The ACPI thermal management code supports slowing down a CPU when it's
overheating. Right now that's done by choosing to run it at 100%, 75%, 50%
or 25% of full speed. However, most CPUs do not allow an arbitrary
frequency to be set and so will run at the first frequency below that value.
This doesn't match the intent of the specification, which is to drop the
frequency state by state until the temperature stabalises. Fix this up
so it uses actual frequencies rather than percentages.

Reported by: Gene Snider <snider6982@comcast.net>
Signed-off-by: Matthew Garrett <mjg@redhat.com>
---
 drivers/acpi/processor_thermal.c |   45 +++++++++++++++++++++----------------
 1 files changed, 25 insertions(+), 20 deletions(-)

diff --git a/drivers/acpi/processor_thermal.c b/drivers/acpi/processor_thermal.c
index 870550d..1c4eb60 100644
--- a/drivers/acpi/processor_thermal.c
+++ b/drivers/acpi/processor_thermal.c
@@ -52,10 +52,8 @@ ACPI_MODULE_NAME("processor_thermal");
  * _any_ cpufreq driver and not only the acpi-cpufreq driver.
  */
 
-#define CPUFREQ_THERMAL_MIN_STEP 0
-#define CPUFREQ_THERMAL_MAX_STEP 3
 
-static DEFINE_PER_CPU(unsigned int, cpufreq_thermal_reduction_pctg);
+static DEFINE_PER_CPU(unsigned int, cpufreq_thermal_limit_state);
 static unsigned int acpi_thermal_cpufreq_is_init = 0;
 
 static int cpu_has_cpufreq(unsigned int cpu)
@@ -70,19 +68,19 @@ static int acpi_thermal_cpufreq_notifier(struct notifier_block *nb,
 					 unsigned long event, void *data)
 {
 	struct cpufreq_policy *policy = data;
-	unsigned long max_freq = 0;
+	int state = per_cpu(cpufreq_thermal_limit_state, policy->cpu);
+	struct cpufreq_frequency_table *table;
 
 	if (event != CPUFREQ_ADJUST)
-		goto out;
+		return 0;
+
+	table = cpufreq_frequency_get_table(policy->cpu);
 
-	max_freq = (
-	    policy->cpuinfo.max_freq *
-	    (100 - per_cpu(cpufreq_thermal_reduction_pctg, policy->cpu) * 20)
-	) / 100;
+	if (!table)
+		return 0;
 
-	cpufreq_verify_within_limits(policy, 0, max_freq);
+	cpufreq_verify_within_limits(policy, 0, table[state].frequency);
 
-      out:
 	return 0;
 }
 
@@ -92,10 +90,21 @@ static struct notifier_block acpi_thermal_cpufreq_notifier_block = {
 
 static int cpufreq_get_max_state(unsigned int cpu)
 {
+	int count = 0;
+	struct cpufreq_frequency_table *table;
+
 	if (!cpu_has_cpufreq(cpu))
 		return 0;
 
-	return CPUFREQ_THERMAL_MAX_STEP;
+	table = cpufreq_frequency_get_table(cpu);
+
+	if (!table)
+		return 0;
+
+	while (table[count].frequency != CPUFREQ_TABLE_END)
+		count++;
+
+	return count;
 }
 
 static int cpufreq_get_cur_state(unsigned int cpu)
@@ -103,7 +112,7 @@ static int cpufreq_get_cur_state(unsigned int cpu)
 	if (!cpu_has_cpufreq(cpu))
 		return 0;
 
-	return per_cpu(cpufreq_thermal_reduction_pctg, cpu);
+	return per_cpu(cpufreq_thermal_limit_state, cpu);
 }
 
 static int cpufreq_set_cur_state(unsigned int cpu, int state)
@@ -111,7 +120,7 @@ static int cpufreq_set_cur_state(unsigned int cpu, int state)
 	if (!cpu_has_cpufreq(cpu))
 		return 0;
 
-	per_cpu(cpufreq_thermal_reduction_pctg, cpu) = state;
+	per_cpu(cpufreq_thermal_limit_state, cpu) = state;
 	cpufreq_update_policy(cpu);
 	return 0;
 }
@@ -122,7 +131,7 @@ void acpi_thermal_cpufreq_init(void)
 
 	for (i = 0; i < nr_cpu_ids; i++)
 		if (cpu_present(i))
-			per_cpu(cpufreq_thermal_reduction_pctg, i) = 0;
+			per_cpu(cpufreq_thermal_limit_state, i) = 0;
 
 	i = cpufreq_register_notifier(&acpi_thermal_cpufreq_notifier_block,
 				      CPUFREQ_POLICY_NOTIFIER);
@@ -170,15 +179,11 @@ int acpi_processor_get_limit_info(struct acpi_processor *pr)
 	return 0;
 }
 
-/* thermal coolign device callbacks */
+/* thermal cooling device callbacks */
 static int acpi_processor_max_state(struct acpi_processor *pr)
 {
 	int max_state = 0;
 
-	/*
-	 * There exists four states according to
-	 * cpufreq_thermal_reduction_ptg. 0, 1, 2, 3
-	 */
 	max_state += cpufreq_get_max_state(pr->id);
 	if (pr->flags.throttling)
 		max_state += (pr->throttling.state_count -1);
-- 
1.7.6.1