Skip to contentSkip to conversation

fix up CPU detection in p4-clockmod

PATCH In review Dominik Brodowski 3 participants 5 messages Jan 10, 2004 – Jan 11, 2004

lore
Share & tools
lore

Dominik Brodowski

author
#0 lore raw

PATCH 1/1 fix up CPU detection in p4-clockmod

Too many users use the p4-clockmod cpufreq driver instead of the more advanced speedstep-centrino, speedstep-ich or even acpi drivers. All of the latter (usually) provide voltage scaling, while the p4-clockmod driver only offers a variant of frequency scaling. So, warn users if they try out this driver instead.

Also, instead of using a local copy, use the speedstep_lib infrastructure for detecting the processor speed. Adding the Pentium-M get_frequency function to that module only costs about 200 bytes in object size.

 arch/i386/kernel/cpu/cpufreq/Kconfig         |   10 ++--
 arch/i386/kernel/cpu/cpufreq/p4-clockmod.c   |   55 ++++++++-------------------
 arch/i386/kernel/cpu/cpufreq/speedstep-lib.c |   23 +++++++++++
 arch/i386/kernel/cpu/cpufreq/speedstep-lib.h |    6 ++
 4 files changed, 52 insertions(+), 42 deletions(-)
arch/i386/kernel/cpu/cpufreq/Kconfig +5 -5
diff -ruN linux-original/arch/i386/kernel/cpu/cpufreq/Kconfig linux/arch/i386/kernel/cpu/cpufreq/Kconfig
@@ -145,11 +145,6 @@
 
 	  If in doubt, say N.
 
-config X86_SPEEDSTEP_LIB
-	tristate
-	depends on (X86_SPEEDSTEP_ICH || X86_SPEEDSTEP_SMI)
-	default (X86_SPEEDSTEP_ICH || X86_SPEEDSTEP_SMI)
-
 config X86_P4_CLOCKMOD
 	tristate "Intel Pentium 4 clock modulation"
 	depends on CPU_FREQ_TABLE

@@ -161,6 +156,11 @@
 
 	  If in doubt, say N.
 
+config X86_SPEEDSTEP_LIB
+	tristate
+	depends on (X86_SPEEDSTEP_ICH || X86_SPEEDSTEP_SMI || X86_P4_CLOCKMOD)
+	default (X86_SPEEDSTEP_ICH || X86_SPEEDSTEP_SMI || X86_P4_CLOCKMOD)
+
 config X86_LONGRUN
 	tristate "Transmeta LongRun"
 	depends on CPU_FREQ
arch/i386/kernel/cpu/cpufreq/p4-clockmod.c +18 -37

81 lines not on this page · raw message

arch/i386/kernel/cpu/cpufreq/speedstep-lib.c +23 -0
diff -ruN linux-original/arch/i386/kernel/cpu/cpufreq/speedstep-lib.c linux/arch/i386/kernel/cpu/cpufreq/speedstep-lib.c
@@ -104,6 +104,26 @@
 }
 
 
+static unsigned int pentiumM_get_frequency(void)
+{
+	u32     msr_lo, msr_tmp;
+
+	rdmsr(MSR_IA32_EBL_CR_POWERON, msr_lo, msr_tmp);
+	dprintk(KERN_DEBUG "speedstep-lib: PM - MSR_IA32_EBL_CR_POWERON: 0x%x 0x%x\n", msr_lo, msr_tmp);
+
+	/* see table B-2 of 24547212.pdf */
+	if (msr_lo & 0x00040000) {
+		printk(KERN_DEBUG "speedstep-lib: PM - invalid FSB: 0x%x 0x%x\n", msr_lo, msr_tmp);
+		return 0;
+	}
+
+	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);
+}
+
+
 static unsigned int pentium4_get_frequency(void)
 {
 	struct cpuinfo_x86 *c = &boot_cpu_data;

@@ -151,6 +171,9 @@
 unsigned int speedstep_get_processor_frequency(unsigned int processor)
 {
 	switch (processor) {
+	case SPEEDSTEP_PROCESSOR_PM:
+		return pentiumM_get_frequency();
+	case SPEEDSTEP_PROCESSOR_P4D:
 	case SPEEDSTEP_PROCESSOR_P4M:
 		return pentium4_get_frequency();
 	case SPEEDSTEP_PROCESSOR_PIII_T:
arch/i386/kernel/cpu/cpufreq/speedstep-lib.h +6 -0
diff -ruN linux-original/arch/i386/kernel/cpu/cpufreq/speedstep-lib.h linux/arch/i386/kernel/cpu/cpufreq/speedstep-lib.h
@@ -17,6 +17,12 @@
 #define SPEEDSTEP_PROCESSOR_PIII_T 		0x00000003  /* Tualatin core */
 #define SPEEDSTEP_PROCESSOR_P4M			0x00000004  /* P4-M  */
 
+/* the following processors are not speedstep-capable and are not auto-detected
+ * in speedstep_detect_processor(). However, their speed can be detected using
+ * the speedstep_get_processor_frequency() call. */
+#define SPEEDSTEP_PROCESSOR_PM			0xFFFFFF03  /* Pentium M  */
+#define SPEEDSTEP_PROCESSOR_P4D			0xFFFFFF04  /* desktop P4  */
+
 /* speedstep states -- only two of them */
 
 #define SPEEDSTEP_HIGH                  0x00000000
#1 lore raw
Quoted 9 lines · Dominik Brodowski

Too many users use the p4-clockmod cpufreq driver instead of the more advanced speedstep-centrino, speedstep-ich or even acpi drivers. All of the latter (usually) provide voltage scaling, while the p4-clockmod driver only offers a variant of frequency scaling. So, warn users if they try out this driver instead.

Also, instead of using a local copy, use the speedstep_lib infrastructure for detecting the processor speed. Adding the Pentium-M get_frequency function to that module only costs about 200 bytes in object size.

another idea might be putting the cpu identify code into a speedstep.ko and have that request_module the right module

Dave.

author
#2 lore raw
Quoted 7 lines · Dave Jones
Quoted 6 lines · Dominik Brodowski

Too many users use the p4-clockmod cpufreq driver instead of the more advanced speedstep-centrino, speedstep-ich or even acpi drivers. All of the latter (usually) provide voltage scaling, while the p4-clockmod driver only offers a variant of frequency scaling. So, warn users if they try out this driver instead.

Quoted 6 lines
Quoted 3 lines

Also, instead of using a local copy, use the speedstep_lib infrastructure for detecting the processor speed. Adding the Pentium-M get_frequency function to that module only costs about 200 bytes in object size.

another idea might be putting the cpu identify code into a speedstep.ko and have that request_module the right module

That's not a valid option as a replacement -- there _are_ different modules possible for the same CPU. However, it's a good idea if it is implemented as an "autoloader".

Dominik

_______________________________________________
Cpufreq mailing list
[email protected]
http://www.linux.org.uk/mailman/listinfo/cpufreq

#3 lore raw
In reply to Dave Jones
Quoted 8 lines · Dave Jones
Quoted 6 lines · Dominik Brodowski

Too many users use the p4-clockmod cpufreq driver instead of the more advanced speedstep-centrino, speedstep-ich or even acpi drivers. All of the latter (usually) provide voltage scaling, while the p4-clockmod driver only offers a variant of frequency scaling. So, warn users if they try out this driver instead.

Quoted 7 lines
Quoted 3 lines

Also, instead of using a local copy, use the speedstep_lib infrastructure for detecting the processor speed. Adding the Pentium-M get_frequency function to that module only costs about 200 bytes in object size.

another idea might be putting the cpu identify code into a speedstep.ko and have that request_module the right module

That seems like a more user-friendly approach to me.

I'll merge Dominik's patch while we think about it.

author
#4 lore raw
Quoted 13 lines · Andrew Morton
Quoted 12 lines · Dave Jones
Quoted 9 lines · Dominik Brodowski

Too many users use the p4-clockmod cpufreq driver instead of the more advanced speedstep-centrino, speedstep-ich or even acpi drivers. All of the latter (usually) provide voltage scaling, while the p4-clockmod driver only offers a variant of frequency scaling. So, warn users if they try out this driver instead.

Also, instead of using a local copy, use the speedstep_lib infrastructure for detecting the processor speed. Adding the Pentium-M get_frequency function to that module only costs about 200 bytes in object size.

Quoted 7 lines
Quoted 3 lines

another idea might be putting the cpu identify code into a speedstep.ko and have that request_module the right module

That seems like a more user-friendly approach to me.

I'll merge Dominik's patch while we think about it.

We still need cpu identification code in the various speedstep-* drivers as they need to do a few things differently on different CPUs...

The "request_modul"ing of the right module could be done more easily in userspace, IMHO. A small sample is attached.

Dominik

_______________________________________________
Cpufreq mailing list
[email protected]
http://www.linux.org.uk/mailman/listinfo/cpufreq