diff options
Diffstat (limited to 'tools/power/x86')
| -rw-r--r-- | tools/power/x86/intel-speed-select/isst-config.c | 25 | ||||
| -rw-r--r-- | tools/power/x86/intel-speed-select/isst-core-mbox.c | 3 | ||||
| -rw-r--r-- | tools/power/x86/intel-speed-select/isst-core-tpmi.c | 12 | ||||
| -rw-r--r-- | tools/power/x86/intel-speed-select/isst-core.c | 7 | ||||
| -rw-r--r-- | tools/power/x86/intel-speed-select/isst-display.c | 30 | ||||
| -rw-r--r-- | tools/power/x86/intel-speed-select/isst.h | 2 | ||||
| -rw-r--r-- | tools/power/x86/turbostat/Makefile | 32 | ||||
| -rw-r--r-- | tools/power/x86/turbostat/turbostat.8 | 169 | ||||
| -rw-r--r-- | tools/power/x86/turbostat/turbostat.c | 5330 | ||||
| -rw-r--r-- | tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c | 1 |
10 files changed, 4833 insertions, 778 deletions
diff --git a/tools/power/x86/intel-speed-select/isst-config.c b/tools/power/x86/intel-speed-select/isst-config.c index d865dc1f89ee..fadfb02b8611 100644 --- a/tools/power/x86/intel-speed-select/isst-config.c +++ b/tools/power/x86/intel-speed-select/isst-config.c @@ -16,9 +16,9 @@ struct process_cmd_struct { int arg; }; -static const char *version_str = "v1.18"; +static const char *version_str = "v1.21"; -static const int supported_api_ver = 2; +static const int supported_api_ver = 3; static struct isst_if_platform_info isst_platform_info; static char *progname; static int debug_flag; @@ -46,6 +46,8 @@ static int force_online_offline; static int auto_mode; static int fact_enable_fail; static int cgroupv2; +static int max_die_id; +static int max_punit_id; /* clos related */ static int current_clos = -1; @@ -562,6 +564,18 @@ void for_each_online_power_domain_in_set(void (*callback)(struct isst_id *, void } for (i = 0; i < MAX_PACKAGE_COUNT; i++) { + if (max_die_id == max_punit_id) { + for (k = 0; k < MAX_PUNIT_PER_DIE && k < MAX_DIE_PER_PACKAGE; k++) { + id.cpu = cpus[i][k][k]; + id.pkg = i; + id.die = k; + id.punit = k; + if (isst_is_punit_valid(&id)) + callback(&id, arg1, arg2, arg3, arg4); + } + continue; + } + for (j = 0; j < MAX_DIE_PER_PACKAGE; j++) { /* * Fix me: @@ -795,6 +809,12 @@ static void create_cpu_map(void) cpu_cnt[pkg_id][die_id][punit_id]++; + if (max_die_id < die_id) + max_die_id = die_id; + + if (max_punit_id < cpu_map[i].punit_id) + max_punit_id = cpu_map[i].punit_id; + debug_printf( "map logical_cpu:%d core: %d die:%d pkg:%d punit:%d punit_cpu:%d punit_core:%d\n", i, cpu_map[i].core_id, cpu_map[i].die_id, @@ -2054,6 +2074,7 @@ static void dump_fact_config_for_cpu(struct isst_id *id, void *arg1, void *arg2, struct isst_fact_info fact_info; int ret; + memset(&fact_info, 0, sizeof(fact_info)); ret = isst_get_fact_info(id, tdp_level, fact_bucket, &fact_info); if (ret) { isst_display_error_info_message(1, "Failed to get turbo-freq info at this level", 1, tdp_level); diff --git a/tools/power/x86/intel-speed-select/isst-core-mbox.c b/tools/power/x86/intel-speed-select/isst-core-mbox.c index 24bea57f4ff5..c81ecd602bcf 100644 --- a/tools/power/x86/intel-speed-select/isst-core-mbox.c +++ b/tools/power/x86/intel-speed-select/isst-core-mbox.c @@ -746,6 +746,7 @@ static int mbox_set_pbf_fact_status(struct isst_id *id, int pbf, int enable) static int _get_fact_bucket_info(struct isst_id *id, int level, struct isst_fact_bucket_info *bucket_info) { + int trl_max_levels = isst_get_trl_max_levels(); unsigned int resp; int i, k, ret; @@ -769,7 +770,7 @@ static int _get_fact_bucket_info(struct isst_id *id, int level, } } - for (k = 0; k < 3; ++k) { + for (k = 0; k < trl_max_levels; ++k) { for (i = 0; i < 2; ++i) { int j; diff --git a/tools/power/x86/intel-speed-select/isst-core-tpmi.c b/tools/power/x86/intel-speed-select/isst-core-tpmi.c index 3458768562e5..da53aaa27fc9 100644 --- a/tools/power/x86/intel-speed-select/isst-core-tpmi.c +++ b/tools/power/x86/intel-speed-select/isst-core-tpmi.c @@ -194,8 +194,14 @@ static int tpmi_get_ctdp_control(struct isst_id *id, int config_index, if (!(info.level_mask & level_mask)) return -1; - ctdp_level->fact_support = info.sst_tf_support; - ctdp_level->pbf_support = info.sst_bf_support; + if (api_version() > 2) { + ctdp_level->fact_support = info.sst_tf_support & BIT(config_index); + ctdp_level->pbf_support = info.sst_bf_support & BIT(config_index); + } else { + ctdp_level->fact_support = info.sst_tf_support; + ctdp_level->pbf_support = info.sst_bf_support; + } + ctdp_level->fact_enabled = !!(info.feature_state & BIT(1)); ctdp_level->pbf_enabled = !!(info.feature_state & BIT(0)); @@ -323,7 +329,7 @@ static int tpmi_get_get_trls(struct isst_id *id, int config_index, return 0; } -static int tpmi_get_get_trl(struct isst_id *id, int level, int config_index, +static int tpmi_get_get_trl(struct isst_id *id, int config_index, int level, int *trl) { struct isst_pkg_ctdp_level_info ctdp_level; diff --git a/tools/power/x86/intel-speed-select/isst-core.c b/tools/power/x86/intel-speed-select/isst-core.c index f55fef4c13a7..e05561d00458 100644 --- a/tools/power/x86/intel-speed-select/isst-core.c +++ b/tools/power/x86/intel-speed-select/isst-core.c @@ -23,6 +23,7 @@ int isst_set_platform_ops(int api_version) isst_ops = mbox_get_platform_ops(); break; case 2: + case 3: isst_ops = tpmi_get_platform_ops(); break; default: @@ -282,6 +283,8 @@ int isst_set_trl(struct isst_id *id, unsigned long long trl) return 0; } +#define MSR_TRL_FREQ_MULTIPLIER 100 + int isst_set_trl_from_current_tdp(struct isst_id *id, unsigned long long trl) { unsigned long long msr_trl; @@ -309,6 +312,10 @@ int isst_set_trl_from_current_tdp(struct isst_id *id, unsigned long long trl) for (i = 0; i < 8; ++i) { unsigned long long _trl = trl[i]; + /* MSR is always in 100 MHz unit */ + if (isst_get_disp_freq_multiplier() == 1) + _trl /= MSR_TRL_FREQ_MULTIPLIER; + msr_trl |= (_trl << (i * 8)); } } diff --git a/tools/power/x86/intel-speed-select/isst-display.c b/tools/power/x86/intel-speed-select/isst-display.c index 14c9b037859a..07ebd08f3202 100644 --- a/tools/power/x86/intel-speed-select/isst-display.c +++ b/tools/power/x86/intel-speed-select/isst-display.c @@ -172,12 +172,19 @@ static int print_package_info(struct isst_id *id, FILE *outf) int level = 1; if (out_format_is_json()) { - if (api_version() > 1) - snprintf(header, sizeof(header), "package-%d:die-%d:powerdomain-%d:cpu-%d", - id->pkg, id->die, id->punit, id->cpu); - else + if (api_version() > 1) { + if (id->cpu < 0) + snprintf(header, sizeof(header), + "package-%d:die-%d:powerdomain-%d:cpu-None", + id->pkg, id->die, id->punit); + else + snprintf(header, sizeof(header), + "package-%d:die-%d:powerdomain-%d:cpu-%d", + id->pkg, id->die, id->punit, id->cpu); + } else { snprintf(header, sizeof(header), "package-%d:die-%d:cpu-%d", id->pkg, id->die, id->cpu); + } format_and_print(outf, level, header, NULL); return 1; } @@ -189,7 +196,12 @@ static int print_package_info(struct isst_id *id, FILE *outf) snprintf(header, sizeof(header), "powerdomain-%d", id->punit); format_and_print(outf, level++, header, NULL); } - snprintf(header, sizeof(header), "cpu-%d", id->cpu); + + if (id->cpu < 0) + snprintf(header, sizeof(header), "cpu-None"); + else + snprintf(header, sizeof(header), "cpu-%d", id->cpu); + format_and_print(outf, level, header, NULL); return level; @@ -199,8 +211,8 @@ static void _isst_pbf_display_information(struct isst_id *id, FILE *outf, int le struct isst_pbf_info *pbf_info, int disp_level) { - char header[256]; - char value[512]; + static char header[256]; + static char value[1024]; snprintf(header, sizeof(header), "speed-select-base-freq-properties"); format_and_print(outf, disp_level, header, NULL); @@ -338,8 +350,8 @@ void isst_ctdp_display_core_info(struct isst_id *id, FILE *outf, char *prefix, void isst_ctdp_display_information(struct isst_id *id, FILE *outf, int tdp_level, struct isst_pkg_ctdp *pkg_dev) { - char header[256]; - char value[512]; + static char header[256]; + static char value[1024]; static int level; int trl_max_levels = isst_get_trl_max_levels(); int i; diff --git a/tools/power/x86/intel-speed-select/isst.h b/tools/power/x86/intel-speed-select/isst.h index 4bddd3c66bf7..39ee75677c2c 100644 --- a/tools/power/x86/intel-speed-select/isst.h +++ b/tools/power/x86/intel-speed-select/isst.h @@ -80,7 +80,7 @@ #define DISP_FREQ_MULTIPLIER 100 #define MAX_PACKAGE_COUNT 32 -#define MAX_DIE_PER_PACKAGE 2 +#define MAX_DIE_PER_PACKAGE 16 #define MAX_PUNIT_PER_DIE 8 /* Unified structure to specific a CPU or a Power Domain */ diff --git a/tools/power/x86/turbostat/Makefile b/tools/power/x86/turbostat/Makefile index 92e139b9c792..3946d5254a1f 100644 --- a/tools/power/x86/turbostat/Makefile +++ b/tools/power/x86/turbostat/Makefile @@ -3,6 +3,8 @@ CC = $(CROSS_COMPILE)gcc BUILD_OUTPUT := $(CURDIR) PREFIX ?= /usr DESTDIR ?= +DAY := $(shell date +%Y.%m.%d) +SNAPSHOT = turbostat-$(DAY) ifeq ("$(origin O)", "command line") BUILD_OUTPUT := $(O) @@ -12,6 +14,7 @@ turbostat : turbostat.c override CFLAGS += -O2 -Wall -Wextra -I../../../include override CFLAGS += -DMSRHEADER='"../../../../arch/x86/include/asm/msr-index.h"' override CFLAGS += -DINTEL_FAMILY_HEADER='"../../../../arch/x86/include/asm/intel-family.h"' +override CFLAGS += -DBUILD_BUG_HEADER='"../../../../include/linux/build_bug.h"' override CFLAGS += -D_FILE_OFFSET_BITS=64 override CFLAGS += -D_FORTIFY_SOURCE=2 @@ -22,9 +25,34 @@ override CFLAGS += -D_FORTIFY_SOURCE=2 .PHONY : clean clean : @rm -f $(BUILD_OUTPUT)/turbostat + @rm -f $(SNAPSHOT).tar.gz install : turbostat - install -d $(DESTDIR)$(PREFIX)/bin + install -d $(DESTDIR)$(PREFIX)/bin install $(BUILD_OUTPUT)/turbostat $(DESTDIR)$(PREFIX)/bin/turbostat - install -d $(DESTDIR)$(PREFIX)/share/man/man8 + install -d $(DESTDIR)$(PREFIX)/share/man/man8 install -m 644 turbostat.8 $(DESTDIR)$(PREFIX)/share/man/man8 + +snapshot: turbostat + @rm -rf $(SNAPSHOT) + @mkdir $(SNAPSHOT) + @cp turbostat Makefile turbostat.c turbostat.8 ../../../../arch/x86/include/asm/intel-family.h $(SNAPSHOT) + + @sed -e 's/^#include <linux\/bits.h>/#include "bits.h"/' ../../../../arch/x86/include/asm/msr-index.h > $(SNAPSHOT)/msr-index.h + @echo '#define ARRAY_SIZE(arr) (sizeof(arr) / sizeof((arr)[0]))' >> $(SNAPSHOT)/msr-index.h + @echo "#define BIT(x) (1 << (x))" > $(SNAPSHOT)/bits.h + @echo "#define BIT_ULL(nr) (1ULL << (nr))" >> $(SNAPSHOT)/bits.h + @echo "#define GENMASK(h, l) (((~0UL) << (l)) & (~0UL >> (sizeof(long) * 8 - 1 - (h))))" >> $(SNAPSHOT)/bits.h + @echo "#define GENMASK_ULL(h, l) (((~0ULL) << (l)) & (~0ULL >> (sizeof(long long) * 8 - 1 - (h))))" >> $(SNAPSHOT)/bits.h + + @echo '#define BUILD_BUG_ON(cond) do { enum { compile_time_check ## __COUNTER__ = 1/(!(cond)) }; } while (0)' > $(SNAPSHOT)/build_bug.h + @echo '#define __must_be_array(arr) 0' >> $(SNAPSHOT)/build_bug.h + + @echo PWD=. > $(SNAPSHOT)/Makefile + @echo "CFLAGS += -DMSRHEADER='\"msr-index.h\"'" >> $(SNAPSHOT)/Makefile + @echo "CFLAGS += -DINTEL_FAMILY_HEADER='\"intel-family.h\"'" >> $(SNAPSHOT)/Makefile + @echo "CFLAGS += -DBUILD_BUG_HEADER='\"build_bug.h\"'" >> $(SNAPSHOT)/Makefile + @sed -e's/.*MSRHEADER.*//' -e's/.*INTEL_FAMILY_HEADER.*//' -e's/.*BUILD_BUG_HEADER.*//' Makefile >> $(SNAPSHOT)/Makefile + + @rm -f $(SNAPSHOT).tar.gz + tar cvzf $(SNAPSHOT).tar.gz $(SNAPSHOT) diff --git a/tools/power/x86/turbostat/turbostat.8 b/tools/power/x86/turbostat/turbostat.8 index 8f08c3fd498d..99bf905ade81 100644 --- a/tools/power/x86/turbostat/turbostat.8 +++ b/tools/power/x86/turbostat/turbostat.8 @@ -28,10 +28,16 @@ name as necessary to disambiguate it from others is necessary. Note that option .PP \fB--add attributes\fP add column with counter having specified 'attributes'. The 'location' attribute is required, all others are optional. .nf - location: {\fBmsrDDD\fP | \fBmsr0xXXX\fP | \fB/sys/path...\fP} + location: {\fBmsrDDD\fP | \fBmsr0xXXX\fP | \fB/sys/path...\fP | \fBperf/<device>/<event>\fP} msrDDD is a decimal offset, eg. msr16 msr0xXXX is a hex offset, eg. msr0x10 /sys/path... is an absolute path to a sysfs attribute + <device> is a perf device from /sys/bus/event_source/devices/<device> eg. cstate_core + On Intel hybrid platforms, instead of one "cpu" perf device there are two, "cpu_core" and "cpu_atom" devices for P and E cores respectively. + Turbostat, in this case, allow user to use "cpu" device and will automatically detect the type of a CPU and translate it to "cpu_core" and "cpu_atom" accordingly. + For a complete example see "ADD PERF COUNTER EXAMPLE #2 (using virtual "cpu" device)". + <event> is a perf event for given device from /sys/bus/event_source/devices/<device>/events/<event> eg. c1-residency + perf/cstate_core/c1-residency would then use /sys/bus/event_source/devices/cstate_core/events/c1-residency scope: {\fBcpu\fP | \fBcore\fP | \fBpackage\fP} sample and print the counter for every cpu, core, or package. @@ -52,6 +58,39 @@ name as necessary to disambiguate it from others is necessary. Note that option as the column header. .fi .PP +\fB--add pmt,[attr_name=attr_value, ...]\fP add column with a PMT (Intel Platform Monitoring Technology) counter in a similar way to --add option above, but require PMT metadata to be supplied to correctly read and display the counter. The metadata can be found in the Intel PMT XML files, hosted at https://github.com/intel/Intel-PMT. For a complete example see "ADD PMT COUNTER EXAMPLE". +.nf + name="name_string" + For column header. + + type={\fBraw\fP} + 'raw' shows the counter contents in hex. + default: raw + + format={\fBraw\fP | \fBdelta\fP} + 'raw' shows the counter contents in hex. + 'delta' shows the difference in values during the measurement interval. + default: raw + + domain={\fBcpu%u\fP | \fBcore%u\fP | \fBpackage%u\fP} + 'cpu' per cpu/thread counter. + 'core' per core counter. + 'package' per package counter. + '%u' denotes id of the domain that the counter is associated with. For example core4 would mean that the counter is associated with core number 4. + + offset=\fB%u\fP + '%u' offset within the PMT MMIO region. + + lsb=\fB%u\fP + '%u' least significant bit within the 64 bit value read from 'offset'. Together with 'msb', used to form a read mask. + + msb=\fB%u\fP + '%u' most significant bit within the 64 bit value read from 'offset'. Together with 'lsb', used to form a read mask. + + guid=\fB%x\fP + '%x' hex identifier of the PMT MMIO region. +.fi +.PP \fB--cpu cpu-set\fP limit output to system summary plus the specified cpu-set. If cpu-set is the string "core", then the system summary plus the first CPU in each core are printed -- eg. subsequent HT siblings are not printed. Or if cpu-set is the string "package", then the system summary plus the first CPU in each package is printed. Otherwise, the system summary plus the specified set of CPUs are printed. The cpu-set is ordered from low to high, comma delimited with ".." and "-" permitted to denote a range. eg. 1,2,8,14..17,21-44 .PP \fB--hide column\fP do not show the specified built-in columns. May be invoked multiple times, or with a comma-separated list of column names. @@ -67,6 +106,10 @@ 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--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. @@ -93,7 +136,7 @@ displays the statistics gathered since it was forked. The system configuration dump (if --quiet is not used) is followed by statistics. The first row of the statistics labels the content of each column (below). The second row of statistics is the system summary line. The system summary line has a '-' in the columns for the Package, Core, and CPU. The contents of the system summary line depends on the type of column. Columns that count items (eg. IRQ) show the sum across all CPUs in the system. Columns that show a percentage show the average across all CPUs in the system. Columns that dump raw MSR values simply show 0 in the summary. After the system summary row, each row describes a specific Package/Core/CPU. Note that if the --cpu parameter is used to limit which specific CPUs are displayed, turbostat will still collect statistics for all CPUs in the system and will still show the system summary for all CPUs in the system. .SH COLUMN DESCRIPTIONS .PP -\fBusec\fP For each CPU, the number of microseconds elapsed during counter collection, including thread migration -- if any. This counter is disabled by default, and is enabled with "--enable usec", or --debug. On the summary row, usec refers to the total elapsed time to collect the counters on all cpus. +\fBusec\fP For each CPU, the number of microseconds elapsed during counter collection, including thread migration -- if any. This counter is disabled by default, and is enabled with "--enable usec", or --debug. On the summary row, usec refers to the total elapsed time to snapshot the procfs/sysfs and collect the counters on all cpus. .PP \fBTime_Of_Day_Seconds\fP For each CPU, the gettimeofday(2) value (seconds.subsec since Epoch) when the counters ending the measurement interval were collected. This column is disabled by default, and can be enabled with "--enable Time_Of_Day_Seconds" or "--debug". On the summary row, Time_Of_Day_Seconds refers to the timestamp following collection of counters on the last CPU. .PP @@ -125,9 +168,17 @@ 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 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 -\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. +\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 @@ -139,11 +190,15 @@ The system configuration dump (if --quiet is not used) is followed by statistics .PP \fBRAMWatt\fP Watts consumed by the DRAM DIMMS -- available only on server processors. .PP +\fBSysWatt\fP Watts consumed by the whole platform (RAPL PSYS). +.PP \fBPKG_%\fP percent of the interval that RAPL throttling was active on the Package. Note that the system summary is the sum of the package throttling time, and thus may be higher than 100% on a multi-package system. Note that the meaning of this field is model specific. For example, some hardware increments this counter when RAPL responds to thermal limits, but does not increment this counter when RAPL responds to power limits. Comparing PkgWatt and PkgTmp to system limits is necessary. .PP \fBRAM_%\fP percent of the interval that RAPL throttling was active on DRAM. .PP -\fBUncMHz\fP uncore MHz, instantaneous sample. +\fBUncMHz\fP per-package uncore MHz, instantaneous sample. +.PP +\fBUMHz1.0\fP per-package uncore MHz for domain=1 and fabric_cluster=0, instantaneous sample. System summary is the average of all packages. .SH TOO MUCH INFORMATION EXAMPLE By default, turbostat dumps all possible information -- a system configuration header, followed by columns for all counters. This is ideal for remote debugging, use the "--out" option to save everything to a text file, and get that file to the expert helping you debug. @@ -306,7 +361,7 @@ available on all processors. Here we limit turbostat to showing just the CPU number for cpu0 - cpu3. We add a counter showing the 32-bit raw value of MSR 0x199 (MSR_IA32_PERF_CTL), labeling it with the column header, "PRF_CTRL", and display it only once, -afte the conclusion of a 0.1 second sleep. +after the conclusion of a 0.1 second sleep. .nf sudo ./turbostat --quiet --cpu 0-3 --show CPU --add msr0x199,u32,raw,PRF_CTRL sleep .1 0.101604 sec @@ -319,6 +374,78 @@ CPU PRF_CTRL .fi +.SH ADD PERF COUNTER EXAMPLE +Here we limit turbostat to showing just the CPU number for cpu0 - cpu3. +We add a counter showing time spent in C1 core cstate, +labeling it with the column header, "pCPU%c1", and display it only once, +after the conclusion of 0.1 second sleep. +We also show CPU%c1 built-in counter that should show similar values. +.nf +sudo ./turbostat --quiet --cpu 0-3 --show CPU,CPU%c1 --add perf/cstate_core/c1-residency,cpu,delta,percent,pCPU%c1 sleep .1 +0.102448 sec +CPU pCPU%c1 CPU%c1 +- 34.89 34.89 +0 45.99 45.99 +1 45.94 45.94 +2 23.83 23.83 +3 23.84 23.84 + +.fi + +.SH ADD PERF COUNTER EXAMPLE #2 (using virtual cpu device) +Here we run on hybrid, Raptor Lake platform. +We limit turbostat to show output for just cpu0 (pcore) and cpu12 (ecore). +We add a counter showing number of L3 cache misses, using virtual "cpu" device, +labeling it with the column header, "VCMISS". +We add a counter showing number of L3 cache misses, using virtual "cpu_core" device, +labeling it with the column header, "PCMISS". This will fail on ecore cpu12. +We add a counter showing number of L3 cache misses, using virtual "cpu_atom" device, +labeling it with the column header, "ECMISS". This will fail on pcore cpu0. +We display it only once, after the conclusion of 0.1 second sleep. +.nf +sudo ./turbostat --quiet --cpu 0,12 --show CPU --add perf/cpu/cache-misses,cpu,delta,raw,VCMISS --add perf/cpu_core/cache-misses,cpu,delta,raw,PCMISS --add perf/cpu_atom/cache-misses,cpu,delta,raw,ECMISS sleep .1 +turbostat: added_perf_counters_init_: perf/cpu_atom/cache-misses: failed to open counter on cpu0 +turbostat: added_perf_counters_init_: perf/cpu_core/cache-misses: failed to open counter on cpu12 +0.104630 sec +CPU ECMISS PCMISS VCMISS +- 0x0000000000000000 0x0000000000000000 0x0000000000000000 +0 0x0000000000000000 0x0000000000007951 0x0000000000007796 +12 0x000000000001137a 0x0000000000000000 0x0000000000011392 + +.fi + +.SH ADD PMT COUNTER EXAMPLE +Here we limit turbostat to showing just the CPU number 0. +We add two counters, showing crystal clock count and the DC6 residency. +All the parameters passed are based on the metadata found in the PMT XML files. + +For the crystal clock count, we +label it with the column header, "XTAL", +we set the type to 'raw', to read the number of clock ticks in hex, +we set the format to 'delta', to display the difference in ticks during the measurement interval, +we set the domain to 'package0', to collect it and associate it with the whole package number 0, +we set the offset to '0', which is a offset of the counter within the PMT MMIO region, +we set the lsb and msb to cover all 64 bits of the read 64 bit value, +and finally we set the guid to '0x1a067102', that identifies the PMT MMIO region to which the 'offset' is applied to read the counter value. + +For the DC6 residency counter, we +label it with the column header, "Die%c6", +we set the type to 'txtal_time', to obtain the percent residency value +we set the format to 'delta', to display the difference in ticks during the measurement interval, +we set the domain to 'package0', to collect it and associate it with the whole package number 0, +we set the offset to '0', which is a offset of the counter within the PMT MMIO region, +we set the lsb and msb to cover all 64 bits of the read 64 bit value, +and finally we set the guid to '0x1a067102', that identifies the PMT MMIO region to which the 'offset' is applied to read the counter value. + +.nf +sudo ./turbostat --quiet --cpu 0 --show CPU --add pmt,name=XTAL,type=raw,format=delta,domain=package0,offset=0,lsb=0,msb=63,guid=0x1a067102 --add pmt,name=Die%c6,type=txtal_time,format=delta,domain=package0,offset=120,lsb=0,msb=63,guid=0x1a067102 +0.104352 sec +CPU XTAL Die%c6 +- 0x0000006d4d957ca7 0.00 +0 0x0000006d4d957ca7 0.00 +0.102448 sec +.fi + .SH INPUT For interval-mode, turbostat will immediately end the current interval @@ -370,7 +497,7 @@ 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. @@ -389,14 +516,40 @@ that they count at TSC rate, which is true on all processors tested to date. Volume 3B: System Programming Guide" https://www.intel.com/products/processor/manuals/ +.SH RUN THE LATEST VERSION +If turbostat complains that it doesn't recognize your processor, +please try the latest version. + +The latest version of turbostat does not require the latest version of the Linux kernel. +However, some features, such as perf(1) counters, do require kernel support. + +The latest turbostat release is available in the upstream Linux Kernel source tree. +eg. "git pull https://git.kernel.org/pub/scm/linux/kernel/git/torvalds/linux.git" +and run make in tools/power/x86/turbostat/. + +n.b. "make install" will update your system manually, but a distro update may subsequently downgrade your turbostat to an older version. +For this reason, manually installing to /usr/local/bin may be what you want. + +Note that turbostat/Makefile has a "make snapshot" target, which will create a tar file +that can build without a local kernel source tree. + +If the upstream version isn't new enough, the development tree can be found here: +"git pull https://git.kernel.org/pub/scm/linux/kernel/git/lenb/linux.git turbostat" + +If the development tree doesn't work, please contact the author via chat, +or via email with the word "turbostat" on the Subject line. + .SH FILES .ta .nf +/sys/bus/event_source/devices/ /dev/cpu/*/msr +/sys/class/intel_pmt/ +/sys/devices/system/cpu/ .fi .SH "SEE ALSO" -msr(4), vmstat(8) +perf(1), msr(4), vmstat(8) .PP .SH AUTHOR .nf diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c index 7a334377f92b..8d5011a0bf60 100644 --- a/tools/power/x86/turbostat/turbostat.c +++ b/tools/power/x86/turbostat/turbostat.c @@ -3,13 +3,41 @@ * turbostat -- show CPU frequency and C-state residency * on modern Intel and AMD processors. * - * Copyright (c) 2023 Intel Corporation. + * Copyright (c) 2025 Intel Corporation. * Len Brown <len.brown@intel.com> */ #define _GNU_SOURCE #include MSRHEADER + +// copied from arch/x86/include/asm/cpu_device_id.h +#define VFM_MODEL_BIT 0 +#define VFM_FAMILY_BIT 8 +#define VFM_VENDOR_BIT 16 +#define VFM_RSVD_BIT 24 + +#define VFM_MODEL_MASK GENMASK(VFM_FAMILY_BIT - 1, VFM_MODEL_BIT) +#define VFM_FAMILY_MASK GENMASK(VFM_VENDOR_BIT - 1, VFM_FAMILY_BIT) +#define VFM_VENDOR_MASK GENMASK(VFM_RSVD_BIT - 1, VFM_VENDOR_BIT) + +#define VFM_MODEL(vfm) (((vfm) & VFM_MODEL_MASK) >> VFM_MODEL_BIT) +#define VFM_FAMILY(vfm) (((vfm) & VFM_FAMILY_MASK) >> VFM_FAMILY_BIT) +#define VFM_VENDOR(vfm) (((vfm) & VFM_VENDOR_MASK) >> VFM_VENDOR_BIT) + +#define VFM_MAKE(_vendor, _family, _model) ( \ + ((_model) << VFM_MODEL_BIT) | \ + ((_family) << VFM_FAMILY_BIT) | \ + ((_vendor) << VFM_VENDOR_BIT) \ +) +// end copied section + +#define CPUID_LEAF_MODEL_ID 0x1A +#define CPUID_LEAF_MODEL_ID_CORE_TYPE_SHIFT 24 + +#define X86_VENDOR_INTEL 0 + #include INTEL_FAMILY_HEADER +#include BUILD_BUG_HEADER #include <stdarg.h> #include <stdio.h> #include <err.h> @@ -19,6 +47,7 @@ #include <sys/stat.h> #include <sys/select.h> #include <sys/resource.h> +#include <sys/mman.h> #include <fcntl.h> #include <signal.h> #include <sys/time.h> @@ -36,6 +65,8 @@ #include <linux/perf_event.h> #include <asm/unistd.h> #include <stdbool.h> +#include <assert.h> +#include <linux/kernel.h> #define UNUSED(x) (void)(x) @@ -52,15 +83,55 @@ */ #define NAME_BYTES 20 #define PATH_BYTES 128 +#define PERF_NAME_BYTES 128 + +#define MAX_NOFILE 0x8000 + +#define COUNTER_KIND_PERF_PREFIX "perf/" +#define COUNTER_KIND_PERF_PREFIX_LEN strlen(COUNTER_KIND_PERF_PREFIX) +#define PERF_DEV_NAME_BYTES 32 +#define PERF_EVT_NAME_BYTES 32 + +#define INTEL_ECORE_TYPE 0x20 +#define INTEL_PCORE_TYPE 0x40 + +#define ROUND_UP_TO_PAGE_SIZE(n) (((n) + 0x1000UL-1UL) & ~(0x1000UL-1UL)) 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 counter_type { COUNTER_ITEMS, COUNTER_CYCLES, COUNTER_SECONDS, COUNTER_USEC, COUNTER_K2M }; +enum counter_format { FORMAT_RAW, FORMAT_DELTA, FORMAT_PERCENT, FORMAT_AVERAGE }; +enum counter_source { COUNTER_SOURCE_NONE, COUNTER_SOURCE_PERF, COUNTER_SOURCE_MSR }; + +struct perf_counter_info { + struct perf_counter_info *next; + + /* How to open the counter / What counter it is. */ + char device[PERF_DEV_NAME_BYTES]; + char event[PERF_EVT_NAME_BYTES]; + + /* How to show/format the counter. */ + char name[PERF_NAME_BYTES]; + unsigned int width; + enum counter_scope scope; + enum counter_type type; + enum counter_format format; + double scale; + + /* For reading the counter. */ + int *fd_perf_per_domain; + size_t num_domains; +}; + +struct sysfs_path { + char path[PATH_BYTES]; + int id; + struct sysfs_path *next; +}; struct msr_counter { unsigned int msr_num; char name[NAME_BYTES]; - char path[PATH_BYTES]; + struct sysfs_path *sp; unsigned int width; enum counter_type type; enum counter_format format; @@ -72,61 +143,69 @@ struct msr_counter { }; struct msr_counter bic[] = { - { 0x0, "usec", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Time_Of_Day_Seconds", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Package", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Node", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Avg_MHz", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Busy%", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Bzy_MHz", "", 0, 0, 0, NULL, 0 }, - { 0x0, "TSC_MHz", "", 0, 0, 0, NULL, 0 }, - { 0x0, "IRQ", "", 0, 0, 0, NULL, 0 }, - { 0x0, "SMI", "", 32, 0, FORMAT_DELTA, NULL, 0 }, - { 0x0, "sysfs", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CPU%c1", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CPU%c3", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CPU%c6", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CPU%c7", "", 0, 0, 0, NULL, 0 }, - { 0x0, "ThreadC", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CoreTmp", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CoreCnt", "", 0, 0, 0, NULL, 0 }, - { 0x0, "PkgTmp", "", 0, 0, 0, NULL, 0 }, - { 0x0, "GFX%rc6", "", 0, 0, 0, NULL, 0 }, - { 0x0, "GFXMHz", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pkg%pc2", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pkg%pc3", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pkg%pc6", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pkg%pc7", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pkg%pc8", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pkg%pc9", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pk%pc10", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CPU%LPI", "", 0, 0, 0, NULL, 0 }, - { 0x0, "SYS%LPI", "", 0, 0, 0, NULL, 0 }, - { 0x0, "PkgWatt", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CorWatt", "", 0, 0, 0, NULL, 0 }, - { 0x0, "GFXWatt", "", 0, 0, 0, NULL, 0 }, - { 0x0, "PkgCnt", "", 0, 0, 0, NULL, 0 }, - { 0x0, "RAMWatt", "", 0, 0, 0, NULL, 0 }, - { 0x0, "PKG_%", "", 0, 0, 0, NULL, 0 }, - { 0x0, "RAM_%", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Pkg_J", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Cor_J", "", 0, 0, 0, NULL, 0 }, - { 0x0, "GFX_J", "", 0, 0, 0, NULL, 0 }, - { 0x0, "RAM_J", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Mod%c6", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Totl%C0", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Any%C0", "", 0, 0, 0, NULL, 0 }, - { 0x0, "GFX%C0", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CPUGFX%", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Core", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CPU", "", 0, 0, 0, NULL, 0 }, - { 0x0, "APIC", "", 0, 0, 0, NULL, 0 }, - { 0x0, "X2APIC", "", 0, 0, 0, NULL, 0 }, - { 0x0, "Die", "", 0, 0, 0, NULL, 0 }, - { 0x0, "GFXAMHz", "", 0, 0, 0, NULL, 0 }, - { 0x0, "IPC", "", 0, 0, 0, NULL, 0 }, - { 0x0, "CoreThr", "", 0, 0, 0, NULL, 0 }, - { 0x0, "UncMHz", "", 0, 0, 0, NULL, 0 }, + { 0x0, "usec", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Time_Of_Day_Seconds", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Package", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Node", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Avg_MHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Busy%", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Bzy_MHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "TSC_MHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "IRQ", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "SMI", NULL, 32, 0, FORMAT_DELTA, NULL, 0 }, + { 0x0, "sysfs", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU%c1", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU%c3", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU%c6", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU%c7", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "ThreadC", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CoreTmp", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CoreCnt", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "PkgTmp", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "GFX%rc6", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "GFXMHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pkg%pc2", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pkg%pc3", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pkg%pc6", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pkg%pc7", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pkg%pc8", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pkg%pc9", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pk%pc10", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU%LPI", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "SYS%LPI", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "PkgWatt", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CorWatt", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "GFXWatt", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "PkgCnt", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "RAMWatt", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "PKG_%", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "RAM_%", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Pkg_J", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Cor_J", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "GFX_J", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "RAM_J", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Mod%c6", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Totl%C0", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Any%C0", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "GFX%C0", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPUGFX%", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Core", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "APIC", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "X2APIC", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Die", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "GFXAMHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "IPC", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CoreThr", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "UncMHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "SAM%mc6", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "SAMMHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "SAMAMHz", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Die%c6", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "SysWatt", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "Sys_J", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "NMI", NULL, 0, 0, 0, NULL, 0 }, + { 0x0, "CPU%c1e", NULL, 0, 0, 0, NULL, 0 }, }; #define MAX_BIC (sizeof(bic) / sizeof(struct msr_counter)) @@ -185,12 +264,20 @@ struct msr_counter bic[] = { #define BIC_IPC (1ULL << 52) #define BIC_CORE_THROT_CNT (1ULL << 53) #define BIC_UNCORE_MHZ (1ULL << 54) - -#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_OTHER ( BIC_IRQ | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) +#define BIC_SAM_mc6 (1ULL << 55) +#define BIC_SAMMHz (1ULL << 56) +#define BIC_SAMACTMHz (1ULL << 57) +#define BIC_Diec6 (1ULL << 58) +#define BIC_SysWatt (1ULL << 59) +#define BIC_Sys_J (1ULL << 60) +#define BIC_NMI (1ULL << 61) +#define BIC_CPU_c1e (1ULL << 62) + +#define BIC_TOPOLOGY (BIC_Package | BIC_Node | BIC_CoreCnt | BIC_PkgCnt | BIC_Core | BIC_CPU | BIC_Die) +#define BIC_THERMAL_PWR (BIC_CoreTmp | BIC_PkgTmp | BIC_PkgWatt | BIC_CorWatt | BIC_GFXWatt | BIC_RAMWatt | BIC_PKG__ | BIC_RAM__ | BIC_SysWatt) +#define BIC_FREQUENCY (BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_SAMMHz | BIC_SAMACTMHz | BIC_UNCORE_MHZ) +#define BIC_IDLE (BIC_Busy | BIC_sysfs | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_SAM_mc6 | BIC_Diec6) +#define BIC_OTHER (BIC_IRQ | BIC_NMI | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC) #define BIC_DISABLED_BY_DEFAULT (BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC) @@ -204,6 +291,30 @@ 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) +/* + * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit: + * If you change the values, note they are used both in comparisons + * (>= PCL__7) and to index pkg_cstate_limit_strings[]. + */ +#define PCLUKN 0 /* Unknown */ +#define PCLRSV 1 /* Reserved */ +#define PCL__0 2 /* PC0 */ +#define PCL__1 3 /* PC1 */ +#define PCL__2 4 /* PC2 */ +#define PCL__3 5 /* PC3 */ +#define PCL__4 6 /* PC4 */ +#define PCL__6 7 /* PC6 */ +#define PCL_6N 8 /* PC6 No Retention */ +#define PCL_6R 9 /* PC6 Retention */ +#define PCL__7 10 /* PC7 */ +#define PCL_7S 11 /* PC7 Shrink */ +#define PCL__8 12 /* PC8 */ +#define PCL__9 13 /* PC9 */ +#define PCL_10 14 /* PC10 */ +#define PCLUNL 15 /* Unlimited */ + +struct amperf_group_fd; + char *proc_stat = "/proc/stat"; FILE *outf; int *fd_percpu; @@ -221,7 +332,9 @@ unsigned int rapl_joules; unsigned int summary_only; unsigned int list_header_only; unsigned int dump_only; +unsigned int force_load; unsigned int has_aperf; +unsigned int has_aperf_access; unsigned int has_epb; unsigned int has_turbo; unsigned int is_hybrid; @@ -242,15 +355,12 @@ 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; -double rapl_dram_energy_units, rapl_energy_units; +double rapl_dram_energy_units, rapl_energy_units, rapl_psys_energy_units; double rapl_joule_counter_range; unsigned int crystal_hz; unsigned long long tsc_hz; @@ -262,9 +372,35 @@ unsigned int has_hwp_activity_window; /* IA32_HWP_REQUEST[bits 41:32] */ 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; + +static struct timeval procsysfs_tv_begin; + int ignore_stdin; +bool no_msr; +bool no_perf; + +enum gfx_sysfs_idx { + GFX_rc6, + GFX_MHz, + GFX_ACTMHz, + SAM_mc6, + SAM_MHz, + SAM_ACTMHz, + GFX_MAX +}; + +struct gfx_sysfs_info { + FILE *fp; + unsigned int val; + unsigned long long val_ull; +}; + +static struct gfx_sysfs_info gfx_info[GFX_MAX]; int get_msr(int cpu, off_t offset, unsigned long long *msr); +int add_counter(unsigned int msr_num, char *path, char *name, + unsigned int width, enum counter_scope scope, + enum counter_type type, enum counter_format format, int flags, int package_num); /* Model specific support Start */ @@ -293,6 +429,7 @@ struct platform_features { bool has_per_core_rapl; /* Indicates cores energy collection is per-core, not per-package. AMD specific for now */ bool has_rapl_divisor; /* Divisor for Energy unit raw value from MSR_RAPL_POWER_UNIT */ bool has_fixed_rapl_unit; /* Fixed Energy Unit used for DRAM RAPL Domain */ + bool has_fixed_rapl_psys_unit; /* Fixed Energy Unit used for PSYS RAPL Domain */ int rapl_quirk_tdp; /* Hardcoded TDP value when cannot be retrieved from hardware */ int tcc_offset_bits; /* TCC Offset bits in MSR_IA32_TEMPERATURE_TARGET */ bool enable_tsc_tweak; /* Use CPU Base freq instead of TSC freq for aperf/mperf counter */ @@ -300,7 +437,7 @@ struct platform_features { }; struct platform_data { - unsigned int model; + unsigned int vfm; const struct platform_features *features; }; @@ -385,12 +522,15 @@ enum rapl_msrs { RAPL_AMD_PWR_UNIT = BIT(14), /* 0xc0010299 MSR_AMD_RAPL_POWER_UNIT */ RAPL_AMD_CORE_ENERGY_STAT = BIT(15), /* 0xc001029a MSR_AMD_CORE_ENERGY_STATUS */ RAPL_AMD_PKG_ENERGY_STAT = BIT(16), /* 0xc001029b MSR_AMD_PKG_ENERGY_STATUS */ + RAPL_PLATFORM_ENERGY_LIMIT = BIT(17), /* 0x64c MSR_PLATFORM_ENERGY_LIMIT */ + RAPL_PLATFORM_ENERGY_STATUS = BIT(18), /* 0x64d MSR_PLATFORM_ENERGY_STATUS */ }; #define RAPL_PKG (RAPL_PKG_ENERGY_STATUS | RAPL_PKG_POWER_LIMIT) #define RAPL_DRAM (RAPL_DRAM_ENERGY_STATUS | RAPL_DRAM_POWER_LIMIT) #define RAPL_CORE (RAPL_CORE_ENERGY_STATUS | RAPL_CORE_POWER_LIMIT) #define RAPL_GFX (RAPL_GFX_POWER_LIMIT | RAPL_GFX_ENERGY_STATUS) +#define RAPL_PSYS (RAPL_PLATFORM_ENERGY_STATUS | RAPL_PLATFORM_ENERGY_LIMIT) #define RAPL_PKG_ALL (RAPL_PKG | RAPL_PKG_PERF_STATUS | RAPL_PKG_POWER_INFO) #define RAPL_DRAM_ALL (RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_DRAM_POWER_INFO) @@ -591,7 +731,7 @@ static const struct platform_features skl_features = { .has_ext_cst_msrs = 1, .trl_msrs = TRL_BASE, .tcc_offset_bits = 6, - .rapl_msrs = RAPL_PKG_ALL | RAPL_CORE_ALL | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_GFX, + .rapl_msrs = RAPL_PKG_ALL | RAPL_CORE_ALL | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_GFX | RAPL_PSYS, .enable_tsc_tweak = 1, }; @@ -608,25 +748,44 @@ static const struct platform_features cnl_features = { .has_ext_cst_msrs = 1, .trl_msrs = TRL_BASE, .tcc_offset_bits = 6, - .rapl_msrs = RAPL_PKG_ALL | RAPL_CORE_ALL | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_GFX, + .rapl_msrs = RAPL_PKG_ALL | RAPL_CORE_ALL | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_GFX | RAPL_PSYS, .enable_tsc_tweak = 1, }; +/* Copied from cnl_features, with PC7/PC9 removed */ static const struct platform_features adl_features = { - .has_msr_misc_feature_control = 1, - .has_msr_misc_pwr_mgmt = 1, - .has_nhm_msrs = 1, - .has_config_tdp = 1, - .bclk_freq = BCLK_100MHZ, - .supported_cstates = CC1 | CC6 | CC7 | PC2 | PC3 | PC6 | PC8 | PC10, - .cst_limit = CST_LIMIT_HSW, - .has_irtl_msrs = 1, - .has_msr_core_c1_res = 1, - .has_ext_cst_msrs = 1, - .trl_msrs = TRL_BASE, - .tcc_offset_bits = 6, - .rapl_msrs = RAPL_PKG_ALL | RAPL_CORE_ALL | RAPL_DRAM | RAPL_DRAM_PERF_STATUS | RAPL_GFX, - .enable_tsc_tweak = 1, + .has_msr_misc_feature_control = cnl_features.has_msr_misc_feature_control, + .has_msr_misc_pwr_mgmt = cnl_features.has_msr_misc_pwr_mgmt, + .has_nhm_msrs = cnl_features.has_nhm_msrs, + .has_config_tdp = cnl_features.has_config_tdp, + .bclk_freq = cnl_features.bclk_freq, + .supported_cstates = CC1 | CC6 | CC7 | PC2 | PC3 | PC6 | PC8 | PC10, + .cst_limit = cnl_features.cst_limit, + .has_irtl_msrs = cnl_features.has_irtl_msrs, + .has_msr_core_c1_res = cnl_features.has_msr_core_c1_res, + .has_ext_cst_msrs = cnl_features.has_ext_cst_msrs, + .trl_msrs = cnl_features.trl_msrs, + .tcc_offset_bits = cnl_features.tcc_offset_bits, + .rapl_msrs = cnl_features.rapl_msrs, + .enable_tsc_tweak = cnl_features.enable_tsc_tweak, +}; + +/* Copied from adl_features, with PC3/PC8 removed */ +static const struct platform_features lnl_features = { + .has_msr_misc_feature_control = adl_features.has_msr_misc_feature_control, + .has_msr_misc_pwr_mgmt = adl_features.has_msr_misc_pwr_mgmt, + .has_nhm_msrs = adl_features.has_nhm_msrs, + .has_config_tdp = adl_features.has_config_tdp, + .bclk_freq = adl_features.bclk_freq, + .supported_cstates = CC1 | CC6 | CC7 | PC2 | PC6 | PC10, + .cst_limit = adl_features.cst_limit, + .has_irtl_msrs = adl_features.has_irtl_msrs, + .has_msr_core_c1_res = adl_features.has_msr_core_c1_res, + .has_ext_cst_msrs = adl_features.has_ext_cst_msrs, + .trl_msrs = adl_features.trl_msrs, + .tcc_offset_bits = adl_features.tcc_offset_bits, + .rapl_msrs = adl_features.rapl_msrs, + .enable_tsc_tweak = adl_features.enable_tsc_tweak, }; static const struct platform_features skx_features = { @@ -652,10 +811,11 @@ 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, - .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL, + .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL | RAPL_PSYS, .has_fixed_rapl_unit = 1, }; @@ -670,8 +830,9 @@ static const struct platform_features spr_features = { .has_msr_core_c1_res = 1, .has_irtl_msrs = 1, .has_cst_prewake_bit = 1, + .has_fixed_rapl_psys_unit = 1, .trl_msrs = TRL_BASE | TRL_CORECOUNT, - .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL, + .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL | RAPL_PSYS, }; static const struct platform_features srf_features = { @@ -687,7 +848,7 @@ static const struct platform_features srf_features = { .has_irtl_msrs = 1, .has_cst_prewake_bit = 1, .trl_msrs = TRL_BASE | TRL_CORECOUNT, - .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL, + .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL | RAPL_PSYS, }; static const struct platform_features grr_features = { @@ -703,7 +864,7 @@ static const struct platform_features grr_features = { .has_irtl_msrs = 1, .has_cst_prewake_bit = 1, .trl_msrs = TRL_BASE | TRL_CORECOUNT, - .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL, + .rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL | RAPL_PSYS, }; static const struct platform_features slv_features = { @@ -825,73 +986,78 @@ static const struct platform_features amd_features_with_rapl = { }; static const struct platform_data turbostat_pdata[] = { - { INTEL_FAM6_NEHALEM, &nhm_features }, - { INTEL_FAM6_NEHALEM_G, &nhm_features }, - { INTEL_FAM6_NEHALEM_EP, &nhm_features }, - { INTEL_FAM6_NEHALEM_EX, &nhx_features }, - { INTEL_FAM6_WESTMERE, &nhm_features }, - { INTEL_FAM6_WESTMERE_EP, &nhm_features }, - { INTEL_FAM6_WESTMERE_EX, &nhx_features }, - { INTEL_FAM6_SANDYBRIDGE, &snb_features }, - { INTEL_FAM6_SANDYBRIDGE_X, &snx_features }, - { INTEL_FAM6_IVYBRIDGE, &ivb_features }, - { INTEL_FAM6_IVYBRIDGE_X, &ivx_features }, - { INTEL_FAM6_HASWELL, &hsw_features }, - { INTEL_FAM6_HASWELL_X, &hsx_features }, - { INTEL_FAM6_HASWELL_L, &hswl_features }, - { INTEL_FAM6_HASWELL_G, &hswg_features }, - { INTEL_FAM6_BROADWELL, &bdw_features }, - { INTEL_FAM6_BROADWELL_G, &bdwg_features }, - { INTEL_FAM6_BROADWELL_X, &bdx_features }, - { INTEL_FAM6_BROADWELL_D, &bdx_features }, - { INTEL_FAM6_SKYLAKE_L, &skl_features }, - { INTEL_FAM6_SKYLAKE, &skl_features }, - { INTEL_FAM6_SKYLAKE_X, &skx_features }, - { INTEL_FAM6_KABYLAKE_L, &skl_features }, - { INTEL_FAM6_KABYLAKE, &skl_features }, - { INTEL_FAM6_COMETLAKE, &skl_features }, - { INTEL_FAM6_COMETLAKE_L, &skl_features }, - { INTEL_FAM6_CANNONLAKE_L, &cnl_features }, - { INTEL_FAM6_ICELAKE_X, &icx_features }, - { INTEL_FAM6_ICELAKE_D, &icx_features }, - { INTEL_FAM6_ICELAKE_L, &cnl_features }, - { INTEL_FAM6_ICELAKE_NNPI, &cnl_features }, - { INTEL_FAM6_ROCKETLAKE, &cnl_features }, - { INTEL_FAM6_TIGERLAKE_L, &cnl_features }, - { INTEL_FAM6_TIGERLAKE, &cnl_features }, - { INTEL_FAM6_SAPPHIRERAPIDS_X, &spr_features }, - { INTEL_FAM6_EMERALDRAPIDS_X, &spr_features }, - { INTEL_FAM6_GRANITERAPIDS_X, &spr_features }, - { INTEL_FAM6_LAKEFIELD, &cnl_features }, - { INTEL_FAM6_ALDERLAKE, &adl_features }, - { INTEL_FAM6_ALDERLAKE_L, &adl_features }, - { INTEL_FAM6_RAPTORLAKE, &adl_features }, - { INTEL_FAM6_RAPTORLAKE_P, &adl_features }, - { INTEL_FAM6_RAPTORLAKE_S, &adl_features }, - { INTEL_FAM6_METEORLAKE, &cnl_features }, - { INTEL_FAM6_METEORLAKE_L, &cnl_features }, - { INTEL_FAM6_ARROWLAKE, &cnl_features }, - { INTEL_FAM6_LUNARLAKE_M, &cnl_features }, - { INTEL_FAM6_ATOM_SILVERMONT, &slv_features }, - { INTEL_FAM6_ATOM_SILVERMONT_D, &slvd_features }, - { INTEL_FAM6_ATOM_AIRMONT, &amt_features }, - { INTEL_FAM6_ATOM_GOLDMONT, &gmt_features }, - { INTEL_FAM6_ATOM_GOLDMONT_D, &gmtd_features }, - { INTEL_FAM6_ATOM_GOLDMONT_PLUS, &gmtp_features }, - { INTEL_FAM6_ATOM_TREMONT_D, &tmtd_features }, - { INTEL_FAM6_ATOM_TREMONT, &tmt_features }, - { INTEL_FAM6_ATOM_TREMONT_L, &tmt_features }, - { INTEL_FAM6_ATOM_GRACEMONT, &adl_features }, - { INTEL_FAM6_ATOM_CRESTMONT_X, &srf_features }, - { INTEL_FAM6_ATOM_CRESTMONT, &grr_features }, - { INTEL_FAM6_XEON_PHI_KNL, &knl_features }, - { INTEL_FAM6_XEON_PHI_KNM, &knl_features }, + { INTEL_NEHALEM, &nhm_features }, + { INTEL_NEHALEM_G, &nhm_features }, + { INTEL_NEHALEM_EP, &nhm_features }, + { INTEL_NEHALEM_EX, &nhx_features }, + { INTEL_WESTMERE, &nhm_features }, + { INTEL_WESTMERE_EP, &nhm_features }, + { INTEL_WESTMERE_EX, &nhx_features }, + { INTEL_SANDYBRIDGE, &snb_features }, + { INTEL_SANDYBRIDGE_X, &snx_features }, + { INTEL_IVYBRIDGE, &ivb_features }, + { INTEL_IVYBRIDGE_X, &ivx_features }, + { INTEL_HASWELL, &hsw_features }, + { INTEL_HASWELL_X, &hsx_features }, + { INTEL_HASWELL_L, &hswl_features }, + { INTEL_HASWELL_G, &hswg_features }, + { INTEL_BROADWELL, &bdw_features }, + { INTEL_BROADWELL_G, &bdwg_features }, + { INTEL_BROADWELL_X, &bdx_features }, + { INTEL_BROADWELL_D, &bdx_features }, + { INTEL_SKYLAKE_L, &skl_features }, + { INTEL_SKYLAKE, &skl_features }, + { INTEL_SKYLAKE_X, &skx_features }, + { INTEL_KABYLAKE_L, &skl_features }, + { INTEL_KABYLAKE, &skl_features }, + { INTEL_COMETLAKE, &skl_features }, + { INTEL_COMETLAKE_L, &skl_features }, + { INTEL_CANNONLAKE_L, &cnl_features }, + { INTEL_ICELAKE_X, &icx_features }, + { INTEL_ICELAKE_D, &icx_features }, + { INTEL_ICELAKE_L, &cnl_features }, + { INTEL_ICELAKE_NNPI, &cnl_features }, + { INTEL_ROCKETLAKE, &cnl_features }, + { INTEL_TIGERLAKE_L, &cnl_features }, + { INTEL_TIGERLAKE, &cnl_features }, + { INTEL_SAPPHIRERAPIDS_X, &spr_features }, + { INTEL_EMERALDRAPIDS_X, &spr_features }, + { INTEL_GRANITERAPIDS_X, &spr_features }, + { INTEL_GRANITERAPIDS_D, &spr_features }, + { INTEL_LAKEFIELD, &cnl_features }, + { INTEL_ALDERLAKE, &adl_features }, + { INTEL_ALDERLAKE_L, &adl_features }, + { INTEL_RAPTORLAKE, &adl_features }, + { INTEL_RAPTORLAKE_P, &adl_features }, + { INTEL_RAPTORLAKE_S, &adl_features }, + { INTEL_METEORLAKE, &adl_features }, + { INTEL_METEORLAKE_L, &adl_features }, + { INTEL_ARROWLAKE_H, &adl_features }, + { INTEL_ARROWLAKE_U, &adl_features }, + { INTEL_ARROWLAKE, &adl_features }, + { INTEL_LUNARLAKE_M, &lnl_features }, + { INTEL_PANTHERLAKE_L, &lnl_features }, + { INTEL_ATOM_SILVERMONT, &slv_features }, + { INTEL_ATOM_SILVERMONT_D, &slvd_features }, + { INTEL_ATOM_AIRMONT, &amt_features }, + { INTEL_ATOM_GOLDMONT, &gmt_features }, + { INTEL_ATOM_GOLDMONT_D, &gmtd_features }, + { INTEL_ATOM_GOLDMONT_PLUS, &gmtp_features }, + { INTEL_ATOM_TREMONT_D, &tmtd_features }, + { INTEL_ATOM_TREMONT, &tmt_features }, + { INTEL_ATOM_TREMONT_L, &tmt_features }, + { INTEL_ATOM_GRACEMONT, &adl_features }, + { INTEL_ATOM_CRESTMONT_X, &srf_features }, + { INTEL_ATOM_CRESTMONT, &grr_features }, + { INTEL_ATOM_DARKMONT_X, &srf_features }, + { INTEL_XEON_PHI_KNL, &knl_features }, + { INTEL_XEON_PHI_KNM, &knl_features }, /* * Missing support for - * INTEL_FAM6_ICELAKE - * INTEL_FAM6_ATOM_SILVERMONT_MID - * INTEL_FAM6_ATOM_AIRMONT_MID - * INTEL_FAM6_ATOM_AIRMONT_NP + * INTEL_ICELAKE + * INTEL_ATOM_SILVERMONT_MID + * INTEL_ATOM_AIRMONT_MID + * INTEL_ATOM_AIRMONT_NP */ { 0, NULL }, }; @@ -902,9 +1068,10 @@ void probe_platform_features(unsigned int family, unsigned int model) { int i; - platform = &default_features; if (authentic_amd || hygon_genuine) { + /* fallback to default features on unsupported models */ + force_load++; if (max_extended_level >= 0x80000007) { unsigned int eax, ebx, ecx, edx; @@ -913,18 +1080,31 @@ void probe_platform_features(unsigned int family, unsigned int model) if ((edx & (1 << 14)) && family >= 0x17) platform = &amd_features_with_rapl; } - return; + goto end; } - if (!genuine_intel || family != 6) - return; + if (!genuine_intel) + goto end; for (i = 0; turbostat_pdata[i].features; i++) { - if (turbostat_pdata[i].model == model) { + if (VFM_FAMILY(turbostat_pdata[i].vfm) == family && VFM_MODEL(turbostat_pdata[i].vfm) == model) { platform = turbostat_pdata[i].features; return; } } + +end: + if (force_load && !platform) { + fprintf(outf, "Forced to run on unsupported platform!\n"); + platform = &default_features; + } + + if (platform) + return; + + fprintf(stderr, "Unsupported platform detected.\n" + "\tSee RUN THE LATEST VERSION on turbostat(8)\n"); + exit(1); } /* Model specific support End */ @@ -942,12 +1122,612 @@ int backwards_count; char *progname; #define CPU_SUBSET_MAXCPUS 1024 /* need to use before probe... */ -cpu_set_t *cpu_present_set, *cpu_effective_set, *cpu_allowed_set, *cpu_affinity_set, *cpu_subset; -size_t cpu_present_setsize, cpu_effective_setsize, cpu_allowed_setsize, cpu_affinity_setsize, cpu_subset_size; -#define MAX_ADDED_COUNTERS 8 +cpu_set_t *cpu_present_set, *cpu_possible_set, *cpu_effective_set, *cpu_allowed_set, *cpu_affinity_set, *cpu_subset; +size_t cpu_present_setsize, cpu_possible_setsize, cpu_effective_setsize, cpu_allowed_setsize, cpu_affinity_setsize, cpu_subset_size; #define MAX_ADDED_THREAD_COUNTERS 24 +#define MAX_ADDED_CORE_COUNTERS 8 +#define MAX_ADDED_PACKAGE_COUNTERS 16 +#define PMT_MAX_ADDED_THREAD_COUNTERS 24 +#define PMT_MAX_ADDED_CORE_COUNTERS 8 +#define PMT_MAX_ADDED_PACKAGE_COUNTERS 16 #define BITMASK_SIZE 32 +#define ZERO_ARRAY(arr) (memset(arr, 0, sizeof(arr)) + __must_be_array(arr)) + +/* 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, + RAPL_RCI_INDEX_ENERGY_PLATFORM = 7, + 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 counter_source source[NUM_RAPL_COUNTERS]; + unsigned long long flags[NUM_RAPL_COUNTERS]; + double scale[NUM_RAPL_COUNTERS]; + enum rapl_unit unit[NUM_RAPL_COUNTERS]; + 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; +unsigned int rapl_counter_info_perdomain_size; + +#define RAPL_COUNTER_FLAG_PLATFORM_COUNTER (1u << 0) +#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, + }, + { + .feature_mask = RAPL_PSYS, + .perf_subsys = "power", + .perf_name = "energy-psys", + .msr = MSR_PLATFORM_ENERGY_STATUS, + .msr_mask = 0x00000000FFFFFFFF, + .msr_shift = 0, + .platform_rapl_msr_scale = &rapl_psys_energy_units, + .rci_index = RAPL_RCI_INDEX_ENERGY_PLATFORM, + .bic = BIC_SysWatt | BIC_Sys_J, + .compat_scale = 1.0, + .flags = RAPL_COUNTER_FLAG_PLATFORM_COUNTER | RAPL_COUNTER_FLAG_USE_MSR_SUM, + }, +}; + +struct rapl_counter { + unsigned long long raw_value; + enum rapl_unit unit; + double scale; +}; + +/* Indexes used to map data read from perf and MSRs into global variables */ +enum ccstate_rci_index { + CCSTATE_RCI_INDEX_C1_RESIDENCY = 0, + CCSTATE_RCI_INDEX_C3_RESIDENCY = 1, + CCSTATE_RCI_INDEX_C6_RESIDENCY = 2, + CCSTATE_RCI_INDEX_C7_RESIDENCY = 3, + PCSTATE_RCI_INDEX_C2_RESIDENCY = 4, + PCSTATE_RCI_INDEX_C3_RESIDENCY = 5, + PCSTATE_RCI_INDEX_C6_RESIDENCY = 6, + PCSTATE_RCI_INDEX_C7_RESIDENCY = 7, + PCSTATE_RCI_INDEX_C8_RESIDENCY = 8, + PCSTATE_RCI_INDEX_C9_RESIDENCY = 9, + PCSTATE_RCI_INDEX_C10_RESIDENCY = 10, + NUM_CSTATE_COUNTERS, +}; + +struct cstate_counter_info_t { + unsigned long long data[NUM_CSTATE_COUNTERS]; + enum counter_source source[NUM_CSTATE_COUNTERS]; + unsigned long long msr[NUM_CSTATE_COUNTERS]; + int fd_perf_core; + int fd_perf_pkg; +}; + +struct cstate_counter_info_t *ccstate_counter_info; +unsigned int ccstate_counter_info_size; + +#define CSTATE_COUNTER_FLAG_COLLECT_PER_CORE (1u << 0) +#define CSTATE_COUNTER_FLAG_COLLECT_PER_THREAD ((1u << 1) | CSTATE_COUNTER_FLAG_COLLECT_PER_CORE) +#define CSTATE_COUNTER_FLAG_SOFT_C1_DEPENDENCY (1u << 2) + +struct cstate_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 int rci_index; /* Maps data from perf counters to global variables */ + unsigned long long bic; + unsigned long long flags; + int pkg_cstate_limit; +}; + +static struct cstate_counter_arch_info ccstate_counter_arch_infos[] = { + { + .feature_mask = CC1, + .perf_subsys = "cstate_core", + .perf_name = "c1-residency", + .msr = MSR_CORE_C1_RES, + .rci_index = CCSTATE_RCI_INDEX_C1_RESIDENCY, + .bic = BIC_CPU_c1, + .flags = CSTATE_COUNTER_FLAG_COLLECT_PER_THREAD, + .pkg_cstate_limit = 0, + }, + { + .feature_mask = CC3, + .perf_subsys = "cstate_core", + .perf_name = "c3-residency", + .msr = MSR_CORE_C3_RESIDENCY, + .rci_index = CCSTATE_RCI_INDEX_C3_RESIDENCY, + .bic = BIC_CPU_c3, + .flags = CSTATE_COUNTER_FLAG_COLLECT_PER_CORE | CSTATE_COUNTER_FLAG_SOFT_C1_DEPENDENCY, + .pkg_cstate_limit = 0, + }, + { + .feature_mask = CC6, + .perf_subsys = "cstate_core", + .perf_name = "c6-residency", + .msr = MSR_CORE_C6_RESIDENCY, + .rci_index = CCSTATE_RCI_INDEX_C6_RESIDENCY, + .bic = BIC_CPU_c6, + .flags = CSTATE_COUNTER_FLAG_COLLECT_PER_CORE | CSTATE_COUNTER_FLAG_SOFT_C1_DEPENDENCY, + .pkg_cstate_limit = 0, + }, + { + .feature_mask = CC7, + .perf_subsys = "cstate_core", + .perf_name = "c7-residency", + .msr = MSR_CORE_C7_RESIDENCY, + .rci_index = CCSTATE_RCI_INDEX_C7_RESIDENCY, + .bic = BIC_CPU_c7, + .flags = CSTATE_COUNTER_FLAG_COLLECT_PER_CORE | CSTATE_COUNTER_FLAG_SOFT_C1_DEPENDENCY, + .pkg_cstate_limit = 0, + }, + { + .feature_mask = PC2, + .perf_subsys = "cstate_pkg", + .perf_name = "c2-residency", + .msr = MSR_PKG_C2_RESIDENCY, + .rci_index = PCSTATE_RCI_INDEX_C2_RESIDENCY, + .bic = BIC_Pkgpc2, + .flags = 0, + .pkg_cstate_limit = PCL__2, + }, + { + .feature_mask = PC3, + .perf_subsys = "cstate_pkg", + .perf_name = "c3-residency", + .msr = MSR_PKG_C3_RESIDENCY, + .rci_index = PCSTATE_RCI_INDEX_C3_RESIDENCY, + .bic = BIC_Pkgpc3, + .flags = 0, + .pkg_cstate_limit = PCL__3, + }, + { + .feature_mask = PC6, + .perf_subsys = "cstate_pkg", + .perf_name = "c6-residency", + .msr = MSR_PKG_C6_RESIDENCY, + .rci_index = PCSTATE_RCI_INDEX_C6_RESIDENCY, + .bic = BIC_Pkgpc6, + .flags = 0, + .pkg_cstate_limit = PCL__6, + }, + { + .feature_mask = PC7, + .perf_subsys = "cstate_pkg", + .perf_name = "c7-residency", + .msr = MSR_PKG_C7_RESIDENCY, + .rci_index = PCSTATE_RCI_INDEX_C7_RESIDENCY, + .bic = BIC_Pkgpc7, + .flags = 0, + .pkg_cstate_limit = PCL__7, + }, + { + .feature_mask = PC8, + .perf_subsys = "cstate_pkg", + .perf_name = "c8-residency", + .msr = MSR_PKG_C8_RESIDENCY, + .rci_index = PCSTATE_RCI_INDEX_C8_RESIDENCY, + .bic = BIC_Pkgpc8, + .flags = 0, + .pkg_cstate_limit = PCL__8, + }, + { + .feature_mask = PC9, + .perf_subsys = "cstate_pkg", + .perf_name = "c9-residency", + .msr = MSR_PKG_C9_RESIDENCY, + .rci_index = PCSTATE_RCI_INDEX_C9_RESIDENCY, + .bic = BIC_Pkgpc9, + .flags = 0, + .pkg_cstate_limit = PCL__9, + }, + { + .feature_mask = PC10, + .perf_subsys = "cstate_pkg", + .perf_name = "c10-residency", + .msr = MSR_PKG_C10_RESIDENCY, + .rci_index = PCSTATE_RCI_INDEX_C10_RESIDENCY, + .bic = BIC_Pkgpc10, + .flags = 0, + .pkg_cstate_limit = PCL_10, + }, +}; + +/* Indexes used to map data read from perf and MSRs into global variables */ +enum msr_rci_index { + MSR_RCI_INDEX_APERF = 0, + MSR_RCI_INDEX_MPERF = 1, + MSR_RCI_INDEX_SMI = 2, + NUM_MSR_COUNTERS, +}; + +struct msr_counter_info_t { + unsigned long long data[NUM_MSR_COUNTERS]; + enum counter_source source[NUM_MSR_COUNTERS]; + unsigned long long msr[NUM_MSR_COUNTERS]; + unsigned long long msr_mask[NUM_MSR_COUNTERS]; + int fd_perf; +}; + +struct msr_counter_info_t *msr_counter_info; +unsigned int msr_counter_info_size; + +struct msr_counter_arch_info { + const char *perf_subsys; + const char *perf_name; + unsigned long long msr; + unsigned long long msr_mask; + unsigned int rci_index; /* Maps data from perf counters to global variables */ + bool needed; + bool present; +}; + +enum msr_arch_info_index { + MSR_ARCH_INFO_APERF_INDEX = 0, + MSR_ARCH_INFO_MPERF_INDEX = 1, + MSR_ARCH_INFO_SMI_INDEX = 2, +}; + +static struct msr_counter_arch_info msr_counter_arch_infos[] = { + [MSR_ARCH_INFO_APERF_INDEX] = { + .perf_subsys = "msr", + .perf_name = "aperf", + .msr = MSR_IA32_APERF, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .rci_index = MSR_RCI_INDEX_APERF, + }, + + [MSR_ARCH_INFO_MPERF_INDEX] = { + .perf_subsys = "msr", + .perf_name = "mperf", + .msr = MSR_IA32_MPERF, + .msr_mask = 0xFFFFFFFFFFFFFFFF, + .rci_index = MSR_RCI_INDEX_MPERF, + }, + + [MSR_ARCH_INFO_SMI_INDEX] = { + .perf_subsys = "msr", + .perf_name = "smi", + .msr = MSR_SMI_COUNT, + .msr_mask = 0xFFFFFFFF, + .rci_index = MSR_RCI_INDEX_SMI, + }, +}; + +/* Can be redefined when compiling, useful for testing. */ +#ifndef SYSFS_TELEM_PATH +#define SYSFS_TELEM_PATH "/sys/class/intel_pmt" +#endif + +#define PMT_COUNTER_MTL_DC6_OFFSET 120 +#define PMT_COUNTER_MTL_DC6_LSB 0 +#define PMT_COUNTER_MTL_DC6_MSB 63 +#define PMT_MTL_DC6_GUID 0x1a067102 +#define PMT_MTL_DC6_SEQ 0 + +#define PMT_COUNTER_CWF_MC1E_OFFSET_BASE 20936 +#define PMT_COUNTER_CWF_MC1E_OFFSET_INCREMENT 24 +#define PMT_COUNTER_CWF_MC1E_NUM_MODULES_PER_FILE 12 +#define PMT_COUNTER_CWF_CPUS_PER_MODULE 4 +#define PMT_COUNTER_CWF_MC1E_LSB 0 +#define PMT_COUNTER_CWF_MC1E_MSB 63 +#define PMT_CWF_MC1E_GUID 0x14421519 + +unsigned long long tcore_clock_freq_hz = 800000000; + +#define PMT_COUNTER_NAME_SIZE_BYTES 16 +#define PMT_COUNTER_TYPE_NAME_SIZE_BYTES 32 + +struct pmt_mmio { + struct pmt_mmio *next; + + unsigned int guid; + unsigned int size; + + /* Base pointer to the mmaped memory. */ + void *mmio_base; + + /* + * Offset to be applied to the mmio_base + * to get the beginning of the PMT counters for given GUID. + */ + unsigned long pmt_offset; +} *pmt_mmios; + +enum pmt_datatype { + PMT_TYPE_RAW, + PMT_TYPE_XTAL_TIME, + PMT_TYPE_TCORE_CLOCK, +}; + +struct pmt_domain_info { + /* + * Pointer to the MMIO obtained by applying a counter offset + * to the mmio_base of the mmaped region for the given GUID. + * + * This is where to read the raw value of the counter from. + */ + unsigned long *pcounter; +}; + +struct pmt_counter { + struct pmt_counter *next; + + /* PMT metadata */ + char name[PMT_COUNTER_NAME_SIZE_BYTES]; + enum pmt_datatype type; + enum counter_scope scope; + unsigned int lsb; + unsigned int msb; + + /* BIC-like metadata */ + enum counter_format format; + + unsigned int num_domains; + struct pmt_domain_info *domains; +}; + +/* + * PMT telemetry directory iterator. + * Used to iterate telemetry files in sysfs in correct order. + */ +struct pmt_diriter_t { + DIR *dir; + struct dirent **namelist; + unsigned int num_names; + unsigned int current_name_idx; +}; + +int pmt_telemdir_filter(const struct dirent *e) +{ + unsigned int dummy; + + return sscanf(e->d_name, "telem%u", &dummy); +} + +int pmt_telemdir_sort(const struct dirent **a, const struct dirent **b) +{ + unsigned int aidx = 0, bidx = 0; + + sscanf((*a)->d_name, "telem%u", &aidx); + sscanf((*b)->d_name, "telem%u", &bidx); + + return aidx >= bidx; +} + +const struct dirent *pmt_diriter_next(struct pmt_diriter_t *iter) +{ + const struct dirent *ret = NULL; + + if (!iter->dir) + return NULL; + + if (iter->current_name_idx >= iter->num_names) + return NULL; + + ret = iter->namelist[iter->current_name_idx]; + ++iter->current_name_idx; + + return ret; +} + +const struct dirent *pmt_diriter_begin(struct pmt_diriter_t *iter, const char *pmt_root_path) +{ + int num_names = iter->num_names; + + if (!iter->dir) { + iter->dir = opendir(pmt_root_path); + if (iter->dir == NULL) + return NULL; + + num_names = scandir(pmt_root_path, &iter->namelist, pmt_telemdir_filter, pmt_telemdir_sort); + if (num_names == -1) + return NULL; + } + + iter->current_name_idx = 0; + iter->num_names = num_names; + + return pmt_diriter_next(iter); +} + +void pmt_diriter_init(struct pmt_diriter_t *iter) +{ + memset(iter, 0, sizeof(*iter)); +} + +void pmt_diriter_remove(struct pmt_diriter_t *iter) +{ + if (iter->namelist) { + for (unsigned int i = 0; i < iter->num_names; i++) { + free(iter->namelist[i]); + iter->namelist[i] = NULL; + } + } + + free(iter->namelist); + iter->namelist = NULL; + iter->num_names = 0; + iter->current_name_idx = 0; + + closedir(iter->dir); + iter->dir = NULL; +} + +unsigned int pmt_counter_get_width(const struct pmt_counter *p) +{ + return (p->msb - p->lsb) + 1; +} + +void pmt_counter_resize_(struct pmt_counter *pcounter, unsigned int new_size) +{ + struct pmt_domain_info *new_mem; + + new_mem = (struct pmt_domain_info *)reallocarray(pcounter->domains, new_size, sizeof(*pcounter->domains)); + if (!new_mem) { + fprintf(stderr, "%s: failed to allocate memory for PMT counters\n", __func__); + exit(1); + } + + /* Zero initialize just allocated memory. */ + const size_t num_new_domains = new_size - pcounter->num_domains; + + memset(&new_mem[pcounter->num_domains], 0, num_new_domains * sizeof(*pcounter->domains)); + + pcounter->num_domains = new_size; + pcounter->domains = new_mem; +} + +void pmt_counter_resize(struct pmt_counter *pcounter, unsigned int new_size) +{ + /* + * Allocate more memory ahead of time. + * + * Always allocate space for at least 8 elements + * and double the size when growing. + */ + if (new_size < 8) + new_size = 8; + new_size = MAX(new_size, pcounter->num_domains * 2); + + pmt_counter_resize_(pcounter, new_size); +} + struct thread_data { struct timeval tv_begin; struct timeval tv_end; @@ -958,6 +1738,7 @@ struct thread_data { unsigned long long c1; unsigned long long instr_count; unsigned long long irq_count; + unsigned long long nmi_count; unsigned int smi_count; unsigned int cpu_id; unsigned int apic_id; @@ -965,6 +1746,8 @@ struct thread_data { unsigned int flags; bool is_atom; unsigned long long counter[MAX_ADDED_THREAD_COUNTERS]; + unsigned long long perf_counter[MAX_ADDED_THREAD_COUNTERS]; + unsigned long long pmt_counter[PMT_MAX_ADDED_THREAD_COUNTERS]; } *thread_even, *thread_odd; struct core_data { @@ -974,10 +1757,12 @@ 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]; + unsigned long long counter[MAX_ADDED_CORE_COUNTERS]; + unsigned long long perf_counter[MAX_ADDED_CORE_COUNTERS]; + unsigned long long pmt_counter[PMT_MAX_ADDED_CORE_COUNTERS]; } *core_even, *core_odd; struct pkg_data { @@ -989,8 +1774,8 @@ 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; @@ -998,16 +1783,22 @@ 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; - 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]; + unsigned long long die_c6; + unsigned long long counter[MAX_ADDED_PACKAGE_COUNTERS]; + unsigned long long perf_counter[MAX_ADDED_PACKAGE_COUNTERS]; + unsigned long long pmt_counter[PMT_MAX_ADDED_PACKAGE_COUNTERS]; } *package_even, *package_odd; #define ODD_COUNTERS thread_odd, core_odd, package_odd @@ -1041,6 +1832,7 @@ enum { IDX_PP1_ENERGY, IDX_PKG_PERF, IDX_DRAM_PERF, + IDX_PSYS_ENERGY, IDX_COUNT, }; @@ -1085,6 +1877,9 @@ off_t idx_to_offset(int idx) case IDX_DRAM_PERF: offset = MSR_DRAM_PERF_STATUS; break; + case IDX_PSYS_ENERGY: + offset = MSR_PLATFORM_ENERGY_STATUS; + break; default: offset = -1; } @@ -1115,6 +1910,9 @@ int offset_to_idx(off_t offset) case MSR_DRAM_PERF_STATUS: idx = IDX_DRAM_PERF; break; + case MSR_PLATFORM_ENERGY_STATUS: + idx = IDX_PSYS_ENERGY; + break; default: idx = -1; } @@ -1136,26 +1934,83 @@ int idx_valid(int idx) return platform->rapl_msrs & RAPL_PKG_PERF_STATUS; case IDX_DRAM_PERF: return platform->rapl_msrs & RAPL_DRAM_PERF_STATUS; + case IDX_PSYS_ENERGY: + return platform->rapl_msrs & RAPL_PSYS; default: return 0; } } struct sys_counters { + /* MSR added counters */ unsigned int added_thread_counters; unsigned int added_core_counters; unsigned int added_package_counters; struct msr_counter *tp; struct msr_counter *cp; struct msr_counter *pp; + + /* perf added counters */ + unsigned int added_thread_perf_counters; + unsigned int added_core_perf_counters; + unsigned int added_package_perf_counters; + struct perf_counter_info *perf_tp; + struct perf_counter_info *perf_cp; + struct perf_counter_info *perf_pp; + + struct pmt_counter *pmt_tp; + struct pmt_counter *pmt_cp; + struct pmt_counter *pmt_pp; } sys; +static size_t free_msr_counters_(struct msr_counter **pp) +{ + struct msr_counter *p = NULL; + size_t num_freed = 0; + + while (*pp) { + p = *pp; + + if (p->msr_num != 0) { + *pp = p->next; + + free(p); + ++num_freed; + + continue; + } + + pp = &p->next; + } + + return num_freed; +} + +/* + * Free all added counters accessed via msr. + */ +static void free_sys_msr_counters(void) +{ + /* Thread counters */ + sys.added_thread_counters -= free_msr_counters_(&sys.tp); + + /* Core counters */ + sys.added_core_counters -= free_msr_counters_(&sys.cp); + + /* Package counters */ + sys.added_package_counters -= free_msr_counters_(&sys.pp); +} + struct system_summary { struct thread_data threads; struct core_data cores; struct pkg_data packages; } average; +struct platform_counters { + struct rapl_counter energy_psys; /* MSR_PLATFORM_ENERGY_STATUS */ +} platform_counters_odd, platform_counters_even; + struct cpu_topology { int physical_package_id; int die_id; @@ -1164,6 +2019,7 @@ struct cpu_topology { int logical_node_id; /* 0-based count within the package */ int physical_core_id; int thread_id; + int type; cpu_set_t *put_ids; /* Processing Unit/Thread IDs */ } *cpus; @@ -1176,6 +2032,9 @@ struct topo_params { int allowed_cpus; int allowed_cores; int max_cpu_num; + int max_core_id; + int max_package_id; + int max_die_id; int max_node_num; int nodes_per_pkg; int cores_per_node; @@ -1186,6 +2045,7 @@ struct timeval tv_even, tv_odd, tv_delta; int *irq_column_2_cpu; /* /proc/interrupts column numbers */ int *irqs_per_cpu; /* indexed by cpu_num */ +int *nmi_per_cpu; /* indexed by cpu_num */ void setup_all_buffers(bool startup); @@ -1213,6 +2073,8 @@ int for_all_cpus(int (func) (struct thread_data *, struct core_data *, struct pk { int retval, pkg_no, core_no, thread_no, node_no; + retval = 0; + for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { for (node_no = 0; node_no < topo.nodes_per_pkg; node_no++) { for (core_no = 0; core_no < topo.cores_per_node; ++core_no) { @@ -1228,14 +2090,12 @@ int for_all_cpus(int (func) (struct thread_data *, struct core_data *, struct pk c = GET_CORE(core_base, core_no, node_no, pkg_no); p = GET_PKG(pkg_base, pkg_no); - retval = func(t, c, p); - if (retval) - return retval; + retval |= func(t, c, p); } } } } - return 0; + return retval; } int is_cpu_first_thread_in_core(struct thread_data *t, struct core_data *c, struct pkg_data *p) @@ -1280,34 +2140,49 @@ 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_Mod_c6 | BIC_CoreTmp | + BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_PkgTmp; + + bic_enabled &= ~bic_msrs; + + free_sys_msr_counters(); +} + 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); } -static int perf_instr_count_open(int cpu_num) +static long open_perf_counter(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; + assert(!no_perf); - /* 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); - } + memset(&attr, 0, sizeof(struct perf_event_attr)); + + 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); return fd; } @@ -1317,7 +2192,7 @@ int get_instr_count_fd(int cpu) if (fd_instr_count_percpu[cpu]) return fd_instr_count_percpu[cpu]; - fd_instr_count_percpu[cpu] = perf_instr_count_open(cpu); + fd_instr_count_percpu[cpu] = open_perf_counter(cpu, PERF_TYPE_HARDWARE, PERF_COUNT_HW_INSTRUCTIONS, -1, 0); return fd_instr_count_percpu[cpu]; } @@ -1326,6 +2201,8 @@ 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) @@ -1334,6 +2211,42 @@ 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 value; + + assert(!no_msr); + + retval = pread(get_msr_fd(cpu), &value, sizeof(value), offset); + + /* + * Expect MSRs to accumulate some non-zero value since the system was powered on. + * Treat zero as a read failure. + */ + if (retval != sizeof(value) || value == 0) + return 1; + + return 0; +} + +/* Convert CPU ID to domain ID for given added perf counter. */ +unsigned int cpu_to_domain(const struct perf_counter_info *pc, int cpu) +{ + switch (pc->scope) { + case SCOPE_CPU: + return cpu; + + case SCOPE_CORE: + return cpus[cpu].physical_core_id; + + case SCOPE_PACKAGE: + return cpus[cpu].physical_package_id; + } + + __builtin_unreachable(); +} + #define MAX_DEFERRED 16 char *deferred_add_names[MAX_DEFERRED]; char *deferred_skip_names[MAX_DEFERRED]; @@ -1355,36 +2268,54 @@ void help(void) "when COMMAND completes.\n" "If no COMMAND is specified, turbostat wakes every 5-seconds\n" "to print statistics, until interrupted.\n" - " -a, --add add a counter\n" + " -a, --add counter\n" + " add a counter\n" " eg. --add msr0x10,u64,cpu,delta,MY_TSC\n" - " -c, --cpu cpu-set limit output to summary plus cpu-set:\n" + " eg. --add perf/cstate_pkg/c2-residency,package,delta,percent,perfPC2\n" + " eg. --add pmt,name=XTAL,type=raw,domain=package0,offset=0,lsb=0,msb=63,guid=0x1a067102\n" + " -c, --cpu cpu-set\n" + " limit output to summary plus cpu-set:\n" " {core | package | j,k,l..m,n-p }\n" - " -d, --debug displays usec, Time_Of_Day_Seconds and more debugging\n" - " -D, --Dump displays the raw counter values\n" - " -e, --enable [all | column]\n" + " -d, --debug\n" + " displays usec, Time_Of_Day_Seconds and more debugging\n" + " debug messages are printed to stderr\n" + " -D, --Dump\n" + " displays the raw counter values\n" + " -e, --enable [all | column]\n" " shows all or the specified disabled column\n" - " -H, --hide [column|column,column,...]\n" + " -f, --force\n" + " force load turbostat with minimum default features on unsupported platforms.\n" + " -H, --hide [column | column,column,...]\n" " hide the specified column(s)\n" " -i, --interval sec.subsec\n" - " Override default 5-second measurement interval\n" - " -J, --Joules displays energy in Joules instead of Watts\n" - " -l, --list list column headers only\n" + " override default 5-second measurement interval\n" + " -J, --Joules\n" + " displays energy in Joules instead of Watts\n" + " -l, --list\n" + " list column headers only\n" + " -M, --no-msr\n" + " disable all uses of the MSR driver\n" + " -P, --no-perf\n" + " disable all uses of the perf API\n" " -n, --num_iterations num\n" " number of the measurement iterations\n" " -N, --header_iterations num\n" " print header every num iterations\n" " -o, --out file\n" " create or truncate \"file\" for all output\n" - " -q, --quiet skip decoding system configuration header\n" - " -s, --show [column|column,column,...]\n" + " -q, --quiet\n" + " skip decoding system configuration header\n" + " -s, --show [column | column,column,...]\n" " show only the specified column(s)\n" " -S, --Summary\n" " limits output to 1-line system summary per interval\n" " -T, --TCC temperature\n" " sets the Thermal Control Circuit temperature in\n" " degrees Celsius\n" - " -h, --help print this help message\n" - " -v, --version print version information\n" "\n" "For more help, run \"man turbostat\"\n"); + " -h, --help\n" + " print this help message\n" + " -v, --version\n" + " print version information\n\nFor more help, run \"man turbostat\"\n"); } /* @@ -1464,6 +2395,8 @@ unsigned long long bic_lookup(char *name_list, enum show_hide_mode mode) void print_header(char *delim) { struct msr_counter *mp; + struct perf_counter_info *pp; + struct pmt_counter *ppmt; int printed = 0; if (DO_BIC(BIC_USEC)) @@ -1502,6 +2435,12 @@ void print_header(char *delim) else outp += sprintf(outp, "%sIRQ", (printed++ ? delim : "")); } + if (DO_BIC(BIC_NMI)) { + if (sums_need_wide_columns) + outp += sprintf(outp, "%s NMI", (printed++ ? delim : "")); + else + outp += sprintf(outp, "%sNMI", (printed++ ? delim : "")); + } if (DO_BIC(BIC_SMI)) outp += sprintf(outp, "%sSMI", (printed++ ? delim : "")); @@ -1521,6 +2460,41 @@ void print_header(char *delim) } } + for (pp = sys.perf_tp; pp; pp = pp->next) { + + if (pp->format == FORMAT_RAW) { + if (pp->width == 64) + outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), pp->name); + else + outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), pp->name); + } else { + if ((pp->type == COUNTER_ITEMS) && sums_need_wide_columns) + outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), pp->name); + else + outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), pp->name); + } + } + + ppmt = sys.pmt_tp; + while (ppmt) { + switch (ppmt->type) { + case PMT_TYPE_RAW: + if (pmt_counter_get_width(ppmt) <= 32) + outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), ppmt->name); + else + outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), ppmt->name); + + break; + + case PMT_TYPE_XTAL_TIME: + case PMT_TYPE_TCORE_CLOCK: + outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), ppmt->name); + break; + } + + ppmt = ppmt->next; + } + if (DO_BIC(BIC_CPU_c1)) outp += sprintf(outp, "%sCPU%%c1", (printed++ ? delim : "")); if (DO_BIC(BIC_CPU_c3)) @@ -1561,6 +2535,41 @@ void print_header(char *delim) } } + for (pp = sys.perf_cp; pp; pp = pp->next) { + + if (pp->format == FORMAT_RAW) { + if (pp->width == 64) + outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), pp->name); + else + outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), pp->name); + } else { + if ((pp->type == COUNTER_ITEMS) && sums_need_wide_columns) + outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), pp->name); + else + outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), pp->name); + } + } + + ppmt = sys.pmt_cp; + while (ppmt) { + switch (ppmt->type) { + case PMT_TYPE_RAW: + if (pmt_counter_get_width(ppmt) <= 32) + outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), ppmt->name); + else + outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), ppmt->name); + + break; + + case PMT_TYPE_XTAL_TIME: + case PMT_TYPE_TCORE_CLOCK: + outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), ppmt->name); + break; + } + + ppmt = ppmt->next; + } + if (DO_BIC(BIC_PkgTmp)) outp += sprintf(outp, "%sPkgTmp", (printed++ ? delim : "")); @@ -1573,6 +2582,15 @@ 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)) @@ -1596,6 +2614,8 @@ void print_header(char *delim) outp += sprintf(outp, "%sPkg%%pc9", (printed++ ? delim : "")); if (DO_BIC(BIC_Pkgpc10)) outp += sprintf(outp, "%sPk%%pc10", (printed++ ? delim : "")); + if (DO_BIC(BIC_Diec6)) + outp += sprintf(outp, "%sDie%%c6", (printed++ ? delim : "")); if (DO_BIC(BIC_CPU_LPI)) outp += sprintf(outp, "%sCPU%%LPI", (printed++ ? delim : "")); if (DO_BIC(BIC_SYS_LPI)) @@ -1635,16 +2655,58 @@ void print_header(char *delim) if (mp->format == FORMAT_RAW) { if (mp->width == 64) outp += sprintf(outp, "%s%18.18s", delim, mp->name); - else + else if (mp->width == 32) outp += sprintf(outp, "%s%10.10s", delim, mp->name); + else + outp += sprintf(outp, "%s%7.7s", delim, mp->name); } else { if ((mp->type == COUNTER_ITEMS) && sums_need_wide_columns) outp += sprintf(outp, "%s%8s", delim, mp->name); else - outp += sprintf(outp, "%s%s", delim, mp->name); + outp += sprintf(outp, "%s%7.7s", delim, mp->name); + } + } + + for (pp = sys.perf_pp; pp; pp = pp->next) { + + if (pp->format == FORMAT_RAW) { + if (pp->width == 64) + outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), pp->name); + else + outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), pp->name); + } else { + if ((pp->type == COUNTER_ITEMS) && sums_need_wide_columns) + outp += sprintf(outp, "%s%8s", (printed++ ? delim : ""), pp->name); + else + outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), pp->name); + } + } + + ppmt = sys.pmt_pp; + while (ppmt) { + switch (ppmt->type) { + case PMT_TYPE_RAW: + if (pmt_counter_get_width(ppmt) <= 32) + outp += sprintf(outp, "%s%10.10s", (printed++ ? delim : ""), ppmt->name); + else + outp += sprintf(outp, "%s%18.18s", (printed++ ? delim : ""), ppmt->name); + + break; + + case PMT_TYPE_XTAL_TIME: + case PMT_TYPE_TCORE_CLOCK: + outp += sprintf(outp, "%s%s", (printed++ ? delim : ""), ppmt->name); + break; } + + ppmt = ppmt->next; } + if (DO_BIC(BIC_SysWatt)) + outp += sprintf(outp, "%sSysWatt", (printed++ ? delim : "")); + if (DO_BIC(BIC_Sys_J)) + outp += sprintf(outp, "%sSys_J", (printed++ ? delim : "")); + outp += sprintf(outp, "\n"); } @@ -1652,6 +2714,7 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p { int i; struct msr_counter *mp; + struct platform_counters *pplat_cnt = p == package_odd ? &platform_counters_odd : &platform_counters_even; outp += sprintf(outp, "t %p, c %p, p %p\n", t, c, p); @@ -1667,30 +2730,41 @@ int dump_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p if (DO_BIC(BIC_IRQ)) outp += sprintf(outp, "IRQ: %lld\n", t->irq_count); + if (DO_BIC(BIC_NMI)) + outp += sprintf(outp, "IRQ: %lld\n", t->nmi_count); if (DO_BIC(BIC_SMI)) 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->sp->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); 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 += 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->sp->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); @@ -1710,16 +2784,19 @@ 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, "Joules PSYS: %0llX\n", pplat_cnt->energy_psys.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) { - 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->sp->path); } } @@ -1728,18 +2805,45 @@ 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 */ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) { + static int count; + + struct platform_counters *pplat_cnt = NULL; double interval_float, tsc; char *fmt8; int i; struct msr_counter *mp; + struct perf_counter_info *pp; + struct pmt_counter *ppmt; char *delim = "\t"; int printed = 0; + if (t == &average.threads) { + pplat_cnt = count & 1 ? &platform_counters_odd : &platform_counters_even; + ++count; + } + /* if showing only 1st thread in core and this isn't one, bail out */ if (show_core_only && !is_cpu_first_thread_in_core(t, c, p)) return 0; @@ -1847,6 +2951,14 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->irq_count); } + /* NMI */ + if (DO_BIC(BIC_NMI)) { + if (sums_need_wide_columns) + outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->nmi_count); + else + outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->nmi_count); + } + /* SMI */ if (DO_BIC(BIC_SMI)) outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), t->smi_count); @@ -1874,6 +2986,55 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data } } + /* Added perf counters */ + for (i = 0, pp = sys.perf_tp; pp; ++i, pp = pp->next) { + if (pp->format == FORMAT_RAW) { + if (pp->width == 32) + outp += + sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), + (unsigned int)t->perf_counter[i]); + else + outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->perf_counter[i]); + } else if (pp->format == FORMAT_DELTA) { + if ((pp->type == COUNTER_ITEMS) && sums_need_wide_columns) + outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), t->perf_counter[i]); + else + outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), t->perf_counter[i]); + } else if (pp->format == FORMAT_PERCENT) { + if (pp->type == COUNTER_USEC) + outp += + sprintf(outp, "%s%.2f", (printed++ ? delim : ""), + t->perf_counter[i] / interval_float / 10000); + else + outp += + sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->perf_counter[i] / tsc); + } + } + + for (i = 0, ppmt = sys.pmt_tp; ppmt; i++, ppmt = ppmt->next) { + const unsigned long value_raw = t->pmt_counter[i]; + double value_converted; + switch (ppmt->type) { + case PMT_TYPE_RAW: + if (pmt_counter_get_width(ppmt) <= 32) + outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), + (unsigned int)t->pmt_counter[i]); + else + outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), t->pmt_counter[i]); + + break; + + case PMT_TYPE_XTAL_TIME: + value_converted = 100.0 * value_raw / crystal_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); + break; + + case PMT_TYPE_TCORE_CLOCK: + value_converted = 100.0 * value_raw / tcore_clock_freq_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); + } + } + /* C1 */ if (DO_BIC(BIC_CPU_c1)) outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * t->c1 / tsc); @@ -1917,13 +3078,57 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data } } + for (i = 0, pp = sys.perf_cp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) { + if (pp->width == 32) + outp += + sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), + (unsigned int)c->perf_counter[i]); + else + outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->perf_counter[i]); + } else if (pp->format == FORMAT_DELTA) { + if ((pp->type == COUNTER_ITEMS) && sums_need_wide_columns) + outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), c->perf_counter[i]); + else + outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), c->perf_counter[i]); + } else if (pp->format == FORMAT_PERCENT) { + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * c->perf_counter[i] / tsc); + } + } + + for (i = 0, ppmt = sys.pmt_cp; ppmt; i++, ppmt = ppmt->next) { + const unsigned long value_raw = c->pmt_counter[i]; + double value_converted; + switch (ppmt->type) { + case PMT_TYPE_RAW: + if (pmt_counter_get_width(ppmt) <= 32) + outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), + (unsigned int)c->pmt_counter[i]); + else + outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), c->pmt_counter[i]); + + break; + + case PMT_TYPE_XTAL_TIME: + value_converted = 100.0 * value_raw / crystal_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); + break; + + case PMT_TYPE_TCORE_CLOCK: + value_converted = 100.0 * value_raw / tcore_clock_freq_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); + } + } + fmt8 = "%s%.2f"; 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)) @@ -1951,6 +3156,24 @@ 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); @@ -1976,43 +3199,63 @@ 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)) + if (DO_BIC(BIC_Diec6)) 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); + sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->die_c6 / crystal_hz / 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 += - 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); @@ -2031,9 +3274,62 @@ int format_counters(struct thread_data *t, struct core_data *c, struct pkg_data outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->counter[i]); } else if (mp->format == FORMAT_PERCENT) { outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->counter[i] / tsc); + } else if (mp->type == COUNTER_K2M) + outp += sprintf(outp, "%s%d", (printed++ ? delim : ""), (unsigned int)p->counter[i] / 1000); + } + + for (i = 0, pp = sys.perf_pp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) { + if (pp->width == 32) + outp += + sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), + (unsigned int)p->perf_counter[i]); + else + outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->perf_counter[i]); + } else if (pp->format == FORMAT_DELTA) { + if ((pp->type == COUNTER_ITEMS) && sums_need_wide_columns) + outp += sprintf(outp, "%s%8lld", (printed++ ? delim : ""), p->perf_counter[i]); + else + outp += sprintf(outp, "%s%lld", (printed++ ? delim : ""), p->perf_counter[i]); + } else if (pp->format == FORMAT_PERCENT) { + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), 100.0 * p->perf_counter[i] / tsc); + } else if (pp->type == COUNTER_K2M) { + outp += + sprintf(outp, "%s%d", (printed++ ? delim : ""), (unsigned int)p->perf_counter[i] / 1000); } } + for (i = 0, ppmt = sys.pmt_pp; ppmt; i++, ppmt = ppmt->next) { + const unsigned long value_raw = p->pmt_counter[i]; + double value_converted; + switch (ppmt->type) { + case PMT_TYPE_RAW: + if (pmt_counter_get_width(ppmt) <= 32) + outp += sprintf(outp, "%s0x%08x", (printed++ ? delim : ""), + (unsigned int)p->pmt_counter[i]); + else + outp += sprintf(outp, "%s0x%016llx", (printed++ ? delim : ""), p->pmt_counter[i]); + + break; + + case PMT_TYPE_XTAL_TIME: + value_converted = 100.0 * value_raw / crystal_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); + break; + + case PMT_TYPE_TCORE_CLOCK: + value_converted = 100.0 * value_raw / tcore_clock_freq_hz / interval_float; + outp += sprintf(outp, "%s%.2f", (printed++ ? delim : ""), value_converted); + } + } + + if (DO_BIC(BIC_SysWatt) && (t == &average.threads)) + outp += sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&pplat_cnt->energy_psys, RAPL_UNIT_WATTS, interval_float)); + if (DO_BIC(BIC_Sys_J) && (t == &average.threads)) + outp += sprintf(outp, fmt8, (printed++ ? delim : ""), + rapl_counter_get_value(&pplat_cnt->energy_psys, RAPL_UNIT_JOULES, interval_float)); + done: if (*(outp - 1) != '\n') outp += sprintf(outp, "\n"); @@ -2087,6 +3383,8 @@ int delta_package(struct pkg_data *new, struct pkg_data *old) { int i; struct msr_counter *mp; + struct perf_counter_info *pp; + struct pmt_counter *ppmt; if (DO_BIC(BIC_Totl_c0)) old->pkg_wtd_core_c0 = new->pkg_wtd_core_c0 - old->pkg_wtd_core_c0; @@ -2107,6 +3405,7 @@ int delta_package(struct pkg_data *new, struct pkg_data *old) old->pc8 = new->pc8 - old->pc8; old->pc9 = new->pc9 - old->pc9; old->pc10 = new->pc10 - old->pc10; + old->die_c6 = new->die_c6 - old->die_c6; old->cpu_lpi = new->cpu_lpi - old->cpu_lpi; old->sys_lpi = new->sys_lpi - old->sys_lpi; old->pkg_temp_c = new->pkg_temp_c; @@ -2121,20 +3420,48 @@ 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; + /* 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; + 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) old->counter[i] = new->counter[i]; + else if (mp->format == FORMAT_AVERAGE) + old->counter[i] = new->counter[i]; else old->counter[i] = new->counter[i] - old->counter[i]; } + for (i = 0, pp = sys.perf_pp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + old->perf_counter[i] = new->perf_counter[i]; + else if (pp->format == FORMAT_AVERAGE) + old->perf_counter[i] = new->perf_counter[i]; + else + old->perf_counter[i] = new->perf_counter[i] - old->perf_counter[i]; + } + + for (i = 0, ppmt = sys.pmt_pp; ppmt; i++, ppmt = ppmt->next) { + if (ppmt->format == FORMAT_RAW) + old->pmt_counter[i] = new->pmt_counter[i]; + else + old->pmt_counter[i] = new->pmt_counter[i] - old->pmt_counter[i]; + } + return 0; } @@ -2142,6 +3469,8 @@ void delta_core(struct core_data *new, struct core_data *old) { int i; struct msr_counter *mp; + struct perf_counter_info *pp; + struct pmt_counter *ppmt; old->c3 = new->c3 - old->c3; old->c6 = new->c6 - old->c6; @@ -2150,7 +3479,7 @@ 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) @@ -2158,6 +3487,20 @@ void delta_core(struct core_data *new, struct core_data *old) else old->counter[i] = new->counter[i] - old->counter[i]; } + + for (i = 0, pp = sys.perf_cp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + old->perf_counter[i] = new->perf_counter[i]; + else + old->perf_counter[i] = new->perf_counter[i] - old->perf_counter[i]; + } + + for (i = 0, ppmt = sys.pmt_cp; ppmt; i++, ppmt = ppmt->next) { + if (ppmt->format == FORMAT_RAW) + old->pmt_counter[i] = new->pmt_counter[i]; + else + old->pmt_counter[i] = new->pmt_counter[i] - old->pmt_counter[i]; + } } int soft_c1_residency_display(int bic) @@ -2175,6 +3518,8 @@ int delta_thread(struct thread_data *new, struct thread_data *old, struct core_d { int i; struct msr_counter *mp; + struct perf_counter_info *pp; + struct pmt_counter *ppmt; /* we run cpuid just the 1st time, copy the results */ if (DO_BIC(BIC_APIC)) @@ -2244,6 +3589,9 @@ int delta_thread(struct thread_data *new, struct thread_data *old, struct core_d if (DO_BIC(BIC_IRQ)) old->irq_count = new->irq_count - old->irq_count; + if (DO_BIC(BIC_NMI)) + old->nmi_count = new->nmi_count - old->nmi_count; + if (DO_BIC(BIC_SMI)) old->smi_count = new->smi_count - old->smi_count; @@ -2253,6 +3601,21 @@ int delta_thread(struct thread_data *new, struct thread_data *old, struct core_d else old->counter[i] = new->counter[i] - old->counter[i]; } + + for (i = 0, pp = sys.perf_tp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + old->perf_counter[i] = new->perf_counter[i]; + else + old->perf_counter[i] = new->perf_counter[i] - old->perf_counter[i]; + } + + for (i = 0, ppmt = sys.pmt_tp; ppmt; i++, ppmt = ppmt->next) { + if (ppmt->format == FORMAT_RAW) + old->pmt_counter[i] = new->pmt_counter[i]; + else + old->pmt_counter[i] = new->pmt_counter[i] - old->pmt_counter[i]; + } + return 0; } @@ -2267,16 +3630,26 @@ int delta_cpu(struct thread_data *t, struct core_data *c, /* always calculate thread delta */ retval = delta_thread(t, t2, c2); /* c2 is core delta */ - if (retval) - return retval; /* calculate package delta only for 1st core in package */ if (is_cpu_first_core_in_package(t, c, p)) - retval = delta_package(p, p2); + retval |= delta_package(p, p2); return retval; } +void delta_platform(struct platform_counters *new, struct platform_counters *old) +{ + old->energy_psys.raw_value = new->energy_psys.raw_value - old->energy_psys.raw_value; +} + +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; @@ -2297,6 +3670,7 @@ void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data t->instr_count = 0; t->irq_count = 0; + t->nmi_count = 0; t->smi_count = 0; c->c3 = 0; @@ -2304,7 +3678,7 @@ 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; @@ -2322,21 +3696,25 @@ void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data p->pc8 = 0; p->pc9 = 0; p->pc10 = 0; + p->die_c6 = 0; 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; 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; @@ -2345,12 +3723,36 @@ void clear_counters(struct thread_data *t, struct core_data *c, struct pkg_data for (i = 0, mp = sys.pp; mp; i++, mp = mp->next) p->counter[i] = 0; + + memset(&t->perf_counter[0], 0, sizeof(t->perf_counter)); + memset(&c->perf_counter[0], 0, sizeof(c->perf_counter)); + memset(&p->perf_counter[0], 0, sizeof(p->perf_counter)); + + memset(&t->pmt_counter[0], 0, ARRAY_SIZE(t->pmt_counter)); + memset(&c->pmt_counter[0], 0, ARRAY_SIZE(c->pmt_counter)); + memset(&p->pmt_counter[0], 0, ARRAY_SIZE(p->pmt_counter)); +} + +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; struct msr_counter *mp; + struct perf_counter_info *pp; + struct pmt_counter *ppmt; /* copy un-changing apic_id's */ if (DO_BIC(BIC_APIC)) @@ -2360,7 +3762,7 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) /* remember first tv_begin */ if (average.threads.tv_begin.tv_sec == 0) - average.threads.tv_begin = t->tv_begin; + average.threads.tv_begin = procsysfs_tv_begin; /* remember last tv_end */ average.threads.tv_end = t->tv_end; @@ -2373,6 +3775,7 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.threads.instr_count += t->instr_count; average.threads.irq_count += t->irq_count; + average.threads.nmi_count += t->nmi_count; average.threads.smi_count += t->smi_count; for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { @@ -2381,6 +3784,16 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.threads.counter[i] += t->counter[i]; } + for (i = 0, pp = sys.perf_tp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + continue; + average.threads.perf_counter[i] += t->perf_counter[i]; + } + + for (i = 0, ppmt = sys.pmt_tp; ppmt; i++, ppmt = ppmt->next) { + average.threads.pmt_counter[i] += t->pmt_counter[i]; + } + /* sum per-core values only for 1st thread in core */ if (!is_cpu_first_thread_in_core(t, c, p)) return 0; @@ -2393,7 +3806,7 @@ 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) @@ -2401,6 +3814,16 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.cores.counter[i] += c->counter[i]; } + for (i = 0, pp = sys.perf_cp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + continue; + average.cores.perf_counter[i] += c->perf_counter[i]; + } + + for (i = 0, ppmt = sys.pmt_cp; ppmt; i++, ppmt = ppmt->next) { + average.cores.pmt_counter[i] += c->pmt_counter[i]; + } + /* sum per-pkg values only for 1st core in pkg */ if (!is_cpu_first_core_in_package(t, c, p)) return 0; @@ -2424,30 +3847,47 @@ int sum_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) average.packages.pc8 += p->pc8; average.packages.pc9 += p->pc9; average.packages.pc10 += p->pc10; + average.packages.die_c6 += p->die_c6; 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; 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); - 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) - 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]; } + + for (i = 0, pp = sys.perf_pp; pp; i++, pp = pp->next) { + if ((pp->format == FORMAT_RAW) && (topo.num_packages == 0)) + average.packages.perf_counter[i] = p->perf_counter[i]; + else + average.packages.perf_counter[i] += p->perf_counter[i]; + } + + for (i = 0, ppmt = sys.pmt_pp; ppmt; i++, ppmt = ppmt->next) { + average.packages.pmt_counter[i] += p->pmt_counter[i]; + } + return 0; } @@ -2459,6 +3899,8 @@ void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data { int i; struct msr_counter *mp; + struct perf_counter_info *pp; + struct pmt_counter *ppmt; clear_counters(&average.threads, &average.cores, &average.packages); @@ -2475,6 +3917,9 @@ void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data if (average.threads.irq_count > 9999999) sums_need_wide_columns = 1; + if (average.threads.nmi_count > 9999999) + sums_need_wide_columns = 1; + average.cores.c3 /= topo.allowed_cores; average.cores.c6 /= topo.allowed_cores; @@ -2501,6 +3946,7 @@ void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data average.packages.pc8 /= topo.allowed_packages; average.packages.pc9 /= topo.allowed_packages; average.packages.pc10 /= topo.allowed_packages; + average.packages.die_c6 /= topo.allowed_packages; for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { if (mp->format == FORMAT_RAW) @@ -2530,6 +3976,45 @@ void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data } average.packages.counter[i] /= topo.allowed_packages; } + + for (i = 0, pp = sys.perf_tp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + continue; + if (pp->type == COUNTER_ITEMS) { + if (average.threads.perf_counter[i] > 9999999) + sums_need_wide_columns = 1; + continue; + } + average.threads.perf_counter[i] /= topo.allowed_cpus; + } + for (i = 0, pp = sys.perf_cp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + continue; + if (pp->type == COUNTER_ITEMS) { + if (average.cores.perf_counter[i] > 9999999) + sums_need_wide_columns = 1; + } + average.cores.perf_counter[i] /= topo.allowed_cores; + } + for (i = 0, pp = sys.perf_pp; pp; i++, pp = pp->next) { + if (pp->format == FORMAT_RAW) + continue; + if (pp->type == COUNTER_ITEMS) { + if (average.packages.perf_counter[i] > 9999999) + sums_need_wide_columns = 1; + } + average.packages.perf_counter[i] /= topo.allowed_packages; + } + + for (i = 0, ppmt = sys.pmt_tp; ppmt; i++, ppmt = ppmt->next) { + average.threads.pmt_counter[i] /= topo.allowed_cpus; + } + for (i = 0, ppmt = sys.pmt_cp; ppmt; i++, ppmt = ppmt->next) { + average.cores.pmt_counter[i] /= topo.allowed_cores; + } + for (i = 0, ppmt = sys.pmt_pp; ppmt; i++, ppmt = ppmt->next) { + average.packages.pmt_counter[i] /= topo.allowed_packages; + } } static unsigned long long rdtsc(void) @@ -2575,34 +4060,50 @@ unsigned long long snapshot_sysfs_counter(char *path) return counter; } -int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp) +int get_mp(int cpu, struct msr_counter *mp, unsigned long long *counterp, char *counter_path) { if (mp->msr_num != 0) { + assert(!no_msr); if (get_msr(cpu, mp->msr_num, counterp)) return -1; } else { char path[128 + PATH_BYTES]; if (mp->flags & SYSFS_PERCPU) { - sprintf(path, "/sys/devices/system/cpu/cpu%d/%s", cpu, mp->path); + sprintf(path, "/sys/devices/system/cpu/cpu%d/%s", cpu, mp->sp->path); *counterp = snapshot_sysfs_counter(path); } else { - *counterp = snapshot_sysfs_counter(mp->path); + *counterp = snapshot_sysfs_counter(counter_path); } } return 0; } -unsigned long long get_uncore_mhz(int package, int die) +unsigned long long get_legacy_uncore_mhz(int package) { char path[128]; + int die; + static int warn_once; - sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/current_freq_khz", package, - die); + /* + * for this package, use the first die_id that exists + */ + for (die = 0; die <= topo.max_die_id; ++die) { - return (snapshot_sysfs_counter(path) / 1000); + sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_%02d_die_%02d/current_freq_khz", + package, die); + + if (access(path, R_OK) == 0) + return (snapshot_sysfs_counter(path) / 1000); + } + if (!warn_once) { + warnx("BUG: %s: No %s", __func__, path); + warn_once = 1; + } + + return 0; } int get_epb(int cpu) @@ -2627,6 +4128,9 @@ 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; @@ -2700,137 +4204,613 @@ int get_core_throt_cnt(int cpu, unsigned long long *cnt) return 0; } -/* - * get_counters(...) - * migrate to cpu - * acquire and record local counters for that cpu - */ -int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) +struct amperf_group_fd { + int aperf; /* Also the group descriptor */ + int mperf; +}; + +static int read_perf_counter_info(const char *const path, const char *const parse_format, void *value_ptr) { - int cpu = t->cpu_id; - unsigned long long msr; - int aperf_mperf_retry_count = 0; - struct msr_counter *mp; - int i; + int fdmt; + int bytes_read; + char buf[64]; + int ret = -1; - if (cpu_migrate(cpu)) { - fprintf(outf, "get_counters: Could not migrate to CPU %d\n", cpu); + fdmt = open(path, O_RDONLY, 0); + if (fdmt == -1) { + if (debug) + fprintf(stderr, "Failed to parse perf counter info %s\n", path); + ret = -1; + goto cleanup_and_exit; + } + + 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[bytes_read] = '\0'; + + 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; +} + +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"; + char path[128]; + + snprintf(path, sizeof(path), path_format, subsys); + + return read_perf_counter_info_n(path, format); +} + +static unsigned int read_perf_config(const char *subsys, const char *event_name) +{ + const char *const path_format = "/sys/bus/event_source/devices/%s/events/%s"; + FILE *fconfig = NULL; + char path[128]; + char config_str[64]; + unsigned int config; + unsigned int umask; + bool has_config = false; + bool has_umask = false; + unsigned int ret = -1; + + snprintf(path, sizeof(path), path_format, subsys, event_name); + + fconfig = fopen(path, "r"); + if (!fconfig) return -1; + + if (fgets(config_str, ARRAY_SIZE(config_str), fconfig) != config_str) + goto cleanup_and_exit; + + for (char *pconfig_str = &config_str[0]; pconfig_str;) { + if (sscanf(pconfig_str, "event=%x", &config) == 1) { + has_config = true; + goto next; + } + + if (sscanf(pconfig_str, "umask=%x", &umask) == 1) { + has_umask = true; + goto next; + } + +next: + pconfig_str = strchr(pconfig_str, ','); + if (pconfig_str) { + *pconfig_str = '\0'; + ++pconfig_str; + } } - gettimeofday(&t->tv_begin, (struct timezone *)NULL); + if (!has_umask) + umask = 0; - if (first_counter_read) - get_apic_id(t); -retry: - t->tsc = rdtsc(); /* we are running on local CPU of interest */ + if (has_config) + ret = (umask << 8) | config; - 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; +cleanup_and_exit: + fclose(fconfig); + return ret; +} - /* - * 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. - */ +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"; + char path[128]; + char unit_buffer[16]; - /* - * This initial dummy APERF read has been seen to - * reduce jitter in the subsequent reads. - */ + snprintf(path, sizeof(path), path_format, subsys, event_name); - if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) - return -3; + read_perf_counter_info(path, format, &unit_buffer); + if (strcmp("Joules", unit_buffer) == 0) + return RAPL_UNIT_JOULES; - t->tsc = rdtsc(); /* re-read close to APERF */ + return RAPL_UNIT_INVALID; +} - tsc_before = t->tsc; +static double read_perf_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; - if (get_msr(cpu, MSR_IA32_APERF, &t->aperf)) - return -3; + snprintf(path, sizeof(path), path_format, subsys, event_name); - tsc_between = rdtsc(); + if (read_perf_counter_info(path, format, &scale)) + return 0.0; - if (get_msr(cpu, MSR_IA32_MPERF, &t->mperf)) - return -4; + return scale; +} - tsc_after = rdtsc(); +size_t rapl_counter_info_count_perf(const struct rapl_counter_info_t *rci) +{ + size_t ret = 0; - aperf_time = tsc_between - tsc_before; - mperf_time = tsc_after - tsc_between; + for (int i = 0; i < NUM_RAPL_COUNTERS; ++i) + if (rci->source[i] == COUNTER_SOURCE_PERF) + ++ret; - /* - * 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; + return ret; +} + +static size_t cstate_counter_info_count_perf(const struct cstate_counter_info_t *cci) +{ + size_t ret = 0; + + for (int i = 0; i < NUM_CSTATE_COUNTERS; ++i) + if (cci->source[i] == COUNTER_SOURCE_PERF) + ++ret; + + return ret; +} + +void write_rapl_counter(struct rapl_counter *rc, struct rapl_counter_info_t *rci, unsigned int idx) +{ + if (rci->source[idx] == COUNTER_SOURCE_NONE) + return; + + rc->raw_value = rci->data[idx]; + rc->unit = rci->unit[idx]; + rc->scale = rci->scale[idx]; +} + +int get_rapl_counters(int cpu, unsigned int domain, struct core_data *c, struct pkg_data *p) +{ + struct platform_counters *pplat_cnt = p == package_odd ? &platform_counters_odd : &platform_counters_even; + unsigned long long perf_data[NUM_RAPL_COUNTERS + 1]; + struct rapl_counter_info_t *rci; + + if (debug >= 2) + fprintf(stderr, "%s: cpu%d domain%d\n", __func__, cpu, domain); + + assert(rapl_counter_info_perdomain); + assert(domain < rapl_counter_info_perdomain_size); + + rci = &rapl_counter_info_perdomain[domain]; + + /* + * 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, "%s: failed to read perf_data (%zu %zu)", __func__, expected_read_size, + actual_read_size); + } + + for (unsigned int i = 0, pi = 1; i < NUM_RAPL_COUNTERS; ++i) { + switch (rci->source[i]) { + case COUNTER_SOURCE_NONE: + rci->data[i] = 0; + break; + + case COUNTER_SOURCE_PERF: + assert(pi < ARRAY_SIZE(perf_data)); + assert(rci->fd_perf != -1); + + if (debug >= 2) + 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 COUNTER_SOURCE_MSR: + if (debug >= 2) + 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 - warnx("cpu%d jitter %lld %lld", cpu, aperf_time, mperf_time); + rci->data[i] <<= abs(rci->msr_shift[i]); + + break; } - aperf_mperf_retry_count = 0; + } + + BUILD_BUG_ON(NUM_RAPL_COUNTERS != 8); + 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); + write_rapl_counter(&pplat_cnt->energy_psys, rci, RAPL_RCI_INDEX_ENERGY_PLATFORM); + + return 0; +} + +char *find_sysfs_path_by_id(struct sysfs_path *sp, int id) +{ + while (sp) { + if (sp->id == id) + return (sp->path); + sp = sp->next; + } + if (debug) + warnx("%s: id%d not found", __func__, id); + return NULL; +} + +int get_cstate_counters(unsigned int cpu, struct thread_data *t, struct core_data *c, struct pkg_data *p) +{ + /* + * Overcommit memory a little bit here, + * but skip calculating exact sizes for the buffers. + */ + unsigned long long perf_data[NUM_CSTATE_COUNTERS]; + unsigned long long perf_data_core[NUM_CSTATE_COUNTERS + 1]; + unsigned long long perf_data_pkg[NUM_CSTATE_COUNTERS + 1]; + + struct cstate_counter_info_t *cci; + + if (debug >= 2) + fprintf(stderr, "%s: cpu%d\n", __func__, cpu); + + assert(ccstate_counter_info); + assert(cpu <= ccstate_counter_info_size); + + ZERO_ARRAY(perf_data); + ZERO_ARRAY(perf_data_core); + ZERO_ARRAY(perf_data_pkg); + + cci = &ccstate_counter_info[cpu]; + + /* + * If we have any perf counters to read, read them all now, in bulk + */ + const size_t num_perf_counters = cstate_counter_info_count_perf(cci); + ssize_t expected_read_size = num_perf_counters * sizeof(unsigned long long); + ssize_t actual_read_size_core = 0, actual_read_size_pkg = 0; + + if (cci->fd_perf_core != -1) { + /* Each descriptor read begins with number of counters read. */ + expected_read_size += sizeof(unsigned long long); + + actual_read_size_core = read(cci->fd_perf_core, &perf_data_core[0], sizeof(perf_data_core)); - t->aperf = t->aperf * aperf_mperf_multiplier; - t->mperf = t->mperf * aperf_mperf_multiplier; + if (actual_read_size_core <= 0) + err(-1, "%s: read perf %s: %ld", __func__, "core", actual_read_size_core); } + if (cci->fd_perf_pkg != -1) { + /* Each descriptor read begins with number of counters read. */ + expected_read_size += sizeof(unsigned long long); + + actual_read_size_pkg = read(cci->fd_perf_pkg, &perf_data_pkg[0], sizeof(perf_data_pkg)); + + if (actual_read_size_pkg <= 0) + err(-1, "%s: read perf %s: %ld", __func__, "pkg", actual_read_size_pkg); + } + + const ssize_t actual_read_size_total = actual_read_size_core + actual_read_size_pkg; + + if (actual_read_size_total != expected_read_size) + err(-1, "%s: failed to read perf_data (%zu %zu)", __func__, expected_read_size, actual_read_size_total); + + /* + * Copy ccstate and pcstate data into unified buffer. + * + * Skip first element from core and pkg buffers. + * Kernel puts there how many counters were read. + */ + const size_t num_core_counters = perf_data_core[0]; + const size_t num_pkg_counters = perf_data_pkg[0]; + + assert(num_perf_counters == num_core_counters + num_pkg_counters); + + /* Copy ccstate perf data */ + memcpy(&perf_data[0], &perf_data_core[1], num_core_counters * sizeof(unsigned long long)); + + /* Copy pcstate perf data */ + memcpy(&perf_data[num_core_counters], &perf_data_pkg[1], num_pkg_counters * sizeof(unsigned long long)); + + for (unsigned int i = 0, pi = 0; i < NUM_CSTATE_COUNTERS; ++i) { + switch (cci->source[i]) { + case COUNTER_SOURCE_NONE: + break; + + case COUNTER_SOURCE_PERF: + assert(pi < ARRAY_SIZE(perf_data)); + assert(cci->fd_perf_core != -1 || cci->fd_perf_pkg != -1); + + if (debug >= 2) + fprintf(stderr, "cstate via %s %u: %llu\n", "perf", i, perf_data[pi]); + + cci->data[i] = perf_data[pi]; + + ++pi; + break; + + case COUNTER_SOURCE_MSR: + assert(!no_msr); + if (get_msr(cpu, cci->msr[i], &cci->data[i])) + return -13 - i; + + if (debug >= 2) + fprintf(stderr, "cstate via %s0x%llx %u: %llu\n", "msr", cci->msr[i], i, cci->data[i]); + + break; + } + } + + /* + * Helper to write the data only if the source of + * the counter for the current cpu is not none. + * + * Otherwise we would overwrite core data with 0 (default value), + * when invoked for the thread sibling. + */ +#define PERF_COUNTER_WRITE_DATA(out_counter, index) do { \ + if (cci->source[index] != COUNTER_SOURCE_NONE) \ + out_counter = cci->data[index]; \ +} while (0) + + BUILD_BUG_ON(NUM_CSTATE_COUNTERS != 11); + + PERF_COUNTER_WRITE_DATA(t->c1, CCSTATE_RCI_INDEX_C1_RESIDENCY); + PERF_COUNTER_WRITE_DATA(c->c3, CCSTATE_RCI_INDEX_C3_RESIDENCY); + PERF_COUNTER_WRITE_DATA(c->c6, CCSTATE_RCI_INDEX_C6_RESIDENCY); + PERF_COUNTER_WRITE_DATA(c->c7, CCSTATE_RCI_INDEX_C7_RESIDENCY); + + PERF_COUNTER_WRITE_DATA(p->pc2, PCSTATE_RCI_INDEX_C2_RESIDENCY); + PERF_COUNTER_WRITE_DATA(p->pc3, PCSTATE_RCI_INDEX_C3_RESIDENCY); + PERF_COUNTER_WRITE_DATA(p->pc6, PCSTATE_RCI_INDEX_C6_RESIDENCY); + PERF_COUNTER_WRITE_DATA(p->pc7, PCSTATE_RCI_INDEX_C7_RESIDENCY); + PERF_COUNTER_WRITE_DATA(p->pc8, PCSTATE_RCI_INDEX_C8_RESIDENCY); + PERF_COUNTER_WRITE_DATA(p->pc9, PCSTATE_RCI_INDEX_C9_RESIDENCY); + PERF_COUNTER_WRITE_DATA(p->pc10, PCSTATE_RCI_INDEX_C10_RESIDENCY); + +#undef PERF_COUNTER_WRITE_DATA + + return 0; +} + +size_t msr_counter_info_count_perf(const struct msr_counter_info_t *mci) +{ + size_t ret = 0; + + for (int i = 0; i < NUM_MSR_COUNTERS; ++i) + if (mci->source[i] == COUNTER_SOURCE_PERF) + ++ret; + + return ret; +} + +int get_smi_aperf_mperf(unsigned int cpu, struct thread_data *t) +{ + unsigned long long perf_data[NUM_MSR_COUNTERS + 1]; + + struct msr_counter_info_t *mci; + + if (debug >= 2) + fprintf(stderr, "%s: cpu%d\n", __func__, cpu); + + assert(msr_counter_info); + assert(cpu <= msr_counter_info_size); + + mci = &msr_counter_info[cpu]; + + ZERO_ARRAY(perf_data); + ZERO_ARRAY(mci->data); + + if (mci->fd_perf != -1) { + const size_t num_perf_counters = msr_counter_info_count_perf(mci); + const ssize_t expected_read_size = (num_perf_counters + 1) * sizeof(unsigned long long); + const ssize_t actual_read_size = read(mci->fd_perf, &perf_data[0], sizeof(perf_data)); + + if (actual_read_size != expected_read_size) + err(-1, "%s: failed to read perf_data (%zu %zu)", __func__, expected_read_size, + actual_read_size); + } + + for (unsigned int i = 0, pi = 1; i < NUM_MSR_COUNTERS; ++i) { + switch (mci->source[i]) { + case COUNTER_SOURCE_NONE: + break; + + case COUNTER_SOURCE_PERF: + assert(pi < ARRAY_SIZE(perf_data)); + assert(mci->fd_perf != -1); + + if (debug >= 2) + fprintf(stderr, "Reading msr counter via perf at %u: %llu\n", i, perf_data[pi]); + + mci->data[i] = perf_data[pi]; + + ++pi; + break; + + case COUNTER_SOURCE_MSR: + assert(!no_msr); + + if (get_msr(cpu, mci->msr[i], &mci->data[i])) + return -2 - i; + + mci->data[i] &= mci->msr_mask[i]; + + if (debug >= 2) + fprintf(stderr, "Reading msr counter via msr at %u: %llu\n", i, mci->data[i]); + + break; + } + } + + BUILD_BUG_ON(NUM_MSR_COUNTERS != 3); + t->aperf = mci->data[MSR_RCI_INDEX_APERF]; + t->mperf = mci->data[MSR_RCI_INDEX_MPERF]; + t->smi_count = mci->data[MSR_RCI_INDEX_SMI]; + + return 0; +} + +int perf_counter_info_read_values(struct perf_counter_info *pp, int cpu, unsigned long long *out, size_t out_size) +{ + unsigned int domain; + unsigned long long value; + int fd_counter; + + for (size_t i = 0; pp; ++i, pp = pp->next) { + domain = cpu_to_domain(pp, cpu); + assert(domain < pp->num_domains); + + fd_counter = pp->fd_perf_per_domain[domain]; + + if (fd_counter == -1) + continue; + + if (read(fd_counter, &value, sizeof(value)) != sizeof(value)) + return 1; + + assert(i < out_size); + out[i] = value * pp->scale; + } + + return 0; +} + +unsigned long pmt_gen_value_mask(unsigned int lsb, unsigned int msb) +{ + unsigned long mask; + + if (msb == 63) + mask = 0xffffffffffffffff; + else + mask = ((1 << (msb + 1)) - 1); + + mask -= (1 << lsb) - 1; + + return mask; +} + +unsigned long pmt_read_counter(struct pmt_counter *ppmt, unsigned int domain_id) +{ + if (domain_id >= ppmt->num_domains) + return 0; + + const unsigned long *pmmio = ppmt->domains[domain_id].pcounter; + const unsigned long value = pmmio ? *pmmio : 0; + const unsigned long value_mask = pmt_gen_value_mask(ppmt->lsb, ppmt->msb); + const unsigned long value_shift = ppmt->lsb; + + return (value & value_mask) >> value_shift; +} + +/* + * get_counters(...) + * migrate to cpu + * acquire and record local counters for that cpu + */ +int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p) +{ + int cpu = t->cpu_id; + unsigned long long msr; + struct msr_counter *mp; + struct pmt_counter *pp; + int i; + int status; + + if (cpu_migrate(cpu)) { + fprintf(outf, "%s: Could not migrate to CPU %d\n", __func__, cpu); + return -1; + } + + gettimeofday(&t->tv_begin, (struct timezone *)NULL); + + if (first_counter_read) + get_apic_id(t); + + t->tsc = rdtsc(); /* we are running on local CPU of interest */ + + get_smi_aperf_mperf(cpu, t); + 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)) { - if (get_msr(cpu, MSR_SMI_COUNT, &msr)) - return -5; - t->smi_count = msr & 0xFFFFFFFF; - } - if (DO_BIC(BIC_CPU_c1) && platform->has_msr_core_c1_res) { - if (get_msr(cpu, MSR_CORE_C1_RES, &t->c1)) - return -6; - } + if (DO_BIC(BIC_NMI)) + t->nmi_count = nmi_per_cpu[cpu]; + + get_cstate_counters(cpu, t, c, p); for (i = 0, mp = sys.tp; mp; i++, mp = mp->next) { - if (get_mp(cpu, mp, &t->counter[i])) + if (get_mp(cpu, mp, &t->counter[i], mp->sp->path)) return -10; } + if (perf_counter_info_read_values(sys.perf_tp, cpu, t->perf_counter, MAX_ADDED_THREAD_COUNTERS)) + return -10; + + for (i = 0, pp = sys.pmt_tp; pp; i++, pp = pp->next) + t->pmt_counter[i] = pmt_read_counter(pp, t->cpu_id); + /* collect core counters only for 1st thread in core */ if (!is_cpu_first_thread_in_core(t, c, p)) goto done; - 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; + 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_c6) || soft_c1_residency_display(BIC_CPU_c6)) && !platform->has_msr_knl_core_c6_residency) { - if (get_msr(cpu, MSR_CORE_C6_RESIDENCY, &c->c6)) - return -7; - } else if (platform->has_msr_knl_core_c6_residency && soft_c1_residency_display(BIC_CPU_c6)) { - if (get_msr(cpu, MSR_KNL_CORE_C6_RESIDENCY, &c->c6)) - return -7; - } - - if (DO_BIC(BIC_CPU_c7) || soft_c1_residency_display(BIC_CPU_c7)) { - if (get_msr(cpu, MSR_CORE_C7_RESIDENCY, &c->c7)) - return -8; - else if (t->is_atom) { - /* - * For Atom CPUs that has core cstate deeper than c6, - * MSR_CORE_C6_RESIDENCY returns residency of cc6 and deeper. - * Minus CC7 (and deeper cstates) residency to get - * accturate cc6 residency. - */ - c->c6 -= c->c7; - } + if (DO_BIC(BIC_CPU_c7) && t->is_atom) { + /* + * For Atom CPUs that has core cstate deeper than c6, + * MSR_CORE_C6_RESIDENCY returns residency of cc6 and deeper. + * Minus CC7 (and deeper cstates) residency to get + * accturate cc6 residency. + */ + c->c6 -= c->c7; } if (DO_BIC(BIC_Mod_c6)) @@ -2846,17 +4826,17 @@ 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 (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])) + if (get_mp(cpu, mp, &c->counter[i], mp->sp->path)) return -10; } + if (perf_counter_info_read_values(sys.perf_cp, cpu, c->perf_counter, MAX_ADDED_CORE_COUNTERS)) + return -10; + + for (i = 0, pp = sys.pmt_cp; pp; i++, pp = pp->next) + c->pmt_counter[i] = pmt_read_counter(pp, c->core_id); + /* collect package counters only for 1st core in package */ if (!is_cpu_first_core_in_package(t, c, p)) goto done; @@ -2877,129 +4857,73 @@ retry: if (get_msr(cpu, MSR_PKG_BOTH_CORE_GFXE_C0_RES, &p->pkg_both_core_gfxe_c0)) return -13; } - if (DO_BIC(BIC_Pkgpc3)) - if (get_msr(cpu, MSR_PKG_C3_RESIDENCY, &p->pc3)) - return -9; - if (DO_BIC(BIC_Pkgpc6)) { - if (platform->has_msr_atom_pkg_c6_residency) { - if (get_msr(cpu, MSR_ATOM_PKG_C6_RESIDENCY, &p->pc6)) - return -10; - } else { - if (get_msr(cpu, MSR_PKG_C6_RESIDENCY, &p->pc6)) - return -10; - } - } - - if (DO_BIC(BIC_Pkgpc2)) - if (get_msr(cpu, MSR_PKG_C2_RESIDENCY, &p->pc2)) - return -11; - if (DO_BIC(BIC_Pkgpc7)) - if (get_msr(cpu, MSR_PKG_C7_RESIDENCY, &p->pc7)) - return -12; - if (DO_BIC(BIC_Pkgpc8)) - if (get_msr(cpu, MSR_PKG_C8_RESIDENCY, &p->pc8)) - return -13; - if (DO_BIC(BIC_Pkgpc9)) - if (get_msr(cpu, MSR_PKG_C9_RESIDENCY, &p->pc9)) - return -13; - if (DO_BIC(BIC_Pkgpc10)) - if (get_msr(cpu, MSR_PKG_C10_RESIDENCY, &p->pc10)) - return -13; if (DO_BIC(BIC_CPU_LPI)) p->cpu_lpi = cpuidle_cur_cpu_lpi_us; 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 (!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)) { if (get_msr(cpu, MSR_IA32_PACKAGE_THERM_STATUS, &msr)) return -17; p->pkg_temp_c = tj_max - ((msr >> 16) & 0x7F); } - if (DO_BIC(BIC_GFX_rc6)) - p->gfx_rc6_ms = gfx_cur_rc6_ms; - - /* 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); + p->uncore_mhz = get_legacy_uncore_mhz(p->package_id); + + 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_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; + + 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])) + char *path = NULL; + + if (mp->msr_num == 0) { + path = find_sysfs_path_by_id(mp->sp, p->package_id); + if (path == NULL) { + warnx("%s: package_id %d not found", __func__, p->package_id); + return -10; + } + } + if (get_mp(cpu, mp, &p->counter[i], path)) return -10; } + + if (perf_counter_info_read_values(sys.perf_pp, cpu, p->perf_counter, MAX_ADDED_PACKAGE_COUNTERS)) + return -10; + + for (i = 0, pp = sys.pmt_pp; pp; i++, pp = pp->next) + p->pmt_counter[i] = pmt_read_counter(pp, p->package_id); + done: gettimeofday(&t->tv_end, (struct timezone *)NULL); return 0; } -/* - * MSR_PKG_CST_CONFIG_CONTROL decoding for pkg_cstate_limit: - * If you change the values, note they are used both in comparisons - * (>= PCL__7) and to index pkg_cstate_limit_strings[]. - */ - -#define PCLUKN 0 /* Unknown */ -#define PCLRSV 1 /* Reserved */ -#define PCL__0 2 /* PC0 */ -#define PCL__1 3 /* PC1 */ -#define PCL__2 4 /* PC2 */ -#define PCL__3 5 /* PC3 */ -#define PCL__4 6 /* PC4 */ -#define PCL__6 7 /* PC6 */ -#define PCL_6N 8 /* PC6 No Retention */ -#define PCL_6R 9 /* PC6 Retention */ -#define PCL__7 10 /* PC7 */ -#define PCL_7S 11 /* PC7 Shrink */ -#define PCL__8 12 /* PC8 */ -#define PCL__9 13 /* PC9 */ -#define PCL_10 14 /* PC10 */ -#define PCLUNL 15 /* Unlimited */ - int pkg_cstate_limit = PCLUKN; -char *pkg_cstate_limit_strings[] = { "reserved", "unknown", "pc0", "pc1", "pc2", +char *pkg_cstate_limit_strings[] = { "unknown", "reserved", "pc0", "pc1", "pc2", "pc3", "pc4", "pc6", "pc6n", "pc6r", "pc7", "pc7s", "pc8", "pc9", "pc10", "unlimited" }; @@ -3053,7 +4977,7 @@ 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) { @@ -3097,7 +5021,7 @@ 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); @@ -3115,7 +5039,7 @@ 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); @@ -3321,7 +5245,7 @@ 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); @@ -3393,7 +5317,7 @@ 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) { @@ -3443,12 +5367,112 @@ 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_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_fd_cstate(void) +{ + if (!ccstate_counter_info) + return; + + const int counter_info_num = ccstate_counter_info_size; + + for (int counter_id = 0; counter_id < counter_info_num; ++counter_id) { + if (ccstate_counter_info[counter_id].fd_perf_core != -1) + close(ccstate_counter_info[counter_id].fd_perf_core); + + if (ccstate_counter_info[counter_id].fd_perf_pkg != -1) + close(ccstate_counter_info[counter_id].fd_perf_pkg); + } + + free(ccstate_counter_info); + ccstate_counter_info = NULL; + ccstate_counter_info_size = 0; +} + +void free_fd_msr(void) +{ + if (!msr_counter_info) + return; + + for (int cpu = 0; cpu < topo.max_cpu_num; ++cpu) { + if (msr_counter_info[cpu].fd_perf != -1) + close(msr_counter_info[cpu].fd_perf); + } + + free(msr_counter_info); + msr_counter_info = NULL; + msr_counter_info_size = 0; +} + +void free_fd_rapl_percpu(void) +{ + if (!rapl_counter_info_perdomain) + return; + + const int num_domains = rapl_counter_info_perdomain_size; + + 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); + rapl_counter_info_perdomain = NULL; + rapl_counter_info_perdomain_size = 0; +} + +void free_fd_added_perf_counters_(struct perf_counter_info *pp) +{ + if (!pp) + return; + + if (!pp->fd_perf_per_domain) + return; + + while (pp) { + for (size_t domain = 0; domain < pp->num_domains; ++domain) { + if (pp->fd_perf_per_domain[domain] != -1) { + close(pp->fd_perf_per_domain[domain]); + pp->fd_perf_per_domain[domain] = -1; + } + } + + free(pp->fd_perf_per_domain); + pp->fd_perf_per_domain = NULL; + + pp = pp->next; + } +} + +void free_fd_added_perf_counters(void) +{ + free_fd_added_perf_counters_(sys.perf_tp); + free_fd_added_perf_counters_(sys.perf_cp); + free_fd_added_perf_counters_(sys.perf_pp); } void free_all_buffers(void) @@ -3492,9 +5516,15 @@ void free_all_buffers(void) outp = NULL; free_fd_percpu(); + free_fd_instr_count_percpu(); + free_fd_msr(); + free_fd_rapl_percpu(); + free_fd_cstate(); + free_fd_added_perf_counters(); free(irq_column_2_cpu); free(irqs_per_cpu); + free(nmi_per_cpu); for (i = 0; i <= topo.max_cpu_num; ++i) { if (cpus[i].put_ids) @@ -3619,6 +5649,9 @@ static int parse_cpu_str(char *cpu_str, cpu_set_t *cpu_set, int cpu_set_size) if (*next == '-') /* no negative cpu numbers */ return 1; + if (*next == '\0' || *next == '\n') + break; + start = strtoul(next, &next, 10); if (start >= CPU_SUBSET_MAXCPUS) @@ -3723,6 +5756,8 @@ int for_all_cpus_2(int (func) (struct thread_data *, struct core_data *, { int retval, pkg_no, node_no, core_no, thread_no; + retval = 0; + for (pkg_no = 0; pkg_no < topo.num_packages; ++pkg_no) { for (node_no = 0; node_no < topo.nodes_per_pkg; ++node_no) { for (core_no = 0; core_no < topo.cores_per_node; ++core_no) { @@ -3744,14 +5779,12 @@ int for_all_cpus_2(int (func) (struct thread_data *, struct core_data *, p = GET_PKG(pkg_base, pkg_no); p2 = GET_PKG(pkg_base2, pkg_no); - retval = func(t, c, p, t2, c2, p2); - if (retval) - return retval; + retval |= func(t, c, p, t2, c2, p2); } } } } - return 0; + return retval; } /* @@ -3825,11 +5858,25 @@ 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 msr_perf_init(void); +void rapl_perf_init(void); +void cstate_perf_init(void); +void added_perf_counters_init(void); +void pmt_init(void); + 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); + linux_perf_init(); + msr_perf_init(); + rapl_perf_init(); + cstate_perf_init(); + added_perf_counters_init(); + pmt_init(); + 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) @@ -3876,6 +5923,32 @@ int init_thread_id(int cpu) return 0; } +int set_my_cpu_type(void) +{ + unsigned int eax, ebx, ecx, edx; + unsigned int max_level; + + __cpuid(0, max_level, ebx, ecx, edx); + + if (max_level < CPUID_LEAF_MODEL_ID) + return 0; + + __cpuid(CPUID_LEAF_MODEL_ID, eax, ebx, ecx, edx); + + return (eax >> CPUID_LEAF_MODEL_ID_CORE_TYPE_SHIFT); +} + +int set_cpu_hybrid_type(int cpu) +{ + if (cpu_migrate(cpu)) + return -1; + + int type = set_my_cpu_type(); + + cpus[cpu].type = type; + return 0; +} + /* * snapshot_proc_interrupts() * @@ -3908,31 +5981,37 @@ int snapshot_proc_interrupts(void) irq_column_2_cpu[column] = cpu_number; irqs_per_cpu[cpu_number] = 0; + nmi_per_cpu[cpu_number] = 0; } /* read /proc/interrupt count lines and sum up irqs per cpu */ while (1) { int column; char buf[64]; + int this_row_is_nmi = 0; - retval = fscanf(fp, " %s:", buf); /* flush irq# "N:" */ + retval = fscanf(fp, " %s:", buf); /* irq# "N:" */ if (retval != 1) break; + if (strncmp(buf, "NMI", strlen("NMI")) == 0) + this_row_is_nmi = 1; + /* read the count per cpu */ for (column = 0; column < topo.num_cpus; ++column) { int cpu_number, irq_count; retval = fscanf(fp, " %d", &irq_count); + if (retval != 1) break; cpu_number = irq_column_2_cpu[column]; irqs_per_cpu[cpu_number] += irq_count; - + if (this_row_is_nmi) + nmi_per_cpu[cpu_number] += irq_count; } - while (getc(fp) != '\n') ; /* flush interrupt description */ } @@ -3940,85 +6019,36 @@ 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("/sys/class/drm/card0/power/rc6_residency_ms", "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("/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"); - } else { - rewind(fp); - fflush(fp); - } - - retval = fscanf(fp, "%d", &gfx_cur_mhz); - if (retval != 1) - err(1, "GFX MHz"); - - return 0; -} + rewind(gfx_info[idx].fp); -/* - * 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("/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"); - } else { - rewind(fp); - fflush(fp); + switch (idx) { + case GFX_rc6: + case SAM_mc6: + retval = fscanf(gfx_info[idx].fp, "%lld", &gfx_info[idx].val_ull); + if (retval != 1) + err(1, "rc6"); + return 0; + case GFX_MHz: + case GFX_ACTMHz: + case SAM_MHz: + case SAM_ACTMHz: + 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_act_mhz); - if (retval != 1) - err(1, "GFX ACT MHz"); - - return 0; } /* @@ -4078,18 +6108,29 @@ int snapshot_sys_lpi_us(void) */ int snapshot_proc_sysfs_files(void) { - if (DO_BIC(BIC_IRQ)) + gettimeofday(&procsysfs_tv_begin, (struct timezone *)NULL); + + if (DO_BIC(BIC_IRQ) || DO_BIC(BIC_NMI)) if (snapshot_proc_interrupts()) 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_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(); @@ -4173,6 +6214,8 @@ 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; @@ -4201,6 +6244,8 @@ 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; @@ -4280,7 +6325,8 @@ 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) { @@ -4290,16 +6336,16 @@ 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; } @@ -4314,6 +6360,9 @@ 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); @@ -4361,6 +6410,7 @@ restart: re_initialize(); goto restart; } + delta_platform(&platform_counters_odd, &platform_counters_even); compute_average(EVEN_COUNTERS); format_all_counters(EVEN_COUNTERS); flush_output_stdout(); @@ -4384,6 +6434,7 @@ restart: re_initialize(); goto restart; } + delta_platform(&platform_counters_even, &platform_counters_odd); compute_average(ODD_COUNTERS); format_all_counters(ODD_COUNTERS); flush_output_stdout(); @@ -4399,10 +6450,13 @@ 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; } /* @@ -4414,47 +6468,51 @@ 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"); + return 1; - if (cap_get_flag(caps, CAP_SYS_RAWIO, CAP_EFFECTIVE, &cap_flag_value)) - err(-6, "cap_get\n"); + 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) { - warnx("capget(CAP_SYS_RAWIO) failed," " try \"# setcap cap_sys_rawio=ep %s\"", progname); - return 1; + 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"); + 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) @@ -4462,7 +6520,7 @@ 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) @@ -4502,7 +6560,7 @@ 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) @@ -4564,64 +6622,230 @@ static void dump_sysfs_file(char *path) fprintf(outf, "%s: %s", strrchr(path, '/') + 1, cpuidle_buf); } -static void probe_intel_uncore_frequency(void) +static void probe_intel_uncore_frequency_legacy(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; + for (i = 0; i < topo.num_packages; ++i) { + for (j = 0; j <= topo.max_die_id; ++j) { + int k, l; + char path_base[128]; - /* Cluster level sysfs not supported yet. */ - if (!access("/sys/devices/system/cpu/intel_uncore_frequency/uncore00", R_OK)) - return; + sprintf(path_base, "/sys/devices/system/cpu/intel_uncore_frequency/package_%02d_die_%02d", i, + j); - if (!access("/sys/devices/system/cpu/intel_uncore_frequency/package_00_die_00/current_freq_khz", R_OK)) - BIC_PRESENT(BIC_UNCORE_MHZ); + if (access(path_base, R_OK)) + continue; - if (quiet) - return; + BIC_PRESENT(BIC_UNCORE_MHZ); - for (i = 0; i < topo.num_packages; ++i) { - for (j = 0; j < topo.num_die; ++j) { - int k, l; + if (quiet) + return; - sprintf(path, "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/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/max_freq_khz", - i, j); + sprintf(path, "%s/max_freq_khz", path_base); l = read_sysfs_int(path); - fprintf(outf, "Uncore Frequency pkg%d die%d: %d - %d MHz ", i, j, k / 1000, l / 1000); + fprintf(outf, "Uncore Frequency package%d die%d: %d - %d MHz ", i, j, k / 1000, l / 1000); - sprintf(path, - "/sys/devices/system/cpu/intel_uncore_frequency/package_0%d_die_0%d/initial_min_freq_khz", - i, j); + sprintf(path, "%s/initial_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/initial_max_freq_khz", path_base); l = read_sysfs_int(path); - fprintf(outf, "(%d - %d MHz)\n", k / 1000, l / 1000); + 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_intel_uncore_frequency_cluster(void) +{ + int i, uncore_max_id; + char path[256]; + char path_base[128]; + + if (access("/sys/devices/system/cpu/intel_uncore_frequency/uncore00/current_freq_khz", R_OK)) + return; + + for (uncore_max_id = 0;; ++uncore_max_id) { + + sprintf(path_base, "/sys/devices/system/cpu/intel_uncore_frequency/uncore%02d", uncore_max_id); + + /* uncore## start at 00 and skips no numbers, so stop upon first missing */ + if (access(path_base, R_OK)) { + uncore_max_id -= 1; + break; + } + } + for (i = uncore_max_id; i >= 0; --i) { + int k, l; + int package_id, domain_id, cluster_id; + char name_buf[16]; + + sprintf(path_base, "/sys/devices/system/cpu/intel_uncore_frequency/uncore%02d", i); + + if (access(path_base, R_OK)) + err(1, "%s: %s\n", __func__, path_base); + + 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/current_freq_khz", path_base); + sprintf(name_buf, "UMHz%d.%d", domain_id, cluster_id); + + add_counter(0, path, name_buf, 0, SCOPE_PACKAGE, COUNTER_K2M, FORMAT_AVERAGE, 0, package_id); + + if (quiet) + continue; + + 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_intel_uncore_frequency(void) +{ + if (!genuine_intel) + return; + + if (access("/sys/devices/system/cpu/intel_uncore_frequency/uncore00", R_OK) == 0) + probe_intel_uncore_frequency_cluster(); + else + probe_intel_uncore_frequency_legacy(); +} + +static void set_graphics_fp(char *path, int idx) +{ + if (!access(path, R_OK)) + gfx_info[idx].fp = fopen_or_die(path, "r"); +} + +/* Enlarge this if there are /sys/class/drm/card2 ... */ +#define GFX_MAX_CARDS 2 + static void probe_graphics(void) { - if (!access("/sys/class/drm/card0/power/rc6_residency_ms", R_OK)) - BIC_PRESENT(BIC_GFX_rc6); + char path[PATH_MAX]; + int i; - 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)) - BIC_PRESENT(BIC_GFXMHz); + /* 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; + + 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; + + set_graphics_fp("/sys/class/drm/card0/device/tile0/gt0/gtidle/idle_residency_ms", gt0_is_gt ? GFX_rc6 : SAM_mc6); + + set_graphics_fp("/sys/class/drm/card0/device/tile0/gt0/freq0/cur_freq", gt0_is_gt ? GFX_MHz : SAM_MHz); + + set_graphics_fp("/sys/class/drm/card0/device/tile0/gt0/freq0/act_freq", gt0_is_gt ? GFX_ACTMHz : SAM_ACTMHz); + + set_graphics_fp("/sys/class/drm/card0/device/tile0/gt1/gtidle/idle_residency_ms", gt0_is_gt ? SAM_mc6 : GFX_rc6); + + set_graphics_fp("/sys/class/drm/card0/device/tile0/gt1/freq0/cur_freq", gt0_is_gt ? SAM_MHz : GFX_MHz); + + set_graphics_fp("/sys/class/drm/card0/device/tile0/gt1/freq0/act_freq", gt0_is_gt ? SAM_ACTMHz : GFX_ACTMHz); + + goto end; + } + +next: + /* New i915 graphics sysfs knobs */ + for (i = 0; i < GFX_MAX_CARDS; i++) { + snprintf(path, PATH_MAX, "/sys/class/drm/card%d/gt/gt0/rc6_residency_ms", i); + if (!access(path, R_OK)) + break; + } + + if (i == GFX_MAX_CARDS) + goto legacy_i915; + + snprintf(path, PATH_MAX, "/sys/class/drm/card%d/gt/gt0/rc6_residency_ms", i); + set_graphics_fp(path, GFX_rc6); + + snprintf(path, PATH_MAX, "/sys/class/drm/card%d/gt/gt0/rps_cur_freq_mhz", i); + set_graphics_fp(path, GFX_MHz); + + snprintf(path, PATH_MAX, "/sys/class/drm/card%d/gt/gt0/rps_act_freq_mhz", i); + set_graphics_fp(path, GFX_ACTMHz); + + snprintf(path, PATH_MAX, "/sys/class/drm/card%d/gt/gt1/rc6_residency_ms", i); + set_graphics_fp(path, SAM_mc6); + + snprintf(path, PATH_MAX, "/sys/class/drm/card%d/gt/gt1/rps_cur_freq_mhz", i); + set_graphics_fp(path, SAM_MHz); + + snprintf(path, PATH_MAX, "/sys/class/drm/card%d/gt/gt1/rps_act_freq_mhz", i); + set_graphics_fp(path, SAM_ACTMHz); + + goto end; + +legacy_i915: + /* Fall back to traditional i915 graphics sysfs knobs */ + set_graphics_fp("/sys/class/drm/card0/power/rc6_residency_ms", GFX_rc6); + + set_graphics_fp("/sys/class/drm/card0/gt_cur_freq_mhz", GFX_MHz); + if (!gfx_info[GFX_MHz].fp) + set_graphics_fp("/sys/class/graphics/fb0/device/drm/card0/gt_cur_freq_mhz", GFX_MHz); - 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)) + set_graphics_fp("/sys/class/drm/card0/gt_act_freq_mhz", GFX_ACTMHz); + if (!gfx_info[GFX_ACTMHz].fp) + set_graphics_fp("/sys/class/graphics/fb0/device/drm/card0/gt_act_freq_mhz", GFX_ACTMHz); + +end: + if (gfx_info[GFX_rc6].fp) + BIC_PRESENT(BIC_GFX_rc6); + if (gfx_info[GFX_MHz].fp) + BIC_PRESENT(BIC_GFXMHz); + if (gfx_info[GFX_ACTMHz].fp) BIC_PRESENT(BIC_GFXACTMHz); + if (gfx_info[SAM_mc6].fp) + BIC_PRESENT(BIC_SAM_mc6); + if (gfx_info[SAM_MHz].fp) + BIC_PRESENT(BIC_SAMMHz); + if (gfx_info[SAM_ACTMHz].fp) + BIC_PRESENT(BIC_SAMACTMHz); } static void dump_sysfs_cstate_config(void) @@ -4783,6 +7007,9 @@ 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; @@ -4869,6 +7096,9 @@ 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 */ @@ -4983,31 +7213,18 @@ void rapl_probe_intel(void) unsigned long long msr; unsigned int time_unit; double tdp; + const unsigned long long bic_watt_bits = BIC_SysWatt | BIC_PkgWatt | BIC_CorWatt | BIC_RAMWatt | BIC_GFXWatt; + const unsigned long long bic_joules_bits = BIC_Sys_J | 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)) @@ -5024,6 +7241,11 @@ void rapl_probe_intel(void) else rapl_dram_energy_units = rapl_energy_units; + if (platform->has_fixed_rapl_psys_unit) + rapl_psys_energy_units = 1.0; + else + rapl_psys_energy_units = rapl_energy_units; + time_unit = msr >> 16 & 0xF; if (time_unit == 0) time_unit = 0xA; @@ -5041,14 +7263,13 @@ 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; @@ -5202,7 +7423,7 @@ 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) @@ -5258,7 +7479,7 @@ 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)) @@ -5305,6 +7526,9 @@ 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; @@ -5402,6 +7626,9 @@ 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" : ""); @@ -5411,6 +7638,9 @@ void decode_misc_enable_msr(void) { unsigned long long msr; + if (no_msr) + return; + if (!genuine_intel) return; @@ -5428,6 +7658,9 @@ void decode_misc_feature_control(void) { unsigned long long msr; + if (no_msr) + return; + if (!platform->has_msr_misc_feature_control) return; @@ -5449,6 +7682,9 @@ void decode_misc_pwr_mgmt_msr(void) { unsigned long long msr; + if (no_msr) + return; + if (!platform->has_msr_misc_pwr_mgmt) return; @@ -5468,6 +7704,9 @@ void decode_c6_demotion_policy_msr(void) { unsigned long long msr; + if (no_msr) + return; + if (!platform->has_msr_c6_demotion_policy_config) return; @@ -5489,7 +7728,8 @@ 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; } @@ -5504,66 +7744,455 @@ 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; +} + +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_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 int rapl_type = read_perf_type(cai->perf_subsys); + const unsigned int rapl_energy_pkg_config = read_perf_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 >= 2) + fprintf(stderr, "%s: %d (cpu: %d)\n", __func__, ret, cpu); + + return ret; +} + /* * Linux-perf manages 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"); + 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"); + } +} + +void rapl_perf_init(void) +{ + const unsigned int num_domains = (platform->has_per_core_rapl ? topo.max_core_id : topo.max_package_id) + 1; + 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"); + rapl_counter_info_perdomain_size = num_domains; - BIC_PRESENT(BIC_IPC); + /* + * Initialize rapl_counter_info_percpu + */ + for (unsigned 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] = COUNTER_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; + unsigned 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; + + assert(next_domain < num_domains); + + if (domain_visited[next_domain]) + continue; + + domain_visited[next_domain] = 1; + + if ((cai->flags & RAPL_COUNTER_FLAG_PLATFORM_COUNTER) && (cpu != base_cpu)) + continue; + + 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] = COUNTER_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] = COUNTER_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] != COUNTER_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); } -void probe_cstates(void) +/* Assumes msr_counter_info is populated */ +static int has_amperf_access(void) { - probe_cst_limit(); + return msr_counter_arch_infos[MSR_ARCH_INFO_APERF_INDEX].present && + msr_counter_arch_infos[MSR_ARCH_INFO_MPERF_INDEX].present; +} - if (platform->supported_cstates & CC1) - BIC_PRESENT(BIC_CPU_c1); +int *get_cstate_perf_group_fd(struct cstate_counter_info_t *cci, const char *group_name) +{ + if (strcmp(group_name, "cstate_core") == 0) + return &cci->fd_perf_core; + + if (strcmp(group_name, "cstate_pkg") == 0) + return &cci->fd_perf_pkg; + + return NULL; +} + +int add_cstate_perf_counter_(int cpu, struct cstate_counter_info_t *cci, const struct cstate_counter_arch_info *cai) +{ + if (no_perf) + return -1; + + int *pfd_group = get_cstate_perf_group_fd(cci, cai->perf_subsys); + + if (pfd_group == NULL) + return -1; + + const unsigned int type = read_perf_type(cai->perf_subsys); + const unsigned int config = read_perf_config(cai->perf_subsys, cai->perf_name); + + const int fd_counter = open_perf_counter(cpu, type, config, *pfd_group, PERF_FORMAT_GROUP); + + if (fd_counter == -1) + return -1; + + /* If it's the first counter opened, make it a group descriptor */ + if (*pfd_group == -1) + *pfd_group = fd_counter; + + return fd_counter; +} + +int add_cstate_perf_counter(int cpu, struct cstate_counter_info_t *cci, const struct cstate_counter_arch_info *cai) +{ + int ret = add_cstate_perf_counter_(cpu, cci, cai); + + if (debug >= 2) + fprintf(stderr, "%s: %d (cpu: %d)\n", __func__, ret, cpu); + + return ret; +} + +int add_msr_perf_counter_(int cpu, struct msr_counter_info_t *cci, const struct msr_counter_arch_info *cai) +{ + if (no_perf) + return -1; + + const unsigned int type = read_perf_type(cai->perf_subsys); + const unsigned int config = read_perf_config(cai->perf_subsys, cai->perf_name); + + const int fd_counter = open_perf_counter(cpu, type, config, cci->fd_perf, PERF_FORMAT_GROUP); + + if (fd_counter == -1) + return -1; + + /* If it's the first counter opened, make it a group descriptor */ + if (cci->fd_perf == -1) + cci->fd_perf = fd_counter; + + return fd_counter; +} + +int add_msr_perf_counter(int cpu, struct msr_counter_info_t *cci, const struct msr_counter_arch_info *cai) +{ + int ret = add_msr_perf_counter_(cpu, cci, cai); + + if (debug) + fprintf(stderr, "%s: %s/%s: %d (cpu: %d)\n", __func__, cai->perf_subsys, cai->perf_name, ret, cpu); + + return ret; +} + +void msr_perf_init_(void) +{ + const int mci_num = topo.max_cpu_num + 1; + + msr_counter_info = calloc(mci_num, sizeof(*msr_counter_info)); + if (!msr_counter_info) + err(1, "calloc msr_counter_info"); + msr_counter_info_size = mci_num; + + for (int cpu = 0; cpu < mci_num; ++cpu) + msr_counter_info[cpu].fd_perf = -1; + + for (int cidx = 0; cidx < NUM_MSR_COUNTERS; ++cidx) { + + struct msr_counter_arch_info *cai = &msr_counter_arch_infos[cidx]; + + cai->present = false; + + for (int cpu = 0; cpu < mci_num; ++cpu) { + + struct msr_counter_info_t *const cci = &msr_counter_info[cpu]; + + if (cpu_is_not_allowed(cpu)) + continue; + + if (cai->needed) { + /* Use perf API for this counter */ + if (!no_perf && cai->perf_name && add_msr_perf_counter(cpu, cci, cai) != -1) { + cci->source[cai->rci_index] = COUNTER_SOURCE_PERF; + cai->present = true; + + /* User MSR for this counter */ + } else if (!no_msr && cai->msr && probe_msr(cpu, cai->msr) == 0) { + cci->source[cai->rci_index] = COUNTER_SOURCE_MSR; + cci->msr[cai->rci_index] = cai->msr; + cci->msr_mask[cai->rci_index] = cai->msr_mask; + cai->present = true; + } + } + } + } +} + +/* Initialize data for reading perf counters from the MSR group. */ +void msr_perf_init(void) +{ + bool need_amperf = false, need_smi = false; + const bool need_soft_c1 = (!platform->has_msr_core_c1_res) && (platform->supported_cstates & CC1); + + need_amperf = BIC_IS_ENABLED(BIC_Avg_MHz) || BIC_IS_ENABLED(BIC_Busy) || BIC_IS_ENABLED(BIC_Bzy_MHz) + || BIC_IS_ENABLED(BIC_IPC) || need_soft_c1; + + if (BIC_IS_ENABLED(BIC_SMI)) + need_smi = true; + + /* Enable needed counters */ + msr_counter_arch_infos[MSR_ARCH_INFO_APERF_INDEX].needed = need_amperf; + msr_counter_arch_infos[MSR_ARCH_INFO_MPERF_INDEX].needed = need_amperf; + msr_counter_arch_infos[MSR_ARCH_INFO_SMI_INDEX].needed = need_smi; + + msr_perf_init_(); + + const bool has_amperf = has_amperf_access(); + const bool has_smi = msr_counter_arch_infos[MSR_ARCH_INFO_SMI_INDEX].present; + + has_aperf_access = has_amperf; + + if (has_amperf) { + BIC_PRESENT(BIC_Avg_MHz); + BIC_PRESENT(BIC_Busy); + BIC_PRESENT(BIC_Bzy_MHz); + BIC_PRESENT(BIC_SMI); + } + + if (has_smi) + BIC_PRESENT(BIC_SMI); +} + +void cstate_perf_init_(bool soft_c1) +{ + bool has_counter; + bool *cores_visited = NULL, *pkg_visited = NULL; + const int cores_visited_elems = topo.max_core_id + 1; + const int pkg_visited_elems = topo.max_package_id + 1; + const int cci_num = topo.max_cpu_num + 1; + + ccstate_counter_info = calloc(cci_num, sizeof(*ccstate_counter_info)); + if (!ccstate_counter_info) + err(1, "calloc ccstate_counter_arch_info"); + ccstate_counter_info_size = cci_num; + + cores_visited = calloc(cores_visited_elems, sizeof(*cores_visited)); + if (!cores_visited) + err(1, "calloc cores_visited"); + + pkg_visited = calloc(pkg_visited_elems, sizeof(*pkg_visited)); + if (!pkg_visited) + err(1, "calloc pkg_visited"); + + /* Initialize cstate_counter_info_percpu */ + for (int cpu = 0; cpu < cci_num; ++cpu) { + ccstate_counter_info[cpu].fd_perf_core = -1; + ccstate_counter_info[cpu].fd_perf_pkg = -1; + } + + for (int cidx = 0; cidx < NUM_CSTATE_COUNTERS; ++cidx) { + has_counter = false; + memset(cores_visited, 0, cores_visited_elems * sizeof(*cores_visited)); + memset(pkg_visited, 0, pkg_visited_elems * sizeof(*pkg_visited)); + + const struct cstate_counter_arch_info *cai = &ccstate_counter_arch_infos[cidx]; + + for (int cpu = 0; cpu < cci_num; ++cpu) { + + struct cstate_counter_info_t *const cci = &ccstate_counter_info[cpu]; + + if (cpu_is_not_allowed(cpu)) + continue; + + const int core_id = cpus[cpu].physical_core_id; + const int pkg_id = cpus[cpu].physical_package_id; + + assert(core_id < cores_visited_elems); + assert(pkg_id < pkg_visited_elems); + + const bool per_thread = cai->flags & CSTATE_COUNTER_FLAG_COLLECT_PER_THREAD; + const bool per_core = cai->flags & CSTATE_COUNTER_FLAG_COLLECT_PER_CORE; - if (platform->supported_cstates & CC3) - BIC_PRESENT(BIC_CPU_c3); + if (!per_thread && cores_visited[core_id]) + continue; - if (platform->supported_cstates & CC6) - BIC_PRESENT(BIC_CPU_c6); + if (!per_core && pkg_visited[pkg_id]) + continue; - if (platform->supported_cstates & CC7) - BIC_PRESENT(BIC_CPU_c7); + const bool counter_needed = BIC_IS_ENABLED(cai->bic) || + (soft_c1 && (cai->flags & CSTATE_COUNTER_FLAG_SOFT_C1_DEPENDENCY)); + const bool counter_supported = (platform->supported_cstates & cai->feature_mask); - if (platform->supported_cstates & PC2 && (pkg_cstate_limit >= PCL__2)) - BIC_PRESENT(BIC_Pkgpc2); + if (counter_needed && counter_supported) { + /* Use perf API for this counter */ + if (!no_perf && cai->perf_name && add_cstate_perf_counter(cpu, cci, cai) != -1) { - if (platform->supported_cstates & PC3 && (pkg_cstate_limit >= PCL__3)) - BIC_PRESENT(BIC_Pkgpc3); + cci->source[cai->rci_index] = COUNTER_SOURCE_PERF; - if (platform->supported_cstates & PC6 && (pkg_cstate_limit >= PCL__6)) - BIC_PRESENT(BIC_Pkgpc6); + /* User MSR for this counter */ + } else if (!no_msr && cai->msr && pkg_cstate_limit >= cai->pkg_cstate_limit + && probe_msr(cpu, cai->msr) == 0) { + cci->source[cai->rci_index] = COUNTER_SOURCE_MSR; + cci->msr[cai->rci_index] = cai->msr; + } + } - if (platform->supported_cstates & PC7 && (pkg_cstate_limit >= PCL__7)) - BIC_PRESENT(BIC_Pkgpc7); + if (cci->source[cai->rci_index] != COUNTER_SOURCE_NONE) { + has_counter = true; + cores_visited[core_id] = true; + pkg_visited[pkg_id] = true; + } + } - if (platform->supported_cstates & PC8 && (pkg_cstate_limit >= PCL__8)) - BIC_PRESENT(BIC_Pkgpc8); + /* If any CPU has access to the counter, make it present */ + if (has_counter) + BIC_PRESENT(cai->bic); + } - if (platform->supported_cstates & PC9 && (pkg_cstate_limit >= PCL__9)) - BIC_PRESENT(BIC_Pkgpc9); + free(cores_visited); + free(pkg_visited); +} - if (platform->supported_cstates & PC10 && (pkg_cstate_limit >= PCL_10)) - BIC_PRESENT(BIC_Pkgpc10); +void cstate_perf_init(void) +{ + /* + * If we don't have a C1 residency MSR, we calculate it "in software", + * but we need APERF, MPERF too. + */ + const bool soft_c1 = !platform->has_msr_core_c1_res && has_amperf_access() + && platform->supported_cstates & CC1; + + if (soft_c1) + BIC_PRESENT(BIC_CPU_c1); + + cstate_perf_init_(soft_c1); +} + +void probe_cstates(void) +{ + probe_cst_limit(); 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); @@ -5623,6 +8252,7 @@ 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; @@ -5650,8 +8280,12 @@ void process_cpuid() ecx_flags = ecx; edx_flags = edx; - if (get_msr(sched_getcpu(), MSR_IA32_UCODE_REV, &ucode_patch)) - warnx("get_msr(UCODE)"); + 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. @@ -5662,9 +8296,12 @@ 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" : "-", @@ -5700,11 +8337,6 @@ void process_cpuid() __cpuid(0x6, eax, ebx, ecx, edx); has_aperf = ecx & (1 << 0); - if (has_aperf) { - BIC_PRESENT(BIC_Avg_MHz); - BIC_PRESENT(BIC_Busy); - BIC_PRESENT(BIC_Bzy_MHz); - } do_dts = eax & (1 << 0); if (do_dts) BIC_PRESENT(BIC_CoreTmp); @@ -5786,6 +8418,15 @@ 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); @@ -5795,9 +8436,31 @@ void process_cpuid() aperf_mperf_multiplier = platform->need_perf_multiplier ? 1024 : 1; BIC_PRESENT(BIC_IRQ); + BIC_PRESENT(BIC_NMI); BIC_PRESENT(BIC_TSC_MHz); } +static void counter_info_init(void) +{ + for (int i = 0; i < NUM_CSTATE_COUNTERS; ++i) { + struct cstate_counter_arch_info *const cai = &ccstate_counter_arch_infos[i]; + + if (platform->has_msr_knl_core_c6_residency && cai->msr == MSR_CORE_C6_RESIDENCY) + cai->msr = MSR_KNL_CORE_C6_RESIDENCY; + + if (!platform->has_msr_core_c1_res && cai->msr == MSR_CORE_C1_RES) + cai->msr = 0; + + if (platform->has_msr_atom_pkg_c6_residency && cai->msr == MSR_PKG_C6_RESIDENCY) + cai->msr = MSR_ATOM_PKG_C6_RESIDENCY; + } + + for (int i = 0; i < NUM_MSR_COUNTERS; ++i) { + msr_counter_arch_infos[i].present = false; + msr_counter_arch_infos[i].needed = false; + } +} + void probe_pm_features(void) { probe_pstates(); @@ -5814,7 +8477,7 @@ 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) @@ -5833,19 +8496,45 @@ int dir_filter(const struct dirent *dirp) return 0; } +char *possible_file = "/sys/devices/system/cpu/possible"; +char possible_buf[1024]; + +int initialize_cpu_possible_set(void) +{ + FILE *fp; + + fp = fopen(possible_file, "r"); + if (!fp) { + warn("open %s", possible_file); + return -1; + } + if (fread(possible_buf, sizeof(char), 1024, fp) == 0) { + warn("read %s", possible_file); + goto err; + } + if (parse_cpu_str(possible_buf, cpu_possible_set, cpu_possible_setsize)) { + warnx("%s: cpu str malformat %s\n", possible_file, cpu_effective_str); + goto err; + } + return 0; + +err: + fclose(fp); + return -1; +} + void topology_probe(bool startup) { int i; int max_core_id = 0; int max_package_id = 0; - int max_die_id = 0; int max_siblings = 0; /* Initialize num_cpus, max_cpu_num */ set_max_cpu_num(); topo.num_cpus = 0; for_all_proc_cpus(count_cpus); - if (!summary_only && topo.num_cpus > 1) + if (!summary_only) BIC_PRESENT(BIC_CPU); if (debug > 1) @@ -5866,6 +8555,16 @@ void topology_probe(bool startup) for_all_proc_cpus(mark_cpu_present); /* + * Allocate and initialize cpu_possible_set + */ + cpu_possible_set = CPU_ALLOC((topo.max_cpu_num + 1)); + if (cpu_possible_set == NULL) + err(3, "CPU_ALLOC"); + cpu_possible_setsize = CPU_ALLOC_SIZE((topo.max_cpu_num + 1)); + CPU_ZERO_S(cpu_possible_setsize, cpu_possible_set); + initialize_cpu_possible_set(); + + /* * Allocate and initialize cpu_effective_set */ cpu_effective_set = CPU_ALLOC((topo.max_cpu_num + 1)); @@ -5933,6 +8632,8 @@ void topology_probe(bool startup) for_all_proc_cpus(init_thread_id); + for_all_proc_cpus(set_cpu_hybrid_type); + /* * For online cpus * find max_core_id, max_package_id @@ -5955,8 +8656,8 @@ void topology_probe(bool startup) /* get die information */ cpus[i].die_id = get_die_id(i); - if (cpus[i].die_id > max_die_id) - max_die_id = cpus[i].die_id; + if (cpus[i].die_id > topo.max_die_id) + topo.max_die_id = cpus[i].die_id; /* get numa node information */ cpus[i].physical_node_id = get_physical_node_id(&cpus[i]); @@ -5975,16 +8676,18 @@ void topology_probe(bool startup) if (cpus[i].thread_id == 0) topo.num_cores++; } + topo.max_core_id = max_core_id; + topo.max_package_id = max_package_id; topo.cores_per_node = max_core_id + 1; if (debug > 1) fprintf(outf, "max_core_id %d, sizing for %d cores per package\n", max_core_id, topo.cores_per_node); - if (!summary_only && topo.cores_per_node > 1) + if (!summary_only) BIC_PRESENT(BIC_Core); - topo.num_die = max_die_id + 1; + topo.num_die = topo.max_die_id + 1; if (debug > 1) - fprintf(outf, "max_die_id %d, sizing for %d die\n", max_die_id, topo.num_die); + fprintf(outf, "max_die_id %d, sizing for %d die\n", topo.max_die_id, topo.num_die); if (!summary_only && topo.num_die > 1) BIC_PRESENT(BIC_Die); @@ -6121,7 +8824,11 @@ void allocate_irq_buffers(void) irqs_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int)); if (irqs_per_cpu == NULL) - err(-1, "calloc %d", topo.max_cpu_num + 1); + err(-1, "calloc %d IRQ", topo.max_cpu_num + 1); + + nmi_per_cpu = calloc(topo.max_cpu_num + 1, sizeof(int)); + if (nmi_per_cpu == NULL) + err(-1, "calloc %d NMI", topo.max_cpu_num + 1); } int update_topo(struct thread_data *t, struct core_data *c, struct pkg_data *p) @@ -6142,6 +8849,7 @@ void topology_update(void) topo.allowed_packages = 0; for_all_cpus(update_topo, ODD_COUNTERS); } + void setup_all_buffers(bool startup) { topology_probe(startup); @@ -6169,21 +8877,613 @@ void set_base_cpu(void) err(-ENODEV, "No valid cpus found"); } +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; +} + +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 || !BIC_IS_ENABLED(BIC_IPC) || !has_instr_count_access()) + bic_enabled &= ~BIC_IPC; +} + +bool perf_has_hybrid_devices(void) +{ + /* + * 0: unknown + * 1: has separate perf device for p and e core + * -1: doesn't have separate perf device for p and e core + */ + static int cached; + + if (cached > 0) + return true; + + if (cached < 0) + return false; + + if (access("/sys/bus/event_source/devices/cpu_core", F_OK)) { + cached = -1; + return false; + } + + if (access("/sys/bus/event_source/devices/cpu_atom", F_OK)) { + cached = -1; + return false; + } + + cached = 1; + return true; +} + +int added_perf_counters_init_(struct perf_counter_info *pinfo) +{ + size_t num_domains = 0; + unsigned int next_domain; + bool *domain_visited; + unsigned int perf_type, perf_config; + double perf_scale; + int fd_perf; + + if (!pinfo) + return 0; + + const size_t max_num_domains = MAX(topo.max_cpu_num + 1, MAX(topo.max_core_id + 1, topo.max_package_id + 1)); + + domain_visited = calloc(max_num_domains, sizeof(*domain_visited)); + + while (pinfo) { + switch (pinfo->scope) { + case SCOPE_CPU: + num_domains = topo.max_cpu_num + 1; + break; + + case SCOPE_CORE: + num_domains = topo.max_core_id + 1; + break; + + case SCOPE_PACKAGE: + num_domains = topo.max_package_id + 1; + break; + } + + /* Allocate buffer for file descriptor for each domain. */ + pinfo->fd_perf_per_domain = calloc(num_domains, sizeof(*pinfo->fd_perf_per_domain)); + if (!pinfo->fd_perf_per_domain) + errx(1, "%s: alloc %s", __func__, "fd_perf_per_domain"); + + for (size_t i = 0; i < num_domains; ++i) + pinfo->fd_perf_per_domain[i] = -1; + + pinfo->num_domains = num_domains; + pinfo->scale = 1.0; + + memset(domain_visited, 0, max_num_domains * sizeof(*domain_visited)); + + for (int cpu = 0; cpu < topo.max_cpu_num + 1; ++cpu) { + + next_domain = cpu_to_domain(pinfo, cpu); + + assert(next_domain < num_domains); + + if (cpu_is_not_allowed(cpu)) + continue; + + if (domain_visited[next_domain]) + continue; + + /* + * Intel hybrid platforms expose different perf devices for P and E cores. + * Instead of one, "/sys/bus/event_source/devices/cpu" device, there are + * "/sys/bus/event_source/devices/{cpu_core,cpu_atom}". + * + * This makes it more complicated to the user, because most of the counters + * are available on both and have to be handled manually, otherwise. + * + * Code below, allow user to use the old "cpu" name, which is translated accordingly. + */ + const char *perf_device = pinfo->device; + + if (strcmp(perf_device, "cpu") == 0 && perf_has_hybrid_devices()) { + switch (cpus[cpu].type) { + case INTEL_PCORE_TYPE: + perf_device = "cpu_core"; + break; + + case INTEL_ECORE_TYPE: + perf_device = "cpu_atom"; + break; + + default: /* Don't change, we will probably fail and report a problem soon. */ + break; + } + } + + perf_type = read_perf_type(perf_device); + if (perf_type == (unsigned int)-1) { + warnx("%s: perf/%s/%s: failed to read %s", + __func__, perf_device, pinfo->event, "type"); + continue; + } + + perf_config = read_perf_config(perf_device, pinfo->event); + if (perf_config == (unsigned int)-1) { + warnx("%s: perf/%s/%s: failed to read %s", + __func__, perf_device, pinfo->event, "config"); + continue; + } + + /* Scale is not required, some counters just don't have it. */ + perf_scale = read_perf_scale(perf_device, pinfo->event); + if (perf_scale == 0.0) + perf_scale = 1.0; + + fd_perf = open_perf_counter(cpu, perf_type, perf_config, -1, 0); + if (fd_perf == -1) { + warnx("%s: perf/%s/%s: failed to open counter on cpu%d", + __func__, perf_device, pinfo->event, cpu); + continue; + } + + domain_visited[next_domain] = 1; + pinfo->fd_perf_per_domain[next_domain] = fd_perf; + pinfo->scale = perf_scale; + + if (debug) + fprintf(stderr, "Add perf/%s/%s cpu%d: %d\n", + perf_device, pinfo->event, cpu, pinfo->fd_perf_per_domain[next_domain]); + } + + pinfo = pinfo->next; + } + + free(domain_visited); + + return 0; +} + +void added_perf_counters_init(void) +{ + if (added_perf_counters_init_(sys.perf_tp)) + errx(1, "%s: %s", __func__, "thread"); + + if (added_perf_counters_init_(sys.perf_cp)) + errx(1, "%s: %s", __func__, "core"); + + if (added_perf_counters_init_(sys.perf_pp)) + errx(1, "%s: %s", __func__, "package"); +} + +int parse_telem_info_file(int fd_dir, const char *info_filename, const char *format, unsigned long *output) +{ + int fd_telem_info; + FILE *file_telem_info; + unsigned long value; + + fd_telem_info = openat(fd_dir, info_filename, O_RDONLY); + if (fd_telem_info == -1) + return -1; + + file_telem_info = fdopen(fd_telem_info, "r"); + if (file_telem_info == NULL) { + close(fd_telem_info); + return -1; + } + + if (fscanf(file_telem_info, format, &value) != 1) { + fclose(file_telem_info); + return -1; + } + + fclose(file_telem_info); + + *output = value; + + return 0; +} + +struct pmt_mmio *pmt_mmio_open(unsigned int target_guid) +{ + struct pmt_diriter_t pmt_iter; + const struct dirent *entry; + struct stat st; + int fd_telem_dir, fd_pmt; + unsigned long guid, size, offset; + size_t mmap_size; + void *mmio; + struct pmt_mmio *head = NULL, *last = NULL; + struct pmt_mmio *new_pmt = NULL; + + if (stat(SYSFS_TELEM_PATH, &st) == -1) + return NULL; + + pmt_diriter_init(&pmt_iter); + entry = pmt_diriter_begin(&pmt_iter, SYSFS_TELEM_PATH); + if (!entry) { + pmt_diriter_remove(&pmt_iter); + return NULL; + } + + for ( ; entry != NULL; entry = pmt_diriter_next(&pmt_iter)) { + if (fstatat(dirfd(pmt_iter.dir), entry->d_name, &st, 0) == -1) + break; + + if (!S_ISDIR(st.st_mode)) + continue; + + fd_telem_dir = openat(dirfd(pmt_iter.dir), entry->d_name, O_RDONLY); + if (fd_telem_dir == -1) + break; + + if (parse_telem_info_file(fd_telem_dir, "guid", "%lx", &guid)) { + close(fd_telem_dir); + break; + } + + if (parse_telem_info_file(fd_telem_dir, "size", "%lu", &size)) { + close(fd_telem_dir); + break; + } + + if (guid != target_guid) { + close(fd_telem_dir); + continue; + } + + if (parse_telem_info_file(fd_telem_dir, "offset", "%lu", &offset)) { + close(fd_telem_dir); + break; + } + + assert(offset == 0); + + fd_pmt = openat(fd_telem_dir, "telem", O_RDONLY); + if (fd_pmt == -1) + goto loop_cleanup_and_break; + + mmap_size = ROUND_UP_TO_PAGE_SIZE(size); + mmio = mmap(0, mmap_size, PROT_READ, MAP_SHARED, fd_pmt, 0); + if (mmio != MAP_FAILED) { + if (debug) + fprintf(stderr, "%s: 0x%lx mmaped at: %p\n", __func__, guid, mmio); + + new_pmt = calloc(1, sizeof(*new_pmt)); + + if (!new_pmt) { + fprintf(stderr, "%s: Failed to allocate pmt_mmio\n", __func__); + exit(1); + } + + /* + * Create linked list of mmaped regions, + * but preserve the ordering from sysfs. + * Ordering is important for the user to + * use the seq=%u parameter when adding a counter. + */ + new_pmt->guid = guid; + new_pmt->mmio_base = mmio; + new_pmt->pmt_offset = offset; + new_pmt->size = size; + new_pmt->next = pmt_mmios; + + if (last) + last->next = new_pmt; + else + head = new_pmt; + + last = new_pmt; + } + +loop_cleanup_and_break: + close(fd_pmt); + close(fd_telem_dir); + } + + pmt_diriter_remove(&pmt_iter); + + /* + * If we found something, stick just + * created linked list to the front. + */ + if (head) + pmt_mmios = head; + + return head; +} + +struct pmt_mmio *pmt_mmio_find(unsigned int guid) +{ + struct pmt_mmio *pmmio = pmt_mmios; + + while (pmmio) { + if (pmmio->guid == guid) + return pmmio; + + pmmio = pmmio->next; + } + + return NULL; +} + +void *pmt_get_counter_pointer(struct pmt_mmio *pmmio, unsigned long counter_offset) +{ + char *ret; + + /* Get base of mmaped PMT file. */ + ret = (char *)pmmio->mmio_base; + + /* + * Apply PMT MMIO offset to obtain beginning of the mmaped telemetry data. + * It's not guaranteed that the mmaped memory begins with the telemetry data + * - we might have to apply the offset first. + */ + ret += pmmio->pmt_offset; + + /* Apply the counter offset to get the address to the mmaped counter. */ + ret += counter_offset; + + return ret; +} + +struct pmt_mmio *pmt_add_guid(unsigned int guid, unsigned int seq) +{ + struct pmt_mmio *ret; + + ret = pmt_mmio_find(guid); + if (!ret) + ret = pmt_mmio_open(guid); + + while (ret && seq) { + ret = ret->next; + --seq; + } + + return ret; +} + +enum pmt_open_mode { + PMT_OPEN_TRY, /* Open failure is not an error. */ + PMT_OPEN_REQUIRED, /* Open failure is a fatal error. */ +}; + +struct pmt_counter *pmt_find_counter(struct pmt_counter *pcounter, const char *name) +{ + while (pcounter) { + if (strcmp(pcounter->name, name) == 0) + break; + + pcounter = pcounter->next; + } + + return pcounter; +} + +struct pmt_counter **pmt_get_scope_root(enum counter_scope scope) +{ + switch (scope) { + case SCOPE_CPU: + return &sys.pmt_tp; + case SCOPE_CORE: + return &sys.pmt_cp; + case SCOPE_PACKAGE: + return &sys.pmt_pp; + } + + __builtin_unreachable(); +} + +void pmt_counter_add_domain(struct pmt_counter *pcounter, unsigned long *pmmio, unsigned int domain_id) +{ + /* Make sure the new domain fits. */ + if (domain_id >= pcounter->num_domains) + pmt_counter_resize(pcounter, domain_id + 1); + + assert(pcounter->domains); + assert(domain_id < pcounter->num_domains); + + pcounter->domains[domain_id].pcounter = pmmio; +} + +int pmt_add_counter(unsigned int guid, unsigned int seq, const char *name, enum pmt_datatype type, + unsigned int lsb, unsigned int msb, unsigned int offset, enum counter_scope scope, + enum counter_format format, unsigned int domain_id, enum pmt_open_mode mode) +{ + struct pmt_mmio *mmio; + struct pmt_counter *pcounter; + struct pmt_counter **const pmt_root = pmt_get_scope_root(scope); + bool new_counter = false; + int conflict = 0; + + if (lsb > msb) { + fprintf(stderr, "%s: %s: `%s` must be satisfied\n", __func__, "lsb <= msb", name); + exit(1); + } + + if (msb >= 64) { + fprintf(stderr, "%s: %s: `%s` must be satisfied\n", __func__, "msb < 64", name); + exit(1); + } + + mmio = pmt_add_guid(guid, seq); + if (!mmio) { + if (mode != PMT_OPEN_TRY) { + fprintf(stderr, "%s: failed to map PMT MMIO for guid %x, seq %u\n", __func__, guid, seq); + exit(1); + } + + return 1; + } + + if (offset >= mmio->size) { + if (mode != PMT_OPEN_TRY) { + fprintf(stderr, "%s: offset %u outside of PMT MMIO size %u\n", __func__, offset, mmio->size); + exit(1); + } + + return 1; + } + + pcounter = pmt_find_counter(*pmt_root, name); + if (!pcounter) { + pcounter = calloc(1, sizeof(*pcounter)); + new_counter = true; + } + + if (new_counter) { + strncpy(pcounter->name, name, ARRAY_SIZE(pcounter->name) - 1); + pcounter->type = type; + pcounter->scope = scope; + pcounter->lsb = lsb; + pcounter->msb = msb; + pcounter->format = format; + } else { + conflict += pcounter->type != type; + conflict += pcounter->scope != scope; + conflict += pcounter->lsb != lsb; + conflict += pcounter->msb != msb; + conflict += pcounter->format != format; + } + + if (conflict) { + fprintf(stderr, "%s: conflicting parameters for the PMT counter with the same name %s\n", + __func__, name); + exit(1); + } + + pmt_counter_add_domain(pcounter, pmt_get_counter_pointer(mmio, offset), domain_id); + + if (new_counter) { + pcounter->next = *pmt_root; + *pmt_root = pcounter; + } + + return 0; +} + +void pmt_init(void) +{ + int cpu_num; + unsigned long seq, offset, mod_num; + + if (BIC_IS_ENABLED(BIC_Diec6)) { + pmt_add_counter(PMT_MTL_DC6_GUID, PMT_MTL_DC6_SEQ, "Die%c6", PMT_TYPE_XTAL_TIME, + PMT_COUNTER_MTL_DC6_LSB, PMT_COUNTER_MTL_DC6_MSB, PMT_COUNTER_MTL_DC6_OFFSET, + SCOPE_PACKAGE, FORMAT_DELTA, 0, PMT_OPEN_TRY); + } + + if (BIC_IS_ENABLED(BIC_CPU_c1e)) { + seq = 0; + offset = PMT_COUNTER_CWF_MC1E_OFFSET_BASE; + mod_num = 0; /* Relative module number for current PMT file. */ + + /* Open the counter for each CPU. */ + for (cpu_num = 0; cpu_num < topo.max_cpu_num;) { + + if (cpu_is_not_allowed(cpu_num)) + goto next_loop_iter; + + /* + * Set the scope to CPU, even though CWF report the counter per module. + * CPUs inside the same module will read from the same location, instead of reporting zeros. + * + * CWF with newer firmware might require a PMT_TYPE_XTAL_TIME intead of PMT_TYPE_TCORE_CLOCK. + */ + pmt_add_counter(PMT_CWF_MC1E_GUID, seq, "CPU%c1e", PMT_TYPE_TCORE_CLOCK, + PMT_COUNTER_CWF_MC1E_LSB, PMT_COUNTER_CWF_MC1E_MSB, offset, SCOPE_CPU, + FORMAT_DELTA, cpu_num, PMT_OPEN_TRY); + + /* + * Rather complex logic for each time we go to the next loop iteration, + * so keep it as a label. + */ +next_loop_iter: + /* + * Advance the cpu number and check if we should also advance offset to + * the next counter inside the PMT file. + * + * On Clearwater Forest platform, the counter is reported per module, + * so open the same counter for all of the CPUs inside the module. + * That way, reported table show the correct value for all of the CPUs inside the module, + * instead of zeros. + */ + ++cpu_num; + if (cpu_num % PMT_COUNTER_CWF_CPUS_PER_MODULE == 0) { + offset += PMT_COUNTER_CWF_MC1E_OFFSET_INCREMENT; + ++mod_num; + } + + /* + * There are PMT_COUNTER_CWF_MC1E_NUM_MODULES_PER_FILE in each PMT file. + * + * If that number is reached, seq must be incremented to advance to the next file in a sequence. + * Offset inside that file and a module counter has to be reset. + */ + if (mod_num == PMT_COUNTER_CWF_MC1E_NUM_MODULES_PER_FILE) { + ++seq; + offset = PMT_COUNTER_CWF_MC1E_OFFSET_BASE; + mod_num = 0; + } + } + } +} + void turbostat_init() { setup_all_buffers(true); set_base_cpu(); - check_dev_msr(); - check_permissions(); + check_msr_access(); + check_perf_access(); process_cpuid(); + counter_info_init(); probe_pm_features(); + msr_perf_init(); linux_perf_init(); + rapl_perf_init(); + cstate_perf_init(); + added_perf_counters_init(); + pmt_init(); for_all_cpus(get_cpu_type, ODD_COUNTERS); for_all_cpus(get_cpu_type, EVEN_COUNTERS); - if (DO_BIC(BIC_IPC)) - (void)get_instr_count_fd(base_cpu); + if (BIC_IS_ENABLED(BIC_IPC) && has_aperf_access && get_instr_count_fd(base_cpu) != -1) + BIC_PRESENT(BIC_IPC); + + /* + * 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; + } +} + +void affinitize_child(void) +{ + /* Prefer cpu_possible_set, if available */ + if (sched_setaffinity(0, cpu_possible_setsize, cpu_possible_set)) { + warn("sched_setaffinity cpu_possible_set"); + + /* Otherwise, allow child to run on same cpu set as turbostat */ + if (sched_setaffinity(0, cpu_allowed_setsize, cpu_allowed_set)) + warn("sched_setaffinity cpu_allowed_set"); + } } int fork_it(char **argv) @@ -6201,6 +9501,7 @@ int fork_it(char **argv) child_pid = fork(); if (!child_pid) { /* child */ + affinitize_child(); execvp(argv[0], argv); err(errno, "exec %s", argv[0]); } else { @@ -6227,10 +9528,9 @@ int fork_it(char **argv) timersub(&tv_odd, &tv_even, &tv_delta); if (for_all_cpus_2(delta_cpu, ODD_COUNTERS, EVEN_COUNTERS)) fprintf(outf, "%s: Counter reset detected\n", progname); - else { - compute_average(EVEN_COUNTERS); - format_all_counters(EVEN_COUNTERS); - } + + compute_average(EVEN_COUNTERS); + format_all_counters(EVEN_COUNTERS); fprintf(outf, "%.6f sec\n", tv_delta.tv_sec + tv_delta.tv_usec / 1000000.0); @@ -6259,7 +9559,7 @@ int get_and_dump_counters(void) void print_version() { - fprintf(outf, "turbostat version 2023.11.07 - Len Brown <lenb@kernel.org>\n"); + fprintf(outf, "turbostat version 2025.02.02 - Len Brown <lenb@kernel.org>\n"); } #define COMMAND_LINE_SIZE 2048 @@ -6285,68 +9585,220 @@ void print_bootcmd(void) fclose(fp); } +struct msr_counter *find_msrp_by_name(struct msr_counter *head, char *name) +{ + struct msr_counter *mp; + + for (mp = head; mp; mp = mp->next) { + if (debug) + fprintf(stderr, "%s: %s %s\n", __func__, name, mp->name); + if (!strncmp(name, mp->name, strlen(mp->name))) + return mp; + } + return NULL; +} + int add_counter(unsigned int msr_num, char *path, char *name, unsigned int width, enum counter_scope scope, - enum counter_type type, enum counter_format format, int flags) + enum counter_type type, enum counter_format format, int flags, int id) { struct msr_counter *msrp; - msrp = calloc(1, sizeof(struct msr_counter)); + if (no_msr && msr_num) + errx(1, "Requested MSR counter 0x%x, but in --no-msr mode", msr_num); + + if (debug) + fprintf(stderr, "%s(msr%d, %s, %s, width%d, scope%d, type%d, format%d, flags%x, id%d)\n", + __func__, msr_num, path, name, width, scope, type, format, flags, id); + + switch (scope) { + + case SCOPE_CPU: + msrp = find_msrp_by_name(sys.tp, name); + if (msrp) { + if (debug) + fprintf(stderr, "%s: %s FOUND\n", __func__, name); + break; + } + if (sys.added_thread_counters++ >= MAX_ADDED_THREAD_COUNTERS) { + warnx("ignoring thread counter %s", name); + return -1; + } + break; + case SCOPE_CORE: + msrp = find_msrp_by_name(sys.cp, name); + if (msrp) { + if (debug) + fprintf(stderr, "%s: %s FOUND\n", __func__, name); + break; + } + if (sys.added_core_counters++ >= MAX_ADDED_CORE_COUNTERS) { + warnx("ignoring core counter %s", name); + return -1; + } + break; + case SCOPE_PACKAGE: + msrp = find_msrp_by_name(sys.pp, name); + if (msrp) { + if (debug) + fprintf(stderr, "%s: %s FOUND\n", __func__, name); + break; + } + if (sys.added_package_counters++ >= MAX_ADDED_PACKAGE_COUNTERS) { + warnx("ignoring package counter %s", name); + return -1; + } + break; + default: + warnx("ignoring counter %s with unknown scope", name); + return -1; + } + if (msrp == NULL) { - perror("calloc"); - exit(1); + msrp = calloc(1, sizeof(struct msr_counter)); + if (msrp == NULL) + err(-1, "calloc msr_counter"); + + msrp->msr_num = msr_num; + strncpy(msrp->name, name, NAME_BYTES - 1); + msrp->width = width; + msrp->type = type; + msrp->format = format; + msrp->flags = flags; + + switch (scope) { + case SCOPE_CPU: + msrp->next = sys.tp; + sys.tp = msrp; + break; + case SCOPE_CORE: + msrp->next = sys.cp; + sys.cp = msrp; + break; + case SCOPE_PACKAGE: + msrp->next = sys.pp; + sys.pp = msrp; + break; + } } - msrp->msr_num = msr_num; - strncpy(msrp->name, name, NAME_BYTES - 1); - if (path) - strncpy(msrp->path, path, PATH_BYTES - 1); - msrp->width = width; - msrp->type = type; - msrp->format = format; - msrp->flags = flags; + if (path) { + struct sysfs_path *sp; - switch (scope) { + sp = calloc(1, sizeof(struct sysfs_path)); + if (sp == NULL) { + perror("calloc"); + exit(1); + } + strncpy(sp->path, path, PATH_BYTES - 1); + sp->id = id; + sp->next = msrp->sp; + msrp->sp = sp; + } + + return 0; +} + +/* + * Initialize the fields used for identifying and opening the counter. + * + * Defer the initialization of any runtime buffers for actually reading + * the counters for when we initialize all perf counters, so we can later + * easily call re_initialize(). + */ +struct perf_counter_info *make_perf_counter_info(const char *perf_device, + const char *perf_event, + const char *name, + unsigned int width, + enum counter_scope scope, + enum counter_type type, enum counter_format format) +{ + struct perf_counter_info *pinfo; + + pinfo = calloc(1, sizeof(*pinfo)); + if (!pinfo) + errx(1, "%s: Failed to allocate %s/%s\n", __func__, perf_device, perf_event); + + strncpy(pinfo->device, perf_device, ARRAY_SIZE(pinfo->device) - 1); + strncpy(pinfo->event, perf_event, ARRAY_SIZE(pinfo->event) - 1); + + strncpy(pinfo->name, name, ARRAY_SIZE(pinfo->name) - 1); + pinfo->width = width; + pinfo->scope = scope; + pinfo->type = type; + pinfo->format = format; + + return pinfo; +} +int add_perf_counter(const char *perf_device, const char *perf_event, const char *name_buffer, unsigned int width, + enum counter_scope scope, enum counter_type type, enum counter_format format) +{ + struct perf_counter_info *pinfo; + + switch (scope) { case SCOPE_CPU: - msrp->next = sys.tp; - sys.tp = msrp; - sys.added_thread_counters++; - if (sys.added_thread_counters > MAX_ADDED_THREAD_COUNTERS) { - fprintf(stderr, "exceeded max %d added thread counters\n", MAX_ADDED_COUNTERS); - exit(-1); + if (sys.added_thread_perf_counters >= MAX_ADDED_THREAD_COUNTERS) { + warnx("ignoring thread counter perf/%s/%s", perf_device, perf_event); + return -1; } break; case SCOPE_CORE: - msrp->next = sys.cp; - sys.cp = msrp; - sys.added_core_counters++; - if (sys.added_core_counters > MAX_ADDED_COUNTERS) { - fprintf(stderr, "exceeded max %d added core counters\n", MAX_ADDED_COUNTERS); - exit(-1); + if (sys.added_core_perf_counters >= MAX_ADDED_CORE_COUNTERS) { + warnx("ignoring core counter perf/%s/%s", perf_device, perf_event); + return -1; } break; case SCOPE_PACKAGE: - msrp->next = sys.pp; - sys.pp = msrp; - sys.added_package_counters++; - if (sys.added_package_counters > MAX_ADDED_COUNTERS) { - fprintf(stderr, "exceeded max %d added package counters\n", MAX_ADDED_COUNTERS); - exit(-1); + if (sys.added_package_perf_counters >= MAX_ADDED_PACKAGE_COUNTERS) { + warnx("ignoring package counter perf/%s/%s", perf_device, perf_event); + return -1; } break; } + pinfo = make_perf_counter_info(perf_device, perf_event, name_buffer, width, scope, type, format); + + if (!pinfo) + return -1; + + switch (scope) { + case SCOPE_CPU: + pinfo->next = sys.perf_tp; + sys.perf_tp = pinfo; + ++sys.added_thread_perf_counters; + break; + + case SCOPE_CORE: + pinfo->next = sys.perf_cp; + sys.perf_cp = pinfo; + ++sys.added_core_perf_counters; + break; + + case SCOPE_PACKAGE: + pinfo->next = sys.perf_pp; + sys.perf_pp = pinfo; + ++sys.added_package_perf_counters; + break; + } + + // FIXME: we might not have debug here yet + if (debug) + fprintf(stderr, "%s: %s/%s, name: %s, scope%d\n", + __func__, pinfo->device, pinfo->event, pinfo->name, pinfo->scope); + return 0; } -void parse_add_command(char *add_command) +void parse_add_command_msr(char *add_command) { int msr_num = 0; char *path = NULL; - char name_buffer[NAME_BYTES] = ""; + char perf_device[PERF_DEV_NAME_BYTES] = ""; + char perf_event[PERF_EVT_NAME_BYTES] = ""; + char name_buffer[PERF_NAME_BYTES] = ""; int width = 64; int fail = 0; enum counter_scope scope = SCOPE_CPU; @@ -6361,6 +9813,11 @@ void parse_add_command(char *add_command) if (sscanf(add_command, "msr%d", &msr_num) == 1) goto next; + BUILD_BUG_ON(ARRAY_SIZE(perf_device) <= 31); + BUILD_BUG_ON(ARRAY_SIZE(perf_event) <= 31); + if (sscanf(add_command, "perf/%31[^/]/%31[^,]", &perf_device[0], &perf_event[0]) == 2) + goto next; + if (*add_command == '/') { path = add_command; goto next; @@ -6408,7 +9865,8 @@ void parse_add_command(char *add_command) goto next; } - if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) { /* 18 < NAME_BYTES */ + BUILD_BUG_ON(ARRAY_SIZE(name_buffer) <= 18); + if (sscanf(add_command, "%18s,%*s", name_buffer) == 1) { char *eos; eos = strchr(name_buffer, ','); @@ -6425,21 +9883,33 @@ next: } } - if ((msr_num == 0) && (path == NULL)) { - fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter ) required\n"); + if ((msr_num == 0) && (path == NULL) && (perf_device[0] == '\0' || perf_event[0] == '\0')) { + fprintf(stderr, "--add: (msrDDD | msr0xXXX | /path_to_counter | perf/device/event) required\n"); fail++; } + /* Test for non-empty perf_device and perf_event */ + const bool is_perf_counter = perf_device[0] && perf_event[0]; + /* generate default column header */ if (*name_buffer == '\0') { - if (width == 32) - sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : ""); - else - sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : ""); + if (is_perf_counter) { + snprintf(name_buffer, ARRAY_SIZE(name_buffer), "perf/%s", perf_event); + } else { + if (width == 32) + sprintf(name_buffer, "M0x%x%s", msr_num, format == FORMAT_PERCENT ? "%" : ""); + else + sprintf(name_buffer, "M0X%x%s", msr_num, format == FORMAT_PERCENT ? "%" : ""); + } } - if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0)) - fail++; + if (is_perf_counter) { + if (add_perf_counter(perf_device, perf_event, name_buffer, width, scope, type, format)) + fail++; + } else { + if (add_counter(msr_num, path, name_buffer, width, scope, type, format, 0, 0)) + fail++; + } if (fail) { help(); @@ -6447,6 +9917,308 @@ next: } } +bool starts_with(const char *str, const char *prefix) +{ + return strncmp(prefix, str, strlen(prefix)) == 0; +} + +int pmt_parse_from_path(const char *target_path, unsigned int *out_guid, unsigned int *out_seq) +{ + struct pmt_diriter_t pmt_iter; + const struct dirent *dirname; + struct stat stat, target_stat; + int fd_telem_dir = -1; + int fd_target_dir; + unsigned int seq = 0; + unsigned long guid, target_guid; + int ret = -1; + + fd_target_dir = open(target_path, O_RDONLY | O_DIRECTORY); + if (fd_target_dir == -1) { + return -1; + } + + if (fstat(fd_target_dir, &target_stat) == -1) { + fprintf(stderr, "%s: Failed to stat the target: %s", __func__, strerror(errno)); + exit(1); + } + + if (parse_telem_info_file(fd_target_dir, "guid", "%lx", &target_guid)) { + fprintf(stderr, "%s: Failed to parse the target guid file: %s", __func__, strerror(errno)); + exit(1); + } + + close(fd_target_dir); + + pmt_diriter_init(&pmt_iter); + + for (dirname = pmt_diriter_begin(&pmt_iter, SYSFS_TELEM_PATH); dirname != NULL; + dirname = pmt_diriter_next(&pmt_iter)) { + + fd_telem_dir = openat(dirfd(pmt_iter.dir), dirname->d_name, O_RDONLY | O_DIRECTORY); + if (fd_telem_dir == -1) + continue; + + if (parse_telem_info_file(fd_telem_dir, "guid", "%lx", &guid)) { + fprintf(stderr, "%s: Failed to parse the guid file: %s", __func__, strerror(errno)); + continue; + } + + if (fstat(fd_telem_dir, &stat) == -1) { + fprintf(stderr, "%s: Failed to stat %s directory: %s", __func__, + dirname->d_name, strerror(errno)); + continue; + } + + /* + * If reached the same directory as target, exit the loop. + * Seq has the correct value now. + */ + if (stat.st_dev == target_stat.st_dev && stat.st_ino == target_stat.st_ino) { + ret = 0; + break; + } + + /* + * If reached directory with the same guid, + * but it's not the target directory yet, + * increment seq and continue the search. + */ + if (guid == target_guid) + ++seq; + + close(fd_telem_dir); + fd_telem_dir = -1; + } + + pmt_diriter_remove(&pmt_iter); + + if (fd_telem_dir != -1) + close(fd_telem_dir); + + if (!ret) { + *out_guid = target_guid; + *out_seq = seq; + } + + return ret; +} + +void parse_add_command_pmt(char *add_command) +{ + char *name = NULL; + char *type_name = NULL; + char *format_name = NULL; + char *direct_path = NULL; + static const char direct_path_prefix[] = "path="; + unsigned int offset; + unsigned int lsb; + unsigned int msb; + unsigned int guid; + unsigned int seq = 0; /* By default, pick first file in a sequence with a given GUID. */ + unsigned int domain_id; + enum counter_scope scope = 0; + enum pmt_datatype type = PMT_TYPE_RAW; + enum counter_format format = FORMAT_RAW; + bool has_offset = false; + bool has_lsb = false; + bool has_msb = false; + bool has_format = true; /* Format has a default value. */ + bool has_guid = false; + bool has_scope = false; + bool has_type = true; /* Type has a default value. */ + + /* Consume the "pmt," prefix. */ + add_command = strchr(add_command, ','); + if (!add_command) { + help(); + exit(1); + } + ++add_command; + + while (add_command) { + if (starts_with(add_command, "name=")) { + name = add_command + strlen("name="); + goto next; + } + + if (starts_with(add_command, "type=")) { + type_name = add_command + strlen("type="); + goto next; + } + + if (starts_with(add_command, "domain=")) { + const size_t prefix_len = strlen("domain="); + + if (sscanf(add_command + prefix_len, "cpu%u", &domain_id) == 1) { + scope = SCOPE_CPU; + has_scope = true; + } else if (sscanf(add_command + prefix_len, "core%u", &domain_id) == 1) { + scope = SCOPE_CORE; + has_scope = true; + } else if (sscanf(add_command + prefix_len, "package%u", &domain_id) == 1) { + scope = SCOPE_PACKAGE; + has_scope = true; + } + + if (!has_scope) { + printf("%s: invalid value for scope. Expected cpu%%u, core%%u or package%%u.\n", + __func__); + exit(1); + } + + goto next; + } + + if (starts_with(add_command, "format=")) { + format_name = add_command + strlen("format="); + goto next; + } + + if (sscanf(add_command, "offset=%u", &offset) == 1) { + has_offset = true; + goto next; + } + + if (sscanf(add_command, "lsb=%u", &lsb) == 1) { + has_lsb = true; + goto next; + } + + if (sscanf(add_command, "msb=%u", &msb) == 1) { + has_msb = true; + goto next; + } + + if (sscanf(add_command, "guid=%x", &guid) == 1) { + has_guid = true; + goto next; + } + + if (sscanf(add_command, "seq=%x", &seq) == 1) + goto next; + + if (strncmp(add_command, direct_path_prefix, strlen(direct_path_prefix)) == 0) { + direct_path = add_command + strlen(direct_path_prefix); + goto next; + } +next: + add_command = strchr(add_command, ','); + if (add_command) { + *add_command = '\0'; + add_command++; + } + } + + if (!name) { + printf("%s: missing %s\n", __func__, "name"); + exit(1); + } + + if (strlen(name) >= PMT_COUNTER_NAME_SIZE_BYTES) { + printf("%s: name has to be at most %d characters long\n", __func__, PMT_COUNTER_NAME_SIZE_BYTES); + exit(1); + } + + if (format_name) { + has_format = false; + + if (strcmp("raw", format_name) == 0) { + format = FORMAT_RAW; + has_format = true; + } + + if (strcmp("delta", format_name) == 0) { + format = FORMAT_DELTA; + has_format = true; + } + + if (!has_format) { + fprintf(stderr, "%s: Invalid format %s. Expected raw or delta\n", __func__, format_name); + exit(1); + } + } + + if (type_name) { + has_type = false; + + if (strcmp("raw", type_name) == 0) { + type = PMT_TYPE_RAW; + has_type = true; + } + + if (strcmp("txtal_time", type_name) == 0) { + type = PMT_TYPE_XTAL_TIME; + has_type = true; + } + + if (strcmp("tcore_clock", type_name) == 0) { + type = PMT_TYPE_TCORE_CLOCK; + has_type = true; + } + + if (!has_type) { + printf("%s: invalid %s: %s\n", __func__, "type", type_name); + exit(1); + } + } + + if (!has_offset) { + printf("%s : missing %s\n", __func__, "offset"); + exit(1); + } + + if (!has_lsb) { + printf("%s: missing %s\n", __func__, "lsb"); + exit(1); + } + + if (!has_msb) { + printf("%s: missing %s\n", __func__, "msb"); + exit(1); + } + + if (direct_path && has_guid) { + printf("%s: path and guid+seq parameters are mutually exclusive\n" + "notice: passed guid=0x%x and path=%s\n", __func__, guid, direct_path); + exit(1); + } + + if (direct_path) { + if (pmt_parse_from_path(direct_path, &guid, &seq)) { + printf("%s: failed to parse PMT file from %s\n", __func__, direct_path); + exit(1); + } + + /* GUID was just infered from the direct path. */ + has_guid = true; + } + + if (!has_guid) { + printf("%s: missing %s\n", __func__, "guid or path"); + exit(1); + } + + if (!has_scope) { + printf("%s: missing %s\n", __func__, "scope"); + exit(1); + } + + if (lsb > msb) { + printf("%s: lsb > msb doesn't make sense\n", __func__); + exit(1); + } + + pmt_add_counter(guid, seq, name, type, lsb, msb, offset, scope, format, domain_id, PMT_OPEN_REQUIRED); +} + +void parse_add_command(char *add_command) +{ + if (strncmp(add_command, "pmt", strlen("pmt")) == 0) + return parse_add_command_pmt(add_command); + return parse_add_command_msr(add_command); +} + int is_deferred_add(char *name) { int i; @@ -6503,7 +10275,7 @@ void probe_sysfs(void) if (is_deferred_skip(name_buf)) continue; - add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC, FORMAT_PERCENT, SYSFS_PERCPU); + add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_USEC, FORMAT_PERCENT, SYSFS_PERCPU, 0); } for (state = 10; state >= 0; --state) { @@ -6531,7 +10303,7 @@ void probe_sysfs(void) if (is_deferred_skip(name_buf)) continue; - add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS, FORMAT_DELTA, SYSFS_PERCPU); + add_counter(0, path, name_buf, 64, SCOPE_CPU, COUNTER_ITEMS, FORMAT_DELTA, SYSFS_PERCPU, 0); } } @@ -6585,6 +10357,7 @@ void cmdline(int argc, char **argv) { "Dump", no_argument, 0, 'D' }, { "debug", no_argument, 0, 'd' }, /* internal, not documented */ { "enable", required_argument, 0, 'e' }, + { "force", no_argument, 0, 'f' }, { "interval", required_argument, 0, 'i' }, { "IPC", no_argument, 0, 'I' }, { "num_iterations", required_argument, 0, 'n' }, @@ -6595,6 +10368,8 @@ 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' }, + { "no-perf", no_argument, 0, 'P' }, { "show", required_argument, 0, 's' }, { "Summary", no_argument, 0, 'S' }, { "TCC", required_argument, 0, 'T' }, @@ -6604,7 +10379,25 @@ 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, "+MPn:", long_options, &option_index)) != -1) { + switch (opt) { + case 'M': + no_msr = 1; + break; + case 'P': + no_perf = 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); @@ -6614,11 +10407,20 @@ void cmdline(int argc, char **argv) break; case 'D': dump_only++; + /* + * Force the no_perf early to prevent using it as a source. + * User asks for raw values, but perf returns them relative + * to the opening of the file descriptor. + */ + no_perf = 1; break; case 'e': /* --enable specified counter */ bic_enabled = bic_enabled | bic_lookup(optarg, SHOW_LIST); break; + case 'f': + force_load++; + break; case 'd': debug++; ENABLE_BIC(BIC_DISABLED_BY_DEFAULT); @@ -6662,6 +10464,10 @@ void cmdline(int argc, char **argv) case 'q': quiet = 1; break; + case 'M': + case 'P': + /* Parsed earlier */ + break; case 'n': num_iterations = strtod(optarg, NULL); @@ -6704,6 +10510,22 @@ 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; @@ -6729,9 +10551,13 @@ skip_cgroup_setting: probe_sysfs(); + if (!getuid()) + set_rlimit(); + turbostat_init(); - msr_sum_record(); + if (!no_msr) + msr_sum_record(); /* dump counters and exit */ if (dump_only) diff --git a/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c index 5fd9e594079c..ebda9c366b2b 100644 --- a/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c +++ b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.c @@ -1241,6 +1241,7 @@ unsigned int get_pkg_num(int cpu) retval = fscanf(fp, "%d\n", &pkg); if (retval != 1) errx(1, "%s: failed to parse", pathname); + fclose(fp); return pkg; } |