From: Rik van Riel <riel@redhat.com>
Date: Thu, 13 Dec 2007 14:19:30 -0500
Subject: [xen] kernel: cpu frequency scaling
Message-id: 20071213141930.48df89bd@cuia.boston.redhat.com
O-Subject: [RHEL 5.2 PATCH 2/3] cpu frequency scaling for Xen BZ#251969
Bugzilla: 251969

Linux side of Xen CPU frequency change support.  Based on
upstream changesets:

Backport of powernow-k8 cpufreq changes, required for power management
	215:8a407c41dfb0

cpufreq: Xen support for the ondemand governor in Linux dom0
	287:7aaec9c0a213

cpufreq: minor clean-ups for ondemand governor on Xen.
	304:98de2b149423

cpufreq: Support cpufreq updates on AMD hardware by dom0 kernel.
	15924:2477e94450aa

cpufreq: Fix the ondemand driver for Xen. No 64-bit division allowed
	310:fced90d566f1

[cpufreq] Correctly calculate load
	336:ba918cb2cf75

Acked-by: Bill Burns <bburns@redhat.com>
Acked-by: "Stephen C. Tweedie" <sct@redhat.com>

diff --git a/arch/i386/kernel/cpu/cpufreq/powernow-k8.c b/arch/i386/kernel/cpu/cpufreq/powernow-k8.c
index 2874dc5..9b7578b 100644
--- a/arch/i386/kernel/cpu/cpufreq/powernow-k8.c
+++ b/arch/i386/kernel/cpu/cpufreq/powernow-k8.c
@@ -767,6 +767,7 @@ static int find_psb_table(struct powernow_k8_data *data)
 
 		data->numps = psb->numps;
 		dprintk("numpstates: 0x%x\n", data->numps);
+		data->starting_core_affinity = cpumask_of_cpu(0);
 		return fill_powernow_table(data, (struct pst_s *)(psb+1), maxvid);
 	}
 	/*
@@ -787,15 +788,43 @@ static int find_psb_table(struct powernow_k8_data *data)
 #ifdef CONFIG_X86_POWERNOW_K8_ACPI
 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, unsigned int index)
 {
-	if (!data->acpi_data.state_count || (cpu_family == CPU_HW_PSTATE))
+	if (!data->acpi_data->state_count || (cpu_family == CPU_HW_PSTATE))
 		return;
 
-	data->irt = (data->acpi_data.states[index].control >> IRT_SHIFT) & IRT_MASK;
-	data->rvo = (data->acpi_data.states[index].control >> RVO_SHIFT) & RVO_MASK;
-	data->exttype = (data->acpi_data.states[index].control >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK;
-	data->plllock = (data->acpi_data.states[index].control >> PLL_L_SHIFT) & PLL_L_MASK;
-	data->vidmvs = 1 << ((data->acpi_data.states[index].control >> MVS_SHIFT) & MVS_MASK);
-	data->vstable = (data->acpi_data.states[index].control >> VST_SHIFT) & VST_MASK;
+	data->irt = (data->acpi_data->states[index].control >> IRT_SHIFT) & IRT_MASK;
+	data->rvo = (data->acpi_data->states[index].control >> RVO_SHIFT) & RVO_MASK;
+	data->exttype = (data->acpi_data->states[index].control >> EXT_TYPE_SHIFT) & EXT_TYPE_MASK;
+	data->plllock = (data->acpi_data->states[index].control >> PLL_L_SHIFT) & PLL_L_MASK;
+	data->vidmvs = 1 << ((data->acpi_data->states[index].control >> MVS_SHIFT) & MVS_MASK);
+	data->vstable = (data->acpi_data->states[index].control >> VST_SHIFT) & VST_MASK;
+}
+
+static struct acpi_processor_performance *acpi_perf_data[NR_CPUS];
+static int preregister_valid = 0;
+
+static int powernow_k8_cpu_preinit_acpi()
+{
+	int i; 
+	struct acpi_processor_performance *data;
+	for_each_possible_cpu(i) {
+		data = kzalloc(sizeof(struct acpi_processor_performance),
+				GFP_KERNEL);
+		if (!data) {
+			int j;
+			for_each_possible_cpu(j) {
+				kfree(acpi_perf_data[j]);
+				acpi_perf_data[j] = NULL;
+			}
+			return -ENODEV;
+		}
+		acpi_perf_data[i] = data;
+	}
+
+	if (acpi_processor_preregister_performance(acpi_perf_data))
+		return -ENODEV;
+	else 
+		preregister_valid = 1;
+	return 0;
 }
 
 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
@@ -803,28 +832,29 @@ static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
 	struct cpufreq_frequency_table *powernow_table;
 	int ret_val;
 
-	if (acpi_processor_register_performance(&data->acpi_data, data->cpu)) {
+	data->acpi_data = acpi_perf_data[data->cpu];
+	if (acpi_processor_register_performance(data->acpi_data, data->cpu)) {
 		dprintk("register performance failed: bad ACPI data\n");
 		return -EIO;
 	}
 
 	/* verify the data contained in the ACPI structures */
-	if (data->acpi_data.state_count <= 1) {
+	if (data->acpi_data->state_count <= 1) {
 		dprintk("No ACPI P-States\n");
 		goto err_out;
 	}
 
-	if ((data->acpi_data.control_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
-		(data->acpi_data.status_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
+	if ((data->acpi_data->control_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE) ||
+		(data->acpi_data->status_register.space_id != ACPI_ADR_SPACE_FIXED_HARDWARE)) {
 		dprintk("Invalid control/status registers (%x - %x)\n",
-			data->acpi_data.control_register.space_id,
-			data->acpi_data.status_register.space_id);
+			data->acpi_data->control_register.space_id,
+			data->acpi_data->status_register.space_id);
 		goto err_out;
 	}
 
 	/* fill in data->powernow_table */
 	powernow_table = kmalloc((sizeof(struct cpufreq_frequency_table)
-		* (data->acpi_data.state_count + 1)), GFP_KERNEL);
+		* (data->acpi_data->state_count + 1)), GFP_KERNEL);
 	if (!powernow_table) {
 		dprintk("powernow_table memory alloc failure\n");
 		goto err_out;
@@ -837,12 +867,12 @@ static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
 	if (ret_val)
 		goto err_out_mem;
 
-	powernow_table[data->acpi_data.state_count].frequency = CPUFREQ_TABLE_END;
-	powernow_table[data->acpi_data.state_count].index = 0;
+	powernow_table[data->acpi_data->state_count].frequency = CPUFREQ_TABLE_END;
+	powernow_table[data->acpi_data->state_count].index = 0;
 	data->powernow_table = powernow_table;
 
 	/* fill in data */
-	data->numps = data->acpi_data.state_count;
+	data->numps = data->acpi_data->state_count;
 	if (first_cpu(cpu_core_map[data->cpu]) == data->cpu)
 		print_basics(data);
 	powernow_k8_acpi_pst_values(data, 0);
@@ -850,16 +880,31 @@ static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data)
 	/* notify BIOS that we exist */
 	acpi_processor_notify_smm(THIS_MODULE);
 
+	/* determine affinity, from ACPI if available */
+	if (preregister_valid) {
+		if ((data->acpi_data->shared_type == CPUFREQ_SHARED_TYPE_ALL) ||
+		    (data->acpi_data->shared_type == CPUFREQ_SHARED_TYPE_ANY))
+			data->starting_core_affinity = data->acpi_data->shared_cpu_map;
+		else
+			data->starting_core_affinity = cpumask_of_cpu(data->cpu);
+	} else {
+		/* best guess from family if not */
+		if (cpu_family == CPU_HW_PSTATE)
+			data->starting_core_affinity = cpumask_of_cpu(data->cpu);
+		else
+			data->starting_core_affinity = cpu_core_map[data->cpu];
+	}
+
 	return 0;
 
 err_out_mem:
 	kfree(powernow_table);
 
 err_out:
-	acpi_processor_unregister_performance(&data->acpi_data, data->cpu);
+	acpi_processor_unregister_performance(data->acpi_data, data->cpu);
 
-	/* data->acpi_data.state_count informs us at ->exit() whether ACPI was used */
-	data->acpi_data.state_count = 0;
+	/* data->acpi_data->state_count informs us at ->exit() whether ACPI was used */
+	data->acpi_data->state_count = 0;
 
 	return -ENODEV;
 }
@@ -868,13 +913,13 @@ static int fill_powernow_table_pstate(struct powernow_k8_data *data, struct cpuf
 {
 	int i;
 
-	for (i = 0; i < data->acpi_data.state_count; i++) {
+	for (i = 0; i < data->acpi_data->state_count; i++) {
 		u32 index;
 		u32 hi = 0, lo = 0;
 		u32 fid;
 		u32 did;
 
-		index = data->acpi_data.states[i].control & HW_PSTATE_MASK;
+		index = data->acpi_data->states[i].control & HW_PSTATE_MASK;
 		if (index > MAX_HW_PSTATE) {
 			printk(KERN_ERR PFX "invalid pstate %d - bad value %d.\n", i, index);
 			printk(KERN_ERR PFX "Please report to BIOS manufacturer\n");
@@ -895,10 +940,10 @@ static int fill_powernow_table_pstate(struct powernow_k8_data *data, struct cpuf
 
 		powernow_table[i].frequency = find_khz_freq_from_fiddid(fid, did);
 
-		if (powernow_table[i].frequency != (data->acpi_data.states[i].core_frequency * 1000)) {
+		if (powernow_table[i].frequency != (data->acpi_data->states[i].core_frequency * 1000)) {
 			printk(KERN_INFO PFX "invalid freq entries %u kHz vs. %u kHz\n",
 				powernow_table[i].frequency,
-				(unsigned int) (data->acpi_data.states[i].core_frequency * 1000));
+				(unsigned int) (data->acpi_data->states[i].core_frequency * 1000));
 			powernow_table[i].frequency = CPUFREQ_ENTRY_INVALID;
 			continue;
 		}
@@ -910,16 +955,16 @@ static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpuf
 {
 	int i;
 	int cntlofreq = 0;
-	for (i = 0; i < data->acpi_data.state_count; i++) {
+	for (i = 0; i < data->acpi_data->state_count; i++) {
 		u32 fid;
 		u32 vid;
 
 		if (data->exttype) {
-			fid = data->acpi_data.states[i].status & EXT_FID_MASK;
-			vid = (data->acpi_data.states[i].status >> VID_SHIFT) & EXT_VID_MASK;
+			fid = data->acpi_data->states[i].status & EXT_FID_MASK;
+			vid = (data->acpi_data->states[i].status >> VID_SHIFT) & EXT_VID_MASK;
 		} else {
-			fid = data->acpi_data.states[i].control & FID_MASK;
-			vid = (data->acpi_data.states[i].control >> VID_SHIFT) & VID_MASK;
+			fid = data->acpi_data->states[i].control & FID_MASK;
+			vid = (data->acpi_data->states[i].control >> VID_SHIFT) & VID_MASK;
 		}
 
 		dprintk("   %d : fid 0x%x, vid 0x%x\n", i, fid, vid);
@@ -960,10 +1005,10 @@ static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpuf
 				cntlofreq = i;
 		}
 
-		if (powernow_table[i].frequency != (data->acpi_data.states[i].core_frequency * 1000)) {
+		if (powernow_table[i].frequency != (data->acpi_data->states[i].core_frequency * 1000)) {
 			printk(KERN_INFO PFX "invalid freq entries %u kHz vs. %u kHz\n",
 				powernow_table[i].frequency,
-				(unsigned int) (data->acpi_data.states[i].core_frequency * 1000));
+				(unsigned int) (data->acpi_data->states[i].core_frequency * 1000));
 			powernow_table[i].frequency = CPUFREQ_ENTRY_INVALID;
 			continue;
 		}
@@ -973,14 +1018,15 @@ static int fill_powernow_table_fidvid(struct powernow_k8_data *data, struct cpuf
 
 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data *data)
 {
-	if (data->acpi_data.state_count)
-		acpi_processor_unregister_performance(&data->acpi_data, data->cpu);
+	if (data->acpi_data->state_count)
+		acpi_processor_unregister_performance(data->acpi_data, data->cpu);
 }
 
 #else
 static int powernow_k8_cpu_init_acpi(struct powernow_k8_data *data) { return -ENODEV; }
 static void powernow_k8_cpu_exit_acpi(struct powernow_k8_data *data) { return; }
 static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, unsigned int index) { return; }
+static int powernow_k8_cpu_preinit_acpi() { return -ENODEV; }
 #endif /* CONFIG_X86_POWERNOW_K8_ACPI */
 
 /* Take a frequency, and issue the fid/vid transition command */
@@ -1243,7 +1289,7 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 		 * an UP version, and is deprecated by AMD.
 		 */
 		if (num_online_cpus() != 1) {
-			printk(KERN_ERR PFX "MP systems not supported by PSB BIOS structure\n");
+			printk(KERN_ERR PFX "Your BIOS does not provide _PSS objects.  PowerNow! does not work on SMP systems without _PSS objects.  Complain to your BIOS vendor.\n");
 			kfree(data);
 			return -ENODEV;
 		}
@@ -1283,10 +1329,7 @@ static int __cpuinit powernowk8_cpu_init(struct cpufreq_policy *pol)
 	set_cpus_allowed(current, oldmask);
 
 	pol->governor = CPUFREQ_DEFAULT_GOVERNOR;
-	if (cpu_family == CPU_HW_PSTATE)
-		pol->cpus = cpumask_of_cpu(pol->cpu);
-	else
-		pol->cpus = cpu_core_map[pol->cpu];
+	pol->cpus = data->starting_core_affinity;
 	data->available_cores = &(pol->cpus);
 
 	/* Take a crude guess here.
@@ -1431,6 +1474,7 @@ static int __cpuinit powernowk8_init(void)
 			for(i=0; i < num_possible_cpus(); i++)
 				req_state[i] = 99;
 		}
+		powernow_k8_cpu_preinit_acpi();
 		printk(KERN_INFO PFX "Found %d %s "
 			"processors (%d cpu cores) (" VERSION ")\n", supported_cpus/cpu_data[0].booted_cores,
 			boot_cpu_data.x86_model_id, supported_cpus);
diff --git a/arch/i386/kernel/cpu/cpufreq/powernow-k8.h b/arch/i386/kernel/cpu/cpufreq/powernow-k8.h
index 268af0e..281e52d 100644
--- a/arch/i386/kernel/cpu/cpufreq/powernow-k8.h
+++ b/arch/i386/kernel/cpu/cpufreq/powernow-k8.h
@@ -32,12 +32,13 @@ struct powernow_k8_data {
 #ifdef CONFIG_X86_POWERNOW_K8_ACPI
 	/* the acpi table needs to be kept. it's only available if ACPI was
 	 * used to determine valid frequency/vid/fid states */
-	struct acpi_processor_performance acpi_data;
+	struct acpi_processor_performance *acpi_data;
 #endif
 	/* we need to keep track of associated cores, but let cpufreq
 	 * handle hotplug events - so just point at cpufreq pol->cpus
 	 * structure */
 	cpumask_t *available_cores;
+	cpumask_t starting_core_affinity;
 };
 
 
diff --git a/arch/i386/kernel/time-xen.c b/arch/i386/kernel/time-xen.c
index a43d912..532ad14 100644
--- a/arch/i386/kernel/time-xen.c
+++ b/arch/i386/kernel/time-xen.c
@@ -50,6 +50,7 @@
 #include <linux/percpu.h>
 #include <linux/kernel_stat.h>
 #include <linux/posix-timers.h>
+#include <linux/cpufreq.h>
 
 #include <asm/io.h>
 #include <asm/smp.h>
@@ -1089,6 +1090,45 @@ void local_teardown_timer(unsigned int cpu)
 }
 #endif
 
+#if CONFIG_CPU_FREQ
+static int time_cpufreq_notifier(struct notifier_block *nb, unsigned long val, 
+				void *data)
+{
+	struct cpufreq_freqs *freq = data;
+	dom0_op_t op;
+
+	if (cpu_has(&cpu_data[freq->cpu], X86_FEATURE_CONSTANT_TSC))
+		return 0;
+
+	if (val == CPUFREQ_PRECHANGE)
+		return 0;
+
+	op.cmd = DOM0_change_freq;
+	op.u.change_freq.flags = 0;
+	op.u.change_freq.cpu = freq->cpu;
+	op.u.change_freq.freq = (u64)freq->new * 1000;
+	HYPERVISOR_dom0_op(&op);
+
+	return 0;
+}
+
+static struct notifier_block time_cpufreq_notifier_block = {
+	.notifier_call = time_cpufreq_notifier
+};
+
+static int __init cpufreq_time_setup(void)
+{
+	if (!cpufreq_register_notifier(&time_cpufreq_notifier_block,
+			CPUFREQ_TRANSITION_NOTIFIER)) {
+		printk(KERN_ERR "failed to set up cpufreq notifier\n");
+		return -ENODEV;
+	}
+	return 0;
+}
+
+core_initcall(cpufreq_time_setup);
+#endif
+
 /*
  * /proc/sys/xen: This really belongs in another file. It can stay here for
  * now however.
diff --git a/drivers/cpufreq/cpufreq_ondemand.c b/drivers/cpufreq/cpufreq_ondemand.c
index 9853ef5..5f5a118 100644
--- a/drivers/cpufreq/cpufreq_ondemand.c
+++ b/drivers/cpufreq/cpufreq_ondemand.c
@@ -224,17 +224,14 @@ static struct attribute_group dbs_attr_group = {
 
 /************************** sysfs end ************************/
 
-static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
+#ifndef CONFIG_XEN
+static int dbs_calc_load(struct cpu_dbs_info_s *this_dbs_info)
 {
-	unsigned int idle_ticks, total_ticks;
-	unsigned int load;
-	cputime64_t cur_jiffies;
-
 	struct cpufreq_policy *policy;
+	cputime64_t cur_jiffies;
+	cputime64_t total_ticks, idle_ticks;
 	unsigned int j;
-
-	if (!this_dbs_info->enable)
-		return;
+	int load;
 
 	policy = this_dbs_info->cur_policy;
 	cur_jiffies = jiffies64_to_cputime64(get_jiffies_64());
@@ -242,7 +239,7 @@ static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
 			this_dbs_info->prev_cpu_wall);
 	this_dbs_info->prev_cpu_wall = cur_jiffies;
 	if (!total_ticks)
-		return;
+		return 200;
 	/*
 	 * Every sampling_rate, we check, if current idle time is less
 	 * than 20% (default), then we try to increase frequency
@@ -272,6 +269,83 @@ static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
 			idle_ticks = tmp_idle_ticks;
 	}
 	load = (100 * (total_ticks - idle_ticks)) / total_ticks;
+	return load;
+}
+#else
+
+#include <xen/interface/platform.h>
+static int dbs_calc_load(struct cpu_dbs_info_s *this_dbs_info)
+{
+	int load = 0;
+	dom0_op_t op;
+	uint64_t idletime[NR_CPUS];
+	struct cpufreq_policy *policy;
+	unsigned int j;
+	cpumask_t cpumap;
+
+	policy = this_dbs_info->cur_policy;
+	cpumap = policy->cpus;
+
+	op.cmd = DOM0_getidletime;
+	set_xen_guest_handle(op.u.getidletime.cpumap_bitmap, (uint8_t *) cpus_addr(cpumap));
+	op.u.getidletime.cpumap_nr_cpus = NR_CPUS;
+	set_xen_guest_handle(op.u.getidletime.idletime, idletime);
+	if (HYPERVISOR_dom0_op(&op))
+		return 200;
+
+	for_each_cpu_mask(j, cpumap) {
+		cputime64_t total_idle_nsecs, tmp_idle_nsecs;
+		cputime64_t total_wall_nsecs, tmp_wall_nsecs;
+		struct cpu_dbs_info_s *j_dbs_info;
+		unsigned long tmp_load, tmp_wall_msecs, tmp_idle_msecs;
+
+		j_dbs_info = &per_cpu(cpu_dbs_info, j);
+		total_idle_nsecs = idletime[j];
+		tmp_idle_nsecs = cputime64_sub(total_idle_nsecs,
+				j_dbs_info->prev_cpu_idle);
+		total_wall_nsecs = op.u.getidletime.now;
+		tmp_wall_nsecs = cputime64_sub(total_wall_nsecs,
+				j_dbs_info->prev_cpu_wall);
+
+		if (tmp_wall_nsecs == 0)
+			return 200;
+
+		j_dbs_info->prev_cpu_wall = total_wall_nsecs;
+		j_dbs_info->prev_cpu_idle = total_idle_nsecs;
+
+		/* Convert nsecs to msecs and clamp times to sane values. */
+		do_div(tmp_wall_nsecs, 1000000);
+		tmp_wall_msecs = tmp_wall_nsecs;
+		do_div(tmp_idle_nsecs, 1000000);
+		tmp_idle_msecs = tmp_idle_nsecs;
+		if (tmp_wall_msecs == 0)
+			tmp_wall_msecs = 1;
+		if (tmp_idle_msecs > tmp_wall_msecs)
+			tmp_idle_msecs = tmp_wall_msecs;
+
+		tmp_load = (100 * (tmp_wall_msecs - tmp_idle_msecs)) /
+				tmp_wall_msecs;
+		tmp_wall_msecs = tmp_wall_nsecs;
+		
+		load = max(load, min(100, (int) tmp_load));
+	}
+	return load;
+}
+#endif
+
+static void dbs_check_cpu(struct cpu_dbs_info_s *this_dbs_info)
+{
+	int load;
+
+	struct cpufreq_policy *policy;
+
+	if (!this_dbs_info->enable)
+		return;
+
+	policy = this_dbs_info->cur_policy;
+	load = dbs_calc_load(this_dbs_info);
+	if (load > 100) 
+		return;
 
 	/* Check for frequency increase */
 	if (load > dbs_tuners_ins.up_threshold) {
diff --git a/drivers/cpufreq/cpufreq_stats.c b/drivers/cpufreq/cpufreq_stats.c
index 9926997..86a5603 100644
--- a/drivers/cpufreq/cpufreq_stats.c
+++ b/drivers/cpufreq/cpufreq_stats.c
@@ -292,6 +292,9 @@ cpufreq_stat_notifier_trans (struct notifier_block *nb, unsigned long val,
 	if (old_index == new_index)
 		return 0;
 
+        if ((old_index < 0) || (new_index < 0))
+                return 0;
+
 	spin_lock(&cpufreq_stats_lock);
 	stat->last_index = new_index;
 #ifdef CONFIG_CPU_FREQ_STAT_DETAILS
diff --git a/include/xen/interface/dom0_ops.h b/include/xen/interface/dom0_ops.h
index 4edf430..9f56a80 100644
--- a/include/xen/interface/dom0_ops.h
+++ b/include/xen/interface/dom0_ops.h
@@ -71,6 +71,41 @@ struct dom0_memory_map_entry {
 typedef struct dom0_memory_map_entry dom0_memory_map_entry_t;
 DEFINE_XEN_GUEST_HANDLE(dom0_memory_map_entry_t);
 
+#define DOM0_change_freq         52
+struct dom0_change_freq {
+    /* IN variables */
+    uint32_t flags; /* Must be zero. */
+    uint32_t cpu;   /* Physical cpu. */
+    uint64_t freq;  /* New frequency (Hz). */
+};
+typedef struct dom0_change_freq dom0_change_freq_t;
+DEFINE_XEN_GUEST_HANDLE(dom0_change_freq_t);
+
+/*
+ * Get idle times (nanoseconds since boot) for physical CPUs specified in the
+ * @cpumap_bitmap with range [0..@cpumap_nr_cpus-1]. The @idletime array is
+ * indexed by CPU number; only entries with the corresponding @cpumap_bitmap
+ * bit set are written to. On return, @cpumap_bitmap is modified so that any
+ * non-existent CPUs are cleared. Such CPUs have their @idletime array entry
+ * cleared.
+ */
+#define DOM0_getidletime         53
+struct dom0_getidletime {
+    /* IN/OUT variables */
+    /* IN: CPUs to interrogate; OUT: subset of IN which are present */
+    XEN_GUEST_HANDLE(uint8_t) cpumap_bitmap;
+    /* IN variables */
+    /* Size of cpumap bitmap. */
+    uint32_t cpumap_nr_cpus;
+    /* Must be indexable for every cpu in cpumap_bitmap. */
+    XEN_GUEST_HANDLE(uint64_t) idletime;
+    /* OUT variables */
+    /* System time when the idletime snapshots were taken. */
+    uint64_t now;
+};
+typedef struct dom0_getidletime dom0_getidletime_t;
+DEFINE_XEN_GUEST_HANDLE(dom0_getidletime_t);
+
 struct dom0_op {
     uint32_t cmd;
     uint32_t interface_version; /* DOM0_INTERFACE_VERSION */
@@ -83,6 +118,8 @@ struct dom0_op {
         struct dom0_microcode         microcode;
         struct dom0_platform_quirk    platform_quirk;
         struct dom0_memory_map_entry  physical_memory_map;
+	struct dom0_change_freq       change_freq;
+	struct dom0_getidletime       getidletime;
 #ifdef CONFIG_X86_64
 	struct xenpf_stratus_call     stratus_call;
 #endif
diff --git a/include/xen/interface/platform.h b/include/xen/interface/platform.h
index 79f82da..f20f844 100644
--- a/include/xen/interface/platform.h
+++ b/include/xen/interface/platform.h
@@ -99,6 +99,41 @@ struct xenpf_platform_quirk {
 typedef struct xenpf_platform_quirk xenpf_platform_quirk_t;
 DEFINE_XEN_GUEST_HANDLE(xenpf_platform_quirk_t);
 
+#define XENPF_change_freq         52
+struct xenpf_change_freq {
+    /* IN variables */
+    uint32_t flags; /* Must be zero. */
+    uint32_t cpu;   /* Physical cpu. */
+    uint64_t freq;  /* New frequency (Hz). */
+};
+typedef struct xenpf_change_freq xenpf_change_freq_t;
+DEFINE_XEN_GUEST_HANDLE(xenpf_change_freq_t);
+
+/*
+ * Get idle times (nanoseconds since boot) for physical CPUs specified in the
+ * @cpumap_bitmap with range [0..@cpumap_nr_cpus-1]. The @idletime array is
+ * indexed by CPU number; only entries with the corresponding @cpumap_bitmap
+ * bit set are written to. On return, @cpumap_bitmap is modified so that any
+ * non-existent CPUs are cleared. Such CPUs have their @idletime array entry
+ * cleared.
+ */
+#define XENPF_getidletime         53
+struct xenpf_getidletime {
+    /* IN/OUT variables */
+    /* IN: CPUs to interrogate; OUT: subset of IN which are present */
+    XEN_GUEST_HANDLE(uint8_t) cpumap_bitmap;
+    /* IN variables */
+    /* Size of cpumap bitmap. */
+    uint32_t cpumap_nr_cpus;
+    /* Must be indexable for every cpu in cpumap_bitmap. */
+    XEN_GUEST_HANDLE(uint64_t) idletime;
+    /* OUT variables */
+    /* System time when the idletime snapshots were taken. */
+    uint64_t now;
+};
+typedef struct xenpf_getidletime xenpf_getidletime_t;
+DEFINE_XEN_GUEST_HANDLE(xenpf_getidletime_t);
+
 struct xen_platform_op {
     uint32_t cmd;
     uint32_t interface_version; /* XENPF_INTERFACE_VERSION */
@@ -109,6 +144,8 @@ struct xen_platform_op {
         struct xenpf_read_memtype      read_memtype;
         struct xenpf_microcode_update  microcode;
         struct xenpf_platform_quirk    platform_quirk;
+        struct xenpf_change_freq       change_freq;
+        struct xenpf_getidletime       getidletime;
         uint8_t                        pad[128];
     } u;
 };