Merge bk://linux-dj.bkbits.net/cpufreq

into ppc970.osdl.org:/home/torvalds/v2.6/linux

arch/i386/kernel/cpu/cpufreq/Kconfig

@@ -120,6 +120,21 @@ config X86_SPEEDSTEP_CENTRINO
 	  
 	  If in doubt, say N.
 
+config X86_SPEEDSTEP_CENTRINO_TABLE
+	bool
+	depends on X86_SPEEDSTEP_CENTRINO
+	default y
+
+config X86_SPEEDSTEP_CENTRINO_ACPI
+	bool "Use ACPI tables to decode valid frequency/voltage pairs (EXPERIMENTAL)"
+	depends on EXPERIMENTAL
+	depends on ((X86_SPEEDSTEP_CENTRINO = "m" && ACPI_PROCESSOR) || (X86_SPEEDSTEP_CENTRINO = "y" && ACPI_PROCESSOR = "y"))
+	help
+	  Use primarily the information provided in the BIOS ACPI tables
+	  to determine valid CPU frequency and voltage pairings.
+
+	  If in doubt, say Y.
+
 config X86_SPEEDSTEP_ICH
 	tristate "Intel Speedstep on ICH-M chipsets (ioport interface)"
 	depends on CPU_FREQ_TABLE
@@ -161,6 +176,16 @@ config X86_SPEEDSTEP_LIB
 	depends on (X86_SPEEDSTEP_ICH || X86_SPEEDSTEP_SMI || X86_P4_CLOCKMOD)
 	default (X86_SPEEDSTEP_ICH || X86_SPEEDSTEP_SMI || X86_P4_CLOCKMOD)
 
+config X86_SPEEDSTEP_RELAXED_CAP_CHECK
+	bool "Relaxed speedstep capability checks"
+	depends on (X86_SPEEDSTEP_SMI || X86_SPEEDSTEP_ICH)
+	help
+	  Don't perform all checks for a speedstep capable system which would 
+	  normally be done. Some ancient or strange systems, though speedstep 
+	  capable, don't always indicate that they are speedstep capable. This 
+	  option let's the probing code bypass some of those checks if the
+	  parameter "relaxed_check=1" is passed to the module.
+
 config X86_LONGRUN
 	tristate "Transmeta LongRun"
 	depends on CPU_FREQ

arch/i386/kernel/cpu/cpufreq/elanfreq.c

@@ -165,7 +165,8 @@ static void elanfreq_set_cpu_state (unsigned int state) {
 
 
 /**
- *	elanfreq_validatespeed: test if frequency range is valid 
+ *	elanfreq_validatespeed: test if frequency range is valid
+ *      @policy: the policy to validate
  *
  *	This function checks if a given frequency range in kHz is valid 
  *      for the hardware supported by the driver. 

arch/i386/kernel/cpu/cpufreq/longhaul.c

@@ -458,7 +458,7 @@ static int __init longhaul_cpu_init (struct cpufreq_policy *policy)
 	return 0;
 }
 
-static int longhaul_cpu_exit(struct cpufreq_policy *policy)
+static int __exit longhaul_cpu_exit(struct cpufreq_policy *policy)
 {
 	cpufreq_frequency_table_put_attr(policy->cpu);
 	return 0;

arch/i386/kernel/cpu/cpufreq/longrun.c

@@ -33,7 +33,7 @@ static unsigned int longrun_low_freq, longrun_high_freq;
  * Reads the current LongRun policy by access to MSR_TMTA_LONGRUN_FLAGS
  * and MSR_TMTA_LONGRUN_CTRL
  */
-static void longrun_get_policy(struct cpufreq_policy *policy)
+static void __init longrun_get_policy(struct cpufreq_policy *policy)
 {
 	u32 msr_lo, msr_hi;
 
@@ -57,7 +57,7 @@ static void longrun_get_policy(struct cpufreq_policy *policy)
 
 /**
  * longrun_set_policy - sets a new CPUFreq policy
- * @policy - new policy
+ * @policy: new policy
  *
  * Sets a new CPUFreq policy on LongRun-capable processors. This function
  * has to be called with cpufreq_driver locked.
@@ -106,6 +106,7 @@ static int longrun_set_policy(struct cpufreq_policy *policy)
 
 /**
  * longrun_verify_poliy - verifies a new CPUFreq policy
+ * @policy: the policy to verify
  *
  * Validates a new CPUFreq policy. This function has to be called with 
  * cpufreq_driver locked.
@@ -130,6 +131,8 @@ static int longrun_verify_policy(struct cpufreq_policy *policy)
 
 /**
  * longrun_determine_freqs - determines the lowest and highest possible core frequency
+ * @low_freq: an int to put the lowest frequency into
+ * @high_freq: an int to put the highest frequency into
  *
  * Determines the lowest and highest possible core frequencies on this CPU.
  * This is necessary to calculate the performance percentage according to
@@ -142,6 +145,7 @@ static unsigned int __init longrun_determine_freqs(unsigned int *low_freq,
 	u32 msr_lo, msr_hi;
 	u32 save_lo, save_hi;
 	u32 eax, ebx, ecx, edx;
+	u32 try_hi;
 	struct cpuinfo_x86 *c = cpu_data;
 
 	if (!low_freq || !high_freq)
@@ -184,12 +188,14 @@ static unsigned int __init longrun_determine_freqs(unsigned int *low_freq,
 	 * upper limit to make the calculation more accurate.
 	 */
 	cpuid(0x80860007, &eax, &ebx, &ecx, &edx);
-	if (ecx > 90) {
-		/* set to 0 to 80 perf_pctg */
+	/* try decreasing in 10% steps, some processors react only
+	 * on some barrier values */
+	for (try_hi = 80; try_hi > 0 && ecx > 90; try_hi -=10) {
+		/* set to 0 to try_hi perf_pctg */
 		msr_lo &= 0xFFFFFF80;
 		msr_hi &= 0xFFFFFF80;
 		msr_lo |= 0;
-		msr_hi |= 80;
+		msr_hi |= try_hi;
 		wrmsr(MSR_TMTA_LONGRUN_CTRL, msr_lo, msr_hi);
 
 		/* read out current core MHz and current perf_pctg */

arch/i386/kernel/cpu/cpufreq/p4-clockmod.c

@@ -181,21 +181,24 @@ static unsigned int cpufreq_p4_get_frequency(struct cpuinfo_x86 *c)
 {
 	if ((c->x86 == 0x06) && (c->x86_model == 0x09)) {
 		/* Pentium M */
-		printk(KERN_DEBUG PFX "Warning: Pentium M detected. The speedstep_centrino module\n");
-		printk(KERN_DEBUG PFX "offers voltage scaling in addition of frequency scaling. You\n");
-		printk(KERN_DEBUG PFX "should use that instead of p4-clockmod, if possible.\n");
+		printk(KERN_WARNING PFX "Warning: Pentium M detected. "
+		       "The speedstep_centrino module offers voltage scaling"
+		       " in addition of frequency scaling. You should use "
+		       "that instead of p4-clockmod, if possible.\n");
 		return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_PM);
 	}
 
 	if (c->x86 != 0xF) {
-		printk(KERN_DEBUG PFX "Unknown p4-clockmod-capable CPU. Please send an e-mail to <linux@brodo.de>\n");
+		printk(KERN_WARNING PFX "Unknown p4-clockmod-capable CPU. Please send an e-mail to <linux@brodo.de>\n");
 		return 0;
 	}
 
 	if (speedstep_detect_processor() == SPEEDSTEP_PROCESSOR_P4M) {
-		printk(KERN_DEBUG PFX "Warning: Pentium 4-M detected. The speedstep-ich or acpi cpufreq \n");
-		printk(KERN_DEBUG PFX "modules offers voltage scaling in addition of frequency scaling. You\n");
-		printk(KERN_DEBUG PFX "should use either one instead of p4-clockmod, if possible.\n");
+		printk(KERN_WARNING PFX "Warning: Pentium 4-M detected. "
+		       "The speedstep-ich or acpi cpufreq modules offer "
+		       "voltage scaling in addition of frequency scaling. "
+		       "You should use either one instead of p4-clockmod, "
+		       "if possible.\n");
 		return speedstep_get_processor_frequency(SPEEDSTEP_PROCESSOR_P4M);
 	}
 

arch/i386/kernel/cpu/cpufreq/powernow-k6.c

@@ -118,7 +118,9 @@ static int powernow_k6_verify(struct cpufreq_policy *policy)
 
 /**
  * powernow_k6_setpolicy - sets a new CPUFreq policy
- * @policy - new policy
+ * @policy: new policy
+ * @target_freq: the target frequency
+ * @relation: how that frequency relates to achieved frequency (CPUFREQ_RELATION_L or CPUFREQ_RELATION_H)
  *
  * sets a new CPUFreq policy
  */

arch/i386/kernel/cpu/cpufreq/powernow-k7.c

@@ -98,18 +98,20 @@ static int check_powernow(void)
 		return 0;
 	}
 
-	if ((c->x86_model == 6) && (c->x86_mask == 0)) {
-		printk (KERN_INFO PFX "K7 660[A0] core detected, enabling errata workarounds\n");
-		have_a0 = 1;
-	}
-
 	/* Get maximum capabilities */
 	maxei = cpuid_eax (0x80000000);
 	if (maxei < 0x80000007) {	/* Any powernow info ? */
+#ifdef MODULE
 		printk (KERN_INFO PFX "No powernow capabilities detected\n");
+#endif
 		return 0;
 	}
 
+	if ((c->x86_model == 6) && (c->x86_mask == 0)) {
+		printk (KERN_INFO PFX "K7 660[A0] core detected, enabling errata workarounds\n");
+		have_a0 = 1;
+	}
+
 	cpuid(0x80000007, &eax, &ebx, &ecx, &edx);
 	printk (KERN_INFO PFX "PowerNOW! Technology present. Can scale: ");
 

arch/i386/kernel/cpu/cpufreq/powernow-k8.h

 /*
- *   (c) 2003 Advanced Micro Devices, Inc.
+ *  (c) 2003, 2004 Advanced Micro Devices, Inc.
  *  Your use of this code is subject to the terms and conditions of the
- *  GNU general public license version 2. See "../../../COPYING" or
+ *  GNU general public license version 2. See "COPYING" or
  *  http://www.gnu.org/licenses/gpl.html
  */
 
+struct powernow_k8_data {
+	unsigned int cpu;
+
+	u32 numps;  /* number of p-states */
+	u32 batps;  /* number of p-states supported on battery */
+
+	/* these values are constant when the PSB is used to determine
+	 * vid/fid pairings, but are modified during the ->target() call
+	 * when ACPI is used */
+	u32 rvo;     /* ramp voltage offset */
+	u32 irt;     /* isochronous relief time */
+	u32 vidmvs;  /* usable value calculated from mvs */
+	u32 vstable; /* voltage stabilization time, units 20 us */
+	u32 plllock; /* pll lock time, units 1 us */
+
+	/* keep track of the current fid / vid */
+	u32 currvid;
+	u32 currfid;
+
+	/* the powernow_table includes all frequency and vid/fid pairings:
+	 * fid are the lower 8 bits of the index, vid are the upper 8 bits.
+	 * frequency is in kHz */
+	struct cpufreq_frequency_table  *powernow_table;
+
+#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;
+#endif
+};
+
+
 /* processor's cpuid instruction support */
-#define CPUID_PROCESSOR_SIGNATURE             1	/* function 1               */
-#define CPUID_F1_FAM                 0x00000f00	/* family mask              */
-#define CPUID_F1_XFAM                0x0ff00000	/* extended family mask     */
-#define CPUID_F1_MOD                 0x000000f0	/* model mask               */
-#define CPUID_F1_STEP                0x0000000f	/* stepping level mask      */
-#define CPUID_XFAM_MOD               0x0ff00ff0	/* xtended fam, fam + model */
-#define ATHLON64_XFAM_MOD            0x00000f40	/* xtended fam, fam + model */
-#define OPTERON_XFAM_MOD             0x00000f50	/* xtended fam, fam + model */
-#define ATHLON64_REV_C0                       8
+#define CPUID_PROCESSOR_SIGNATURE             1	/* function 1 */
+#define CPUID_XFAM_MOD               0x0ff00ff0	/* extended fam, fam + model */
+#define ATHLON64_XFAM_MOD            0x00000f40	/* extended fam, fam + model */
+#define OPTERON_XFAM_MOD             0x00000f50	/* extended fam, fam + model */
 #define CPUID_GET_MAX_CAPABILITIES   0x80000000
 #define CPUID_FREQ_VOLT_CAPABILITIES 0x80000007
 #define P_STATE_TRANSITION_CAPABLE            6
@@ -47,10 +74,23 @@
 #define MSR_S_HI_MAX_WORKING_VID  0x001f0000
 #define MSR_S_HI_START_VID        0x00001f00
 #define MSR_S_HI_CURRENT_VID      0x0000001f
+#define MSR_C_HI_STP_GNT_BENIGN   0x00000001
+
+/*
+ * There are restrictions frequencies have to follow:
+ * - only 1 entry in the low fid table ( <=1.4GHz )
+ * - lowest entry in the high fid table must be >= 2 * the entry in the
+ *   low fid table
+ * - lowest entry in the high fid table must be a <= 200MHz + 2 * the entry
+ *   in the low fid table
+ * - the parts can only step at 200 MHz intervals, so 1.9 GHz is never valid
+ * - lowest frequency must be >= interprocessor hypertransport link speed
+ *   (only applies to MP systems obviously)
+ */
 
 /* fids (frequency identifiers) are arranged in 2 tables - lo and hi */
-#define LO_FID_TABLE_TOP     6
-#define HI_FID_TABLE_BOTTOM  8
+#define LO_FID_TABLE_TOP     6	/* fid values marking the boundary    */
+#define HI_FID_TABLE_BOTTOM  8	/* between the low and high tables    */
 
 #define LO_VCOFREQ_TABLE_TOP    1400	/* corresponding vco frequency values */
 #define HI_VCOFREQ_TABLE_BOTTOM 1600
@@ -58,14 +98,12 @@
 #define MIN_FREQ_RESOLUTION  200 /* fids jump by 2 matching freq jumps by 200 */
 
 #define MAX_FID 0x2a	/* Spec only gives FID values as far as 5 GHz */
-
 #define LEAST_VID 0x1e	/* Lowest (numerically highest) useful vid value */
 
 #define MIN_FREQ 800	/* Min and max freqs, per spec */
 #define MAX_FREQ 5000
 
 #define INVALID_FID_MASK 0xffffffc1  /* not a valid fid if these bits are set */
-
 #define INVALID_VID_MASK 0xffffffe0  /* not a valid vid if these bits are set */
 
 #define STOP_GRANT_5NS 1 /* min poss memory access latency for voltage change */
@@ -73,18 +111,35 @@
 #define PLL_LOCK_CONVERSION (1000/5) /* ms to ns, then divide by clock period */
 
 #define MAXIMUM_VID_STEPS 1  /* Current cpus only allow a single step of 25mV */
-
 #define VST_UNITS_20US 20   /* Voltage Stabalization Time is in units of 20us */
 
 /*
-Version 1.4 of the PSB table. This table is constructed by BIOS and is
-to tell the OS's power management driver which VIDs and FIDs are
-supported by this particular processor. This information is obtained from
-the data sheets for each processor model by the system vendor and
-incorporated into the BIOS.
-If the data in the PSB / PST is wrong, then this driver will program the
-wrong values into hardware, which is very likely to lead to a crash.
-*/
+ * Most values of interest are enocoded in a single field of the _PSS
+ * entries: the "control" value.
+ */
+                                                                                                    
+#define IRT_SHIFT      30
+#define RVO_SHIFT      28
+#define PLL_L_SHIFT    20
+#define MVS_SHIFT      18
+#define VST_SHIFT      11
+#define VID_SHIFT       6
+#define IRT_MASK        3
+#define RVO_MASK        3
+#define PLL_L_MASK   0x7f
+#define MVS_MASK        3
+#define VST_MASK     0x7f
+#define VID_MASK     0x1f
+#define FID_MASK     0x3f
+
+
+/*
+ * Version 1.4 of the PSB table. This table is constructed by BIOS and is
+ * to tell the OS's power management driver which VIDs and FIDs are
+ * supported by this particular processor.
+ * If the data in the PSB / PST is wrong, then this driver will program the
+ * wrong values into hardware, which is very likely to lead to a crash.
+ */
 
 #define PSB_ID_STRING      "AMDK7PNOW!"
 #define PSB_ID_STRING_LEN  10
@@ -117,6 +172,8 @@ struct pst_s {
 #define dprintk(msg...) do { } while(0)
 #endif
 
-static inline int core_voltage_pre_transition(u32 reqvid);
-static inline int core_voltage_post_transition(u32 reqvid);
-static inline int core_frequency_transition(u32 reqfid);
+static int core_voltage_pre_transition(struct powernow_k8_data *data, u32 reqvid);
+static int core_voltage_post_transition(struct powernow_k8_data *data, u32 reqvid);
+static int core_frequency_transition(struct powernow_k8_data *data, u32 reqfid);
+
+static void powernow_k8_acpi_pst_values(struct powernow_k8_data *data, unsigned int index);

arch/i386/kernel/cpu/cpufreq/speedstep-ich.c

@@ -67,6 +67,7 @@ static struct cpufreq_frequency_table speedstep_freqs[] = {
 /**
  * speedstep_set_state - set the SpeedStep state
  * @state: new processor frequency state (SPEEDSTEP_LOW or SPEEDSTEP_HIGH)
+ * @notify: whether to call cpufreq_notify_transition for CPU speed changes
  *
  *   Tries to change the SpeedStep state. 
  */
@@ -239,8 +240,10 @@ static unsigned int speedstep_detect_chipset (void)
 
 
 /**
- * speedstep_setpolicy - set a new CPUFreq policy
+ * speedstep_target - set a new CPUFreq policy
  * @policy: new policy
+ * @target_freq: the target frequency
+ * @relation: how that frequency relates to achieved frequency (CPUFREQ_RELATION_L or CPUFREQ_RELATION_H)
  *
  * Sets a new CPUFreq policy.
  */
@@ -261,7 +264,7 @@ static int speedstep_target (struct cpufreq_policy *policy,
 
 /**
  * speedstep_verify - verifies a new CPUFreq policy
- * @freq: new policy
+ * @policy: new policy
  *
  * Limit must be within speedstep_low_freq and speedstep_high_freq, with
  * at least one border included.

arch/i386/kernel/cpu/cpufreq/speedstep-lib.c

@@ -10,6 +10,7 @@
 
 #include <linux/kernel.h>
 #include <linux/module.h> 
+#include <linux/moduleparam.h>
 #include <linux/init.h>
 #include <linux/cpufreq.h>
 #include <linux/pci.h>
@@ -30,6 +31,12 @@
 #define dprintk(msg...) do { } while(0)
 #endif
 
+#ifdef CONFIG_X86_SPEEDSTEP_RELAXED_CAP_CHECK
+static int relaxed_check = 0;
+#else
+#define relaxed_check 0
+#endif
+
 /*********************************************************************
  *                   GET PROCESSOR CORE SPEED IN KHZ                 *
  *********************************************************************/
@@ -120,7 +127,7 @@ static unsigned int pentiumM_get_frequency(void)
 	msr_tmp = (msr_lo >> 22) & 0x1f;
 	dprintk(KERN_DEBUG "speedstep-lib: bits 22-26 are 0x%x\n", msr_tmp);
 
-	return (msr_tmp * 100 * 10000);
+	return (msr_tmp * 100 * 1000);
 }
 
 
@@ -210,7 +217,7 @@ unsigned int speedstep_detect_processor (void)
 		ebx = cpuid_ebx(0x00000001);
 		ebx &= 0x000000FF;
 
-		dprintk(KERN_INFO "ebx value is %x, x86_mask is %x\n", ebx, c->86_mask);
+		dprintk(KERN_INFO "ebx value is %x, x86_mask is %x\n", ebx, c->x86_mask);
 
 		switch (c->x86_mask) {
 		case 4: 
@@ -265,6 +272,7 @@ unsigned int speedstep_detect_processor (void)
 		ebx = cpuid_ebx(0x00000001);
 
 		ebx &= 0x000000FF;
+
 		if (ebx != 0x06)
 			return 0;
 
@@ -292,7 +300,7 @@ unsigned int speedstep_detect_processor (void)
 		 */
 		rdmsr(MSR_IA32_PLATFORM_ID, msr_lo, msr_hi);
 		dprintk(KERN_DEBUG "cpufreq: Coppermine: MSR_IA32_PLATFORM ID is 0x%x, 0x%x\n", msr_lo, msr_hi);
-		if ((msr_hi & (1<<18)) && (msr_hi & (3<<24))) {
+		if ((msr_hi & (1<<18)) && (relaxed_check ? 1 : (msr_hi & (3<<24)))) {
 			if (c->x86_mask == 0x01)
 				return SPEEDSTEP_PROCESSOR_PIII_C_EARLY;
 			else
@@ -362,6 +370,11 @@ unsigned int speedstep_get_freqs(unsigned int processor,
 }
 EXPORT_SYMBOL_GPL(speedstep_get_freqs);
 
+#ifdef CONFIG_X86_SPEEDSTEP_RELAXED_CAP_CHECK
+module_param(relaxed_check, int, 0444);
+MODULE_PARM_DESC(relaxed_check, "Don't do all checks for speedstep capability.");
+#endif
+
 MODULE_AUTHOR ("Dominik Brodowski <linux@brodo.de>");
 MODULE_DESCRIPTION ("Library for Intel SpeedStep 1 or 2 cpufreq drivers.");
 MODULE_LICENSE ("GPL");

arch/i386/kernel/cpu/cpufreq/speedstep-smi.c

@@ -90,10 +90,12 @@ static int speedstep_smi_ownership (void)
 
 /**
  * speedstep_smi_get_freqs - get SpeedStep preferred & current freq.
+ * @low: the low frequency value is placed here
+ * @high: the high frequency value is placed here
+ *
  * Only available on later SpeedStep-enabled systems, returns false results or
  * even hangs [cf. bugme.osdl.org # 1422] on earlier systems. Empirical testing
  * shows that the latter occurs if !(ist_info.event & 0xFFFF).
- *
  */
 static int speedstep_smi_get_freqs (unsigned int *low, unsigned int *high)
 {
@@ -141,6 +143,7 @@ static int speedstep_get_state (void)
 /**
  * speedstep_set_state - set the SpeedStep state
  * @state: new processor frequency state (SPEEDSTEP_LOW or SPEEDSTEP_HIGH)
+ * @notify: whether to call cpufreq_notify_transition
  *
  */
 static void speedstep_set_state (unsigned int state, unsigned int notify)
@@ -224,7 +227,7 @@ static int speedstep_target (struct cpufreq_policy *policy,
 
 /**
  * speedstep_verify - verifies a new CPUFreq policy
- * @freq: new policy
+ * @policy: new policy
  *
  * Limit must be within speedstep_low_freq and speedstep_high_freq, with
  * at least one border included.

drivers/cpufreq/cpufreq.c

@@ -504,19 +504,16 @@ static int cpufreq_resume(struct sys_device * sysdev)
 	if (cpufreq_driver->setpolicy)
 		ret = cpufreq_driver->setpolicy(cpu_policy);
 	else
-	/* CPUFREQ_RELATION_H or CPUFREQ_RELATION_L have the same effect here, as cpu_policy->cur is known
-	 * to be a valid and exact target frequency
-	 */
+		/* CPUFREQ_RELATION_H or CPUFREQ_RELATION_L have the same effect here, as cpu_policy->cur is known
+		 * to be a valid and exact target frequency
+		 */
 		ret = cpufreq_driver->target(cpu_policy, cpu_policy->cur, CPUFREQ_RELATION_H);
 
-	if (ret) {
+	if (ret)
 		printk(KERN_ERR "cpufreq: resume failed in ->setpolicy/target step on CPU %u\n", cpu_policy->cpu);
-		goto out;
-	}
 
- out:
+out:
 	cpufreq_cpu_put(cpu_policy);
-
 	return ret;
 }
 
@@ -963,6 +960,7 @@ EXPORT_SYMBOL_GPL(cpufreq_notify_transition);
 int cpufreq_register_driver(struct cpufreq_driver *driver_data)
 {
 	unsigned long flags;
+	int ret;
 
 	if (!driver_data || !driver_data->verify || !driver_data->init ||
 	    ((!driver_data->setpolicy) && (!driver_data->target)))
@@ -976,7 +974,28 @@ int cpufreq_register_driver(struct cpufreq_driver *driver_data)
 	cpufreq_driver = driver_data;
 	spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
 
-	return sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
+	ret = sysdev_driver_register(&cpu_sysdev_class,&cpufreq_sysdev_driver);
+
+	if ((!ret) && !(cpufreq_driver->flags & CPUFREQ_STICKY)) {
+		int i;
+		ret = -ENODEV;
+
+		/* check for at least one working CPU */
+		for (i=0; i<NR_CPUS; i++)
+			if (cpufreq_cpu_data[i])
+				ret = 0;
+
+		/* if all ->init() calls failed, unregister */
+		if (ret) {
+			sysdev_driver_unregister(&cpu_sysdev_class, &cpufreq_sysdev_driver);
+
+			spin_lock_irqsave(&cpufreq_driver_lock, flags);
+			cpufreq_driver = NULL;
+			spin_unlock_irqrestore(&cpufreq_driver_lock, flags);
+		}
+	}
+
+	return (ret);
 }
 EXPORT_SYMBOL_GPL(cpufreq_register_driver);
 

include/linux/cpufreq.h

@@ -175,6 +175,7 @@ struct freq_attr;
 struct cpufreq_driver {
 	struct module           *owner;
 	char			name[CPUFREQ_NAME_LEN];
+	u8			flags;
 
 	/* needed by all drivers */
 	int	(*init)		(struct cpufreq_policy *policy);
@@ -192,6 +193,11 @@ struct cpufreq_driver {
 	struct freq_attr	**attr;
 };
 
+/* flags */
+
+#define CPUFREQ_STICKY	0x01	/* the driver isn't removed even if 
+				   all ->init() calls failed */
+
 int cpufreq_register_driver(struct cpufreq_driver *driver_data);
 int cpufreq_unregister_driver(struct cpufreq_driver *driver_data);