#include <sys/capability.h>
#include <errno.h>
#include <math.h>
+#include <linux/perf_event.h>
+#include <asm/unistd.h>
char *proc_stat = "/proc/stat";
FILE *outf;
int *fd_percpu;
+int *fd_instr_count_percpu;
struct timeval interval_tv = {5, 0};
struct timespec interval_ts = {5, 0};
unsigned int num_iterations;
unsigned int do_rapl;
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 long long aperf;
unsigned long long mperf;
unsigned long long c1;
+ unsigned long long instr_count;
unsigned long long irq_count;
unsigned int smi_count;
unsigned int cpu_id;
switch (idx) {
case IDX_PKG_ENERGY:
- offset = MSR_PKG_ENERGY_STATUS;
+ if (do_rapl & RAPL_AMD_F17H)
+ offset = MSR_PKG_ENERGY_STAT;
+ else
+ offset = MSR_PKG_ENERGY_STATUS;
break;
case IDX_DRAM_ENERGY:
offset = MSR_DRAM_ENERGY_STATUS;
switch (offset) {
case MSR_PKG_ENERGY_STATUS:
+ case MSR_PKG_ENERGY_STAT:
idx = IDX_PKG_ENERGY;
break;
case MSR_DRAM_ENERGY_STATUS:
{
switch (idx) {
case IDX_PKG_ENERGY:
- return do_rapl & RAPL_PKG;
+ return do_rapl & (RAPL_PKG | RAPL_AMD_F17H);
case IDX_DRAM_ENERGY:
return do_rapl & RAPL_DRAM;
case IDX_PP0_ENERGY:
return fd;
}
+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);
+}
+
+static int perf_instr_count_open(int cpu_num)
+{
+ struct perf_event_attr pea;
+ int fd;
+
+ 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;
+
+ /* counter for cpu_num, including user + kernel and all processes */
+ fd = perf_event_open(&pea, -1, cpu_num, -1, 0);
+ if (fd == -1)
+ err(-1, "cpu%d: perf instruction counter\n", cpu_num);
+
+ return fd;
+}
+
+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);
+
+ return fd_instr_count_percpu[cpu];
+}
+
int get_msr(int cpu, off_t offset, unsigned long long *msr)
{
ssize_t retval;
{ 0x0, "X2APIC" },
{ 0x0, "Die" },
{ 0x0, "GFXAMHz" },
+ { 0x0, "IPC" },
};
#define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter))
#define BIC_X2APIC (1ULL << 49)
#define BIC_Die (1ULL << 50)
#define BIC_GFXACTMHz (1ULL << 51)
+#define BIC_IPC (1ULL << 52)
#define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC)
#define ENABLE_BIC(COUNTER_NAME) (bic_enabled |= COUNTER_NAME)
#define BIC_PRESENT(COUNTER_BIT) (bic_present |= COUNTER_BIT)
#define BIC_NOT_PRESENT(COUNTER_BIT) (bic_present &= ~COUNTER_BIT)
+#define BIC_IS_ENABLED(COUNTER_BIT) (bic_enabled & COUNTER_BIT)
#define MAX_DEFERRED 16
if (DO_BIC(BIC_TSC_MHz))
outp += sprintf(outp, "%sTSC_MHz", (printed++ ? delim : ""));
+ if (DO_BIC(BIC_IPC))
+ outp += sprintf(outp, "%sIPC", (printed++ ? delim : ""));
+
if (DO_BIC(BIC_IRQ)) {
if (sums_need_wide_columns)
outp += sprintf(outp, "%s IRQ", (printed++ ? delim : ""));
outp += sprintf(outp, "mperf: %016llX\n", t->mperf);
outp += sprintf(outp, "c1: %016llX\n", t->c1);
+ if (DO_BIC(BIC_IPC))
+ outp += sprintf(outp, "IPC: %lld\n", t->instr_count);
+
if (DO_BIC(BIC_IRQ))
outp += sprintf(outp, "IRQ: %lld\n", t->irq_count);
if (DO_BIC(BIC_SMI))
if (DO_BIC(BIC_TSC_MHz))
outp += sprintf(outp, "%s%.0f", (printed++ ? delim : ""), 1.0 * t->tsc/units/interval_float);
+ if (DO_BIC(BIC_IPC))
+ outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 1.0 * t->instr_count / t->aperf);
+
/* IRQ */
if (DO_BIC(BIC_IRQ)) {
if (sums_need_wide_columns)
old->mperf = 1; /* divide by 0 protection */
}
+ if (DO_BIC(BIC_IPC))
+ old->instr_count = new->instr_count - old->instr_count;
+
if (DO_BIC(BIC_IRQ))
old->irq_count = new->irq_count - old->irq_count;
t->mperf = 0;
t->c1 = 0;
+ t->instr_count = 0;
+
t->irq_count = 0;
t->smi_count = 0;
average.threads.mperf += t->mperf;
average.threads.c1 += t->c1;
+ average.threads.instr_count += t->instr_count;
+
average.threads.irq_count += t->irq_count;
average.threads.smi_count += t->smi_count;
average.threads.tsc /= topo.num_cpus;
average.threads.aperf /= topo.num_cpus;
average.threads.mperf /= topo.num_cpus;
+ average.threads.instr_count /= topo.num_cpus;
average.threads.c1 /= topo.num_cpus;
if (average.threads.irq_count > 9999999)
t->mperf = t->mperf * aperf_mperf_multiplier;
}
+ if (DO_BIC(BIC_IPC))
+ if (read(get_instr_count_fd(cpu), &t->instr_count, sizeof(long long)) != sizeof(long long))
+ return -4;
+
if (DO_BIC(BIC_IRQ))
t->irq_count = irqs_per_cpu[cpu];
if (DO_BIC(BIC_SMI)) {
switch (model) {
case INTEL_FAM6_HASWELL_X: /* HSX */
case INTEL_FAM6_BROADWELL_X: /* BDX */
+ case INTEL_FAM6_SKYLAKE_X: /* SKX */
case INTEL_FAM6_XEON_PHI_KNL: /* KNL */
return (rapl_dram_energy_units = 15.3 / 1000000);
default:
* below this value, including the Digital Thermal Sensor (DTS),
* Package Thermal Management Sensor (PTM), and thermal event thresholds.
*/
-int read_tcc_activation_temp()
+int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
{
unsigned long long msr;
- unsigned int tcc, target_c, offset_c;
-
- /* Temperature Target MSR is Nehalem and newer only */
- if (!do_nhm_platform_info)
- return 0;
-
- if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
- return 0;
-
- target_c = (msr >> 16) & 0xFF;
-
- offset_c = (msr >> 24) & 0xF;
-
- tcc = target_c - offset_c;
-
- if (!quiet)
- fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C) (%d default - %d offset)\n",
- base_cpu, msr, tcc, target_c, offset_c);
-
- return tcc;
-}
+ unsigned int target_c_local;
+ int cpu;
-int set_temperature_target(struct thread_data *t, struct core_data *c, struct pkg_data *p)
-{
/* tcc_activation_temp is used only for dts or ptm */
if (!(do_dts || do_ptm))
return 0;
if (!(t->flags & CPU_IS_FIRST_THREAD_IN_CORE) || !(t->flags & CPU_IS_FIRST_CORE_IN_PACKAGE))
return 0;
+ cpu = t->cpu_id;
+ if (cpu_migrate(cpu)) {
+ fprintf(outf, "Could not migrate to CPU %d\n", cpu);
+ return -1;
+ }
+
if (tcc_activation_temp_override != 0) {
tcc_activation_temp = tcc_activation_temp_override;
- fprintf(outf, "Using cmdline TCC Target (%d C)\n", tcc_activation_temp);
+ fprintf(outf, "cpu%d: Using cmdline TCC Target (%d C)\n",
+ cpu, tcc_activation_temp);
return 0;
}
- tcc_activation_temp = read_tcc_activation_temp();
- if (tcc_activation_temp)
- return 0;
+ /* Temperature Target MSR is Nehalem and newer only */
+ if (!do_nhm_platform_info)
+ goto guess;
+
+ if (get_msr(base_cpu, MSR_IA32_TEMPERATURE_TARGET, &msr))
+ goto guess;
+
+ target_c_local = (msr >> 16) & 0xFF;
+ if (!quiet)
+ fprintf(outf, "cpu%d: MSR_IA32_TEMPERATURE_TARGET: 0x%08llx (%d C)\n",
+ cpu, msr, target_c_local);
+
+ if (!target_c_local)
+ goto guess;
+
+ tcc_activation_temp = target_c_local;
+
+ return 0;
+
+guess:
tcc_activation_temp = TJMAX_DEFAULT;
- fprintf(outf, "Guessing tjMax %d C, Please use -T to specify\n", tcc_activation_temp);
+ fprintf(outf, "cpu%d: Guessing tjMax %d C, Please use -T to specify\n",
+ cpu, tcc_activation_temp);
return 0;
}
case INTEL_FAM6_ROCKETLAKE:
case INTEL_FAM6_LAKEFIELD:
case INTEL_FAM6_ALDERLAKE:
+ case INTEL_FAM6_ALDERLAKE_L:
return INTEL_FAM6_CANNONLAKE_L;
case INTEL_FAM6_ATOM_TREMONT_L:
return INTEL_FAM6_ATOM_TREMONT;
case INTEL_FAM6_ICELAKE_X:
+ case INTEL_FAM6_ICELAKE_D:
case INTEL_FAM6_SAPPHIRERAPIDS_X:
return INTEL_FAM6_SKYLAKE_X;
}
close(fd);
}
+
+/*
+ * Linux-perf manages the the HW instructions-retired counter
+ * by enabling when requested, and hiding rollover
+ */
+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");
+
+ BIC_PRESENT(BIC_IPC);
+}
+
void process_cpuid()
{
unsigned int eax, ebx, ecx, edx;
unsigned int fms, family, model, stepping, ecx_flags, edx_flags;
unsigned int has_turbo;
+ unsigned long long ucode_patch = 0;
eax = ebx = ecx = edx = 0;
hygon_genuine = 1;
if (!quiet)
- fprintf(outf, "CPUID(0): %.4s%.4s%.4s ",
- (char *)&ebx, (char *)&edx, (char *)&ecx);
+ fprintf(outf, "CPUID(0): %.4s%.4s%.4s 0x%x CPUID levels\n",
+ (char *)&ebx, (char *)&edx, (char *)&ecx, max_level);
__cpuid(1, fms, ebx, ecx, edx);
family = (fms >> 8) & 0xf;
ecx_flags = ecx;
edx_flags = edx;
+ if (get_msr(sched_getcpu(), MSR_IA32_UCODE_REV, &ucode_patch))
+ warnx("get_msr(UCODE)\n");
+
/*
* check max extended function levels of CPUID.
* This is needed to check for invariant TSC.
__cpuid(0x80000000, max_extended_level, ebx, ecx, edx);
if (!quiet) {
- fprintf(outf, "0x%x CPUID levels; 0x%x xlevels; family:model:stepping 0x%x:%x:%x (%d:%d:%d)\n",
- max_level, max_extended_level, family, model, stepping, family, model, stepping);
+ 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(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" : "-",
ecx_flags & (1 << 3) ? "MONITOR" : "-",
check_dev_msr();
check_permissions();
process_cpuid();
+ linux_perf_init();
if (!quiet)
}
void print_version() {
- fprintf(outf, "turbostat version 20.09.30"
+ fprintf(outf, "turbostat version 21.03.12"
" - Len Brown <lenb@kernel.org>\n");
}
{"debug", no_argument, 0, 'd'}, /* internal, not documented */
{"enable", required_argument, 0, 'e'},
{"interval", required_argument, 0, 'i'},
+ {"IPC", no_argument, 0, 'I'},
{"num_iterations", required_argument, 0, 'n'},
{"help", no_argument, 0, 'h'},
{"hide", required_argument, 0, 'H'}, // meh, -h taken by --help