[PATCH 0/3] cpupower: Improve arm64 frequency and CPPC reporting

[PATCH 0/3] cpupower: Improve arm64 frequency and CPPC reporting

[Jeremy Linton]

On x86, cpupower can report both kernel requested frequency
information and hardware reported frequency information, along with
CPPC related performance limits when available.

Arm64 platforms using ACPI CPPC, expose similar standardized
information, but cpupower does not currently make it visible.

This set adds a generic CPPC performance display based on the
standardized ACPI CPPC sysfs fields, wires it up for non AMD
processors, and updates debug frequency output so cpupower prints both
kernel reported and hardware reported current frequencies when both
are available.

Jeremy Linton (3):
  cpupower: Add generic CPPC performance display
  cpupower: Build and call CPPC information on non-AMD processors
  cpupower: Print kernel and hardware frequency information

 tools/power/cpupower/Makefile                |  2 +-
 tools/power/cpupower/utils/cpufreq-info.c    | 11 ++--
 tools/power/cpupower/utils/helpers/cppc.c    | 56 ++++++++++++++++++++
 tools/power/cpupower/utils/helpers/helpers.h |  2 +
 4 files changed, 65 insertions(+), 6 deletions(-)
 create mode 100644 tools/power/cpupower/utils/helpers/cppc.c

-- 
2.54.0

[PATCH 1/3] cpupower: Add generic CPPC performance display

[Jeremy Linton]

Arm64 machines, and possibly others, use the standard ACPI defined CPPC
infrastructure. cpupower has CPPC support, but it's largely written
around the intricacies of AMD processors, using platform MSRs to avoid
shortcomings in the specification.

Add a generic CPPC display that depends only on standardized fields. The
computed frequency values are best effort and rely on the FW providing
optional values that can be used to derive a meaningful frequency at a
given unique performance level.

Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 tools/power/cpupower/utils/helpers/cppc.c | 56 +++++++++++++++++++++++
 1 file changed, 56 insertions(+)
 create mode 100644 tools/power/cpupower/utils/helpers/cppc.c

diff --git a/tools/power/cpupower/utils/helpers/cppc.c b/tools/power/cpupower/utils/helpers/cppc.c
new file mode 100644
index 000000000000..3493ce8551ea
--- /dev/null
+++ b/tools/power/cpupower/utils/helpers/cppc.c
@@ -0,0 +1,56 @@
+// SPDX-License-Identifier: GPL-2.0
+
+#include <stdio.h>
+#include <errno.h>
+#include <stdlib.h>
+#include <string.h>
+#include <math.h>
+
+#include "helpers/helpers.h"
+#include "cpufreq.h"
+#include "acpi_cppc.h"
+
+#define cppc_to_frequency(perf) (roundf(slope * (perf) + intercept))
+
+void cppc_show_perf_and_freq(unsigned int cpu, int no_rounding)
+{
+	int64_t nominal = acpi_cppc_get_data(cpu, NOMINAL_PERF);
+	int64_t nominal_freq = acpi_cppc_get_data(cpu, NOMINAL_FREQ) * 1000;
+	int64_t lowest = acpi_cppc_get_data(cpu, LOWEST_PERF);
+	int64_t lowest_freq = acpi_cppc_get_data(cpu, LOWEST_FREQ) * 1000;
+	unsigned long non_linear = acpi_cppc_get_data(cpu, LOWEST_NONLINEAR_PERF);
+	unsigned long highest = acpi_cppc_get_data(cpu, HIGHEST_PERF);
+	float slope, intercept;
+
+	/* do the optional freq fields look invalid? */
+	if (!nominal_freq || !lowest_freq || nominal == lowest)
+		return;
+
+	slope = (float)(nominal_freq - lowest_freq) / (nominal - lowest);
+	intercept = lowest_freq - slope * lowest;
+
+	printf(_("  CPPC limits:\n"));
+	printf(_("    Highest Performance: %lu. Maximum Frequency: "),
+	       highest);
+	/*
+	 * If boost isn't active, the cpuinfo_max doesn't indicate real max
+	 * frequency.
+	 */
+	print_speed(cppc_to_frequency(highest), no_rounding);
+	printf(".\n");
+
+	printf(_("    Nominal Performance: %lu. Nominal Frequency: "),
+	       acpi_cppc_get_data(cpu, NOMINAL_PERF));
+	print_speed(nominal_freq,  no_rounding);
+	printf(".\n");
+
+	printf(_("    Lowest Non-linear Performance: %lu. Lowest Non-linear Frequency: "),
+	       non_linear);
+	print_speed(cppc_to_frequency(non_linear), no_rounding);
+	printf(".\n");
+
+	printf(_("    Lowest Performance: %lu. Lowest Frequency: "),
+	       acpi_cppc_get_data(cpu, LOWEST_PERF));
+	print_speed(lowest_freq, no_rounding);
+	printf(".\n");
+}
-- 
2.54.0

[PATCH 2/3] cpupower: Build and call CPPC information on non-AMD processors

[Jeremy Linton]

Now that we have a generic CPPC printout, call it on !AMD
processors. If it fails to detect CPPC, or the registers don't look
reasonable then it will exit without printing anything.

Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 tools/power/cpupower/Makefile                | 2 +-
 tools/power/cpupower/utils/cpufreq-info.c    | 3 ++-
 tools/power/cpupower/utils/helpers/helpers.h | 2 ++
 3 files changed, 5 insertions(+), 2 deletions(-)

diff --git a/tools/power/cpupower/Makefile b/tools/power/cpupower/Makefile
index 969716dfe8de..cd8b7315fe74 100644
--- a/tools/power/cpupower/Makefile
+++ b/tools/power/cpupower/Makefile
@@ -131,7 +131,7 @@ override CFLAGS += -DVERSION=\"$(VERSION)\" -DPACKAGE=\"$(PACKAGE)\" \
 
 UTIL_OBJS =  utils/helpers/amd.o utils/helpers/msr.o \
 	utils/helpers/sysfs.o utils/helpers/misc.o utils/helpers/cpuid.o \
-	utils/helpers/pci.o utils/helpers/bitmask.o \
+	utils/helpers/pci.o utils/helpers/bitmask.o utils/helpers/cppc.o \
 	utils/idle_monitor/nhm_idle.o utils/idle_monitor/snb_idle.o \
 	utils/idle_monitor/hsw_ext_idle.o \
 	utils/idle_monitor/amd_fam14h_idle.o utils/idle_monitor/cpuidle_sysfs.o \
diff --git a/tools/power/cpupower/utils/cpufreq-info.c b/tools/power/cpupower/utils/cpufreq-info.c
index 5a242b491a9d..105f06e690cc 100644
--- a/tools/power/cpupower/utils/cpufreq-info.c
+++ b/tools/power/cpupower/utils/cpufreq-info.c
@@ -477,12 +477,13 @@ static int get_latency(unsigned int cpu, unsigned int human)
 }
 
 /* --performance / -c */
-
 static int get_perf_cap(unsigned int cpu)
 {
 	if (cpupower_cpu_info.vendor == X86_VENDOR_AMD &&
 	    cpupower_cpu_info.caps & CPUPOWER_CAP_AMD_PSTATE)
 		amd_pstate_show_perf_and_freq(cpu, no_rounding);
+	else
+		cppc_show_perf_and_freq(cpu, no_rounding);
 
 	return 0;
 }
diff --git a/tools/power/cpupower/utils/helpers/helpers.h b/tools/power/cpupower/utils/helpers/helpers.h
index a3ad80b9c2c2..9c5126b63966 100644
--- a/tools/power/cpupower/utils/helpers/helpers.h
+++ b/tools/power/cpupower/utils/helpers/helpers.h
@@ -221,4 +221,6 @@ void print_online_cpus(void);
 void print_offline_cpus(void);
 void print_speed(unsigned long speed, int no_rounding);
 
+void cppc_show_perf_and_freq(unsigned int cpu, int no_rounding);
+
 #endif /* __CPUPOWERUTILS_HELPERS__ */
-- 
2.54.0

[PATCH 3/3] cpupower: Print kernel and hardware frequency information

[Jeremy Linton]

The kernel asserted frequency from scaling_cur_freq may not always match
the hardware reported frequency from cpuinfo_cur_freq.

Print both values when they are available, and only print the unavailable
message on x86 when the hardware frequency can't be read.

Signed-off-by: Jeremy Linton <jeremy.linton@arm.com>
---
 tools/power/cpupower/utils/cpufreq-info.c | 8 ++++----
 1 file changed, 4 insertions(+), 4 deletions(-)

diff --git a/tools/power/cpupower/utils/cpufreq-info.c b/tools/power/cpupower/utils/cpufreq-info.c
index 105f06e690cc..11629ae49f98 100644
--- a/tools/power/cpupower/utils/cpufreq-info.c
+++ b/tools/power/cpupower/utils/cpufreq-info.c
@@ -270,10 +270,10 @@ static int get_freq_hardware(unsigned int cpu, unsigned int human)
 {
 	unsigned long freq;
 
-	if (!(cpupower_cpu_info.caps & CPUPOWER_CAP_APERF))
+	freq = cpufreq_get_freq_hardware(cpu);
+	if (!(cpupower_cpu_info.caps & CPUPOWER_CAP_APERF) && !freq)
 		return -EINVAL;
 
-	freq = cpufreq_get_freq_hardware(cpu);
 	printf(_("  current CPU frequency: "));
 	if (!freq) {
 		printf("Unable to call hardware\n");
@@ -514,8 +514,8 @@ static void debug_output_one(unsigned int cpu)
 
 	get_available_governors(cpu);
 	get_policy(cpu);
-	if (get_freq_hardware(cpu, 1) < 0)
-		get_freq_kernel(cpu, 1);
+	get_freq_hardware(cpu, 1);
+	get_freq_kernel(cpu, 1);
 	get_boost_mode(cpu);
 	get_perf_cap(cpu);
 }
-- 
2.54.0

Re: [PATCH 0/3] cpupower: Improve arm64 frequency and CPPC reporting

[Shuah]

On 7/9/26 16:13, Jeremy Linton wrote:
> On x86, cpupower can report both kernel requested frequency
> information and hardware reported frequency information, along with
> CPPC related performance limits when available.
> 
> Arm64 platforms using ACPI CPPC, expose similar standardized
> information, but cpupower does not currently make it visible.
> 
> This set adds a generic CPPC performance display based on the
> standardized ACPI CPPC sysfs fields, wires it up for non AMD
> processors, and updates debug frequency output so cpupower prints both
> kernel reported and hardware reported current frequencies when both
> are available.
> 
> Jeremy Linton (3):
>    cpupower: Add generic CPPC performance display
>    cpupower: Build and call CPPC information on non-AMD processors
>    cpupower: Print kernel and hardware frequency information
> 
>   tools/power/cpupower/Makefile                |  2 +-
>   tools/power/cpupower/utils/cpufreq-info.c    | 11 ++--
>   tools/power/cpupower/utils/helpers/cppc.c    | 56 ++++++++++++++++++++
>   tools/power/cpupower/utils/helpers/helpers.h |  2 +
>   4 files changed, 65 insertions(+), 6 deletions(-)
>   create mode 100644 tools/power/cpupower/utils/helpers/cppc.c
> 

Applied to https://web.git.kernel.org/pub/scm/linux/kernel/git/shuah/linux.git/log/?h=cpupower
for Linux 7.3-rc1.

thanks,
-- Shuah

Re: [PATCH 1/3] cpupower: Add generic CPPC performance display

[Thorsten Leemhuis]

On 7/10/26 00:13, Jeremy Linton wrote:
> Arm64 machines, and possibly others, use the standard ACPI defined CPPC
> infrastructure. cpupower has CPPC support, but it's largely written
> around the intricacies of AMD processors, using platform MSRs to avoid
> shortcomings in the specification.
> 
> [...]
> +#include <stdio.h>
> +#include <errno.h>
> +#include <stdlib.h>
> +#include <string.h>
> +#include <math.h>
> +
> +#include "helpers/helpers.h"
> +#include "cpufreq.h"
> +#include "acpi_cppc.h"
> +
> +#define cppc_to_frequency(perf) (roundf(slope * (perf) + intercept))
FYI, this patch showed up in -next yesterday and I suspect it caused
this build error for me on x86_64:

"""  CC       cpupower
/usr/bin/ld.bfd: /tmp/ccjOeWLF.ltrans0.ltrans.o: in function
`get_perf_cap.isra.0':
/builddir/build/BUILD/kernel-7.2.0-build/kernel-next-20260717/linux-7.2.0-0.0.next.20260717.330.vanilla.fc44.x86_64/tools/power/cpupower/utils/helpers/cppc.c:39:(.text+0x45cc):
undefined reference to `roundf'
/usr/bin/ld.bfd:
/tmp/ccjOeWLF.ltrans0.ltrans.o:/builddir/build/BUILD/kernel-7.2.0-build/kernel-next-20260717/linux-7.2.0-0.0.next.20260717.330.vanilla.fc44.x86_64/tools/power/cpupower/utils/helpers/cppc.c:49:(.text+0x46a2):
undefined reference to `roundf'
collect2: error: ld returned 1 exit status
make: *** [Makefile:238: cpupower] Error 1
"""
> [...]

Full log:
https://download.copr.fedorainfracloud.org/results/@kernel-vanilla/next/fedora-44-x86_64/10739303-next-next-all/builder-live.log.gz

Ciao, Thorsten

Re: [PATCH 1/3] cpupower: Add generic CPPC performance display

[Jeremy Linton]

On 7/18/26 1:12 AM, Thorsten Leemhuis wrote:
> On 7/10/26 00:13, Jeremy Linton wrote:
>> Arm64 machines, and possibly others, use the standard ACPI defined CPPC
>> infrastructure. cpupower has CPPC support, but it's largely written
>> around the intricacies of AMD processors, using platform MSRs to avoid
>> shortcomings in the specification.
>>
>> [...]
>> +#include <stdio.h>
>> +#include <errno.h>
>> +#include <stdlib.h>
>> +#include <string.h>
>> +#include <math.h>
>> +
>> +#include "helpers/helpers.h"
>> +#include "cpufreq.h"
>> +#include "acpi_cppc.h"
>> +
>> +#define cppc_to_frequency(perf) (roundf(slope * (perf) + intercept))
> FYI, this patch showed up in -next yesterday and I suspect it caused
> this build error for me on x86_64:
> 
> """  CC       cpupower
> /usr/bin/ld.bfd: /tmp/ccjOeWLF.ltrans0.ltrans.o: in function
> `get_perf_cap.isra.0':
> /builddir/build/BUILD/kernel-7.2.0-build/kernel-next-20260717/linux-7.2.0-0.0.next.20260717.330.vanilla.fc44.x86_64/tools/power/cpupower/utils/helpers/cppc.c:39:(.text+0x45cc):
> undefined reference to `roundf'
> /usr/bin/ld.bfd:
> /tmp/ccjOeWLF.ltrans0.ltrans.o:/builddir/build/BUILD/kernel-7.2.0-build/kernel-next-20260717/linux-7.2.0-0.0.next.20260717.330.vanilla.fc44.x86_64/tools/power/cpupower/utils/helpers/cppc.c:49:(.text+0x46a2):
> undefined reference to `roundf'
> collect2: error: ld returned 1 exit status
> make: *** [Makefile:238: cpupower] Error 1

Right, when its not inline it needs -lm, I will toss a patch against it 
in a moment.


> """
>> [...]
> 
> Full log:
> https://download.copr.fedorainfracloud.org/results/@kernel-vanilla/next/fedora-44-x86_64/10739303-next-next-all/builder-live.log.gz
> 
> Ciao, Thorsten

Re: [PATCH 1/3] cpupower: Add generic CPPC performance display

[Jeremy Linton]

Hi,

On 7/18/26 1:12 AM, Thorsten Leemhuis wrote:
> On 7/10/26 00:13, Jeremy Linton wrote:
>> Arm64 machines, and possibly others, use the standard ACPI defined CPPC
>> infrastructure. cpupower has CPPC support, but it's largely written
>> around the intricacies of AMD processors, using platform MSRs to avoid
>> shortcomings in the specification.

Patch to fix it.

https://lkml.org/lkml/2026/7/20/2191

>>
>> [...]
>> +#include <stdio.h>
>> +#include <errno.h>
>> +#include <stdlib.h>
>> +#include <string.h>
>> +#include <math.h>
>> +
>> +#include "helpers/helpers.h"
>> +#include "cpufreq.h"
>> +#include "acpi_cppc.h"
>> +
>> +#define cppc_to_frequency(perf) (roundf(slope * (perf) + intercept))
> FYI, this patch showed up in -next yesterday and I suspect it caused
> this build error for me on x86_64:
> 
> """  CC       cpupower
> /usr/bin/ld.bfd: /tmp/ccjOeWLF.ltrans0.ltrans.o: in function
> `get_perf_cap.isra.0':
> /builddir/build/BUILD/kernel-7.2.0-build/kernel-next-20260717/linux-7.2.0-0.0.next.20260717.330.vanilla.fc44.x86_64/tools/power/cpupower/utils/helpers/cppc.c:39:(.text+0x45cc):
> undefined reference to `roundf'
> /usr/bin/ld.bfd:
> /tmp/ccjOeWLF.ltrans0.ltrans.o:/builddir/build/BUILD/kernel-7.2.0-build/kernel-next-20260717/linux-7.2.0-0.0.next.20260717.330.vanilla.fc44.x86_64/tools/power/cpupower/utils/helpers/cppc.c:49:(.text+0x46a2):
> undefined reference to `roundf'
> collect2: error: ld returned 1 exit status
> make: *** [Makefile:238: cpupower] Error 1
> """
>> [...]
> 
> Full log:
> https://download.copr.fedorainfracloud.org/results/@kernel-vanilla/next/fedora-44-x86_64/10739303-next-next-all/builder-live.log.gz
> 
> Ciao, Thorsten

Re: [PATCH 1/3] cpupower: Add generic CPPC performance display

[Shuah Khan]

On 7/20/26 12:33, Jeremy Linton wrote:
> Hi,
> 
> On 7/18/26 1:12 AM, Thorsten Leemhuis wrote:
>> On 7/10/26 00:13, Jeremy Linton wrote:
>>> Arm64 machines, and possibly others, use the standard ACPI defined CPPC
>>> infrastructure. cpupower has CPPC support, but it's largely written
>>> around the intricacies of AMD processors, using platform MSRs to avoid
>>> shortcomings in the specification.
> 
> Patch to fix it.
> 
> https://lkml.org/lkml/2026/7/20/2191
> 

Thank you both. Applied the fix to

https://web.git.kernel.org/pub/scm/linux/kernel/git/shuah/linux.git/log/?h=cpupower

thanks,
-- Shuah