580,13 @@ int delta_cpu(struct thread_data *t, struct core_data *c, return retval; } +void rapl_counter_clear(struct rapl_counter *c) +{ + c->raw_value = 0; + c->scale = 0.0; + c->unit = RAPL_UNIT_INVALID; +} + void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { int i; @@ -2391,7 +2614,7 @@ void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data c->c7 = 0; c->mc6_us = 0; c->core_temp_c = 0; - c->core_energy = 0; + rapl_counter_clear(&c->core_energy); c->core_throt_cnt = 0; p->pkg_wtd_core_c0 = 0; @@ -2412,12 +2635,12 @@ void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data p->cpu_lpi = 0; p->sys_lpi = 0; - p->energy_pkg = 0; - p->energy_dram = 0; - p->energy_cores = 0; - p->energy_gfx = 0; - p->rapl_pkg_perf_status = 0; - p->rapl_dram_perf_status = 0; + rapl_counter_clear(&p->energy_pkg); + rapl_counter_clear(&p->energy_dram); + rapl_counter_clear(&p->energy_cores); + rapl_counter_clear(&p->energy_gfx); + rapl_counter_clear(&p->rapl_pkg_perf_status); + rapl_counter_clear(&p->rapl_dram_perf_status); p->pkg_temp_c = 0; p->gfx_rc6_ms = 0; @@ -2434,6 +2657,20 @@ void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data p->counter[i] = 0; } +void rapl_counter_accumulate(struct rapl_counter *dst, const struct rapl_counter *src) +{ + /* Copy unit and scale from src if dst is not initialized */ + if (dst->unit == RAPL_UNIT_INVALID) { + dst->unit = src->unit; + dst->scale = src->scale; + } + + assert(dst->unit == src->unit); + assert(dst->scale == src->scale); + + dst->raw_value += src->raw_value; +} + int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { int i; @@ -2480,7 +2717,7 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.cores.core_temp_c = MAX(average.cores.core_temp_c, c->core_temp_c); average.cores.core_throt_cnt = MAX(average.cores.core_throt_cnt, c->core_throt_cnt); - average.cores.core_energy += c->core_energy; + rapl_counter_accumulate(&average.cores.core_energy, &c->core_energy); for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { if (mp->format == FORMAT_RAW) @@ -2515,10 +2752,10 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.packages.cpu_lpi = p->cpu_lpi; average.packages.sys_lpi = p->sys_lpi; - average.packages.energy_pkg += p->energy_pkg; - average.packages.energy_dram += p->energy_dram; - average.packages.energy_cores += p->energy_cores; - average.packages.energy_gfx += p->energy_gfx; + rapl_counter_accumulate(&average.packages.energy_pkg, &p->energy_pkg); + rapl_counter_accumulate(&average.packages.energy_dram, &p->energy_dram); + rapl_counter_accumulate(&average.packages.energy_cores, &p->energy_cores); + rapl_counter_accumulate(&average.packages.energy_gfx, &p->energy_gfx); average.packages.gfx_rc6_ms = p->gfx_rc6_ms; average.packages.uncore_mhz = p->uncore_mhz; @@ -2527,8 +2764,8 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c); - average.packages.rapl_pkg_perf_status += p->rapl_pkg_perf_status; - average.packages.rapl_dram_perf_status += p->rapl_dram_perf_status; + rapl_counter_accumulate(&average.packages.rapl_pkg_perf_status, &p->rapl_pkg_perf_status); + rapl_counter_accumulate(&average.packages.rapl_dram_perf_status, &p->rapl_dram_perf_status); for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { if ((mp->format == FORMAT_RAW) && (topo.num_packages == 0)) @@ -2797,25 +3034,53 @@ struct amperf_group_fd { int mperf; }; -static unsigned int read_perf_counter_info(const char *const path, const char *const parse_format) +static int read_perf_counter_info(const char *const path, const char *const parse_format, void *value_ptr) { int fdmt; - char buf[16]; - unsigned int v; + int bytes_read; + char buf[64]; + int ret = -1; fdmt = open(path, O_RDONLY, 0); - if (fdmt == -1) - errx(1, "Failed to read perf counter info %s\n", path); + if (fdmt == -1) { + if (debug) + fprintf(stderr, "Failed to parse perf counter info %s\n", path); + ret = -1; + goto cleanup_and_exit; + } - if (read(fdmt, buf, sizeof(buf)) <= 0) - return 0; + bytes_read = read(fdmt, buf, sizeof(buf) - 1); + if (bytes_read <= 0 || bytes_read >= (int)sizeof(buf)) { + if (debug) + fprintf(stderr, "Failed to parse perf counter info %s\n", path); + ret = -1; + goto cleanup_and_exit; + } - buf[sizeof(buf) - 1] = '\0'; + buf[bytes_read] = '\0'; - if (sscanf(buf, parse_format, &v) != 1) - errx(1, "Failed to parse perf counter info %s\n", path); + if (sscanf(buf, parse_format, value_ptr) != 1) { + if (debug) + fprintf(stderr, "Failed to parse perf counter info %s\n", path); + ret = -1; + goto cleanup_and_exit; + } + ret = 0; + +cleanup_and_exit: close(fdmt); + return ret; +} + +static unsigned int read_perf_counter_info_n(const char *const path, const char *const parse_format) +{ + unsigned int v; + int status; + + status = read_perf_counter_info(path, parse_format, &v); + if (status) + v = -1; return v; } @@ -2825,7 +3090,7 @@ static unsigned read_msr_type(void) const char *const path = "/sys/bus/event_source/devices/msr/type"; const char *const format = "%u"; - return read_perf_counter_info(path, format); + return read_perf_counter_info_n(path, format); } static unsigned read_aperf_config(void) @@ -2833,7 +3098,7 @@ static unsigned read_aperf_config(void) const char *const path = "/sys/bus/event_source/devices/msr/events/aperf"; const char *const format = "event=%x"; - return read_perf_counter_info(path, format); + return read_perf_counter_info_n(path, format); } static unsigned read_mperf_config(void) @@ -2841,7 +3106,60 @@ static unsigned read_mperf_config(void) const char *const path = "/sys/bus/event_source/devices/msr/events/mperf"; const char *const format = "event=%x"; - return read_perf_counter_info(path, format); + return read_perf_counter_info_n(path, format); +} + +static unsigned read_perf_type(const char *subsys) +{ + const char *const path_format = "/sys/bus/event_source/devices/%s/type"; + const char *const format = "%u"; + char path[128]; + + snprintf(path, sizeof(path), path_format, subsys); + + return read_perf_counter_info_n(path, format); +} + +static unsigned read_rapl_config(const char *subsys, const char *event_name) +{ + const char *const path_format = "/sys/bus/event_source/devices/%s/events/%s"; + const char *const format = "event=%x"; + char path[128]; + + snprintf(path, sizeof(path), path_format, subsys, event_name); + + return read_perf_counter_info_n(path, format); +} + +static unsigned read_perf_rapl_unit(const char *subsys, const char *event_name) +{ + const char *const path_format = "/sys/bus/event_source/devices/%s/events/%s.unit"; + const char *const format = "%s"; + char path[128]; + char unit_buffer[16]; + + snprintf(path, sizeof(path), path_format, subsys, event_name); + + read_perf_counter_info(path, format, &unit_buffer); + if (strcmp("Joules", unit_buffer) == 0) + return RAPL_UNIT_JOULES; + + return RAPL_UNIT_INVALID; +} + +static double read_perf_rapl_scale(const char *subsys, const char *event_name) +{ + const char *const path_format = "/sys/bus/event_source/devices/%s/events/%s.scale"; + const char *const format = "%lf"; + char path[128]; + double scale; + + snprintf(path, sizeof(path), path_format, subsys, event_name); + + if (read_perf_counter_info(path, format, &scale)) + return 0.0; + + return scale; } static struct amperf_group_fd open_amperf_fd(int cpu) @@ -2961,6 +3279,99 @@ retry: return 0; } +size_t rapl_counter_info_count_perf(const struct rapl_counter_info_t *rci) +{ + size_t ret = 0; + + for (int i = 0; i < NUM_RAPL_COUNTERS; ++i) + if (rci->source[i] == RAPL_SOURCE_PERF) + ++ret; + + return ret; +} + +void write_rapl_counter(struct rapl_counter *rc, struct rapl_counter_info_t *rci, unsigned int idx) +{ + rc->raw_value = rci->data[idx]; + rc->unit = rci->unit[idx]; + rc->scale = rci->scale[idx]; +} + +int get_rapl_counters(int cpu, int domain, struct core_data *c, struct pkg_data *p) +{ + unsigned long long perf_data[NUM_RAPL_COUNTERS + 1]; + struct rapl_counter_info_t *rci = &rapl_counter_info_perdomain[domain]; + + if (debug) + fprintf(stderr, "get_rapl_counters: cpu%d domain%d\n", cpu, domain); + + assert(rapl_counter_info_perdomain); + + /* + * If we have any perf counters to read, read them all now, in bulk + */ + if (rci->fd_perf != -1) { + size_t num_perf_counters = rapl_counter_info_count_perf(rci); + const ssize_t expected_read_size = (num_perf_counters + 1) * sizeof(unsigned long long); + const ssize_t actual_read_size = read(rci->fd_perf, &perf_data[0], sizeof(perf_data)); + if (actual_read_size != expected_read_size) + err(-1, "get_rapl_counters: failed to read perf_data (%zu %zu)", expected_read_size, + actual_read_size); + } + + for (unsigned int i = 0, pi = 1; i < NUM_RAPL_COUNTERS; ++i) { + switch (rci->source[i]) { + case RAPL_SOURCE_NONE: + break; + + case RAPL_SOURCE_PERF: + assert(pi < ARRAY_SIZE(perf_data)); + assert(rci->fd_perf != -1); + + if (debug) + fprintf(stderr, "Reading rapl counter via perf at %u (%llu %e %lf)\n", + i, perf_data[pi], rci->scale[i], perf_data[pi] * rci->scale[i]); + + rci->data[i] = perf_data[pi]; + + ++pi; + break; + + case RAPL_SOURCE_MSR: + if (debug) + fprintf(stderr, "Reading rapl counter via msr at %u\n", i); + + assert(!no_msr); + if (rci->flags[i] & RAPL_COUNTER_FLAG_USE_MSR_SUM) { + if (get_msr_sum(cpu, rci->msr[i], &rci->data[i])) + return -13 - i; + } else { + if (get_msr(cpu, rci->msr[i], &rci->data[i])) + return -13 - i; + } + + rci->data[i] &= rci->msr_mask[i]; + if (rci->msr_shift[i] >= 0) + rci->data[i] >>= abs(rci->msr_shift[i]); + else + rci->data[i] <<= abs(rci->msr_shift[i]); + + break; + } + } + + _Static_assert(NUM_RAPL_COUNTERS == 7); + write_rapl_counter(&p->energy_pkg, rci, RAPL_RCI_INDEX_ENERGY_PKG); + write_rapl_counter(&p->energy_cores, rci, RAPL_RCI_INDEX_ENERGY_CORES); + write_rapl_counter(&p->energy_dram, rci, RAPL_RCI_INDEX_DRAM); + write_rapl_counter(&p->energy_gfx, rci, RAPL_RCI_INDEX_GFX); + write_rapl_counter(&p->rapl_pkg_perf_status, rci, RAPL_RCI_INDEX_PKG_PERF_STATUS); + write_rapl_counter(&p->rapl_dram_perf_status, rci, RAPL_RCI_INDEX_DRAM_PERF_STATUS); + write_rapl_counter(&c->core_energy, rci, RAPL_RCI_INDEX_CORE_ENERGY); + + return 0; +} + /* * get_counters(...) * migrate to cpu @@ -2972,6 +3383,7 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) unsigned long long msr; struct msr_counter *mp; int i; + int status; if (cpu_migrate(cpu)) { fprintf(outf, "get_counters: Could not migrate to CPU %d\n", cpu); @@ -3029,6 +3441,12 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) if (!is_cpu_first_thread_in_core(t, c, p)) goto done; + if (platform->has_per_core_rapl) { + status = get_rapl_counters(cpu, c->core_id, c, p); + if (status != 0) + return status; + } + if (DO_BIC(BIC_CPU_c3) || soft_c1_residency_display(BIC_CPU_c3)) { if (get_msr(cpu, MSR_CORE_C3_RESIDENCY, &c->c3)) return -6; @@ -3069,12 +3487,6 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) if (DO_BIC(BIC_CORE_THROT_CNT)) get_core_throt_cnt(cpu, &c->core_throt_cnt); - if ((platform->rapl_msrs & RAPL_AMD_F17H) && !no_msr) { - if (get_msr(cpu, MSR_CORE_ENERGY_STAT, &msr)) - return -14; - c->core_energy = msr & 0xFFFFFFFF; - } - for (i = 0, mp = sys.cp; mp; i++, mp = mp->next) { if (get_mp(cpu, mp, &c->counter[i])) return -10; @@ -3134,42 +3546,10 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) if (DO_BIC(BIC_SYS_LPI)) p->sys_lpi = cpuidle_cur_sys_lpi_us; - if (!no_msr) { - if (platform->rapl_msrs & RAPL_PKG) { - if (get_msr_sum(cpu, MSR_PKG_ENERGY_STATUS, &msr)) - return -13; - p->energy_pkg = msr; - } - if (platform->rapl_msrs & RAPL_CORE_ENERGY_STATUS) { - if (get_msr_sum(cpu, MSR_PP0_ENERGY_STATUS, &msr)) - return -14; - p->energy_cores = msr; - } - if (platform->rapl_msrs & RAPL_DRAM) { - if (get_msr_sum(cpu, MSR_DRAM_ENERGY_STATUS, &msr)) - return -15; - p->energy_dram = msr; - } - if (platform->rapl_msrs & RAPL_GFX) { - if (get_msr_sum(cpu, MSR_PP1_ENERGY_STATUS, &msr)) - return -16; - p->energy_gfx = msr; - } - if (platform->rapl_msrs & RAPL_PKG_PERF_STATUS) { - if (get_msr_sum(cpu, MSR_PKG_PERF_STATUS, &msr)) - return -16; - p->rapl_pkg_perf_status = msr; - } - if (platform->rapl_msrs & RAPL_DRAM_PERF_STATUS) { - if (get_msr_sum(cpu, MSR_DRAM_PERF_STATUS, &msr)) - return -16; - p->rapl_dram_perf_status = msr; - } - if (platform->rapl_msrs & RAPL_AMD_F17H) { - if (get_msr_sum(cpu, MSR_PKG_ENERGY_STAT, &msr)) - return -13; - p->energy_pkg = msr; - } + if (!platform->has_per_core_rapl) { + status = get_rapl_counters(cpu, p->package_id, c, p); + if (status != 0) + return status; } if (DO_BIC(BIC_PkgTmp)) { @@ -3714,6 +4094,21 @@ void free_fd_instr_count_percpu(void) fd_instr_count_percpu = NULL; } +void free_fd_rapl_percpu(void) +{ + if (!rapl_counter_info_perdomain) + return; + + const int num_domains = platform->has_per_core_rapl ? topo.num_cores : topo.num_packages; + + for (int domain_id = 0; domain_id < num_domains; ++domain_id) { + if (rapl_counter_info_perdomain[domain_id].fd_perf != -1) + close(rapl_counter_info_perdomain[domain_id].fd_perf); + } + + free(rapl_counter_info_perdomain); +} + void free_all_buffers(void) { int i; @@ -3757,6 +4152,7 @@ void free_all_buffers(void) free_fd_percpu(); free_fd_instr_count_percpu(); free_fd_amperf_percpu(); + free_fd_rapl_percpu(); free(irq_column_2_cpu); free(irqs_per_cpu); @@ -4091,12 +4487,14 @@ static void update_effective_set(bool startup) } void linux_perf_init(void); +void rapl_perf_init(void); void re_initialize(void) { free_all_buffers(); setup_all_buffers(false); linux_perf_init(); + rapl_perf_init(); fprintf(outf, "turbostat: re-initialized with num_cpus %d, allowed_cpus %d\n", topo.num_cpus, topo.allowed_cpus); } @@ -5315,31 +5713,18 @@ void rapl_probe_intel(void) unsigned long long msr; unsigned int time_unit; double tdp; + const unsigned long long bic_watt_bits = BIC_PkgWatt | BIC_CorWatt | BIC_RAMWatt | BIC_GFXWatt; + const unsigned long long bic_joules_bits = BIC_Pkg_J | BIC_Cor_J | BIC_RAM_J | BIC_GFX_J; - if (rapl_joules) { - if (platform->rapl_msrs & RAPL_PKG_ENERGY_STATUS) - BIC_PRESENT(BIC_Pkg_J); - if (platform->rapl_msrs & RAPL_CORE_ENERGY_STATUS) - BIC_PRESENT(BIC_Cor_J); - if (platform->rapl_msrs & RAPL_DRAM_ENERGY_STATUS) - BIC_PRESENT(BIC_RAM_J); - if (platform->rapl_msrs & RAPL_GFX_ENERGY_STATUS) - BIC_PRESENT(BIC_GFX_J); - } else { - if (platform->rapl_msrs & RAPL_PKG_ENERGY_STATUS) - BIC_PRESENT(BIC_PkgWatt); - if (platform->rapl_msrs & RAPL_CORE_ENERGY_STATUS) - BIC_PRESENT(BIC_CorWatt); - if (platform->rapl_msrs & RAPL_DRAM_ENERGY_STATUS) - BIC_PRESENT(BIC_RAMWatt); - if (platform->rapl_msrs & RAPL_GFX_ENERGY_STATUS) - BIC_PRESENT(BIC_GFXWatt); - } + if (rapl_joules) + bic_enabled &= ~bic_watt_bits; + else + bic_enabled &= ~bic_joules_bits; - if (platform->rapl_msrs & RAPL_PKG_PERF_STATUS) - BIC_PRESENT(BIC_PKG__); - if (platform->rapl_msrs & RAPL_DRAM_PERF_STATUS) - BIC_PRESENT(BIC_RAM__); + if (!(platform->rapl_msrs & RAPL_PKG_PERF_STATUS)) + bic_enabled &= ~BIC_PKG__; + if (!(platform->rapl_msrs & RAPL_DRAM_PERF_STATUS)) + bic_enabled &= ~BIC_RAM__; /* units on package 0, verify later other packages match */ if (get_msr(base_cpu, MSR_RAPL_POWER_UNIT, &msr)) @@ -5373,14 +5758,13 @@ void rapl_probe_amd(void) { unsigned long long msr; double tdp; + const unsigned long long bic_watt_bits = BIC_PkgWatt | BIC_CorWatt; + const unsigned long long bic_joules_bits = BIC_Pkg_J | BIC_Cor_J; - if (rapl_joules) { - BIC_PRESENT(BIC_Pkg_J); - BIC_PRESENT(BIC_Cor_J); - } else { - BIC_PRESENT(BIC_PkgWatt); - BIC_PRESENT(BIC_CorWatt); - } + if (rapl_joules) + bic_enabled &= ~bic_watt_bits; + else + bic_enabled &= ~bic_joules_bits; if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr)) return; @@ -5885,6 +6269,48 @@ bool is_aperf_access_required(void) || BIC_IS_ENABLED(BIC_IPC); } +int add_rapl_perf_counter_(int cpu, struct rapl_counter_info_t *rci, const struct rapl_counter_arch_info *cai, + double *scale_, enum rapl_unit *unit_) +{ + if (no_perf) + return -1; + + const double scale = read_perf_rapl_scale(cai->perf_subsys, cai->perf_name); + if (scale == 0.0) + return -1; + + const enum rapl_unit unit = read_perf_rapl_unit(cai->perf_subsys, cai->perf_name); + if (unit == RAPL_UNIT_INVALID) + return -1; + + const unsigned rapl_type = read_perf_type(cai->perf_subsys); + const unsigned rapl_energy_pkg_config = read_rapl_config(cai->perf_subsys, cai->perf_name); + + const int fd_counter = + open_perf_counter(cpu, rapl_type, rapl_energy_pkg_config, rci->fd_perf, PERF_FORMAT_GROUP); + if (fd_counter == -1) + return -1; + + /* If it's the first counter opened, make it a group descriptor */ + if (rci->fd_perf == -1) + rci->fd_perf = fd_counter; + + *scale_ = scale; + *unit_ = unit; + return fd_counter; +} + +int add_rapl_perf_counter(int cpu, struct rapl_counter_info_t *rci, const struct rapl_counter_arch_info *cai, + double *scale, enum rapl_unit *unit) +{ + int ret = add_rapl_perf_counter_(cpu, rci, cai, scale, unit); + + if (debug) + fprintf(stderr, "add_rapl_perf_counter: %d (cpu: %d)\n", ret, cpu); + + return ret; +} + /* * Linux-perf manages the HW instructions-retired counter * by enabling when requested, and hiding rollover @@ -5908,17 +6334,101 @@ void linux_perf_init(void) } } +void rapl_perf_init(void) +{ + const int num_domains = platform->has_per_core_rapl ? topo.num_cores : topo.num_packages; + bool *domain_visited = calloc(num_domains, sizeof(bool)); + + rapl_counter_info_perdomain = calloc(num_domains, sizeof(*rapl_counter_info_perdomain)); + if (rapl_counter_info_perdomain == NULL) + err(-1, "calloc rapl_counter_info_percpu"); + + /* + * Initialize rapl_counter_info_percpu + */ + for (int domain_id = 0; domain_id < num_domains; ++domain_id) { + struct rapl_counter_info_t *rci = &rapl_counter_info_perdomain[domain_id]; + rci->fd_perf = -1; + for (size_t i = 0; i < NUM_RAPL_COUNTERS; ++i) { + rci->data[i] = 0; + rci->source[i] = RAPL_SOURCE_NONE; + } + } + + /* + * Open/probe the counters + * If can't get it via perf, fallback to MSR + */ + for (size_t i = 0; i < ARRAY_SIZE(rapl_counter_arch_infos); ++i) { + + const struct rapl_counter_arch_info *const cai = &rapl_counter_arch_infos[i]; + bool has_counter = 0; + double scale; + enum rapl_unit unit; + int next_domain; + + memset(domain_visited, 0, num_domains * sizeof(*domain_visited)); + + for (int cpu = 0; cpu < topo.max_cpu_num + 1; ++cpu) { + + if (cpu_is_not_allowed(cpu)) + continue; + + /* Skip already seen and handled RAPL domains */ + next_domain = + platform->has_per_core_rapl ? cpus[cpu].physical_core_id : cpus[cpu].physical_package_id; + + if (domain_visited[next_domain]) + continue; + + domain_visited[next_domain] = 1; + + struct rapl_counter_info_t *rci = &rapl_counter_info_perdomain[next_domain]; + + /* Check if the counter is enabled and accessible */ + if (BIC_IS_ENABLED(cai->bic) && (platform->rapl_msrs & cai->feature_mask)) { + + /* Use perf API for this counter */ + if (!no_perf && cai->perf_name + && add_rapl_perf_counter(cpu, rci, cai, &scale, &unit) != -1) { + rci->source[cai->rci_index] = RAPL_SOURCE_PERF; + rci->scale[cai->rci_index] = scale * cai->compat_scale; + rci->unit[cai->rci_index] = unit; + rci->flags[cai->rci_index] = cai->flags; + + /* Use MSR for this counter */ + } else if (!no_msr && cai->msr && probe_msr(cpu, cai->msr) == 0) { + rci->source[cai->rci_index] = RAPL_SOURCE_MSR; + rci->msr[cai->rci_index] = cai->msr; + rci->msr_mask[cai->rci_index] = cai->msr_mask; + rci->msr_shift[cai->rci_index] = cai->msr_shift; + rci->unit[cai->rci_index] = RAPL_UNIT_JOULES; + rci->scale[cai->rci_index] = *cai->platform_rapl_msr_scale * cai->compat_scale; + rci->flags[cai->rci_index] = cai->flags; + } + } + + if (rci->source[cai->rci_index] != RAPL_SOURCE_NONE) + has_counter = 1; + } + + /* If any CPU has access to the counter, make it present */ + if (has_counter) + BIC_PRESENT(cai->bic); + } + + free(domain_visited); +} + static int has_amperf_access_via_msr(void) { - unsigned long long dummy; - if (no_msr) return 0; - if (get_msr(base_cpu, MSR_IA32_APERF, &dummy)) + if (probe_msr(base_cpu, MSR_IA32_APERF)) return 0; - if (get_msr(base_cpu, MSR_IA32_MPERF, &dummy)) + if (probe_msr(base_cpu, MSR_IA32_MPERF)) return 0; return 1; @@ -6696,6 +7206,7 @@ bool is_msr_access_required(void) || BIC_IS_ENABLED(BIC_Pkgpc8) || BIC_IS_ENABLED(BIC_Pkgpc9) || BIC_IS_ENABLED(BIC_Pkgpc10) + /* TODO: Multiplex access with perf */ || BIC_IS_ENABLED(BIC_CorWatt) || BIC_IS_ENABLED(BIC_Cor_J) || BIC_IS_ENABLED(BIC_PkgWatt) @@ -6749,6 +7260,7 @@ void turbostat_init() probe_pm_features(); set_amperf_source(); linux_perf_init(); + rapl_perf_init(); for_all_cpus(get_cpu_type, ODD_COUNTERS); for_all_cpus(get_cpu_type, EVEN_COUNTERS); From 17d1ea136be86f53be0461b0c33daf6b58e6cbf7 Mon Sep 17 00:00:00 2001 From: Patryk Wlazlyn Date: Thu, 7 Mar 2024 17:00:35 +0100 Subject: [PATCH 18/26] tools/power turbostat: Add selftests Signed-off-by: Patryk Wlazlyn Signed-off-by: Len Brown --- .../testing/selftests/turbostat/defcolumns.py | 59 +++++++++++++++++++ 1 file changed, 59 insertions(+) create mode 100755 tools/testing/selftests/turbostat/defcolumns.py diff --git a/tools/testing/selftests/turbostat/defcolumns.py b/tools/testing/selftests/turbostat/defcolumns.py new file mode 100755 index 000000000000..70d3b7780311 --- /dev/null +++ b/tools/testing/selftests/turbostat/defcolumns.py @@ -0,0 +1,59 @@ +#!/bin/env python3 + +import subprocess +from shutil import which + +turbostat = which('turbostat') +if turbostat is None: + print('Could not find turbostat binary') + exit(1) + +timeout = which('timeout') +if timeout is None: + print('Could not find timeout binary') + exit(1) + +proc_turbostat = subprocess.run([turbostat, '--list'], capture_output = True) +if proc_turbostat.returncode != 0: + print(f'turbostat failed with {proc_turbostat.returncode}') + exit(1) + +# +# By default --list reports also "usec" and "Time_Of_Day_Seconds" columns +# which are only visible when running with --debug. +# +expected_columns_debug = proc_turbostat.stdout.replace(b',', b'\t').strip() +expected_columns = expected_columns_debug.replace(b'usec\t', b'').replace(b'Time_Of_Day_Seconds\t', b'').replace(b'X2APIC\t', b'').replace(b'APIC\t', b'') + +# +# Run turbostat with no options for 10 seconds and send SIGINT +# +timeout_argv = [timeout, '--preserve-status', '-s', 'SIGINT', '-k', '3', '1s'] +turbostat_argv = [turbostat, '-i', '0.250'] + +print(f'Running turbostat with {turbostat_argv=}... ', end = '', flush = True) +proc_turbostat = subprocess.run(timeout_argv + turbostat_argv, capture_output = True) +if proc_turbostat.returncode != 0: + print(f'turbostat failed with {proc_turbostat.returncode}') + exit(1) +actual_columns = proc_turbostat.stdout.split(b'\n')[0] +if expected_columns != actual_columns: + print(f'turbostat column check failed\n{expected_columns=}\n{actual_columns=}') + exit(1) +print('OK') + +# +# Same, but with --debug +# +turbostat_argv.append('--debug') + +print(f'Running turbostat with {turbostat_argv=}... ', end = '', flush = True) +proc_turbostat = subprocess.run(timeout_argv + turbostat_argv, capture_output = True) +if proc_turbostat.returncode != 0: + print(f'turbostat failed with {proc_turbostat.returncode}') + exit(1) +actual_columns = proc_turbostat.stdout.split(b'\n')[0] +if expected_columns_debug != actual_columns: + print(f'turbostat column check failed\n{expected_columns_debug=}\n{actual_columns=}') + exit(1) +print('OK') From bb5db22c13125b38b0740e19c18ae94f8e5a0eb6 Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Tue, 12 Mar 2024 11:19:15 +0800 Subject: [PATCH 19/26] tools/power/turbostat: Enable MSR_CORE_C1_RES support for ICX Enable Core C1 hardware residency counter (MSR_CORE_C1_RES) on ICX. 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 283dffb987b5..372f67a70d8a 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -664,6 +664,7 @@ static const struct platform_features icx_features = { .bclk_freq = BCLK_100MHZ, .supported_cstates = CC1 | CC6 | PC2 | PC6, .cst_limit = CST_LIMIT_ICX, + .has_msr_core_c1_res = 1, .has_irtl_msrs = 1, .has_cst_prewake_bit = 1, .trl_msrs = TRL_BASE | TRL_CORECOUNT, From 4e2bbbf78cf7144204214fd0bd7cca309acd8f89 Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Tue, 12 Mar 2024 14:23:37 +0800 Subject: [PATCH 20/26] tools/power/turbostat: Cache graphics sysfs path Graphics drivers (i915/Xe) have different sysfs knobs on different platforms, and it is possible that different sysfs knobs fit into the same turbostat columns. Instead of specifying different sysfs knobs every time, detect them once and cache the path for future use. No functional change. Signed-off-by: Zhang Rui Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 45 +++++++++++++++++++-------- 1 file changed, 32 insertions(+), 13 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 372f67a70d8a..78db4f65a237 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -276,6 +276,19 @@ bool no_msr; bool no_perf; enum amperf_source amperf_source; +enum gfx_sysfs_idx { + GFX_rc6, + GFX_MHz, + GFX_ACTMHz, + GFX_MAX +}; + +struct gfx_sysfs_info { + const char *path; +}; + +static struct gfx_sysfs_info gfx_info[GFX_MAX]; + int get_msr(int cpu, off_t offset, unsigned long long *msr); /* Model specific support Start */ @@ -4620,7 +4633,7 @@ int snapshot_gfx_rc6_ms(void) FILE *fp; int retval; - fp = fopen_or_die("/sys/class/drm/card0/power/rc6_residency_ms", "r"); + fp = fopen_or_die(gfx_info[GFX_rc6].path, "r"); retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms); if (retval != 1) @@ -4645,9 +4658,7 @@ int snapshot_gfx_mhz(void) int retval; if (fp == NULL) { - fp = fopen("/sys/class/drm/card0/gt_cur_freq_mhz", "r"); - if (!fp) - fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", "r"); + fp = fopen_or_die(gfx_info[GFX_MHz].path, "r"); } else { rewind(fp); fflush(fp); @@ -4674,9 +4685,7 @@ int snapshot_gfx_act_mhz(void) int retval; if (fp == NULL) { - fp = fopen("/sys/class/drm/card0/gt_act_freq_mhz", "r"); - if (!fp) - fp = fopen_or_die("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", "r"); + fp = fopen_or_die(gfx_info[GFX_ACTMHz].path, "r"); } else { rewind(fp); fflush(fp); @@ -5338,14 +5347,24 @@ probe_cluster: static void probe_graphics(void) { if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK)) + gfx_info[GFX_rc6].path = "/sys/class/drm/card0/power/rc6_residency_ms"; + + if (!access("/sys/class/drm/card0/gt_cur_freq_mhz", R_OK)) + gfx_info[GFX_MHz].path = "/sys/class/drm/card0/gt_cur_freq_mhz"; + else if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK)) + gfx_info[GFX_MHz].path = "/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz"; + + + if (!access("/sys/class/drm/card0/gt_act_freq_mhz", R_OK)) + gfx_info[GFX_ACTMHz].path = "/sys/class/drm/card0/gt_act_freq_mhz"; + else if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", R_OK)) + gfx_info[GFX_ACTMHz].path = "/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz"; + + if (gfx_info[GFX_rc6].path) BIC_PRESENT(BIC_GFX_rc6); - - if (!access("/sys/class/drm/card0/gt_cur_freq_mhz", R_OK) || - !access("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", R_OK)) + if (gfx_info[GFX_MHz].path) BIC_PRESENT(BIC_GFXMHz); - - if (!access("/sys/class/drm/card0/gt_act_freq_mhz", R_OK) || - !access("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", R_OK)) + if (gfx_info[GFX_ACTMHz].path) BIC_PRESENT(BIC_GFXACTMHz); } From de39d38c06eb047954c5ad20a3f9acb6d3c78498 Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Wed, 13 Mar 2024 10:12:19 +0800 Subject: [PATCH 21/26] tools/power/turbostat: Unify graphics sysfs snapshots Graphics sysfs snapshots share similar logic. Combine them into one function to avoid code duplication. No functional change. Signed-off-by: Zhang Rui Signed-off-by: Len Brown --- tools/power/x86/turbostat/turbostat.c | 109 ++++++++------------------ 1 file changed, 34 insertions(+), 75 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 78db4f65a237..e5b6161fef48 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -251,11 +251,8 @@ char *output_buffer, *outp; unsigned int do_dts; unsigned int do_ptm; unsigned int do_ipc; -unsigned long long gfx_cur_rc6_ms; unsigned long long cpuidle_cur_cpu_lpi_us; unsigned long long cpuidle_cur_sys_lpi_us; -unsigned int gfx_cur_mhz; -unsigned int gfx_act_mhz; unsigned int tj_max; unsigned int tj_max_override; double rapl_power_units, rapl_time_units; @@ -285,6 +282,9 @@ enum gfx_sysfs_idx { struct gfx_sysfs_info { const char *path; + FILE *fp; + unsigned int val; + unsigned long long val_ull; }; static struct gfx_sysfs_info gfx_info[GFX_MAX]; @@ -3573,17 +3573,17 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) } if (DO_BIC(BIC_GFX_rc6)) - p->gfx_rc6_ms = gfx_cur_rc6_ms; + p->gfx_rc6_ms = gfx_info[GFX_rc6].val_ull; /* n.b. assume die0 uncore frequency applies to whole package */ if (DO_BIC(BIC_UNCORE_MHZ)) p->uncore_mhz = get_uncore_mhz(p->package_id, 0); if (DO_BIC(BIC_GFXMHz)) - p->gfx_mhz = gfx_cur_mhz; + p->gfx_mhz = gfx_info[GFX_MHz].val; if (DO_BIC(BIC_GFXACTMHz)) - p->gfx_act_mhz = gfx_act_mhz; + p->gfx_act_mhz = gfx_info[GFX_ACTMHz].val; for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { if (get_mp(cpu, mp, &p->counter[i])) @@ -4621,81 +4621,40 @@ int snapshot_proc_interrupts(void) } /* - * snapshot_gfx_rc6_ms() + * snapshot_graphics() * - * record snapshot of - * /sys/class/drm/card0/power/rc6_residency_ms + * record snapshot of specified graphics sysfs knob * * return 1 if config change requires a restart, else return 0 */ -int snapshot_gfx_rc6_ms(void) +int snapshot_graphics(int idx) { FILE *fp; int retval; - fp = fopen_or_die(gfx_info[GFX_rc6].path, "r"); - - retval = fscanf(fp, "%lld", &gfx_cur_rc6_ms); - if (retval != 1) - err(1, "GFX rc6"); - - fclose(fp); - - return 0; -} - -/* - * snapshot_gfx_mhz() - * - * fall back to /sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz - * when /sys/class/drm/card0/gt_cur_freq_mhz is not available. - * - * return 1 if config change requires a restart, else return 0 - */ -int snapshot_gfx_mhz(void) -{ - static FILE *fp; - int retval; - - if (fp == NULL) { - fp = fopen_or_die(gfx_info[GFX_MHz].path, "r"); - } else { - rewind(fp); - fflush(fp); + switch (idx) { + case GFX_rc6: + fp = fopen_or_die(gfx_info[idx].path, "r"); + retval = fscanf(fp, "%lld", &gfx_info[idx].val_ull); + if (retval != 1) + err(1, "rc6"); + fclose(fp); + return 0; + case GFX_MHz: + case GFX_ACTMHz: + if (gfx_info[idx].fp == NULL) { + gfx_info[idx].fp = fopen_or_die(gfx_info[idx].path, "r"); + } else { + rewind(gfx_info[idx].fp); + fflush(gfx_info[idx].fp); + } + retval = fscanf(gfx_info[idx].fp, "%d", &gfx_info[idx].val); + if (retval != 1) + err(1, "MHz"); + return 0; + default: + return -EINVAL; } - - retval = fscanf(fp, "%d", &gfx_cur_mhz); - if (retval != 1) - err(1, "GFX MHz"); - - return 0; -} - -/* - * snapshot_gfx_cur_mhz() - * - * fall back to /sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz - * when /sys/class/drm/card0/gt_act_freq_mhz is not available. - * - * return 1 if config change requires a restart, else return 0 - */ -int snapshot_gfx_act_mhz(void) -{ - static FILE *fp; - int retval; - - if (fp == NULL) { - fp = fopen_or_die(gfx_info[GFX_ACTMHz].path, "r"); - } else { - rewind(fp); - fflush(fp); - } - - retval = fscanf(fp, "%d", &gfx_act_mhz); - if (retval != 1) - err(1, "GFX ACT MHz"); - - return 0; } /* @@ -4760,13 +4719,13 @@ int snapshot_proc_sysfs_files(void) return 1; if (DO_BIC(BIC_GFX_rc6)) - snapshot_gfx_rc6_ms(); + snapshot_graphics(GFX_rc6); if (DO_BIC(BIC_GFXMHz)) - snapshot_gfx_mhz(); + snapshot_graphics(GFX_MHz); if (DO_BIC(BIC_GFXACTMHz)) - snapshot_gfx_act_mhz(); + snapshot_graphics(GFX_ACTMHz); if (DO_BIC(BIC_CPU_LPI)) snapshot_cpu_lpi_us(); From 60add818ab2543b7e4f2bfeaacf2504743c1eb50 Mon Sep 17 00:00:00 2001 From: Justin Ernst Date: Tue, 2 Apr 2024 13:40:29 -0400 Subject: [PATCH 22/26] tools/power/turbostat: Fix uncore frequency file string Running turbostat on a 16 socket HPE Scale-up Compute 3200 (SapphireRapids) fails with: turbostat: /sys/devices/system/cpu/intel_uncore_frequency/package_010_die_00/current_freq_khz: open failed: No such file or directory We observe the sysfs uncore frequency directories named: ... package_09_die_00/ package_10_die_00/ package_11_die_00/ ... package_15_die_00/ The culprit is an incorrect sprintf format string "package_0%d_die_0%d" used with each instance of reading uncore frequency files. uncore-frequency-common.c creates the sysfs directory with the format "package_%02d_die_%02d". Once the package value reaches double digits, the formats diverge. Change each instance of "package_0%d_die_0%d" to "package_%02d_die_%02d". [lenb: deleted the probe part of this patch, as it was already fixed] Signed-off-by: Justin Ernst Reviewed-by: Thomas Renninger 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 e5b6161fef48..016a5c7dc9bf 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -2939,7 +2939,7 @@ unsigned long long get_uncore_mhz(int package, int die) { char path[128]; - sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/current_freq_khz", package, + sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_%02d_die_%02d/current_freq_khz", package, die); return (snapshot_sysfs_counter(path) / 1000); From 3bbb331c1d34fdd5520a050fce35f71579430485 Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Wed, 13 Mar 2024 10:30:04 +0800 Subject: [PATCH 23/26] tools/power/turbostat: Introduce BIC_SAM_mc6/BIC_SAMMHz/BIC_SAMACTMHz Graphics driver (i915/Xe) on mordern platforms splits GFX and SA Media information via different sysfs knobs. Existing BIC_GFX_rc6/BIC_GFXMHz/BIC_GFXACTMHz columns can be reused for GFX. Introduce BIC_SAM_mc6/BIC_SAMMHz/BIC_SAMACTMHz columns for SA Media. Signed-off-by: Zhang Rui --- tools/power/x86/turbostat/turbostat.8 | 12 +++- tools/power/x86/turbostat/turbostat.c | 91 +++++++++++++++++++++++++-- 2 files changed, 96 insertions(+), 7 deletions(-) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 567327b004e6..0d3672e5d9ed 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -129,9 +129,17 @@ 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 -\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. +\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. +\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. +.PP +\fBGFXAMHz\fP Instantaneous snapshot of what sysfs presents at the end of the measurement interval. From /sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz or /sys/class/drm/card0/gt_act_freq_mhz or /sys/class/drm/card0/gt/gt0/rps_act_freq_mhz or /sys/class/drm/card0/device/tile0/gtN/freq0/act_freq depending on the graphics driver being used. +.PP +\fBSAM%mc6\fP The percentage of time the SA Media is in the "module C6" state, mc6, during the measurement interval. From /sys/class/drm/card0/gt/gt1/rc6_residency_ms or /sys/class/drm/card0/device/tile0/gtN/gtidle/idle_residency_ms depending on the graphics driver being used. +.PP +\fBSAMMHz\fP Instantaneous snapshot of what sysfs presents at the end of the measurement interval. From /sys/class/drm/card0/gt/gt1/rps_cur_freq_mhz or /sys/class/drm/card0/device/tile0/gtN/freq0/cur_freq depending on the graphics driver being used. +.PP +\fBSAMAMHz\fP Instantaneous snapshot of what sysfs presents at the end of the measurement interval. From /sys/class/drm/card0/gt/gt1/rps_act_freq_mhz or /sys/class/drm/card0/device/tile0/gtN/freq0/act_freq depending on the graphics driver being used. .PP \fBPkg%pc2, Pkg%pc3, Pkg%pc6, Pkg%pc7\fP percentage residency in hardware package idle states. These numbers are from hardware residency counters. .PP diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 016a5c7dc9bf..4fa2810da1a3 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -133,6 +133,9 @@ struct msr_counter bic[] = { { 0x0, "IPC", "", 0, 0, 0, NULL, 0 }, { 0x0, "CoreThr", "", 0, 0, 0, NULL, 0 }, { 0x0, "UncMHz", "", 0, 0, 0, NULL, 0 }, + { 0x0, "SAM%mc6", "", 0, 0, 0, NULL, 0 }, + { 0x0, "SAMMHz", "", 0, 0, 0, NULL, 0 }, + { 0x0, "SAMAMHz", "", 0, 0, 0, NULL, 0 }, }; #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter)) @@ -191,11 +194,14 @@ struct msr_counter bic[] = { #define BIC_IPC (1ULL << 52) #define BIC_CORE_THROT_CNT (1ULL << 53) #define BIC_UNCORE_MHZ (1ULL << 54) +#define BIC_SAM_mc6 (1ULL << 55) +#define BIC_SAMMHz (1ULL << 56) +#define BIC_SAMACTMHz (1ULL << 57) #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__) -#define BIC_FREQUENCY ( BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_UNCORE_MHZ) -#define BIC_IDLE ( 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) +#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_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) #define BIC_OTHER ( BIC_IRQ | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) #define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC) @@ -277,6 +283,9 @@ enum gfx_sysfs_idx { GFX_rc6, GFX_MHz, GFX_ACTMHz, + SAM_mc6, + SAM_MHz, + SAM_ACTMHz, GFX_MAX }; @@ -1193,6 +1202,9 @@ struct pkg_data { long long gfx_rc6_ms; unsigned int gfx_mhz; unsigned int gfx_act_mhz; + long long sam_mc6_ms; + unsigned int sam_mhz; + unsigned int sam_act_mhz; unsigned int package_id; struct rapl_counter energy_pkg; /* MSR_PKG_ENERGY_STATUS */ struct rapl_counter energy_dram; /* MSR_DRAM_ENERGY_STATUS */ @@ -1844,6 +1856,15 @@ void print_header(char *delim) if (DO_BIC(BIC_GFXACTMHz)) outp += sprintf(outp, "%sGFXAMHz", (printed++ ? delim : "")); + if (DO_BIC(BIC_SAM_mc6)) + outp += sprintf(outp, "%sSAM%%mc6", (printed++ ? delim : "")); + + if (DO_BIC(BIC_SAMMHz)) + outp += sprintf(outp, "%sSAMMHz", (printed++ ? delim : "")); + + if (DO_BIC(BIC_SAMACTMHz)) + outp += sprintf(outp, "%sSAMAMHz", (printed++ ? delim : "")); + if (DO_BIC(BIC_Totl_c0)) outp += sprintf(outp, "%sTotl%%C0", (printed++ ? delim : "")); if (DO_BIC(BIC_Any_c0)) @@ -2251,6 +2272,24 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data if (DO_BIC(BIC_GFXACTMHz)) outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->gfx_act_mhz); + /* SAMmc6 */ + if (DO_BIC(BIC_SAM_mc6)) { + if (p->sam_mc6_ms == -1) { /* detect GFX counter reset */ + outp += sprintf(outp, "%s**.**", (printed++ ? delim : "")); + } else { + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), + p->sam_mc6_ms / 10.0 / interval_float); + } + } + + /* SAMMHz */ + if (DO_BIC(BIC_SAMMHz)) + outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->sam_mhz); + + /* SAMACTMHz */ + if (DO_BIC(BIC_SAMACTMHz)) + outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), p->sam_act_mhz); + /* Totl%C0, Any%C0 GFX%C0 CPUGFX% */ if (DO_BIC(BIC_Totl_c0)) outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->pkg_wtd_core_c0 / tsc); @@ -2437,6 +2476,15 @@ int delta_package(struct pkg_data *new, struct pkg_data *old) old->gfx_mhz = new->gfx_mhz; old->gfx_act_mhz = new->gfx_act_mhz; + /* flag an error when mc6 counter resets/wraps */ + if (old->sam_mc6_ms > new->sam_mc6_ms) + old->sam_mc6_ms = -1; + else + old->sam_mc6_ms = new->sam_mc6_ms - old->sam_mc6_ms; + + old->sam_mhz = new->sam_mhz; + old->sam_act_mhz = new->sam_act_mhz; + old->energy_pkg.raw_value = new->energy_pkg.raw_value - old->energy_pkg.raw_value; old->energy_cores.raw_value = new->energy_cores.raw_value - old->energy_cores.raw_value; old->energy_gfx.raw_value = new->energy_gfx.raw_value - old->energy_gfx.raw_value; @@ -2661,6 +2709,9 @@ void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data p->uncore_mhz = 0; p->gfx_mhz = 0; p->gfx_act_mhz = 0; + p->sam_mc6_ms = 0; + p->sam_mhz = 0; + p->sam_act_mhz = 0; for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) t->counter[i] = 0; @@ -2775,6 +2826,9 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.packages.uncore_mhz = p->uncore_mhz; average.packages.gfx_mhz = p->gfx_mhz; average.packages.gfx_act_mhz = p->gfx_act_mhz; + average.packages.sam_mc6_ms = p->sam_mc6_ms; + average.packages.sam_mhz = p->sam_mhz; + average.packages.sam_act_mhz = p->sam_act_mhz; average.packages.pkg_temp_c = MAX(average.packages.pkg_temp_c, p->pkg_temp_c); @@ -3572,19 +3626,28 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) p->pkg_temp_c = tj_max - ((msr >> 16) & 0x7F); } - if (DO_BIC(BIC_GFX_rc6)) - p->gfx_rc6_ms = gfx_info[GFX_rc6].val_ull; - /* n.b. assume die0 uncore frequency applies to whole package */ if (DO_BIC(BIC_UNCORE_MHZ)) p->uncore_mhz = get_uncore_mhz(p->package_id, 0); + if (DO_BIC(BIC_GFX_rc6)) + p->gfx_rc6_ms = gfx_info[GFX_rc6].val_ull; + if (DO_BIC(BIC_GFXMHz)) p->gfx_mhz = gfx_info[GFX_MHz].val; if (DO_BIC(BIC_GFXACTMHz)) p->gfx_act_mhz = gfx_info[GFX_ACTMHz].val; + if (DO_BIC(BIC_SAM_mc6)) + p->sam_mc6_ms = gfx_info[SAM_mc6].val_ull; + + if (DO_BIC(BIC_SAMMHz)) + p->sam_mhz = gfx_info[SAM_MHz].val; + + if (DO_BIC(BIC_SAMACTMHz)) + p->sam_act_mhz = gfx_info[SAM_ACTMHz].val; + for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) { if (get_mp(cpu, mp, &p->counter[i])) return -10; @@ -4634,6 +4697,7 @@ int snapshot_graphics(int idx) switch (idx) { case GFX_rc6: + case SAM_mc6: fp = fopen_or_die(gfx_info[idx].path, "r"); retval = fscanf(fp, "%lld", &gfx_info[idx].val_ull); if (retval != 1) @@ -4642,6 +4706,8 @@ int snapshot_graphics(int idx) return 0; case GFX_MHz: case GFX_ACTMHz: + case SAM_MHz: + case SAM_ACTMHz: if (gfx_info[idx].fp == NULL) { gfx_info[idx].fp = fopen_or_die(gfx_info[idx].path, "r"); } else { @@ -4727,6 +4793,15 @@ int snapshot_proc_sysfs_files(void) if (DO_BIC(BIC_GFXACTMHz)) snapshot_graphics(GFX_ACTMHz); + if (DO_BIC(BIC_SAM_mc6)) + snapshot_graphics(SAM_mc6); + + if (DO_BIC(BIC_SAMMHz)) + snapshot_graphics(SAM_MHz); + + if (DO_BIC(BIC_SAMACTMHz)) + snapshot_graphics(SAM_ACTMHz); + if (DO_BIC(BIC_CPU_LPI)) snapshot_cpu_lpi_us(); @@ -5325,6 +5400,12 @@ static void probe_graphics(void) BIC_PRESENT(BIC_GFXMHz); if (gfx_info[GFX_ACTMHz].path) BIC_PRESENT(BIC_GFXACTMHz); + if (gfx_info[SAM_mc6].path) + BIC_PRESENT(BIC_SAM_mc6); + if (gfx_info[SAM_MHz].path) + BIC_PRESENT(BIC_SAMMHz); + if (gfx_info[SAM_ACTMHz].path) + BIC_PRESENT(BIC_SAMACTMHz); } static void dump_sysfs_cstate_config(void) From dc02dc937a3ef819c5da10e97084af6977be26bf Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Fri, 22 Mar 2024 09:52:24 +0800 Subject: [PATCH 24/26] tools/power/turbostat: Add support for new i915 sysfs knobs On Meteorlake platform, i915 driver supports the traditional graphics sysfs knobs including /sys/class/drm/card0/power/rc6_residency_ms /sys/class/drm/card0/gt_cur_freq_mhz /sys/class/drm/card0/gt_act_freq_mhz At the same time, it also supports /sys/class/drm/card0/gt/gt0/rc6_residency_ms /sys/class/drm/card0/gt/gt0/rps_cur_freq_mhz /sys/class/drm/card0/gt/gt0/rps_act_freq_mhz /sys/class/drm/card0/gt/gt1/rc6_residency_ms /sys/class/drm/card0/gt/gt1/rps_cur_freq_mhz /sys/class/drm/card0/gt/gt1/rps_act_freq_mhz gt0 is for GFX and gt1 is for SA Media. Enhance turbostat to prefer the i915 new sysfs knobs. Export gt0 via BIC_GFX_rc6/BIC_GFXMHz/BIC_GFXACTMHz. Export gt1 via BIC_SMA_mc6/BIC_SMAMHz/BIC_SMAACTMHz. Signed-off-by: Zhang Rui --- tools/power/x86/turbostat/turbostat.c | 24 ++++++++++++++++++++++++ 1 file changed, 24 insertions(+) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 4fa2810da1a3..feca7f4cb5cd 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -5380,6 +5380,29 @@ probe_cluster: static void probe_graphics(void) { + /* New i915 graphics sysfs knobs */ + if (!access("/sys/class/drm/card0/gt/gt0/rc6_residency_ms", R_OK)) { + gfx_info[GFX_rc6].path = "/sys/class/drm/card0/gt/gt0/rc6_residency_ms"; + + if (!access("/sys/class/drm/card0/gt/gt0/rps_cur_freq_mhz", R_OK)) + gfx_info[GFX_MHz].path = "/sys/class/drm/card0/gt/gt0/rps_cur_freq_mhz"; + + if (!access("/sys/class/drm/card0/gt/gt0/rps_act_freq_mhz", R_OK)) + gfx_info[GFX_ACTMHz].path = "/sys/class/drm/card0/gt/gt0/rps_act_freq_mhz"; + + if (!access("/sys/class/drm/card0/gt/gt1/rc6_residency_ms", R_OK)) + gfx_info[SAM_mc6].path = "/sys/class/drm/card0/gt/gt1/rc6_residency_ms"; + + if (!access("/sys/class/drm/card0/gt/gt1/rps_cur_freq_mhz", R_OK)) + gfx_info[SAM_MHz].path = "/sys/class/drm/card0/gt/gt1/rps_cur_freq_mhz"; + + if (!access("/sys/class/drm/card0/gt/gt1/rps_act_freq_mhz", R_OK)) + gfx_info[SAM_ACTMHz].path = "/sys/class/drm/card0/gt/gt1/rps_act_freq_mhz"; + + goto end; + } + + /* Fall back to traditional i915 graphics sysfs knobs */ if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK)) gfx_info[GFX_rc6].path = "/sys/class/drm/card0/power/rc6_residency_ms"; @@ -5394,6 +5417,7 @@ static void probe_graphics(void) else if (!access("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", R_OK)) gfx_info[GFX_ACTMHz].path = "/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz"; +end: if (gfx_info[GFX_rc6].path) BIC_PRESENT(BIC_GFX_rc6); if (gfx_info[GFX_MHz].path) From 91a91d389543a86963beec148d98d37875154bd4 Mon Sep 17 00:00:00 2001 From: Zhang Rui Date: Tue, 12 Mar 2024 23:56:02 +0800 Subject: [PATCH 25/26] tools/power/turbostat: Add support for Xe sysfs knobs Xe graphics driver uses different graphics sysfs knobs including /sys/class/drm/card0/device/tile0/gt0/gtidle/idle_residency_ms /sys/class/drm/card0/device/tile0/gt0/freq0/cur_freq /sys/class/drm/card0/device/tile0/gt0/freq0/act_freq /sys/class/drm/card0/device/tile0/gt1/gtidle/idle_residency_ms /sys/class/drm/card0/device/tile0/gt1/freq0/cur_freq /sys/class/drm/card0/device/tile0/gt1/freq0/act_freq Plus that, /sys/class/drm/card0/device/tile0/gt/gtidle/name returns either gt-rc or gt-mc. rc is for GFX and mc is SA Media. Enhance turbostat to prefer the Xe sysfs knobs when they are available. Export gt-rc via BIC_GFX_rc6/BIC_GFXMHz/BIC_GFXACTMHz. Export gt-mc via BIC_SMA_mc6/BIC_SMAMHz/BIC_SMAACTMHz. Signed-off-by: Zhang Rui --- tools/power/x86/turbostat/turbostat.c | 51 +++++++++++++++++++++++++++ 1 file changed, 51 insertions(+) diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index feca7f4cb5cd..bc103851df70 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -5380,6 +5380,57 @@ probe_cluster: static void probe_graphics(void) { + /* Xe graphics sysfs knobs */ + if (!access("/sys/class/drm/card0/device/tile0/gt0/gtidle/idle_residency_ms", R_OK)) { + FILE *fp; + char buf[8]; + bool gt0_is_gt; + int idx; + + fp = fopen("/sys/class/drm/card0/device/tile0/gt0/gtidle/name", "r"); + if (!fp) + goto next; + + if (!fread(buf, sizeof(char), 7, fp)) { + fclose(fp); + goto next; + } + fclose(fp); + + if (!strncmp(buf, "gt0-rc", strlen("gt0-rc"))) + gt0_is_gt = true; + else if (!strncmp(buf, "gt0-mc", strlen("gt0-mc"))) + gt0_is_gt = false; + else + goto next; + + idx = gt0_is_gt ? GFX_rc6 : SAM_mc6; + gfx_info[idx].path = "/sys/class/drm/card0/device/tile0/gt0/gtidle/idle_residency_ms"; + + idx = gt0_is_gt ? GFX_MHz : SAM_MHz; + if (!access("/sys/class/drm/card0/device/tile0/gt0/freq0/cur_freq", R_OK)) + gfx_info[idx].path = "/sys/class/drm/card0/device/tile0/gt0/freq0/cur_freq"; + + idx = gt0_is_gt ? GFX_ACTMHz : SAM_ACTMHz; + if (!access("/sys/class/drm/card0/device/tile0/gt0/freq0/act_freq", R_OK)) + gfx_info[idx].path = "/sys/class/drm/card0/device/tile0/gt0/freq0/act_freq"; + + idx = gt0_is_gt ? SAM_mc6 : GFX_rc6; + if (!access("/sys/class/drm/card0/device/tile0/gt1/gtidle/idle_residency_ms", R_OK)) + gfx_info[idx].path = "/sys/class/drm/card0/device/tile0/gt1/gtidle/idle_residency_ms"; + + idx = gt0_is_gt ? SAM_MHz : GFX_MHz; + if (!access("/sys/class/drm/card0/device/tile0/gt1/freq0/cur_freq", R_OK)) + gfx_info[idx].path = "/sys/class/drm/card0/device/tile0/gt1/freq0/cur_freq"; + + idx = gt0_is_gt ? SAM_ACTMHz : GFX_ACTMHz; + if (!access("/sys/class/drm/card0/device/tile0/gt1/freq0/act_freq", R_OK)) + gfx_info[idx].path = "/sys/class/drm/card0/device/tile0/gt1/freq0/act_freq"; + + goto end; + } + +next: /* New i915 graphics sysfs knobs */ if (!access("/sys/class/drm/card0/gt/gt0/rc6_residency_ms", R_OK)) { gfx_info[GFX_rc6].path = "/sys/class/drm/card0/gt/gt0/rc6_residency_ms"; From 3ab7296a7e6aa34634dcc2926af933107a117996 Mon Sep 17 00:00:00 2001 From: Len Brown Date: Mon, 8 Apr 2024 19:32:58 -0400 Subject: [PATCH 26/26] tools/power turbostat: v2024.04.10 Much of turbostat can now run with perf, rather than using the MSR driver Some of turbostat can now run as a regular non-root user. Add some new output columns for some new GFX hardware. [This patch updates the version, but otherwise changes no function; it touches up some checkpatch issues from previous patches] Signed-off-by: Len Brown --- MAINTAINERS | 1 + tools/power/x86/turbostat/turbostat.c | 41 ++++++++++++------- .../testing/selftests/turbostat/defcolumns.py | 1 + 3 files changed, 28 insertions(+), 15 deletions(-) diff --git a/MAINTAINERS b/MAINTAINERS index a7c4cf8201e0..b8582a466128 100644 --- a/MAINTAINERS +++ b/MAINTAINERS @@ -22116,6 +22116,7 @@ Q: https://patchwork.kernel.org/project/linux-pm/list/ B: https://bugzilla.kernel.org T: git git://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux.git turbostat F: tools/power/x86/turbostat/ +F: tools/testing/selftests/turbostat/ TW5864 VIDEO4LINUX DRIVER M: Bluecherry Maintainers diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index bc103851df70..98256468e248 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) 2023 Intel Corporation. + * Copyright (c) 2024 Intel Corporation. * Len Brown */ @@ -1360,6 +1360,7 @@ struct sys_counters { void free_sys_counters(void) { struct msr_counter *p = sys.tp, *pnext = NULL; + while (p) { pnext = p->next; free(p); @@ -1979,6 +1980,7 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p const unsigned long long energy_value = c->core_energy.raw_value * c->core_energy.scale; const double energy_scale = c->core_energy.scale; + if (c->core_energy.unit == RAPL_UNIT_JOULES) outp += sprintf(outp, "Joules: %0llX (scale: %lf)\n", energy_value, energy_scale); @@ -3153,7 +3155,7 @@ static unsigned int read_perf_counter_info_n(const char *const path, const char return v; } -static unsigned read_msr_type(void) +static unsigned int read_msr_type(void) { const char *const path = "/sys/bus/event_source/devices/msr/type"; const char *const format = "%u"; @@ -3161,7 +3163,7 @@ static unsigned read_msr_type(void) return read_perf_counter_info_n(path, format); } -static unsigned read_aperf_config(void) +static unsigned int read_aperf_config(void) { const char *const path = "/sys/bus/event_source/devices/msr/events/aperf"; const char *const format = "event=%x"; @@ -3169,7 +3171,7 @@ static unsigned read_aperf_config(void) return read_perf_counter_info_n(path, format); } -static unsigned read_mperf_config(void) +static unsigned int read_mperf_config(void) { const char *const path = "/sys/bus/event_source/devices/msr/events/mperf"; const char *const format = "event=%x"; @@ -3177,7 +3179,7 @@ static unsigned read_mperf_config(void) return read_perf_counter_info_n(path, format); } -static unsigned read_perf_type(const char *subsys) +static unsigned int read_perf_type(const char *subsys) { const char *const path_format = "/sys/bus/event_source/devices/%s/type"; const char *const format = "%u"; @@ -3188,7 +3190,7 @@ static unsigned read_perf_type(const char *subsys) return read_perf_counter_info_n(path, format); } -static unsigned read_rapl_config(const char *subsys, const char *event_name) +static unsigned int read_rapl_config(const char *subsys, const char *event_name) { const char *const path_format = "/sys/bus/event_source/devices/%s/events/%s"; const char *const format = "event=%x"; @@ -3199,7 +3201,7 @@ static unsigned read_rapl_config(const char *subsys, const char *event_name) return read_perf_counter_info_n(path, format); } -static unsigned read_perf_rapl_unit(const char *subsys, const char *event_name) +static unsigned int read_perf_rapl_unit(const char *subsys, const char *event_name) { const char *const path_format = "/sys/bus/event_source/devices/%s/events/%s.unit"; const char *const format = "%s"; @@ -3235,7 +3237,7 @@ static struct amperf_group_fd open_amperf_fd(int cpu) const unsigned int msr_type = read_msr_type(); const unsigned int aperf_config = read_aperf_config(); const unsigned int mperf_config = read_mperf_config(); - struct amperf_group_fd fds = {.aperf = -1,.mperf = -1 }; + struct amperf_group_fd fds = {.aperf = -1, .mperf = -1 }; fds.aperf = open_perf_counter(cpu, msr_type, aperf_config, -1, PERF_FORMAT_GROUP); fds.mperf = open_perf_counter(cpu, msr_type, mperf_config, fds.aperf, PERF_FORMAT_GROUP); @@ -3277,6 +3279,7 @@ static int read_aperf_mperf_tsc_perf(struct thread_data *t, int cpu) t->tsc = rdtsc(); const int n = read(fd_amperf, &cnt.as_array[0], sizeof(cnt.as_array)); + if (n != sizeof(cnt.as_array)) return -2; @@ -3371,7 +3374,7 @@ int get_rapl_counters(int cpu, int domain, struct core_data *c, struct pkg_data struct rapl_counter_info_t *rci = &rapl_counter_info_perdomain[domain]; if (debug) - fprintf(stderr, "get_rapl_counters: cpu%d domain%d\n", cpu, domain); + fprintf(stderr, "%s: cpu%d domain%d\n", __func__, cpu, domain); assert(rapl_counter_info_perdomain); @@ -3382,8 +3385,9 @@ int get_rapl_counters(int cpu, int domain, struct core_data *c, struct pkg_data size_t num_perf_counters = rapl_counter_info_count_perf(rci); const ssize_t expected_read_size = (num_perf_counters + 1) * sizeof(unsigned long long); const ssize_t actual_read_size = read(rci->fd_perf, &perf_data[0], sizeof(perf_data)); + if (actual_read_size != expected_read_size) - err(-1, "get_rapl_counters: failed to read perf_data (%zu %zu)", expected_read_size, + err(-1, "%s: failed to read perf_data (%zu %zu)", __func__, expected_read_size, actual_read_size); } @@ -3454,7 +3458,7 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) int status; if (cpu_migrate(cpu)) { - fprintf(outf, "get_counters: Could not migrate to CPU %d\n", cpu); + fprintf(outf, "%s: Could not migrate to CPU %d\n", __func__, cpu); return -1; } @@ -6411,15 +6415,17 @@ int add_rapl_perf_counter_(int cpu, struct rapl_counter_info_t *rci, const struc return -1; const double scale = read_perf_rapl_scale(cai->perf_subsys, cai->perf_name); + if (scale == 0.0) return -1; const enum rapl_unit unit = read_perf_rapl_unit(cai->perf_subsys, cai->perf_name); + if (unit == RAPL_UNIT_INVALID) return -1; - const unsigned rapl_type = read_perf_type(cai->perf_subsys); - const unsigned rapl_energy_pkg_config = read_rapl_config(cai->perf_subsys, cai->perf_name); + const unsigned int rapl_type = read_perf_type(cai->perf_subsys); + const unsigned int rapl_energy_pkg_config = read_rapl_config(cai->perf_subsys, cai->perf_name); const int fd_counter = open_perf_counter(cpu, rapl_type, rapl_energy_pkg_config, rci->fd_perf, PERF_FORMAT_GROUP); @@ -6441,7 +6447,7 @@ int add_rapl_perf_counter(int cpu, struct rapl_counter_info_t *rci, const struct int ret = add_rapl_perf_counter_(cpu, rci, cai, scale, unit); if (debug) - fprintf(stderr, "add_rapl_perf_counter: %d (cpu: %d)\n", ret, cpu); + fprintf(stderr, "%s: %d (cpu: %d)\n", __func__, ret, cpu); return ret; } @@ -6462,6 +6468,7 @@ void linux_perf_init(void) } const bool aperf_required = is_aperf_access_required(); + if (aperf_required && has_aperf && amperf_source == AMPERF_SOURCE_PERF) { fd_amperf_percpu = calloc(topo.max_cpu_num + 1, sizeof(*fd_amperf_percpu)); if (fd_amperf_percpu == NULL) @@ -6483,6 +6490,7 @@ void rapl_perf_init(void) */ for (int domain_id = 0; domain_id < num_domains; ++domain_id) { struct rapl_counter_info_t *rci = &rapl_counter_info_perdomain[domain_id]; + rci->fd_perf = -1; for (size_t i = 0; i < NUM_RAPL_COUNTERS; ++i) { rci->data[i] = 0; @@ -7296,6 +7304,7 @@ static void set_amperf_source(void) amperf_source = AMPERF_SOURCE_PERF; const bool aperf_required = is_aperf_access_required(); + if (no_perf || !aperf_required || !has_amperf_access_via_perf()) amperf_source = AMPERF_SOURCE_MSR; @@ -7373,10 +7382,12 @@ void check_msr_access(void) void check_perf_access(void) { const bool intrcount_required = BIC_IS_ENABLED(BIC_IPC); + if (no_perf || !intrcount_required || !has_instr_count_access()) bic_enabled &= ~BIC_IPC; const bool aperf_required = is_aperf_access_required(); + if (!aperf_required || !has_amperf_access()) { bic_enabled &= ~BIC_Avg_MHz; bic_enabled &= ~BIC_Busy; @@ -7486,7 +7497,7 @@ int get_and_dump_counters(void) void print_version() { - fprintf(outf, "turbostat version 2023.11.07 - Len Brown \n"); + fprintf(outf, "turbostat version 2024.04.08 - Len Brown \n"); } #define COMMAND_LINE_SIZE 2048 diff --git a/tools/testing/selftests/turbostat/defcolumns.py b/tools/testing/selftests/turbostat/defcolumns.py index 70d3b7780311..d9b042097da7 100755 --- a/tools/testing/selftests/turbostat/defcolumns.py +++ b/tools/testing/selftests/turbostat/defcolumns.py @@ -1,4 +1,5 @@ #!/bin/env python3 +# SPDX-License-Identifier: GPL-2.0 import subprocess from shutil import which