scaling_cur_freq is broken with acpi-cpufreq driver and ondemand governor

[Guillaume Chazarain <guichaz@yahoo.fr>]

[-- Attachment #1: Type: text/plain, Size: 817 bytes --]

Hi,

As of linux-2.6.20-rc2 using acpi-cpufreq and ondemand on my Pentium M 
(Asus V6VA laptop) seems broken. 
/sys/devices/system/cpu/cpu0/cpufreq/scaling_cur_freq contains most of 
the time 1, sometimes the actual frequency or (rarely) some random value.

The problem seems to come from 
http://www.kernel.org/git/?p=linux/kernel/git/torvalds/linux-2.6.git;a=commitdiff;h=7650b281b091f39f5e97f13b45ab3813b1526b65

The attached patch fixes the problem for me, but it was obtained from 
random mutilations of the aforementioned patch. For instance, I don't 
understand how cmd.val is used, the freqs.old = extract_freq(cmd.val, 
data); I introduced seems to use an uninitialized value from reading the 
code. But it seems to be also the case in cmd.val = (cmd.val & 
~INTEL_MSR_RANGE) | msr;

Thanks.

-- 
Guillaume


[-- Attachment #2: ACPI-CPUFREQ.diff --]
[-- Type: text/x-patch, Size: 763 bytes --]

--- a/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.c
+++ b/arch/i386/kernel/cpu/cpufreq/acpi-cpufreq.c
@@ -400,6 +400,8 @@ static int acpi_cpufreq_target(struct cp
 	online_policy_cpus = policy->cpus;
 #endif
 
+	freqs.old = extract_freq(cmd.val, data);
+	freqs.new = data->freq_table[next_state].frequency;
 	next_perf_state = data->freq_table[next_state].index;
 	if (perf->state == next_perf_state) {
 		if (unlikely(data->resume)) {
@@ -439,8 +441,6 @@ static int acpi_cpufreq_target(struct cp
 	else
 		cpu_set(policy->cpu, cmd.mask);
 
-	freqs.old = data->freq_table[perf->state].frequency;
-	freqs.new = data->freq_table[next_perf_state].frequency;
 	for_each_cpu_mask(i, cmd.mask) {
 		freqs.cpu = i;
 		cpufreq_notify_transition(&freqs, CPUFREQ_PRECHANGE);