Skip to contentSkip to conversation

cpufreq: fix powernow-k7 smp kernel driver on up machines

PATCH In review Thomas Meyer 2 participants 2 messages Feb 27, 2006 – Mar 1, 2006

lore
Share & tools
lore
#0 lore raw

PATCH 1/1 cpufreq: fix powernow-k7 smp kernel driver on up machines

From: Thomas Meyer <[email protected]>

This patch fixes the powernow-k7 cpufreq driver smp kernel on an up machine.

---

This patch is against branch v2.6.16-rc4. Please give me positive and/or negativ feedback about this fixing approach!

With kind regads
Thomas

(quoted diff) +0 -0

b/arch/i386/kernel/cpu/common.c
index 7eb9213..67fcee6 100644

arch/i386/kernel/cpu/common.c +1 -0
@@ -336,6 +336,7 @@ void __devinit identify_cpu(struct cpuin
 	int i;
 
 	c->loops_per_jiffy = loops_per_jiffy;
+	c->cpu_khz = cpu_khz;
 	c->x86_cache_size = -1;
 	c->x86_vendor = X86_VENDOR_UNKNOWN;
 	c->cpuid_level = -1;	/* CPUID not detected */
(quoted diff) +0 -0

b/arch/i386/kernel/cpu/cpufreq/powernow-k7.c
index edcd626..2125026 100644

arch/i386/kernel/cpu/cpufreq/powernow-k7.c +1 -1
@@ -586,7 +586,7 @@ static int __init powernow_cpu_init (str
 	if (result)
 		return result;
 
-	fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.CFID];
+	fsb = (10 * current_cpu_data.cpu_khz) /

fid_codes[fidvidstatus.bits.CFID];
if (!fsb) {
printk(KERN_WARNING PFX "can not determine bus frequency\n"); return -EINVAL;

arch/i386/kernel/smpboot.c +0 -0
(quoted diff) +0 -0

b/arch/i386/kernel/timers/common.c
index 8163fe0..760add2 100644

arch/i386/kernel/timers/common.c +1 -1
@@ -148,7 +148,9 @@ unsigned long read_timer_tsc(void)
 }
 
 
-/* calculate cpu_khz */
+/* calculate cpu_khz for boot_cpu only, because timer_init ->

select_timer
+ * calls init_cpu_khz, before boot_cpu_data is transfered to cpu_data[x]
+ */
void init_cpu_khz(void)
{
if (cpu_has_tsc) {
@@ -170,3 +172,26 @@ void init_cpu_khz(void)
}
}

+void init_cpu_khz_smp(void)
+{
+ if (cpu_has(&(current_cpu_data), X86_FEATURE_TSC)) {
+ unsigned long tsc_quotient = calibrate_tsc();
+
+ if (tsc_quotient) {
+ /* report CPU clock rate in Hz.
+ * The formula is (10^6 * 2^32) / (2^32 * 1 / (clocks/us)) =
+ * clock/second. Our precision is about 100 ppm.
+ */
+ { unsigned long eax=0, edx=1000;
+ __asm__("divl %2"
+ :"=a" (current_cpu_data.cpu_khz), "=d" (edx)
+ :"r" (tsc_quotient),
+ "0" (eax), "1" (edx));
+ printk("Detected %u.%03u MHz processor.\n",
+ current_cpu_data.cpu_khz / 1000,
+ current_cpu_data.cpu_khz % 1000);
+ }
+ }
+ }
+}
+

(quoted diff) +0 -0

b/arch/i386/kernel/timers/timer_tsc.c
index a7f5a2a..0a3e29e 100644

arch/i386/kernel/timers/timer_tsc.c +7 -10
@@ -341,23 +341,20 @@ static inline void cpufreq_delayed_get(v
 
 int recalibrate_cpu_khz(void)
 {
-#ifndef CONFIG_SMP
-	unsigned int cpu_khz_old = cpu_khz;
+	unsigned int cpu_khz_old = current_cpu_data.cpu_khz;
 
-	if (cpu_has_tsc) {
+	if (cpu_has(&(current_cpu_data), X86_FEATURE_TSC)) {
 		local_irq_disable();
-		init_cpu_khz();
+		init_cpu_khz_smp();
 		local_irq_enable();
-		cpu_data[0].loops_per_jiffy =
-		    cpufreq_scale(cpu_data[0].loops_per_jiffy,
+		current_cpu_data.loops_per_jiffy =
+		    cpufreq_scale(current_cpu_data.loops_per_jiffy,
 			          cpu_khz_old,
-				  cpu_khz);
+				  current_cpu_data.cpu_khz);
 		return 0;
 	} else
 		return -ENODEV;
-#else
-	return -ENODEV;
-#endif
+
 }
 EXPORT_SYMBOL(recalibrate_cpu_khz);
 
include/asm-i386/processor.h +1 -0
index feca5d9..c486001 100644
@@ -67,6 +67,7 @@ struct cpuinfo_x86 {
 	char	pad0;
 	int	x86_power;
 	unsigned long loops_per_jiffy;
+	unsigned int  cpu_khz;
 	unsigned char x86_max_cores;	/* cpuid returned max cores value */
 	unsigned char booted_cores;	/* number of cores as seen by OS */
 	unsigned char apicid;
include/asm-i386/timer.h +1 -0
index aed1643..bbec11d 100644
@@ -58,6 +58,7 @@ extern struct init_timer_opts timer_cycl
 extern unsigned long calibrate_tsc(void);
 extern unsigned long read_timer_tsc(void);
 extern void init_cpu_khz(void);
+extern void init_cpu_khz_smp(void);
 extern int recalibrate_cpu_khz(void);
 #ifdef CONFIG_HPET_TIMER
 extern struct init_timer_opts timer_hpet_init;
#1 lore raw
Quoted 3 lines · Thomas Meyer

This patch fixes the powernow-k7 cpufreq driver smp kernel on an up
machine.

- It was wordwrapped.

- Your description doesn't describe what is being fixed, nor how it was fixed.

- That open-coded assembly-language divide should probably be converted to div64(). If that's not possible we'd need to see a description of why.

- Please cc [email protected] on the next submission.