Merge tag 'pm-6.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm

Pull power management updates from Rafael Wysocki:
 "The majority of these are cpufreq changes, which has been a recurring
  pattern for a few recent cycles.

  Those changes include new hardware support (AN7583 SoC support in the
  airoha cpufreq driver, ipq5424 support in the qcom-nvmem cpufreq
  driver, MT8196 support in the mediatek cpufreq driver, AM62D2 support
  in the ti cpufreq driver), DT bindings and Rust code updates, cleanups
  of the core and governors, and multiple driver fixes and cleanups.

  Beyond that, there are hibernation fixes (some remaining 6.16 cycle
  fallout and an issue related to hybrid suspend in the amdgpu driver),
  cleanups of the PM core code, runtime PM documentation update, cpuidle
  and power capping cleanups, and tooling updates.

  Specifics:

   - Rearrange variable declarations involving __free() in the cpufreq
     core and intel_pstate driver to follow common coding style (Rafael
     Wysocki)

   - Fix object lifecycle issue in update_qos_request(), rearrange freq
     QoS updates using __free(), and adjust frequency percentage
     computations in the intel_pstate driver (Rafael Wysocki)

   - Update intel_pstate to allow it to enable HWP without EPP if the
     new DEC (Dynamic Efficiency Control) HW feature is enabled (Rafael
     Wysocki)

   - Use on_each_cpu_mask() in drv_write() in the ACPI cpufreq driver to
     simplify the code (Rafael Wysocki)

   - Use likely() optimization in intel_pstate_sample() (Yaxiong Tian)

   - Remove dead EPB-related code from intel_pstate (Srinivas
     Pandruvada)

   - Use scope-based cleanup for cpufreq policy references in multiple
     cpufreq drivers (Zihuan Zhang)

   - Avoid calling get_governor() for the first policy in the cpufreq
     core to simplify the initial policy path (Zihuan Zhang)

   - Clean up the cpufreq core in multiple places (Zihuan Zhang)

   - Use int type to store negative error codes in the cpufreq core and
     update the speedstep-lib to use int for error codes (Qianfeng Rong)

   - Update the efficient idle check for Intel extended Families in the
     ondemand cpufreq governor (Sohil Mehta)

   - Replace sscanf() with kstrtouint() in the conservative cpufreq
     governor (Kaushlendra Kumar)

   - Rename CpumaskVar::as[_mut]_ref to from_raw[_mut] in the cpumask
     Rust code and mark CpumaskVar as transparent (Alice Ryhl, Baptiste
     Lepers)

   - Update ARef and AlwaysRefCounted imports from sync::aref in the OPP
     Rust code (Shankari Anand)

   - Add support for AN7583 SoC to the airoha cpufreq driver (Christian
     Marangi)

   - Enable cpufreq for ipq5424 in the qcom-nvmem cpufreq driver (Md
     Sadre Alam)

   - Add support for MT8196 to the mediatek-hw cpufreq driver, refactor
     that driver and add mediatek,mt8196-cpufreq-hw DT binding (Nicolas
     Frattaroli)

   - Avoid redundant conditions in the mediatek cpufreq driver (Liao
     Yuanhong)

   - Add support for AM62D2 to the ti cpufreq driver and blocklist
     ti,am62d2 SoC in dt-platdev (Paresh Bhagat)

   - Support more speed grades on AM62Px SoC in the ti cpufreq driver,
     allow all silicon revisions to support OPPs in it, and fix
     supported hardware for 1GHz OPP (Judith Mendez)

   - Add QCS615 compatible to DT bindings for cpufreq-qcom-hw (Taniya
     Das)

   - Minor assorted updates of the scmi, longhaul, CPPC, and armada-37xx
     cpufreq drivers (Akhilesh Patil, BowenYu, Dennis Beier, and Florian
     Fainelli)

   - Remove outdated cpufreq-dt.txt (Frank Li)

   - Fix python gnuplot package names in the amd_pstate_tracer utility
     (Kuan-Wei Chiu)

   - Saravana Kannan will maintain the virtual-cpufreq driver (Saravana
     Kannan)

   - Prevent CPU capacity updates after registering a perf domain from
     failing on a first CPU that is not present (Christian Loehle)

   - Add support for the cases in which frequency alone is not
     sufficient to uniquely identify an OPP (Krishna Chaitanya Chundru)

   - Use to_result() for OPP error handling in Rust (Onur Özkan)

   - Add support for LPDDR5 on Rockhip RK3588 SoC to rockchip-dfi
     devfreq driver (Nicolas Frattaroli)

   - Fix an issue where DDR cycle counts on RK3588/RK3528 with LPDDR4(X)
     are reported as half by adding a cycle multiplier to the DFI driver
     in rockchip-dfi devfreq-event driver (Nicolas Frattaroli)

   - Fix missing error pointer dereference check of regulator instance
     in the mtk-cci devfreq driver probe and remove a redundant
     condition from an if () statement in that driver (Dan Carpenter,
     Liao Yuanhong)

   - Fail cpuidle device registration if there is one already to avoid
     sysfs-related issues (Rafael Wysocki)

   - Use sysfs_emit()/sysfs_emit_at() instead of sprintf()/scnprintf()
     in cpuidle (Vivek Yadav)

   - Fix device and OF node leaks at probe in the qcom-spm cpuidle
     driver and drop unnecessary initialisations from it (Johan Hovold)

   - Remove unnecessary address-of operators from the intel_idle cpuidle
     driver (Kaushlendra Kumar)

   - Rearrange main loop in menu_select() to make the code in that
     funtion easier to follow (Rafael Wysocki)

   - Convert values in microseconds to ktime using us_to_ktime() where
     applicable in the intel_idle power capping driver (Xichao Zhao)

   - Annotate loops walking device links in the power management core
     code as _srcu and add macros for walking device links to reduce the
     likelihood of coding mistakes related to them (Rafael Wysocki)

   - Document time units for *_time functions in the runtime PM API
     (Brian Norris)

   - Clear power.must_resume in noirq suspend error path to avoid
     resuming a dependant device under a suspended parent or supplier
     (Rafael Wysocki)

   - Fix GFP mask handling during hybrid suspend and make the amdgpu
     driver handle hybrid suspend correctly (Mario Limonciello, Rafael
     Wysocki)

   - Fix GFP mask handling after aborted hibernation in platform mode
     and combine exit paths in power_down() to avoid code duplication
     (Rafael Wysocki)

   - Use vmalloc_array() and vcalloc() in the hibernation core to avoid
     open-coded size computations (Qianfeng Rong)

   - Fix typo in hibernation core code comment (Li Jun)

   - Call pm_wakeup_clear() in the same place where other functions that
     do bookkeeping prior to suspend_prepare() are called (Samuel Wu)

   - Fix and clean up the x86_energy_perf_policy utility and update its
     documentation (Len Brown, Kaushlendra Kumar)

   - Fix incorrect sorting of PMT telemetry in turbostat (Kaushlendra
     Kumar)

   - Fix incorrect size in cpuidle_state_disable() and the error return
     value of cpupower_write_sysfs() in cpupower (Kaushlendra Kumar)"

* tag 'pm-6.18-rc1' of git://git.kernel.org/pub/scm/linux/kernel/git/rafael/linux-pm: (86 commits)
  PM: hibernate: Combine return paths in power_down()
  PM: hibernate: Restrict GFP mask in power_down()
  PM: hibernate: Fix pm_hibernation_mode_is_suspend() build breakage
  PM: runtime: Documentation: ABI: Document time units for *_time
  tools/power x86_energy_perf_policy.8: Emphasize preference for SW interfaces
  tools/power x86_energy_perf_policy: Add make snapshot target
  tools/power x86_energy_perf_policy: Prefer driver HWP limits
  tools/power x86_energy_perf_policy: EPB access is only via sysfs
  tools/power x86_energy_perf_policy: Prepare for MSR/sysfs refactoring
  tools/power x86_energy_perf_policy: Enhance HWP enable
  tools/power x86_energy_perf_policy: Enhance HWP enabled check
  tools/power x86_energy_perf_policy: Fix incorrect fopen mode usage
  tools/power turbostat: Fix incorrect sorting of PMT telemetry
  drm/amd: Fix hybrid sleep
  PM: hibernate: Add pm_hibernation_mode_is_suspend()
  PM: hibernate: Fix hybrid-sleep
  tools/cpupower: Fix incorrect size in cpuidle_state_disable()
  tools/power/x86/amd_pstate_tracer: Fix python gnuplot package names
  cpufreq: Replace pointer subtraction with iteration macro
  cpuidle: Fail cpuidle device registration if there is one already
  ...

7 files changed, 134 insertions, 54 deletions

diff --git a/tools/power/cpupower/lib/cpuidle.c b/tools/power/cpupower/lib/cpuidle.c
index 0ecac009273c..f2c1139adf71 100644
--- a/tools/power/cpupower/lib/cpuidle.c
+++ b/tools/power/cpupower/lib/cpuidle.c
@@ -233,6 +233,7 @@ int cpuidle_state_disable(unsigned int cpu,
 {
 	char value[SYSFS_PATH_MAX];
 	int bytes_written;
+	int len;
 
 	if (cpuidle_state_count(cpu) <= idlestate)
 		return -1;
@@ -241,10 +242,10 @@ int cpuidle_state_disable(unsigned int cpu,
 				 idlestate_value_files[IDLESTATE_DISABLE]))
 		return -2;
 
-	snprintf(value, SYSFS_PATH_MAX, "%u", disable);
+	len = snprintf(value, SYSFS_PATH_MAX, "%u", disable);
 
 	bytes_written = cpuidle_state_write_file(cpu, idlestate, "disable",
-						   value, sizeof(disable));
+						   value, len);
 	if (bytes_written)
 		return 0;
 	return -3;
diff --git a/tools/power/cpupower/lib/cpupower.c b/tools/power/cpupower/lib/cpupower.c
index ce8dfb8e46ab..d7f7ec6f151c 100644
--- a/tools/power/cpupower/lib/cpupower.c
+++ b/tools/power/cpupower/lib/cpupower.c
@@ -56,7 +56,7 @@ unsigned int cpupower_write_sysfs(const char *path, char *buf, size_t buflen)
 	if (numwritten < 1) {
 		perror(path);
 		close(fd);
-		return -1;
+		return 0;
 	}
 
 	close(fd);
diff --git a/tools/power/x86/amd_pstate_tracer/amd_pstate_trace.py b/tools/power/x86/amd_pstate_tracer/amd_pstate_trace.py
index feb9f9421c7b..875b086550d1 100755
--- a/tools/power/x86/amd_pstate_tracer/amd_pstate_trace.py
+++ b/tools/power/x86/amd_pstate_tracer/amd_pstate_trace.py
@@ -11,7 +11,7 @@ Prerequisites:
     gnuplot 5.0 or higher
     gnuplot-py 1.8 or higher
     (Most of the distributions have these required packages. They may be called
-     gnuplot-py, phython-gnuplot or phython3-gnuplot, gnuplot-nox, ... )
+     gnuplot-py, python-gnuplot or python3-gnuplot, gnuplot-nox, ... )
 
     Kernel config for Linux trace is enabled
 
diff --git a/tools/power/x86/turbostat/turbostat.c b/tools/power/x86/turbostat/turbostat.c
index 47eb2d4d13a5..f2512d78bcbd 100644
--- a/tools/power/x86/turbostat/turbostat.c
+++ b/tools/power/x86/turbostat/turbostat.c
@@ -1890,7 +1890,7 @@ int pmt_telemdir_sort(const struct dirent **a, const struct dirent **b)
 	sscanf((*a)->d_name, "telem%u", &aidx);
 	sscanf((*b)->d_name, "telem%u", &bidx);
 
-	return aidx >= bidx;
+	return (aidx > bidx) ? 1 : (aidx < bidx) ? -1 : 0;
 }
 
 const struct dirent *pmt_diriter_next(struct pmt_diriter_t *iter)
diff --git a/tools/power/x86/x86_energy_perf_policy/Makefile b/tools/power/x86/x86_energy_perf_policy/Makefile
index 666b325a62a2..d18284667400 100644
--- a/tools/power/x86/x86_energy_perf_policy/Makefile
+++ b/tools/power/x86/x86_energy_perf_policy/Makefile
@@ -1,8 +1,12 @@
 # SPDX-License-Identifier: GPL-2.0
 CC		= $(CROSS_COMPILE)gcc
-BUILD_OUTPUT    := $(CURDIR)
+BUILD_OUTPUT	:= $(CURDIR)
 PREFIX		:= /usr
 DESTDIR		:=
+DAY		:= $(shell date +%Y.%m.%d)
+SNAPSHOT	= x86_energy_perf_policy-$(DAY)
+
+
 
 ifeq ("$(origin O)", "command line")
 	BUILD_OUTPUT := $(O)
@@ -27,3 +31,26 @@ install : x86_energy_perf_policy
 	install -d  $(DESTDIR)$(PREFIX)/share/man/man8
 	install -m 644 x86_energy_perf_policy.8 $(DESTDIR)$(PREFIX)/share/man/man8
 
+snapshot: x86_energy_perf_policy
+	@rm -rf $(SNAPSHOT)
+	@mkdir $(SNAPSHOT)
+	@cp x86_energy_perf_policy Makefile x86_energy_perf_policy.c x86_energy_perf_policy.8 $(SNAPSHOT)
+
+	@sed -e 's/^#include <linux\/bits.h>/#include "bits.h"/' -e 's/u64/unsigned long long/' ../../../../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 +=	-DBUILD_BUG_HEADER='\"build_bug.h\"'" >> $(SNAPSHOT)/Makefile
+	@sed -e's/.*MSRHEADER.*//' Makefile >> $(SNAPSHOT)/Makefile
+
+	@rm -f $(SNAPSHOT).tar.gz
+	tar cvzf $(SNAPSHOT).tar.gz $(SNAPSHOT)
+
diff --git a/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.8 b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.8
index 78c6361898b1..0aa981c18e56 100644
--- a/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.8
+++ b/tools/power/x86/x86_energy_perf_policy/x86_energy_perf_policy.8
@@ -2,7 +2,7 @@
 .\"  Distributed under the GPL, Copyleft 1994.
 .TH X86_ENERGY_PERF_POLICY 8
 .SH NAME
-x86_energy_perf_policy \- Manage Energy vs. Performance Policy via x86 Model Specific Registers
+x86_energy_perf_policy \- Manage Energy vs. Performance Policy
 .SH SYNOPSIS
 .B x86_energy_perf_policy
 .RB "[ options ] [ scope ] [field \ value]"
@@ -19,9 +19,14 @@ x86_energy_perf_policy \- Manage Energy vs. Performance Policy via x86 Model Spe
 .SH DESCRIPTION
 \fBx86_energy_perf_policy\fP
 displays and updates energy-performance policy settings specific to
-Intel Architecture Processors.  Settings are accessed via Model Specific Register (MSR)
-updates, no matter if the Linux cpufreq sub-system is enabled or not.
+Intel Architecture Processors.  It summarizes settings available
+in standard Linux interfaces (eg. cpufreq),
+and also decodes underlying Model Specific Register (MSRs).
+While \fBx86_energy_perf_policy\fP can manage energy-performance policy
+using only MSR access, it prefers standard
+Linux kernel interfaces, when they are available.
 
+.SH BACKGROUND
 Policy in MSR_IA32_ENERGY_PERF_BIAS (EPB)
 may affect a wide range of hardware decisions,
 such as how aggressively the hardware enters and exits CPU idle states (C-states)
@@ -200,7 +205,9 @@ runs only as root.
 .SH FILES
 .ta
 .nf
-/dev/cpu/*/msr
+EPB: /sys/devices/system/cpu/cpu*/power/energy_perf_bias
+EPP: /sys/devices/system/cpu/cpu*/cpufreq/energy_performance_preference
+MSR: /dev/cpu/*/msr
 .fi
 .SH "SEE ALSO"
 .nf
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 ebda9c366b2b..884a4c746f32 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
@@ -4,7 +4,7 @@
  * policy preference bias on recent X86 processors.
  */
 /*
- * Copyright (c) 2010 - 2017 Intel Corporation.
+ * Copyright (c) 2010 - 2025 Intel Corporation.
  * Len Brown <len.brown@intel.com>
  */
 
@@ -62,6 +62,7 @@ unsigned char turbo_update_value;
 unsigned char update_hwp_epp;
 unsigned char update_hwp_min;
 unsigned char update_hwp_max;
+unsigned char hwp_limits_done_via_sysfs;
 unsigned char update_hwp_desired;
 unsigned char update_hwp_window;
 unsigned char update_hwp_use_pkg;
@@ -517,7 +518,7 @@ void for_packages(unsigned long long pkg_set, int (func)(int))
 
 void print_version(void)
 {
-	printf("x86_energy_perf_policy 17.05.11 (C) Len Brown <len.brown@intel.com>\n");
+	printf("x86_energy_perf_policy 2025.9.19 Len Brown <lenb@kernel.org>\n");
 }
 
 void cmdline(int argc, char **argv)
@@ -630,7 +631,7 @@ void cmdline(int argc, char **argv)
  */
 FILE *fopen_or_die(const char *path, const char *mode)
 {
-	FILE *filep = fopen(path, "r");
+	FILE *filep = fopen(path, mode);
 
 	if (!filep)
 		err(1, "%s: open failed", path);
@@ -644,7 +645,7 @@ void err_on_hypervisor(void)
 	char *buffer;
 
 	/* On VMs /proc/cpuinfo contains a "flags" entry for hypervisor */
-	cpuinfo = fopen_or_die("/proc/cpuinfo", "ro");
+	cpuinfo = fopen_or_die("/proc/cpuinfo", "r");
 
 	buffer = malloc(4096);
 	if (!buffer) {
@@ -809,7 +810,7 @@ void print_hwp_request_pkg(int pkg, struct msr_hwp_request *h, char *str)
 		h->hwp_min, h->hwp_max, h->hwp_desired, h->hwp_epp,
 		h->hwp_window, h->hwp_window & 0x7F, (h->hwp_window >> 7) & 0x7);
 }
-void read_hwp_request(int cpu, struct msr_hwp_request *hwp_req, unsigned int msr_offset)
+void read_hwp_request_msr(int cpu, struct msr_hwp_request *hwp_req, unsigned int msr_offset)
 {
 	unsigned long long msr;
 
@@ -823,7 +824,7 @@ void read_hwp_request(int cpu, struct msr_hwp_request *hwp_req, unsigned int msr
 	hwp_req->hwp_use_pkg = (((msr) >> 42) & 0x1);
 }
 
-void write_hwp_request(int cpu, struct msr_hwp_request *hwp_req, unsigned int msr_offset)
+void write_hwp_request_msr(int cpu, struct msr_hwp_request *hwp_req, unsigned int msr_offset)
 {
 	unsigned long long msr = 0;
 
@@ -843,7 +844,7 @@ void write_hwp_request(int cpu, struct msr_hwp_request *hwp_req, unsigned int ms
 	put_msr(cpu, msr_offset, msr);
 }
 
-static int get_epb(int cpu)
+static int get_epb_sysfs(int cpu)
 {
 	char path[SYSFS_PATH_MAX];
 	char linebuf[3];
@@ -865,7 +866,7 @@ static int get_epb(int cpu)
 	return (int)val;
 }
 
-static int set_epb(int cpu, int val)
+static int set_epb_sysfs(int cpu, int val)
 {
 	char path[SYSFS_PATH_MAX];
 	char linebuf[3];
@@ -895,14 +896,14 @@ int print_cpu_msrs(int cpu)
 	struct msr_hwp_cap cap;
 	int epb;
 
-	epb = get_epb(cpu);
+	epb = get_epb_sysfs(cpu);
 	if (epb >= 0)
 		printf("cpu%d: EPB %u\n", cpu, (unsigned int) epb);
 
 	if (!has_hwp)
 		return 0;
 
-	read_hwp_request(cpu, &req, MSR_HWP_REQUEST);
+	read_hwp_request_msr(cpu, &req, MSR_HWP_REQUEST);
 	print_hwp_request(cpu, &req, "");
 
 	read_hwp_cap(cpu, &cap, MSR_HWP_CAPABILITIES);
@@ -919,7 +920,7 @@ int print_pkg_msrs(int pkg)
 	if (!has_hwp)
 		return 0;
 
-	read_hwp_request(first_cpu_in_pkg[pkg], &req, MSR_HWP_REQUEST_PKG);
+	read_hwp_request_msr(first_cpu_in_pkg[pkg], &req, MSR_HWP_REQUEST_PKG);
 	print_hwp_request_pkg(pkg, &req, "");
 
 	if (has_hwp_notify) {
@@ -951,8 +952,10 @@ int ratio_2_sysfs_khz(int ratio)
 }
 /*
  * If HWP is enabled and cpufreq sysfs attribtes are present,
- * then update sysfs, so that it will not become
- * stale when we write to MSRs.
+ * then update via sysfs. The intel_pstate driver may modify (clip)
+ * this request, say, when HWP_CAP is outside of PLATFORM_INFO limits,
+ * and the driver-chosen value takes precidence.
+ *
  * (intel_pstate's max_perf_pct and min_perf_pct will follow cpufreq,
  *  so we don't have to touch that.)
  */
@@ -1007,6 +1010,8 @@ int update_sysfs(int cpu)
 	if (update_hwp_max)
 		update_cpufreq_scaling_freq(1, cpu, req_update.hwp_max);
 
+	hwp_limits_done_via_sysfs = 1;
+
 	return 0;
 }
 
@@ -1074,21 +1079,21 @@ int check_hwp_request_v_hwp_capabilities(int cpu, struct msr_hwp_request *req, s
 	return 0;
 }
 
-int update_hwp_request(int cpu)
+int update_hwp_request_msr(int cpu)
 {
 	struct msr_hwp_request req;
 	struct msr_hwp_cap cap;
 
 	int msr_offset = MSR_HWP_REQUEST;
 
-	read_hwp_request(cpu, &req, msr_offset);
+	read_hwp_request_msr(cpu, &req, msr_offset);
 	if (debug)
 		print_hwp_request(cpu, &req, "old: ");
 
-	if (update_hwp_min)
+	if (update_hwp_min && !hwp_limits_done_via_sysfs)
 		req.hwp_min = req_update.hwp_min;
 
-	if (update_hwp_max)
+	if (update_hwp_max && !hwp_limits_done_via_sysfs)
 		req.hwp_max = req_update.hwp_max;
 
 	if (update_hwp_desired)
@@ -1111,15 +1116,15 @@ int update_hwp_request(int cpu)
 
 	verify_hwp_req_self_consistency(cpu, &req);
 
-	write_hwp_request(cpu, &req, msr_offset);
+	write_hwp_request_msr(cpu, &req, msr_offset);
 
 	if (debug) {
-		read_hwp_request(cpu, &req, msr_offset);
+		read_hwp_request_msr(cpu, &req, msr_offset);
 		print_hwp_request(cpu, &req, "new: ");
 	}
 	return 0;
 }
-int update_hwp_request_pkg(int pkg)
+int update_hwp_request_pkg_msr(int pkg)
 {
 	struct msr_hwp_request req;
 	struct msr_hwp_cap cap;
@@ -1127,7 +1132,7 @@ int update_hwp_request_pkg(int pkg)
 
 	int msr_offset = MSR_HWP_REQUEST_PKG;
 
-	read_hwp_request(cpu, &req, msr_offset);
+	read_hwp_request_msr(cpu, &req, msr_offset);
 	if (debug)
 		print_hwp_request_pkg(pkg, &req, "old: ");
 
@@ -1155,10 +1160,10 @@ int update_hwp_request_pkg(int pkg)
 
 	verify_hwp_req_self_consistency(cpu, &req);
 
-	write_hwp_request(cpu, &req, msr_offset);
+	write_hwp_request_msr(cpu, &req, msr_offset);
 
 	if (debug) {
-		read_hwp_request(cpu, &req, msr_offset);
+		read_hwp_request_msr(cpu, &req, msr_offset);
 		print_hwp_request_pkg(pkg, &req, "new: ");
 	}
 	return 0;
@@ -1166,30 +1171,39 @@ int update_hwp_request_pkg(int pkg)
 
 int enable_hwp_on_cpu(int cpu)
 {
-	unsigned long long msr;
+	unsigned long long old_msr, new_msr;
+
+	get_msr(cpu, MSR_PM_ENABLE, &old_msr);
+
+	if (old_msr & 1)
+		return 0;	/* already enabled */
 
-	get_msr(cpu, MSR_PM_ENABLE, &msr);
-	put_msr(cpu, MSR_PM_ENABLE, 1);
+	new_msr = old_msr | 1;
+	put_msr(cpu, MSR_PM_ENABLE, new_msr);
 
 	if (verbose)
-		printf("cpu%d: MSR_PM_ENABLE old: %d new: %d\n", cpu, (unsigned int) msr, 1);
+		printf("cpu%d: MSR_PM_ENABLE old: %llX new: %llX\n", cpu, old_msr, new_msr);
 
 	return 0;
 }
 
-int update_cpu_msrs(int cpu)
+int update_cpu_epb_sysfs(int cpu)
 {
-	unsigned long long msr;
 	int epb;
 
-	if (update_epb) {
-		epb = get_epb(cpu);
-		set_epb(cpu, new_epb);
+	epb = get_epb_sysfs(cpu);
+	set_epb_sysfs(cpu, new_epb);
 
-		if (verbose)
-			printf("cpu%d: ENERGY_PERF_BIAS old: %d new: %d\n",
-				cpu, epb, (unsigned int) new_epb);
-	}
+	if (verbose)
+		printf("cpu%d: ENERGY_PERF_BIAS old: %d new: %d\n",
+			cpu, epb, (unsigned int) new_epb);
+
+	return 0;
+}
+
+int update_cpu_msrs(int cpu)
+{
+	unsigned long long msr;
 
 	if (update_turbo) {
 		int turbo_is_present_and_disabled;
@@ -1224,7 +1238,7 @@ int update_cpu_msrs(int cpu)
 	if (!hwp_update_enabled())
 		return 0;
 
-	update_hwp_request(cpu);
+	update_hwp_request_msr(cpu);
 	return 0;
 }
 
@@ -1312,6 +1326,17 @@ void for_all_cpus_in_set(size_t set_size, cpu_set_t *cpu_set, int (func)(int))
 		if (CPU_ISSET_S(cpu_num, set_size, cpu_set))
 			func(cpu_num);
 }
+int for_all_cpus_in_set_and(size_t set_size, cpu_set_t *cpu_set, int (func)(int))
+{
+	int cpu_num;
+	int retval = 1;
+
+	for (cpu_num = 0; cpu_num <= max_cpu_num; ++cpu_num)
+		if (CPU_ISSET_S(cpu_num, set_size, cpu_set))
+			retval &= func(cpu_num);
+
+	return retval;
+}
 
 void init_data_structures(void)
 {
@@ -1326,21 +1351,38 @@ void init_data_structures(void)
 	for_all_proc_cpus(mark_cpu_present);
 }
 
-/* clear has_hwp if it is not enable (or being enabled) */
+int is_hwp_enabled_on_cpu(int cpu_num)
+{
+	unsigned long long msr;
+	int retval;
+
+	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
+	get_msr(cpu_num, MSR_PM_ENABLE, &msr);
+	retval = (msr & 1);
+
+	if (verbose)
+		fprintf(stderr, "cpu%d: %sHWP\n", cpu_num, retval ? "" : "No-");
+
+	return retval;
+}
 
+/*
+ * verify_hwp_is_enabled()
+ *
+ * Set (has_hwp=0) if no HWP feature or any of selected CPU set does not have HWP enabled
+ */
 void verify_hwp_is_enabled(void)
 {
-	unsigned long long msr;
+	int retval;
 
 	if (!has_hwp)	/* set in early_cpuid() */
 		return;
 
-	/* MSR_PM_ENABLE[1] == 1 if HWP is enabled and MSRs visible */
-	get_msr(base_cpu, MSR_PM_ENABLE, &msr);
-	if ((msr & 1) == 0) {
+	retval = for_all_cpus_in_set_and(cpu_setsize, cpu_selected_set, is_hwp_enabled_on_cpu);
+
+	if (retval == 0) {
 		fprintf(stderr, "HWP can be enabled using '--hwp-enable'\n");
 		has_hwp = 0;
-		return;
 	}
 }
 
@@ -1551,10 +1593,13 @@ int main(int argc, char **argv)
 
 	/* update CPU set */
 	if (cpu_selected_set) {
+		if (update_epb)
+			for_all_cpus_in_set(cpu_setsize, cpu_selected_set, update_cpu_epb_sysfs);
 		for_all_cpus_in_set(cpu_setsize, cpu_selected_set, update_sysfs);
 		for_all_cpus_in_set(cpu_setsize, cpu_selected_set, update_cpu_msrs);
+
 	} else if (pkg_selected_set)
-		for_packages(pkg_selected_set, update_hwp_request_pkg);
+		for_packages(pkg_selected_set, update_hwp_request_pkg_msr);
 
 	return 0;
 }