acpi-cpufreq: get the cur_freq from acpi_processor_performance states

As the initialized freq_tables maybe different from the p-states values, so the array index is different as well.

p-states value: [2400 2400 2000 ...], while the freq_tables:
[2400 2000 ... CPUFREQ_TABLE_END]. After setted the freqs 2000, the perf->state is 3 while the freqs_table's index should be 2. So when call the get_cur_freq_on_cpu, the freqs value we get is 2400.

So, fix the problem with the correct tables.

Signed-off-by Wang Weidong <[email protected]>

---
 drivers/cpufreq/acpi-cpufreq.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

drivers/cpufreq/acpi-cpufreq.c +1 -0

index b0c18ed..ac93885 100644
@@ -365,6 +365,7 @@ static u32 get_cur_val(const struct cpumask *mask)
 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 {
        struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
+       struct acpi_processor_performance *perf;
        unsigned int freq;
        unsigned int cached_freq;
 

As the initialized freq_tables maybe different from the p-states values, so the array index is different as well.

p-states value: [2400 2400 2000 ...], while the freq_tables:
[2400 2000 ... CPUFREQ_TABLE_END]. After setted the freqs 2000, the perf->state is 3 while the freqs_table's index should be 2. So when call the get_cur_freq_on_cpu, the freqs value we get is 2400.

So, fix the problem with the correct tables.

Signed-off-by Wang Weidong <[email protected]>

---
 drivers/cpufreq/acpi-cpufreq.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

drivers/cpufreq/acpi-cpufreq.c +3 -1

index b0c18ed..ac93885 100644
@@ -365,6 +365,7 @@ static u32 get_cur_val(const struct cpumask *mask)
 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 {
 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
+	struct acpi_processor_performance *perf;
 	unsigned int freq;
 	unsigned int cached_freq;
 

@@ -375,7 +376,8 @@ static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 		return 0;
 	}
 
-	cached_freq = data->freq_table[data->acpi_data->state].frequency;
+	perf = data->acpi_data;
+	cached_freq = perf->states[perf->state].core_frequency * 1000;
 	freq = extract_freq(get_cur_val(cpumask_of(cpu)), data);
 	if (freq != cached_freq) {
 		/*

Quoted 9 lines · Wang Weidong

As the initialized freq_tables maybe different from the p-states values, so the array index is different as well.

p-states value: [2400 2400 2000 ...], while the freq_tables:
[2400 2000 ... CPUFREQ_TABLE_END]. After setted the freqs 2000, the perf->state is 3 while the freqs_table's index should be 2. So when call the get_cur_freq_on_cpu, the freqs value we get is 2400.

Quoted 1 line

So, fix the problem with the correct tables.

What you're saying is basically that freq_table and perf->states diverge at one point. Shouldn't we re-generate freq_table in that case instead of fixing up get_cur_freq_on_cpu() only in a quite indirect way?

Quoted 28 lines

Signed-off-by Wang Weidong <[email protected]>

---
 drivers/cpufreq/acpi-cpufreq.c | 4 +++-
 1 file changed, 3 insertions(+), 1 deletion(-)

drivers/cpufreq/acpi-cpufreq.c +3 -1

index b0c18ed..ac93885 100644
@@ -365,6 +365,7 @@ static u32 get_cur_val(const struct cpumask *mask)
 static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 {
 	struct acpi_cpufreq_data *data = per_cpu(acfreq_data, cpu);
+	struct acpi_processor_performance *perf;
 	unsigned int freq;
 	unsigned int cached_freq;
 

@@ -375,7 +376,8 @@ static unsigned int get_cur_freq_on_cpu(unsigned int cpu)
 		return 0;
 	}
 
-	cached_freq = data->freq_table[data->acpi_data->state].frequency;
+	perf = data->acpi_data;
+	cached_freq = perf->states[perf->state].core_frequency * 1000;
 	freq = extract_freq(get_cur_val(cpumask_of(cpu)), data);
 	if (freq != cached_freq) {
 		/*

Quoted 10 lines · Rafael J. Wysocki

Quoted 9 lines · Wang Weidong

As the initialized freq_tables maybe different from the p-states values, so the array index is different as well.

p-states value: [2400 2400 2000 ...], while the freq_tables:
[2400 2000 ... CPUFREQ_TABLE_END]. After setted the freqs 2000, the perf->state is 3 while the freqs_table's index should be 2. So when call the get_cur_freq_on_cpu, the freqs value we get is 2400.

Quoted 7 lines

Quoted 1 line

So, fix the problem with the correct tables.

What you're saying is basically that freq_table and perf->states diverge at one point. Shouldn't we re-generate freq_table in that case instead of fixing up get_cur_freq_on_cpu() only in a quite indirect way?

Hi Rafael,

Thanks for your reply.

You mean that we should re-generate the freq_table in that case? Could we fix the table init like this:

drivers/cpufreq/acpi-cpufreq.c +1 -1

@@ -779,7 +779,7 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
 
        /* table init */
        for (i = 0; i < perf->state_count; i++) {
-               if (i > 0 && perf->states[i].core_frequency >=
+               if (i > 0 && perf->states[i].core_frequency >
                    data->freq_table[valid_states-1].frequency / 1000)
                        continue;
 

when the value is same, we just keep the value into the freq_table.

Quoted 36 lines · Wang Weidong

Quoted 10 lines · Rafael J. Wysocki

Quoted 9 lines · Wang Weidong

As the initialized freq_tables maybe different from the p-states values, so the array index is different as well.

p-states value: [2400 2400 2000 ...], while the freq_tables:
[2400 2000 ... CPUFREQ_TABLE_END]. After setted the freqs 2000, the perf->state is 3 while the freqs_table's index should be 2. So when call the get_cur_freq_on_cpu, the freqs value we get is 2400.

Quoted 7 lines

Quoted 1 line

So, fix the problem with the correct tables.

What you're saying is basically that freq_table and perf->states diverge at one point. Shouldn't we re-generate freq_table in that case instead of fixing up get_cur_freq_on_cpu() only in a quite indirect way?

Hi Rafael,

Thanks for your reply.

You mean that we should re-generate the freq_table in that case? Could we fix the table init like this:

drivers/cpufreq/acpi-cpufreq.c +1 -1

@@ -779,7 +779,7 @@ static int acpi_cpufreq_cpu_init(struct cpufreq_policy *policy)
 
        /* table init */
        for (i = 0; i < perf->state_count; i++) {
-               if (i > 0 && perf->states[i].core_frequency >=
+               if (i > 0 && perf->states[i].core_frequency >
                    data->freq_table[valid_states-1].frequency / 1000)
                        continue;
 

Quoted 1 line

when the value is same, we just keep the value into the freq_table.

That would only be OK if it is guaranteed that the set of available states hasn't changed, which I'm not sure is the case.

Rafael

.