Skip to contentSkip to conversation

turbostat v2025.06.08

PATCH New Len Brown 1 participant 19 messages Jun 8, 2025 Turbostat utility

lore
Share & tools
lore

Len Brown

author maintainer
#0 lore raw

cover turbostat v2025.06.08

Please let me know if you have troubles with any of these patches.

thanks!
-Len

[PATCH 01/18] tools/power turbostat.8: fix typo: idle_pct should be [PATCH 02/18] tools/power turbostat.8: pm_domain wording fix [PATCH 03/18] tools/power turbostat: Add Android support for MSR [PATCH 04/18] tools/power turbostat: Fix RAPL_GFX_ALL typo [PATCH 05/18] tools/power turbostat: Fix AMD package-energy reporting [PATCH 06/18] tools/power turbostat: Always check rapl_joules flag [PATCH 07/18] tools/power turbostat: Quit early for unsupported RAPL [PATCH 08/18] tools/power turbostat: Remove add_rapl_perf_counter_() [PATCH 09/18] tools/power turbostat: Remove
[PATCH 10/18] tools/power turbostat: Remove add_msr_perf_counter_() [PATCH 11/18] tools/power turbostat: Introduce add_msr_counter() [PATCH 12/18] tools/power turbostat: Clean up add perf/msr counter [PATCH 13/18] tools/power turbostat: Allow probing RAPL with [PATCH 14/18] tools/power turbostat: Avoid probing the same perf [PATCH 15/18] tools/power turbostat: Dump RAPL sysfs info [PATCH 16/18] tools/power turbostat: Add initial support for DMR [PATCH 17/18] tools/power turbostat: Add initial support for [PATCH 18/18] tools/power turbostat: version 2025.06.08

author maintainer
#1 lore raw

PATCH 1/18 tools/power turbostat.8: fix typo: idle_pct should be pct_idle

From: Len Brown <[email protected]>

---
 tools/power/x86/turbostat/turbostat.8 | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/power/x86/turbostat/turbostat.8 +1 -1
index b74ed916057e..183d9c0fb315 100644
@@ -100,7 +100,7 @@ The column name "all" can be used to enable all disabled-by-default built-in cou
 .PP
 \fB--show column\fP show only the specified built-in columns.  May be invoked multiple times, or with a comma-separated list of column names.
 .PP
-\fB--show CATEGORY --hide CATEGORY\fP  Show and hide also accept a single CATEGORY of columns: "all", "topology", "idle", "frequency", "power", "cpuidle", "hwidle", "swidle", "other".  "idle" (enabled by default), includes "hwidle" and "idle_pct".  "cpuidle" (default disabled) includes cpuidle software invocation counters.  "swidle" includes "cpuidle" plus "idle_pct".  "hwidle" includes only hardware based idle residency counters.  Older versions of turbostat used the term "sysfs" for what is now "swidle".
+\fB--show CATEGORY --hide CATEGORY\fP  Show and hide also accept a single CATEGORY of columns: "all", "topology", "idle", "frequency", "power", "cpuidle", "hwidle", "swidle", "other".  "idle" (enabled by default), includes "hwidle" and "pct_idle".  "cpuidle" (default disabled) includes cpuidle software invocation counters.  "swidle" includes "cpuidle" plus "pct_idle".  "hwidle" includes only hardware based idle residency counters.  Older versions of turbostat used the term "sysfs" for what is now "swidle".
 .PP
 \fB--Dump\fP displays the raw counter values.
 .PP
Signature · 1 line
2.45.2
author maintainer
#2 lore raw

PATCH 2/18 tools/power turbostat.8: pm_domain wording fix

From: Len Brown <[email protected]>

turbostat.8: clarify that uncore "domains" are Power Management domains, aka pm_domains.

---
 tools/power/x86/turbostat/turbostat.8 | 4 ++--
 1 file changed, 2 insertions(+), 2 deletions(-)
tools/power/x86/turbostat/turbostat.8 +2 -2
index 183d9c0fb315..fb11108aaf42 100644
@@ -204,8 +204,8 @@ The system configuration dump (if --quiet is not used) is followed by statistics
 .PP
 \fBUncMHz\fP per-package uncore MHz, instantaneous sample.
 .PP
-\fBUMHz1.0\fP per-package uncore MHz for domain=1 and fabric_cluster=0, instantaneous sample.  System summary is the average of all packages.
-Intel Granite Rapids systems use domains 0-2 for CPUs, and 3-4 for IO, with cluster always 0.
+\fBUMHz1.0\fP per-package uncore MHz for pm_domain=1 and fabric_cluster=0, instantaneous sample.  System summary is the average of all packages.
+Intel Granite Rapids systems use pm_domains 0-2 for CPUs, and 3-4 for IO, with cluster always 0.
 For the "--show" and "--hide" options, use "UncMHz" to operate on all UMHz*.* as a group.
 .SH TOO MUCH INFORMATION EXAMPLE
 By default, turbostat dumps all possible information -- a system configuration header, followed by columns for all counters.
Signature · 1 line
2.45.2
author maintainer
#3 lore raw

PATCH 3/18 tools/power turbostat: Add Android support for MSR device handling

In reply to Len Brown

From: Kaushlendra Kumar <[email protected]>

It uses /dev/msrN device paths on Android instead of /dev/cpu/N/msr, updates error messages and permission checks to reflect the Android device path, and wraps platform-specific code with #if defined(ANDROID) to ensure correct behavior on both Android and non-Android systems. These changes improve compatibility and usability of turbostat on Android devices.

---
 tools/power/x86/turbostat/turbostat.c | 20 +++++++++++++++++---
 1 file changed, 17 insertions(+), 3 deletions(-)
tools/power/x86/turbostat/turbostat.c +17 -3
index 0170d3cc6819..3b418cae8812 100644
@@ -2140,13 +2140,20 @@ int get_msr_fd(int cpu)
 
 	if (fd)
 		return fd;
-
+#if defined(ANDROID)
+	sprintf(pathname, "/dev/msr%d", cpu);
+#else
 	sprintf(pathname, "/dev/cpu/%d/msr", cpu);
+#endif
 	fd = open(pathname, O_RDONLY);
 	if (fd < 0)
+#if defined(ANDROID)
+		err(-1, "%s open failed, try chown or chmod +r /dev/msr*, "
+		    "or run with --no-msr, or run as root", pathname);
+#else
 		err(-1, "%s open failed, try chown or chmod +r /dev/cpu/*/msr, "
 		    "or run with --no-msr, or run as root", pathname);
-
+#endif
 	fd_percpu[cpu] = fd;
 
 	return fd;
@@ -6476,8 +6483,11 @@ void check_dev_msr()
 
 	if (no_msr)
 		return;
-
+#if defined(ANDROID)
+	sprintf(pathname, "/dev/msr%d", base_cpu);
+#else
 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
+#endif
 	if (stat(pathname, &sb))
 		if (system("/sbin/modprobe msr > /dev/null 2>&1"))
 			no_msr = 1;
@@ -6527,7 +6537,11 @@ void check_msr_permission(void)
 	failed += check_for_cap_sys_rawio();
 
 	/* test file permissions */
+#if defined(ANDROID)
+	sprintf(pathname, "/dev/msr%d", base_cpu);
+#else
 	sprintf(pathname, "/dev/cpu/%d/msr", base_cpu);
+#endif
 	if (euidaccess(pathname, R_OK)) {
 		failed++;
 	}
Signature · 1 line
2.45.2
author maintainer
#4 lore raw

PATCH 4/18 tools/power turbostat: Fix RAPL_GFX_ALL typo

In reply to Len Brown

From: Kaushlendra Kumar <[email protected]>

Fix typo in the currently unused RAPL_GFX_ALL macro definition.

---
 tools/power/x86/turbostat/turbostat.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)
tools/power/x86/turbostat/turbostat.c +1 -1
index 3b418cae8812..018f0fe96691 100644
@@ -539,7 +539,7 @@ enum rapl_msrs {
 #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)
 #define RAPL_CORE_ALL	(RAPL_CORE | RAPL_CORE_POLICY)
-#define RAPL_GFX_ALL	(RAPL_GFX | RAPL_GFX_POLIGY)
+#define RAPL_GFX_ALL	(RAPL_GFX | RAPL_GFX_POLICY)
 
 #define RAPL_AMD_F17H	(RAPL_AMD_PWR_UNIT | RAPL_AMD_CORE_ENERGY_STAT | RAPL_AMD_PKG_ENERGY_STAT)
 
Signature · 1 line
2.45.2
author maintainer
#5 lore raw

PATCH 5/18 tools/power turbostat: Fix AMD package-energy reporting

In reply to Len Brown

From: "Gautham R. Shenoy" <[email protected]>

commit 05a2f07db888 ("tools/power turbostat: read RAPL counters via perf") that adds support to read RAPL counters via perf defines the notion of a RAPL domain_id which is set to physical_core_id on platforms which support per_core_rapl counters (Eg: AMD processors Family 17h onwards) and is set to the physical_package_id on all the other platforms.

However, the physical_core_id is only unique within a package and on platforms with multiple packages more than one core can have the same physical_core_id and thus the same domain_id. (For eg, the first cores of each package have the physical_core_id = 0). This results in all these cores with the same physical_core_id using the same entry in the rapl_counter_info_perdomain[]. Since rapl_perf_init() skips the perf-initialization for cores whose domain_ids have already been visited, cores that have the same physical_core_id always read the perf file corresponding to the physical_core_id of the first package and thus the package-energy is incorrectly reported to be the same value for different packages.

Note: This issue only arises when RAPL counters are read via perf and not when they are read via MSRs since in the latter case the MSRs are read separately on each core.

Fix this issue by associating each CPU with rapl_core_id which is unique across all the packages in the system.

Fixes 05a2f07db888 ("tools/power turbostat: read RAPL counters via perf")
---
 tools/power/x86/turbostat/turbostat.c | 41 +++++++++++++++++++++++----
 1 file changed, 36 insertions(+), 5 deletions(-)
tools/power/x86/turbostat/turbostat.c +36 -5
index 018f0fe96691..743db19a13c2 100644
@@ -4773,6 +4773,38 @@ unsigned long pmt_read_counter(struct pmt_counter *ppmt, unsigned int domain_id)
 	return (value & value_mask) >> value_shift;
 }
 
+
+/* Rapl domain enumeration helpers */
+static inline int get_rapl_num_domains(void)
+{
+	int num_packages = topo.max_package_id + 1;
+	int num_cores_per_package;
+	int num_cores;
+
+	if (!platform->has_per_core_rapl)
+		return num_packages;
+
+	num_cores_per_package = topo.max_core_id + 1;
+	num_cores = num_cores_per_package * num_packages;
+
+	return num_cores;
+}
+
+static inline int get_rapl_domain_id(int cpu)
+{
+	int nr_cores_per_package = topo.max_core_id + 1;
+	int rapl_core_id;
+
+	if (!platform->has_per_core_rapl)
+		return cpus[cpu].physical_package_id;
+
+	/* Compute the system-wide unique core-id for @cpu */
+	rapl_core_id = cpus[cpu].physical_core_id;
+	rapl_core_id += cpus[cpu].physical_package_id * nr_cores_per_package;
+
+	return rapl_core_id;
+}
+
 /*
  * get_counters(...)
  * migrate to cpu
@@ -4828,7 +4860,7 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
 		goto done;
 
 	if (platform->has_per_core_rapl) {
-		status = get_rapl_counters(cpu, c->core_id, c, p);
+		status = get_rapl_counters(cpu, get_rapl_domain_id(cpu), c, p);
 		if (status != 0)
 			return status;
 	}
@@ -4894,7 +4926,7 @@ int get_counters(struct thread_data *t, struct core_data *c, struct pkg_data *p)
 		p->sys_lpi = cpuidle_cur_sys_lpi_us;
 
 	if (!platform->has_per_core_rapl) {
-		status = get_rapl_counters(cpu, p->package_id, c, p);
+		status = get_rapl_counters(cpu, get_rapl_domain_id(cpu), c, p);
 		if (status != 0)
 			return status;
 	}
@@ -7877,7 +7909,7 @@ void linux_perf_init(void)
 
 void rapl_perf_init(void)
 {
-	const unsigned int num_domains = (platform->has_per_core_rapl ? topo.max_core_id : topo.max_package_id) + 1;
+	const unsigned int num_domains = get_rapl_num_domains();
 	bool *domain_visited = calloc(num_domains, sizeof(bool));
 
 	rapl_counter_info_perdomain = calloc(num_domains, sizeof(*rapl_counter_info_perdomain));
@@ -7918,8 +7950,7 @@ void rapl_perf_init(void)
 				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;
+			next_domain = get_rapl_domain_id(cpu);
 
 			assert(next_domain < num_domains);
 
Signature · 1 line
2.45.2
author maintainer
#6 lore raw

PATCH 6/18 tools/power turbostat: Always check rapl_joules flag

In reply to Len Brown

From: Zhang Rui <[email protected]>

rapl_joules bit should always be checked even if platform_features->rapl_msrs is not set or no_msr flag is used.

---
 tools/power/x86/turbostat/turbostat.c | 12 +++++++++---
 1 file changed, 9 insertions(+), 3 deletions(-)
tools/power/x86/turbostat/turbostat.c +9 -3
index 743db19a13c2..69c19e01b681 100644
@@ -7302,6 +7302,9 @@ void rapl_probe_intel(void)
 	else
 		bic_enabled &= ~bic_joules_bits;
 
+	if (!platform->rapl_msrs || no_msr)
+		return;
+
 	if (!(platform->rapl_msrs & RAPL_PKG_PERF_STATUS))
 		bic_enabled &= ~BIC_PKG__;
 	if (!(platform->rapl_msrs & RAPL_DRAM_PERF_STATUS))
@@ -7352,6 +7355,9 @@ void rapl_probe_amd(void)
 	else
 		bic_enabled &= ~bic_joules_bits;
 
+	if (!platform->rapl_msrs || no_msr)
+		return;
+
 	if (get_msr(base_cpu, MSR_RAPL_PWR_UNIT, &msr))
 		return;
 
@@ -7504,9 +7510,6 @@ int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
  */
 void probe_rapl(void)
 {
-	if (!platform->rapl_msrs || no_msr)
-		return;
-
 	if (genuine_intel)
 		rapl_probe_intel();
 	if (authentic_amd || hygon_genuine)
@@ -7515,6 +7518,9 @@ void probe_rapl(void)
 	if (quiet)
 		return;
 
+	if (!platform->rapl_msrs || no_msr)
+		return;
+
 	for_all_cpus(print_rapl, ODD_COUNTERS);
 }
 
Signature · 1 line
2.45.2
author maintainer
#7 lore raw

PATCH 7/18 tools/power turbostat: Quit early for unsupported RAPL counters

In reply to Len Brown

From: Zhang Rui <[email protected]>

Quit early for unsupported RAPL counters.

No functional change.

---
 tools/power/x86/turbostat/turbostat.c | 5 ++++-
 1 file changed, 4 insertions(+), 1 deletion(-)
tools/power/x86/turbostat/turbostat.c +4 -1
index 69c19e01b681..7e7d25d2362a 100644
@@ -7948,6 +7948,9 @@ void rapl_perf_init(void)
 		enum rapl_unit unit;
 		unsigned int next_domain;
 
+		if (!BIC_IS_ENABLED(cai->bic))
+			continue;
+
 		memset(domain_visited, 0, num_domains * sizeof(*domain_visited));
 
 		for (int cpu = 0; cpu < topo.max_cpu_num + 1; ++cpu) {
@@ -7971,7 +7974,7 @@ void rapl_perf_init(void)
 			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)) {
+			if (platform->rapl_msrs & cai->feature_mask) {
 
 				/* Use perf API for this counter */
 				if (!no_perf && cai->perf_name
Signature · 1 line
2.45.2
author maintainer
#8 lore raw

PATCH 8/18 tools/power turbostat: Remove add_rapl_perf_counter_()

In reply to Len Brown

From: Zhang Rui <[email protected]>

As the only caller of add_rapl_perf_counter_(), add_rapl_perf_counter() just gives extra debug output on top. There is no need to keep both functions.

Remove add_rapl_perf_counter_() and move all the logic to add_rapl_perf_counter().

No functional change.

---
 tools/power/x86/turbostat/turbostat.c | 25 ++++++++++---------------
 1 file changed, 10 insertions(+), 15 deletions(-)
tools/power/x86/turbostat/turbostat.c +10 -15
index 7e7d25d2362a..2651030c506b 100644
@@ -7853,44 +7853,39 @@ static int has_instr_count_access(void)
 	return has_access;
 }
 
-int add_rapl_perf_counter_(int cpu, struct rapl_counter_info_t *rci, const struct rapl_counter_arch_info *cai,
+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 = -1;
+
 	if (no_perf)
 		return -1;
 
 	const double scale = read_perf_scale(cai->perf_subsys, cai->perf_name);
 
 	if (scale == 0.0)
-		return -1;
+		goto end;
 
 	const enum rapl_unit unit = read_perf_rapl_unit(cai->perf_subsys, cai->perf_name);
 
 	if (unit == RAPL_UNIT_INVALID)
-		return -1;
+		goto end;
 
 	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;
+	ret = open_perf_counter(cpu, rapl_type, rapl_energy_pkg_config, rci->fd_perf, PERF_FORMAT_GROUP);
+	if (ret == -1)
+		goto end;
 
 	/* If it's the first counter opened, make it a group descriptor */
 	if (rci->fd_perf == -1)
-		rci->fd_perf = fd_counter;
+		rci->fd_perf = ret;
 
 	*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);
 
+end:
 	if (debug >= 2)
 		fprintf(stderr, "%s: %d (cpu: %d)\n", __func__, ret, cpu);
 
Signature · 1 line
2.45.2
author maintainer
#9 lore raw

PATCH 9/18 tools/power turbostat: Remove add_cstate_perf_counter_()

In reply to Len Brown

From: Zhang Rui <[email protected]>

As the only caller of add_cstate_perf_counter_(), add_cstate_perf_counter() just gives extra debug output on top. There is no need to keep both functions.

Remove add_cstate_perf_counter_() and move all the logic to add_cstate_perf_counter().

No functional change.

---
 tools/power/x86/turbostat/turbostat.c | 22 +++++++++-------------
 1 file changed, 9 insertions(+), 13 deletions(-)
tools/power/x86/turbostat/turbostat.c +9 -13
index 2651030c506b..9af1e13484e2 100644
@@ -8021,35 +8021,31 @@ int *get_cstate_perf_group_fd(struct cstate_counter_info_t *cci, const char *gro
 	return NULL;
 }
 
-int add_cstate_perf_counter_(int cpu, struct cstate_counter_info_t *cci, const struct cstate_counter_arch_info *cai)
+int add_cstate_perf_counter(int cpu, struct cstate_counter_info_t *cci, const struct cstate_counter_arch_info *cai)
 {
+	int ret = -1;
+
 	if (no_perf)
 		return -1;
 
 	int *pfd_group = get_cstate_perf_group_fd(cci, cai->perf_subsys);
 
 	if (pfd_group == NULL)
-		return -1;
+		goto end;
 
 	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);
+	ret = open_perf_counter(cpu, type, config, *pfd_group, PERF_FORMAT_GROUP);
 
-	if (fd_counter == -1)
-		return -1;
+	if (ret == -1)
+		goto end;
 
 	/* 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);
+		*pfd_group = ret;
 
+end:
 	if (debug >= 2)
 		fprintf(stderr, "%s: %d (cpu: %d)\n", __func__, ret, cpu);
 
Signature · 1 line
2.45.2
author maintainer
#10 lore raw

PATCH 10/18 tools/power turbostat: Remove add_msr_perf_counter_()

In reply to Len Brown

From: Zhang Rui <[email protected]>

As the only caller of add_msr_perf_counter_(), add_msr_perf_counter() just gives extra debug output on top. There is no need to keep both functions.

Remove add_msr_perf_counter_() and move all the logic to add_msr_perf_counter().

No functional change.

---
 tools/power/x86/turbostat/turbostat.c | 20 ++++++++------------
 1 file changed, 8 insertions(+), 12 deletions(-)
tools/power/x86/turbostat/turbostat.c +8 -12
index 9af1e13484e2..ef923da00765 100644
@@ -8052,30 +8052,26 @@ int add_cstate_perf_counter(int cpu, struct cstate_counter_info_t *cci, const st
 	return ret;
 }
 
-int add_msr_perf_counter_(int cpu, struct msr_counter_info_t *cci, const struct msr_counter_arch_info *cai)
+int add_msr_perf_counter(int cpu, struct msr_counter_info_t *cci, const struct msr_counter_arch_info *cai)
 {
+	int ret = -1;
+
 	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);
+	ret = open_perf_counter(cpu, type, config, cci->fd_perf, PERF_FORMAT_GROUP);
 
-	if (fd_counter == -1)
-		return -1;
+	if (ret == -1)
+		goto end;
 
 	/* 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);
+		cci->fd_perf = ret;
 
+end:
 	if (debug)
 		fprintf(stderr, "%s: %s/%s: %d (cpu: %d)\n", __func__, cai->perf_subsys, cai->perf_name, ret, cpu);
 
Signature · 1 line
2.45.2
author maintainer
#11 lore raw

PATCH 11/18 tools/power turbostat: Introduce add_msr_counter()

In reply to Len Brown

From: Zhang Rui <[email protected]>

probe_rapl_msr() is reused for probing RAPL MSR counters, cstate MSR counters and MPERF/APERF/SMI MSR counters, thus its name is misleading.

Similar to add_perf_counter(), introduce add_msr_counter() to probe a counter via MSR. Introduce wrapper function add_rapl_msr_counter() at the same time to add extra check for Zero return value for specified RAPL counters.

No functional change intended.

---
 tools/power/x86/turbostat/turbostat.c | 32 +++++++++++++++++++--------
 1 file changed, 23 insertions(+), 9 deletions(-)
tools/power/x86/turbostat/turbostat.c +23 -9
index ef923da00765..a38cb43ff140 100644
@@ -2222,32 +2222,46 @@ int get_msr(int cpu, off_t offset, unsigned long long *msr)
 	return 0;
 }
 
-int probe_rapl_msr(int cpu, off_t offset, int index)
+int add_msr_counter(int cpu, off_t offset)
 {
 	ssize_t retval;
 	unsigned long long value;
 
-	assert(!no_msr);
+	if (no_msr)
+		return -1;
 
 	retval = pread(get_msr_fd(cpu), &value, sizeof(value), offset);
 
 	/* if the read failed, the probe fails */
 	if (retval != sizeof(value))
-		return 1;
+		return -1;
+
+	if (value == 0)
+		return 0;
+
+	return 1;
+}
+
+int add_rapl_msr_counter(int cpu, off_t offset, int index)
+{
+	int ret;
+
+	ret = add_msr_counter(cpu, offset);
+	if (ret < 0)
+		return -1;
 
-	/* If an Energy Status Counter MSR returns 0, the probe fails */
 	switch (index) {
 	case RAPL_RCI_INDEX_ENERGY_PKG:
 	case RAPL_RCI_INDEX_ENERGY_CORES:
 	case RAPL_RCI_INDEX_DRAM:
 	case RAPL_RCI_INDEX_GFX:
 	case RAPL_RCI_INDEX_ENERGY_PLATFORM:
-		if (value == 0)
+		if (ret == 0)
 			return 1;
 	}
 
 	/* PKG,DRAM_PERF_STATUS MSRs, can return any value */
-	return 0;
+	return 1;
 }
 
 /* Convert CPU ID to domain ID for given added perf counter. */
@@ -7980,7 +7994,7 @@ void rapl_perf_init(void)
 					rci->flags[cai->rci_index] = cai->flags;
 
 					/* Use MSR for this counter */
-				} else if (!no_msr && cai->msr && probe_rapl_msr(cpu, cai->msr, cai->rci_index) == 0) {
+				} else if (!no_msr && cai->msr && add_rapl_msr_counter(cpu, cai->msr, cai->rci_index) >= 0) {
 					rci->source[cai->rci_index] = COUNTER_SOURCE_MSR;
 					rci->msr[cai->rci_index] = cai->msr;
 					rci->msr_mask[cai->rci_index] = cai->msr_mask;
@@ -8110,7 +8124,7 @@ void msr_perf_init_(void)
 					cai->present = true;
 
 					/* User MSR for this counter */
-				} else if (!no_msr && cai->msr && probe_rapl_msr(cpu, cai->msr, cai->rci_index) == 0) {
+				} else if (!no_msr && cai->msr && add_msr_counter(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;
@@ -8224,7 +8238,7 @@ void cstate_perf_init_(bool soft_c1)
 
 					/* User MSR for this counter */
 				} else if (!no_msr && cai->msr && pkg_cstate_limit >= cai->pkg_cstate_limit
-					   && probe_rapl_msr(cpu, cai->msr, cai->rci_index) == 0) {
+					   && add_msr_counter(cpu, cai->msr) >= 0) {
 					cci->source[cai->rci_index] = COUNTER_SOURCE_MSR;
 					cci->msr[cai->rci_index] = cai->msr;
 				}
Signature · 1 line
2.45.2
author maintainer
#12 lore raw

PATCH 12/18 tools/power turbostat: Clean up add perf/msr counter logic

In reply to Len Brown

From: Zhang Rui <[email protected]>

Increase the code readability by moving the no_perf/no_msr flag and the cai->perf_name/cai->msr sanity checks into the counter probe functions.

No functional change.

---
 tools/power/x86/turbostat/turbostat.c | 25 ++++++++++++++++++-------
 1 file changed, 18 insertions(+), 7 deletions(-)
tools/power/x86/turbostat/turbostat.c +18 -7
index a38cb43ff140..5a0c526e33c0 100644
@@ -2230,6 +2230,9 @@ int add_msr_counter(int cpu, off_t offset)
 	if (no_msr)
 		return -1;
 
+	if (!offset)
+		return -1;
+
 	retval = pread(get_msr_fd(cpu), &value, sizeof(value), offset);
 
 	/* if the read failed, the probe fails */
@@ -7875,6 +7878,9 @@ int add_rapl_perf_counter(int cpu, struct rapl_counter_info_t *rci, const struct
 	if (no_perf)
 		return -1;
 
+	if (!cai->perf_name)
+		return -1;
+
 	const double scale = read_perf_scale(cai->perf_subsys, cai->perf_name);
 
 	if (scale == 0.0)
@@ -7986,15 +7992,14 @@ void rapl_perf_init(void)
 			if (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) {
+				if (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 && add_rapl_msr_counter(cpu, cai->msr, cai->rci_index) >= 0) {
+				} else if (add_rapl_msr_counter(cpu, cai->msr, cai->rci_index) >= 0) {
 					rci->source[cai->rci_index] = COUNTER_SOURCE_MSR;
 					rci->msr[cai->rci_index] = cai->msr;
 					rci->msr_mask[cai->rci_index] = cai->msr_mask;
@@ -8042,6 +8047,9 @@ int add_cstate_perf_counter(int cpu, struct cstate_counter_info_t *cci, const st
 	if (no_perf)
 		return -1;
 
+	if (!cai->perf_name)
+		return -1;
+
 	int *pfd_group = get_cstate_perf_group_fd(cci, cai->perf_subsys);
 
 	if (pfd_group == NULL)
@@ -8073,6 +8081,9 @@ int add_msr_perf_counter(int cpu, struct msr_counter_info_t *cci, const struct m
 	if (no_perf)
 		return -1;
 
+	if (!cai->perf_name)
+		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);
 
@@ -8119,12 +8130,12 @@ void msr_perf_init_(void)
 
 			if (cai->needed) {
 				/* Use perf API for this counter */
-				if (!no_perf && cai->perf_name && add_msr_perf_counter(cpu, cci, cai) != -1) {
+				if (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 && add_msr_counter(cpu, cai->msr) >= 0) {
+				} else if (add_msr_counter(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;
@@ -8232,12 +8243,12 @@ void cstate_perf_init_(bool soft_c1)
 
 			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 (add_cstate_perf_counter(cpu, cci, cai) != -1) {
 
 					cci->source[cai->rci_index] = COUNTER_SOURCE_PERF;
 
 					/* User MSR for this counter */
-				} else if (!no_msr && cai->msr && pkg_cstate_limit >= cai->pkg_cstate_limit
+				} else if (pkg_cstate_limit >= cai->pkg_cstate_limit
 					   && add_msr_counter(cpu, cai->msr) >= 0) {
 					cci->source[cai->rci_index] = COUNTER_SOURCE_MSR;
 					cci->msr[cai->rci_index] = cai->msr;
Signature · 1 line
2.45.2
author maintainer
#13 lore raw

PATCH 13/18 tools/power turbostat: Allow probing RAPL with platform_features->rapl_msrs cleared

In reply to Len Brown

From: Zhang Rui <[email protected]>

platform_features->rapl_msrs describes the RAPL MSRs supported. While RAPL Perf counters can be exposed from different kernel backend drivers, e.g. RAPL MSR I/F driver, or RAPL TPMI I/F driver.

Thus, turbostat should first blindly probe all the available RAPL Perf counters, and falls back to the RAPL MSR counters if they are listed in platform_features->rapl_msrs.

With this, platforms that don't have RAPL MSRs can clear the platform_features->rapl_msrs bits and use RAPL Perf counters only.

---
 tools/power/x86/turbostat/turbostat.c | 49 +++++++++++++--------------
 1 file changed, 24 insertions(+), 25 deletions(-)
tools/power/x86/turbostat/turbostat.c +24 -25
index 5a0c526e33c0..6f91ec3f3f14 100644
@@ -2245,15 +2245,18 @@ int add_msr_counter(int cpu, off_t offset)
 	return 1;
 }
 
-int add_rapl_msr_counter(int cpu, off_t offset, int index)
+int add_rapl_msr_counter(int cpu, const struct rapl_counter_arch_info *cai)
 {
 	int ret;
 
-	ret = add_msr_counter(cpu, offset);
+	if (!(platform->rapl_msrs & cai->feature_mask))
+		return -1;
+
+	ret = add_msr_counter(cpu, cai->msr);
 	if (ret < 0)
 		return -1;
 
-	switch (index) {
+	switch (cai->rci_index) {
 	case RAPL_RCI_INDEX_ENERGY_PKG:
 	case RAPL_RCI_INDEX_ENERGY_CORES:
 	case RAPL_RCI_INDEX_DRAM:
@@ -2668,7 +2671,7 @@ void print_header(char *delim)
 	if (DO_BIC(BIC_SYS_LPI))
 		outp += sprintf(outp, "%sSYS%%LPI", (printed++ ? delim : ""));
 
-	if (platform->rapl_msrs && !rapl_joules) {
+	if (!rapl_joules) {
 		if (DO_BIC(BIC_PkgWatt))
 			outp += sprintf(outp, "%sPkgWatt", (printed++ ? delim : ""));
 		if (DO_BIC(BIC_CorWatt) && !platform->has_per_core_rapl)
@@ -2681,7 +2684,7 @@ void print_header(char *delim)
 			outp += sprintf(outp, "%sPKG_%%", (printed++ ? delim : ""));
 		if (DO_BIC(BIC_RAM__))
 			outp += sprintf(outp, "%sRAM_%%", (printed++ ? delim : ""));
-	} else if (platform->rapl_msrs && rapl_joules) {
+	} else {
 		if (DO_BIC(BIC_Pkg_J))
 			outp += sprintf(outp, "%sPkg_J", (printed++ ? delim : ""));
 		if (DO_BIC(BIC_Cor_J) && !platform->has_per_core_rapl)
@@ -7988,26 +7991,22 @@ void rapl_perf_init(void)
 
 			struct rapl_counter_info_t *rci = &rapl_counter_info_perdomain[next_domain];
 
-			/* Check if the counter is enabled and accessible */
-			if (platform->rapl_msrs & cai->feature_mask) {
-
-				/* Use perf API for this counter */
-				if (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 (add_rapl_msr_counter(cpu, cai->msr, cai->rci_index) >= 0) {
-					rci->source[cai->rci_index] = COUNTER_SOURCE_MSR;
-					rci->msr[cai->rci_index] = cai->msr;
-					rci->msr_mask[cai->rci_index] = cai->msr_mask;
-					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;
-				}
+			/* Use perf API for this counter */
+			if (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 (add_rapl_msr_counter(cpu, cai) >= 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)
Signature · 1 line
2.45.2
author maintainer
#14 lore raw

PATCH 14/18 tools/power turbostat: Avoid probing the same perf counters

In reply to Len Brown

From: Zhang Rui <[email protected]>

For the RAPL package energy status counter, Intel and AMD share the same perf_subsys and perf_name, but with different MSR addresses.

Both rapl_counter_arch_infos[0] and rapl_counter_arch_infos[1] are introduced to describe this counter for different Vendors.

As a result, the perf counter is probed twice, and causes a failure in in get_rapl_counters() because expected_read_size and actual_read_size don't match.

Fix the problem by skipping the already probed counter.

Note, this is not a perfect fix. For example, if different vendors/platforms use the same MSR value for different purpose, the code can be fooled when it probes a rapl_counter_arch_infos[] entry that does not belong to the running Vendor/Platform.

In a long run, better to put rapl_counter_arch_infos[] into the platform_features so that this becomes Vendor/Platform specific.

---
 tools/power/x86/turbostat/turbostat.c | 15 +++++++++++++++
 1 file changed, 15 insertions(+)
tools/power/x86/turbostat/turbostat.c +15 -0
index 6f91ec3f3f14..8deb6a23c7dd 100644
@@ -7991,6 +7991,21 @@ void rapl_perf_init(void)
 
 			struct rapl_counter_info_t *rci = &rapl_counter_info_perdomain[next_domain];
 
+			/*
+			 * rapl_counter_arch_infos[] can have multiple entries describing the same
+			 * counter, due to the difference from different platforms/Vendors.
+			 * E.g. rapl_counter_arch_infos[0] and rapl_counter_arch_infos[1] share the
+			 * same perf_subsys and perf_name, but with different MSR address.
+			 * rapl_counter_arch_infos[0] is for Intel and rapl_counter_arch_infos[1]
+			 * is for AMD.
+			 * In this case, it is possible that multiple rapl_counter_arch_infos[]
+			 * entries are probed just because their perf/msr is duplicate and valid.
+			 *
+			 * Thus need a check to avoid re-probe the same counters.
+			 */
+			if (rci->source[cai->rci_index] != COUNTER_SOURCE_NONE)
+				break;
+
 			/* Use perf API for this counter */
 			if (add_rapl_perf_counter(cpu, rci, cai, &scale, &unit) != -1) {
 				rci->source[cai->rci_index] = COUNTER_SOURCE_PERF;
Signature · 1 line
2.45.2
author maintainer
#15 lore raw

PATCH 15/18 tools/power turbostat: Dump RAPL sysfs info

In reply to Len Brown

From: Zhang Rui <[email protected]>

for example:

intel-rapl:1: psys 28.0s:100W 976.0us:100W intel-rapl:0: package-0 28.0s:57W,max:15W 2.4ms:57W intel-rapl:0/intel-rapl:0:0: core disabled intel-rapl:0/intel-rapl:0:1: uncore disabled intel-rapl-mmio:0: package-0 28.0s:28W,max:15W 2.4ms:57W

[lenb: simplified format]

squish me

---
 tools/power/x86/turbostat/turbostat.c | 156 ++++++++++++++++++++++++++
 1 file changed, 156 insertions(+)
tools/power/x86/turbostat/turbostat.c +156 -0
index 8deb6a23c7dd..30faa214676f 100644
@@ -7404,6 +7404,160 @@ void print_power_limit_msr(int cpu, unsigned long long msr, char *label)
 	return;
 }
 
+static int fread_int(char *path, int *val)
+{
+	FILE *filep;
+	int ret;
+
+	filep = fopen (path, "r");
+	if (!filep)
+		return -1;
+
+	ret = fscanf(filep, "%d", val);
+	fclose(filep);
+	return ret;
+}
+
+static int fread_ull(char *path, unsigned long long *val)
+{
+	FILE *filep;
+	int ret;
+
+	filep = fopen (path, "r");
+	if (!filep)
+		return -1;
+
+	ret = fscanf(filep, "%llu", val);
+	fclose(filep);
+	return ret;
+}
+
+static int fread_str(char *path, char *buf, int size)
+{
+	FILE *filep;
+	int ret;
+	char *cp;
+
+	filep = fopen (path, "r");
+	if (!filep)
+		return -1;
+
+	ret = fread(buf, 1, size, filep);
+	fclose(filep);
+
+	/* replace '\n' with '\0' */
+	cp = strchr(buf, '\n');
+	if (cp != NULL)
+		*cp = '\0';
+
+	return ret;
+}
+
+#define PATH_RAPL_SYSFS	"/sys/class/powercap"
+
+static int dump_one_domain(char *domain_path)
+{
+	char path[PATH_MAX];
+	char str[PATH_MAX];
+	unsigned long long val;
+	int constraint;
+	int enable;
+	int ret;
+
+	snprintf(path, PATH_MAX, "%s/name", domain_path);
+	ret = fread_str(path, str, PATH_MAX);
+	if (ret <= 0)
+		return -1;
+
+	fprintf(outf, "%s: %s", domain_path + strlen(PATH_RAPL_SYSFS) + 1, str);
+
+	snprintf(path, PATH_MAX, "%s/enabled", domain_path);
+	ret = fread_int(path, &enable);
+	if (ret <= 0)
+		return -1;
+
+	if (!enable) {
+		fputs(" disabled\n", outf);
+		return 0;
+	}
+
+	for (constraint = 0;; constraint++)
+	{
+		snprintf(path, PATH_MAX, "%s/constraint_%d_time_window_us", domain_path, constraint);
+		ret = fread_ull(path, &val);
+		if (ret <= 0)
+			break;
+
+		if (val > 1000000)
+			fprintf(outf, " %0.1fs", (double)val/1000000);
+		else if (val > 1000)
+			fprintf(outf, " %0.1fms", (double)val/1000);
+		else
+			fprintf(outf, " %0.1fus", (double)val);
+
+		snprintf(path, PATH_MAX, "%s/constraint_%d_power_limit_uw", domain_path, constraint);
+		ret = fread_ull(path, &val);
+		if (ret > 0 && val)
+			fprintf(outf, ":%lluW", val / 1000000);
+
+		snprintf(path, PATH_MAX, "%s/constraint_%d_max_power_uw", domain_path, constraint);
+		ret = fread_ull(path, &val);
+		if (ret > 0 && val)
+			fprintf(outf, ",max:%lluW", val / 1000000);
+	}
+	fputc('\n', outf);
+
+	return 0;
+}
+
+static int print_rapl_sysfs(void)
+{
+	DIR *dir, *cdir;
+	struct dirent *entry, *centry;
+	char path[PATH_MAX];
+	char str[PATH_MAX];
+
+	if ((dir = opendir(PATH_RAPL_SYSFS)) == NULL) {
+		warn("open %s failed", PATH_RAPL_SYSFS);
+		return 1;
+	}
+
+	while ((entry = readdir (dir)) != NULL) {
+		if (strlen(entry->d_name) > 100)
+                        continue;
+
+		if (strncmp(entry->d_name, "intel-rapl", strlen("intel-rapl")))
+                        continue;
+
+		snprintf(path, PATH_MAX, "%s/%s/name", PATH_RAPL_SYSFS, entry->d_name);
+
+		/* Parse top level domains first, including package and psys */
+		fread_str(path, str, PATH_MAX);
+		if (strncmp(str, "package", strlen("package")) &&
+		    strncmp(str, "psys", strlen("psys")))
+			continue;
+
+		snprintf(path, PATH_MAX, "%s/%s", PATH_RAPL_SYSFS, entry->d_name);
+		if ((cdir = opendir(path)) == NULL) {
+			perror ("opendir() error");
+			return 1;
+		}
+
+		dump_one_domain(path);
+
+		while ((centry = readdir (cdir)) != NULL) {
+			if (strncmp(centry->d_name, "intel-rapl", strlen("intel-rapl")))
+				continue;
+			snprintf(path, PATH_MAX, "%s/%s/%s", PATH_RAPL_SYSFS, entry->d_name, centry->d_name);
+			dump_one_domain(path);
+		}
+		closedir(cdir);
+	}
+
+	closedir(dir);
+	return 0;
+}
+
 int print_rapl(struct thread_data *t, struct core_data *c, struct pkg_data *p)
 {
 	unsigned long long msr;
@@ -7538,6 +7692,8 @@ void probe_rapl(void)
 	if (quiet)
 		return;
 
+	print_rapl_sysfs();
+
 	if (!platform->rapl_msrs || no_msr)
 		return;
 
Signature · 1 line
2.45.2
author maintainer
#16 lore raw

PATCH 16/18 tools/power turbostat: Add initial support for DMR

In reply to Len Brown

From: Zhang Rui <[email protected]>

Add initial support for DMR.

---
 tools/power/x86/turbostat/turbostat.c | 18 ++++++++++++++++++
 1 file changed, 18 insertions(+)
tools/power/x86/turbostat/turbostat.c +18 -0
index 30faa214676f..dca0753a76fb 100644
@@ -839,6 +839,23 @@ static const struct platform_features spr_features = {
 	.rapl_msrs = RAPL_PKG_ALL | RAPL_DRAM_ALL | RAPL_PSYS,
 };
 
+static const struct platform_features dmr_features = {
+	.has_msr_misc_feature_control = spr_features.has_msr_misc_feature_control,
+	.has_msr_misc_pwr_mgmt = spr_features.has_msr_misc_pwr_mgmt,
+	.has_nhm_msrs = spr_features.has_nhm_msrs,
+	.has_config_tdp = spr_features.has_config_tdp,
+	.bclk_freq = spr_features.bclk_freq,
+	.supported_cstates = spr_features.supported_cstates,
+	.cst_limit = spr_features.cst_limit,
+	.has_msr_core_c1_res = spr_features.has_msr_core_c1_res,
+	.has_msr_module_c6_res_ms = 1,	/* DMR has Dual Core Module and MC6 MSR */
+	.has_irtl_msrs = spr_features.has_irtl_msrs,
+	.has_cst_prewake_bit = spr_features.has_cst_prewake_bit,
+	.has_fixed_rapl_psys_unit = spr_features.has_fixed_rapl_psys_unit,
+	.trl_msrs = spr_features.trl_msrs,
+	.rapl_msrs = 0,		/* DMR does not have RAPL MSRs */
+};
+
 static const struct platform_features srf_features = {
 	.has_msr_misc_feature_control = 1,
 	.has_msr_misc_pwr_mgmt = 1,
@@ -1028,6 +1045,7 @@ static const struct platform_data turbostat_pdata[] = {
 	{ INTEL_EMERALDRAPIDS_X, &spr_features },
 	{ INTEL_GRANITERAPIDS_X, &spr_features },
 	{ INTEL_GRANITERAPIDS_D, &spr_features },
+	{ INTEL_PANTHERCOVE_X, &dmr_features },
 	{ INTEL_LAKEFIELD, &cnl_features },
 	{ INTEL_ALDERLAKE, &adl_features },
 	{ INTEL_ALDERLAKE_L, &adl_features },
Signature · 1 line
2.45.2
author maintainer
#17 lore raw

PATCH 17/18 tools/power turbostat: Add initial support for BartlettLake

In reply to Len Brown

From: Zhang Rui <[email protected]>

Add initial support for BartlettLake.

---
 tools/power/x86/turbostat/turbostat.c | 1 +
 1 file changed, 1 insertion(+)
tools/power/x86/turbostat/turbostat.c +1 -0
index dca0753a76fb..db0134adc8ce 100644
@@ -1052,6 +1052,7 @@ static const struct platform_data turbostat_pdata[] = {
 	{ INTEL_RAPTORLAKE, &adl_features },
 	{ INTEL_RAPTORLAKE_P, &adl_features },
 	{ INTEL_RAPTORLAKE_S, &adl_features },
+	{ INTEL_BARTLETTLAKE, &adl_features },
 	{ INTEL_METEORLAKE, &adl_features },
 	{ INTEL_METEORLAKE_L, &adl_features },
 	{ INTEL_ARROWLAKE_H, &adl_features },
Signature · 1 line
2.45.2
author maintainer
#18 lore raw

PATCH 18/18 tools/power turbostat: version 2025.06.08

In reply to Len Brown

From: Len Brown <[email protected]>

Add initial DMR support, which required smarter RAPL probe Fix AMD MSR RAPL energy reporting
Add RAPL power limit configuration output
Minor fixes

---
 tools/power/x86/turbostat/turbostat.c | 73 +++++++++++++--------------
 1 file changed, 36 insertions(+), 37 deletions(-)
tools/power/x86/turbostat/turbostat.c +36 -37
index db0134adc8ce..5230e072e414 100644
@@ -280,7 +280,7 @@ struct msr_counter bic[] = {
 #define BIC_GROUP_FREQUENCY (BIC_Avg_MHz | BIC_Busy | BIC_Bzy_MHz | BIC_TSC_MHz | BIC_GFXMHz | BIC_GFXACTMHz | BIC_SAMMHz | BIC_SAMACTMHz | BIC_UNCORE_MHZ)
 #define BIC_GROUP_HW_IDLE (BIC_Busy | BIC_CPU_c1 | BIC_CPU_c3 | BIC_CPU_c6 | BIC_CPU_c7 | BIC_GFX_rc6 | BIC_Pkgpc2 | BIC_Pkgpc3 | BIC_Pkgpc6 | BIC_Pkgpc7 | BIC_Pkgpc8 | BIC_Pkgpc9 | BIC_Pkgpc10 | BIC_CPU_LPI | BIC_SYS_LPI | BIC_Mod_c6 | BIC_Totl_c0 | BIC_Any_c0 | BIC_GFX_c0 | BIC_CPUGFX | BIC_SAM_mc6 | BIC_Diec6)
 #define BIC_GROUP_SW_IDLE (BIC_Busy | BIC_cpuidle | BIC_pct_idle )
-#define BIC_GROUP_IDLE (BIC_GROUP_HW_IDLE | BIC_pct_idle) 
+#define BIC_GROUP_IDLE (BIC_GROUP_HW_IDLE | BIC_pct_idle)
 #define BIC_OTHER (BIC_IRQ | BIC_NMI | BIC_SMI | BIC_ThreadC | BIC_CoreTmp | BIC_IPC)
 
 #define BIC_DISABLED_BY_DEFAULT	(BIC_USEC | BIC_TOD | BIC_APIC | BIC_X2APIC | BIC_cpuidle)
@@ -1091,7 +1091,6 @@ void probe_platform_features(unsigned int family, unsigned int model)
 {
 	int i;
 
-
 	if (authentic_amd || hygon_genuine) {
 		/* fallback to default features on unsupported models */
 		force_load++;
@@ -1125,8 +1124,7 @@ void probe_platform_features(unsigned int family, unsigned int model)
 	if (platform)
 		return;
 
-	fprintf(stderr, "Unsupported platform detected.\n"
-		"\tSee RUN THE LATEST VERSION on turbostat(8)\n");
+	fprintf(stderr, "Unsupported platform detected.\n\tSee RUN THE LATEST VERSION on turbostat(8)\n");
 	exit(1);
 }
 
@@ -1146,7 +1144,8 @@ char *progname;
 
 #define CPU_SUBSET_MAXCPUS	8192	/* need to use before probe... */
 cpu_set_t *cpu_present_set, *cpu_possible_set, *cpu_effective_set, *cpu_allowed_set, *cpu_affinity_set, *cpu_subset;
-size_t cpu_present_setsize, cpu_possible_setsize, cpu_effective_setsize, cpu_allowed_setsize, cpu_affinity_setsize, cpu_subset_size;
+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
@@ -2373,8 +2372,7 @@ void help(void)
 		"		  degrees Celsius\n"
 		"  -h, --help\n"
 		"		print this help message\n"
-		"  -v, --version\n"
-		"		print version information\n\nFor more help, run \"man turbostat\"\n");
+		"  -v, --version\n\t\tprint version information\n\nFor more help, run \"man turbostat\"\n");
 }
 
 /*
@@ -3989,7 +3987,6 @@ void compute_average(struct thread_data *t, struct core_data *c, struct pkg_data
 	if (average.threads.nmi_count > 9999999)
 		sums_need_wide_columns = 1;
 
-
 	average.cores.c3 /= topo.allowed_cores;
 	average.cores.c6 /= topo.allowed_cores;
 	average.cores.c7 /= topo.allowed_cores;
@@ -4812,7 +4809,6 @@ unsigned long pmt_read_counter(struct pmt_counter *ppmt, unsigned int domain_id)
 	return (value & value_mask) >> value_shift;
 }
 
-
 /* Rapl domain enumeration helpers */
 static inline int get_rapl_num_domains(void)
 {
@@ -6822,8 +6818,10 @@ static void probe_intel_uncore_frequency_cluster(void)
 		 * This allows "--show/--hide UncMHz" to be effective for
 		 * the clustered MHz counters, as a group.
 		 */
-		if BIC_IS_ENABLED(BIC_UNCORE_MHZ)
-			add_counter(0, path, name_buf, 0, SCOPE_PACKAGE, COUNTER_K2M, FORMAT_AVERAGE, 0, package_id);
+		if BIC_IS_ENABLED
+			(BIC_UNCORE_MHZ)
+			    add_counter(0, path, name_buf, 0, SCOPE_PACKAGE, COUNTER_K2M, FORMAT_AVERAGE, 0,
+					package_id);
 
 		if (quiet)
 			continue;
@@ -6895,17 +6893,21 @@ static void probe_graphics(void)
 		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/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/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/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);
+		set_graphics_fp("/sys/class/drm/card0/device/tile0/gt1/freq0/act_freq",
+				gt0_is_gt ? SAM_ACTMHz : GFX_ACTMHz);
 
 		goto end;
 	}
@@ -7428,7 +7430,7 @@ static int fread_int(char *path, int *val)
 	FILE *filep;
 	int ret;
 
-	filep = fopen (path, "r");
+	filep = fopen(path, "r");
 	if (!filep)
 		return -1;
 
@@ -7442,7 +7444,7 @@ static int fread_ull(char *path, unsigned long long *val)
 	FILE *filep;
 	int ret;
 
-	filep = fopen (path, "r");
+	filep = fopen(path, "r");
 	if (!filep)
 		return -1;
 
@@ -7457,7 +7459,7 @@ static int fread_str(char *path, char *buf, int size)
 	int ret;
 	char *cp;
 
-	filep = fopen (path, "r");
+	filep = fopen(path, "r");
 	if (!filep)
 		return -1;
 
@@ -7500,17 +7502,16 @@ static int dump_one_domain(char *domain_path)
 		return 0;
 	}
 
-	for (constraint = 0;; constraint++)
-	{
+	for (constraint = 0;; constraint++) {
 		snprintf(path, PATH_MAX, "%s/constraint_%d_time_window_us", domain_path, constraint);
 		ret = fread_ull(path, &val);
 		if (ret <= 0)
 			break;
 
 		if (val > 1000000)
-			fprintf(outf, " %0.1fs", (double)val/1000000);
+			fprintf(outf, " %0.1fs", (double)val / 1000000);
 		else if (val > 1000)
-			fprintf(outf, " %0.1fms", (double)val/1000);
+			fprintf(outf, " %0.1fms", (double)val / 1000);
 		else
 			fprintf(outf, " %0.1fus", (double)val);
 
@@ -7541,30 +7542,29 @@ static int print_rapl_sysfs(void)
 		return 1;
 	}
 
-	while ((entry = readdir (dir)) != NULL) {
+	while ((entry = readdir(dir)) != NULL) {
 		if (strlen(entry->d_name) > 100)
-                        continue;
+			continue;
 
 		if (strncmp(entry->d_name, "intel-rapl", strlen("intel-rapl")))
-                        continue;
+			continue;
 
 		snprintf(path, PATH_MAX, "%s/%s/name", PATH_RAPL_SYSFS, entry->d_name);
 
 		/* Parse top level domains first, including package and psys */
 		fread_str(path, str, PATH_MAX);
-		if (strncmp(str, "package", strlen("package")) &&
-		    strncmp(str, "psys", strlen("psys")))
+		if (strncmp(str, "package", strlen("package")) && strncmp(str, "psys", strlen("psys")))
 			continue;
 
 		snprintf(path, PATH_MAX, "%s/%s", PATH_RAPL_SYSFS, entry->d_name);
 		if ((cdir = opendir(path)) == NULL) {
-			perror ("opendir() error");
+			perror("opendir() error");
 			return 1;
 		}
 
 		dump_one_domain(path);
 
-		while ((centry = readdir (cdir)) != NULL) {
+		while ((centry = readdir(cdir)) != NULL) {
 			if (strncmp(centry->d_name, "intel-rapl", strlen("intel-rapl")))
 				continue;
 			snprintf(path, PATH_MAX, "%s/%s/%s", PATH_RAPL_SYSFS, entry->d_name, centry->d_name);
@@ -8049,7 +8049,7 @@ static int has_instr_count_access(void)
 }
 
 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_)
+			  double *scale_, enum rapl_unit *unit_)
 {
 	int ret = -1;
 
@@ -8188,7 +8188,7 @@ void rapl_perf_init(void)
 				rci->unit[cai->rci_index] = unit;
 				rci->flags[cai->rci_index] = cai->flags;
 
-			/* Use MSR for this counter */
+				/* Use MSR for this counter */
 			} else if (add_rapl_msr_counter(cpu, cai) >= 0) {
 				rci->source[cai->rci_index] = COUNTER_SOURCE_MSR;
 				rci->msr[cai->rci_index] = cai->msr;
@@ -9299,15 +9299,14 @@ int added_perf_counters_init_(struct perf_counter_info *pinfo)
 					perf_device = "cpu_atom";
 					break;
 
-				default: /* Don't change, we will probably fail and report a problem soon. */
+				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");
+				warnx("%s: perf/%s/%s: failed to read %s", __func__, perf_device, pinfo->event, "type");
 				continue;
 			}
 
@@ -9409,7 +9408,7 @@ struct pmt_mmio *pmt_mmio_open(unsigned int target_guid)
 		return NULL;
 	}
 
-	for ( ; entry != NULL; entry = pmt_diriter_next(&pmt_iter)) {
+	for (; entry != NULL; entry = pmt_diriter_next(&pmt_iter)) {
 		if (fstatat(dirfd(pmt_iter.dir), entry->d_name, &st, 0) == -1)
 			break;
 
@@ -9849,7 +9848,7 @@ int get_and_dump_counters(void)
 
 void print_version()
 {
-	fprintf(outf, "turbostat version 2025.04.06 - Len Brown <[email protected]>\n");
+	fprintf(outf, "turbostat version 2025.06.08 - Len Brown <[email protected]>\n");
 }
 
 #define COMMAND_LINE_SIZE 2048
@@ -10305,7 +10304,7 @@ void parse_add_command_pmt(char *add_command)
 	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 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;
Signature · 1 line
2.45.2