powernow-k7: wrong FSB frequency detection

powernow-k7: wrong FSB frequency detection

[Sergey Vlasov]

[-- Attachment #1.1: Type: text/plain, Size: 3872 bytes --]

Hello!

On the MaxSelect A7W notebook the powernow-k7 driver in kernel 2.6.7
basically works, but shows completely wrong frequency values:

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
4355876 3266907 2800206 2489072 2177938 1555670

The CPU is Mobile AMD Athlon(tm) XP 2500+ - 1867 MHz max.

powernow-k7 is built as module; it gives these messages when loading:

powernow: PowerNOW! Technology present. Can scale: frequency and voltage.
powernow: FSB: 311.134 MHz
powernow: Found PSB header at c00f1960
powernow: Table version: 0x12
powernow: Flags: 0x0 (Mobile voltage regulator)
powernow: Settling Time: 100 microseconds.
powernow: Has 1 PST tables. (Only dumping ones relevant to this CPU).
powernow: No PST tables match this cpuid (0x7a0)
powernow: This is indicative of a broken BIOS.
powernow: Trying ACPI perflib
powernow: acpi:  P0: 1867 MHz 24000 mW 125 uS control 009c4176 SGTC 10000
powernow:    FID: 0x16 (14.0x [4355MHz])        VID: 0xb (1.450V)
powernow: acpi:  P1: 1400 MHz 14486 mW 125 uS control 009c41cf SGTC 10000
powernow:    FID: 0xf (10.5x [3266MHz]) VID: 0xe (1.300V)
powernow: acpi:  P2: 1200 MHz 14486 mW 125 uS control 009c41cc SGTC 10000
powernow:    FID: 0xc (9.0x [2800MHz])  VID: 0xe (1.300V)
powernow: acpi:  P3: 1067 MHz 14486 mW 125 uS control 009c426a SGTC 10000
powernow:    FID: 0xa (8.0x [2489MHz])  VID: 0x13 (1.200V)
powernow: acpi:  P4: 933 MHz 14486 mW 125 uS control 009c4268 SGTC 10000
powernow:    FID: 0x8 (7.0x [2177MHz])  VID: 0x13 (1.200V)
powernow: acpi:  P5: 668 MHz 14486 mW 125 uS control 009c42e4 SGTC 10000
powernow:    FID: 0x4 (5.0x [1555MHz])  VID: 0x17 (1.100V)
powernow: Minimum speed 1555 MHz. Maximum speed 4355 MHz.

The ACPI frequency values are correct, but powernow-k7 calculates its own
values based on the frequency multiplier and FSB, and the FSB value is
obviously wrong.

/proc/cpuinfo before loading powernow-k7:

processor       : 0
vendor_id       : AuthenticAMD
cpu family      : 6
model           : 10
model name      : Mobile AMD Athlon(tm) XP 2500+
stepping        : 0
cpu MHz         : 1866.808
cache size      : 512 KB
fdiv_bug        : no
hlt_bug         : no
f00f_bug        : no
coma_bug        : no
fpu             : yes
fpu_exception   : yes
cpuid level     : 1
wp              : yes
flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mp mmxext 3dnowext 3dnow
bogomips        : 3710.97

However, I tried to run this simple program to find the actual CPU speed:

#include <stdio.h>
#include <unistd.h>
#include <asm/msr.h>

int main(void)
{
	unsigned long long tsc1, tsc2;
	rdtscll(tsc1);
	sleep(1);
	rdtscll(tsc2);
	printf("TSC frequency: %llu Hz\n", tsc2 - tsc1);
	return 0;
}

And I got TSC frequency: 800811489 Hz (this was _before_ loading
powernow-k8).  This explains the calculated FSB value: the CPU actually
has 133 MHz FSB and multiplier of 6.0x (which gives approx. 800 Mhz), but
at some time before (during the cpu_khz initialization) the CPU frequency
multiplier was set to 14.0x (which gave 1866.808 MHz).  powernow-k7 does
not know about this change and happily uses wrong cpu_khz for its
calculations.

Another strange thing is that 6.0x does not appear in the ACPI powernow
table at all.

Just in case, here is the PSB dump from this machine:

PSB version: 12 flags: 0 settling time 100us res1 0 num pst 1
  PST 1  cpuid 780 fsb 133 mfid 3 svid a numberstates 8
    0   798000kHz  FID 06 ( 6.0)  VID 0a (1500mV)
    1   931000kHz  FID 08 ( 7.0)  VID 0a (1500mV)
    2  1064000kHz  FID 0a ( 8.0)  VID 0a (1500mV)
    3  1197000kHz  FID 0c ( 9.0)  VID 0a (1500mV)
    4  1330000kHz  FID 0e (10.0)  VID 0a (1500mV)
    5  1463000kHz  FID 00 (11.0)  VID 0a (1500mV)
    6  1596000kHz  FID 02 (12.0)  VID 0a (1500mV)
    7  1662500kHz  FID 03 (12.5)  VID 0a (1500mV)

-- 
Sergey Vlasov

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: powernow-k7: wrong FSB frequency detection

[Bruno Ducrot]

On Thu, Jul 01, 2004 at 04:25:55PM +0400, Sergey Vlasov wrote:
> Hello!
> 
> On the MaxSelect A7W notebook the powernow-k7 driver in kernel 2.6.7
> basically works, but shows completely wrong frequency values:
> 
> # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
> 4355876 3266907 2800206 2489072 2177938 1555670
> 
> The CPU is Mobile AMD Athlon(tm) XP 2500+ - 1867 MHz max.
> 
> powernow-k7 is built as module; it gives these messages when loading:
> 
> powernow: PowerNOW! Technology present. Can scale: frequency and voltage.
> powernow: FSB: 311.134 MHz
> powernow: Found PSB header at c00f1960
> powernow: Table version: 0x12
> powernow: Flags: 0x0 (Mobile voltage regulator)
> powernow: Settling Time: 100 microseconds.
> powernow: Has 1 PST tables. (Only dumping ones relevant to this CPU).
> powernow: No PST tables match this cpuid (0x7a0)
> powernow: This is indicative of a broken BIOS.
> powernow: Trying ACPI perflib
> powernow: acpi:  P0: 1867 MHz 24000 mW 125 uS control 009c4176 SGTC 10000
> powernow:    FID: 0x16 (14.0x [4355MHz])        VID: 0xb (1.450V)
> powernow: acpi:  P1: 1400 MHz 14486 mW 125 uS control 009c41cf SGTC 10000
> powernow:    FID: 0xf (10.5x [3266MHz]) VID: 0xe (1.300V)
> powernow: acpi:  P2: 1200 MHz 14486 mW 125 uS control 009c41cc SGTC 10000
> powernow:    FID: 0xc (9.0x [2800MHz])  VID: 0xe (1.300V)
> powernow: acpi:  P3: 1067 MHz 14486 mW 125 uS control 009c426a SGTC 10000
> powernow:    FID: 0xa (8.0x [2489MHz])  VID: 0x13 (1.200V)
> powernow: acpi:  P4: 933 MHz 14486 mW 125 uS control 009c4268 SGTC 10000
> powernow:    FID: 0x8 (7.0x [2177MHz])  VID: 0x13 (1.200V)
> powernow: acpi:  P5: 668 MHz 14486 mW 125 uS control 009c42e4 SGTC 10000
> powernow:    FID: 0x4 (5.0x [1555MHz])  VID: 0x17 (1.100V)
> powernow: Minimum speed 1555 MHz. Maximum speed 4355 MHz.
> 
> The ACPI frequency values are correct, but powernow-k7 calculates its own
> values based on the frequency multiplier and FSB, and the FSB value is
> obviously wrong.
> 
> /proc/cpuinfo before loading powernow-k7:
> 
> processor       : 0
> vendor_id       : AuthenticAMD
> cpu family      : 6
> model           : 10
> model name      : Mobile AMD Athlon(tm) XP 2500+
> stepping        : 0
> cpu MHz         : 1866.808
> cache size      : 512 KB
> fdiv_bug        : no
> hlt_bug         : no
> f00f_bug        : no
> coma_bug        : no
> fpu             : yes
> fpu_exception   : yes
> cpuid level     : 1
> wp              : yes
> flags           : fpu vme de pse tsc msr pae mce cx8 apic sep mtrr pge mca cmov pat pse36 mmx fxsr sse syscall mp mmxext 3dnowext 3dnow
> bogomips        : 3710.97
> 
> However, I tried to run this simple program to find the actual CPU speed:
> 
> #include <stdio.h>
> #include <unistd.h>
> #include <asm/msr.h>
> 
> int main(void)
> {
> 	unsigned long long tsc1, tsc2;
> 	rdtscll(tsc1);
> 	sleep(1);
> 	rdtscll(tsc2);
> 	printf("TSC frequency: %llu Hz\n", tsc2 - tsc1);
> 	return 0;
> }
> 
> And I got TSC frequency: 800811489 Hz (this was _before_ loading
> powernow-k8).  This explains the calculated FSB value: the CPU actually
> has 133 MHz FSB and multiplier of 6.0x (which gives approx. 800 Mhz), but
> at some time before (during the cpu_khz initialization) the CPU frequency
> multiplier was set to 14.0x (which gave 1866.808 MHz).  powernow-k7 does
> not know about this change and happily uses wrong cpu_khz for its
> calculations.
> 
> Another strange thing is that 6.0x does not appear in the ACPI powernow
> table at all.
> 
> Just in case, here is the PSB dump from this machine:
> 
> PSB version: 12 flags: 0 settling time 100us res1 0 num pst 1
>   PST 1  cpuid 780 fsb 133 mfid 3 svid a numberstates 8
>     0   798000kHz  FID 06 ( 6.0)  VID 0a (1500mV)
>     1   931000kHz  FID 08 ( 7.0)  VID 0a (1500mV)
>     2  1064000kHz  FID 0a ( 8.0)  VID 0a (1500mV)
>     3  1197000kHz  FID 0c ( 9.0)  VID 0a (1500mV)
>     4  1330000kHz  FID 0e (10.0)  VID 0a (1500mV)
>     5  1463000kHz  FID 00 (11.0)  VID 0a (1500mV)
>     6  1596000kHz  FID 02 (12.0)  VID 0a (1500mV)
>     7  1662500kHz  FID 03 (12.5)  VID 0a (1500mV)
> 

Could you please try this little debug stuff ?  That will not help for
now, but I want to be sure.

(I think the real problem is at:
linux-2.6/drivers/acpi/hardware/hwacpi.c in function acpi_hw_set_mode()
when we tell the bios that we go to ACPI mode, though I'm not sure at
100% yet, just this is the only possibility I see for now).


--- powernow-k7.c	2004/07/01 15:01:46	1.1
+++ powernow-k7.c	2004/07/01 15:14:16
@@ -566,6 +566,8 @@ static int __init powernow_cpu_init (str
 
 	/* A K7 with powernow technology is set to max frequency by BIOS */
 	fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.CFID];
+	dprintk(KERN_INFO PFX "cpu_khz %d    fid current %x\n", cpu_khz,
+			fidvidstatus.bits.CFID);
 	if (!fsb) {
 		printk(KERN_WARNING PFX "can not determine bus frequency\n");
 		return -EINVAL;


-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.

Re: powernow-k7: wrong FSB frequency detection

[Sergey Vlasov]

[-- Attachment #1.1: Type: text/plain, Size: 2742 bytes --]

On Thu, Jul 01, 2004 at 05:28:03PM +0200, Bruno Ducrot wrote:
> On Thu, Jul 01, 2004 at 04:25:55PM +0400, Sergey Vlasov wrote:
> > On the MaxSelect A7W notebook the powernow-k7 driver in kernel 2.6.7
> > basically works, but shows completely wrong frequency values:
> > 
> > # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
> > 4355876 3266907 2800206 2489072 2177938 1555670
> > 
[skip]
> Could you please try this little debug stuff ?  That will not help for
> now, but I want to be sure.

Here is the output:

powernow: PowerNOW! Technology present. Can scale: frequency and voltage.
powernow: cpu_khz 1866830    fid current 6
powernow: FSB: 311.138 MHz
powernow: Found PSB header at c00f1960
powernow: Table version: 0x12
powernow: Flags: 0x0 (Mobile voltage regulator)
powernow: Settling Time: 100 microseconds.
powernow: Has 1 PST tables. (Only dumping ones relevant to this CPU).
powernow: No PST tables match this cpuid (0x7a0)
powernow: This is indicative of a broken BIOS.
powernow: Trying ACPI perflib
powernow: acpi:  P0: 1867 MHz 24000 mW 125 uS control 009c4176 SGTC 10000
powernow:    FID: 0x16 (14.0x [4355MHz])        VID: 0xb (1.450V)
powernow: acpi:  P1: 1400 MHz 14486 mW 125 uS control 009c41cf SGTC 10000
powernow:    FID: 0xf (10.5x [3266MHz]) VID: 0xe (1.300V)
powernow: acpi:  P2: 1200 MHz 14486 mW 125 uS control 009c41cc SGTC 10000
powernow:    FID: 0xc (9.0x [2800MHz])  VID: 0xe (1.300V)
powernow: acpi:  P3: 1067 MHz 14486 mW 125 uS control 009c426a SGTC 10000
powernow:    FID: 0xa (8.0x [2489MHz])  VID: 0x13 (1.200V)
powernow: acpi:  P4: 933 MHz 14486 mW 125 uS control 009c4268 SGTC 10000
powernow:    FID: 0x8 (7.0x [2177MHz])  VID: 0x13 (1.200V)
powernow: acpi:  P5: 668 MHz 14486 mW 125 uS control 009c42e4 SGTC 10000
powernow:    FID: 0x4 (5.0x [1555MHz])  VID: 0x17 (1.100V)
powernow: Minimum speed 1555 MHz. Maximum speed 4355 MHz.


> (I think the real problem is at:
> linux-2.6/drivers/acpi/hardware/hwacpi.c in function acpi_hw_set_mode()
> when we tell the bios that we go to ACPI mode, though I'm not sure at
> 100% yet, just this is the only possibility I see for now).
> 
> 
> --- powernow-k7.c	2004/07/01 15:01:46	1.1
> +++ powernow-k7.c	2004/07/01 15:14:16
> @@ -566,6 +566,8 @@ static int __init powernow_cpu_init (str
>  
>  	/* A K7 with powernow technology is set to max frequency by BIOS */
>  	fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.CFID];
> +	dprintk(KERN_INFO PFX "cpu_khz %d    fid current %x\n", cpu_khz,
> +			fidvidstatus.bits.CFID);
>  	if (!fsb) {
>  		printk(KERN_WARNING PFX "can not determine bus frequency\n");
>  		return -EINVAL;
> 
> 
> -- 
> Bruno Ducrot
> 
> --  Which is worse:  ignorance or apathy?
> --  Don't know.  Don't care.

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq

Re: powernow-k7: wrong FSB frequency detection

[Bruno Ducrot]

On Thu, Jul 01, 2004 at 08:35:49PM +0400, Sergey Vlasov wrote:
> On Thu, Jul 01, 2004 at 05:28:03PM +0200, Bruno Ducrot wrote:
> > On Thu, Jul 01, 2004 at 04:25:55PM +0400, Sergey Vlasov wrote:
> > > On the MaxSelect A7W notebook the powernow-k7 driver in kernel 2.6.7
> > > basically works, but shows completely wrong frequency values:
> > > 
> > > # cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
> > > 4355876 3266907 2800206 2489072 2177938 1555670
> > > 
> [skip]
> > Could you please try this little debug stuff ?  That will not help for
> > now, but I want to be sure.
> 

Thanks.  IMHO we should restore the old behaviour, that is using the
max fid as per powernow spec, then provide something for people who
unfortunately do have borken bios (those who put the processor not in
max fid before booting OS).

Dave, could you please apply this patch?  I will provide the workaround
for the unfortunate tomorow, after searching who do have the borken
bios, and probably yet another powernow boot option :(

--- linux-2.6/arch/i386/kernel/cpu/cpufreq/powernow-k7.c	2004/07/01 15:01:46	1.1
+++ linux-2.6/arch/i386/kernel/cpu/cpufreq/powernow-k7.c	2004/07/01 17:56:12
@@ -565,7 +565,7 @@ static int __init powernow_cpu_init (str
 	rdmsrl (MSR_K7_FID_VID_STATUS, fidvidstatus.val);
 
 	/* A K7 with powernow technology is set to max frequency by BIOS */
-	fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.CFID];
+	fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.MFID];
 	if (!fsb) {
 		printk(KERN_WARNING PFX "can not determine bus frequency\n");
 		return -EINVAL;

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.

Re: powernow-k7: wrong FSB frequency detection

[Dominik Brodowski]

On Thu, Jul 01, 2004 at 08:02:15PM +0200, Bruno Ducrot wrote:
> Dave, could you please apply this patch?  I will provide the workaround
> for the unfortunate tomorow, after searching who do have the borken
> bios, and probably yet another powernow boot option :(

and/or DMI parsing?

	Dominik

Re: powernow-k7: wrong FSB frequency detection

[Bruno Ducrot]

On Fri, Jul 02, 2004 at 10:58:37AM +0200, Dominik Brodowski wrote:
> On Thu, Jul 01, 2004 at 08:02:15PM +0200, Bruno Ducrot wrote:
> > Dave, could you please apply this patch?  I will provide the workaround
> > for the unfortunate tomorow, after searching who do have the borken
> > bios, and probably yet another powernow boot option :(
> 
> and/or DMI parsing?

That why I need to search the borken bios ;)

-- 
Bruno Ducrot

--  Which is worse:  ignorance or apathy?
--  Don't know.  Don't care.

Re: powernow-k7: wrong FSB frequency detection

[Sergey Vlasov]

[-- Attachment #1.1: Type: text/plain, Size: 2707 bytes --]

On Thu, Jul 01, 2004 at 08:02:15PM +0200, Bruno Ducrot wrote:
> > > Could you please try this little debug stuff ?  That will not help for
> > > now, but I want to be sure.
> > 
> 
> Thanks.  IMHO we should restore the old behaviour, that is using the
> max fid as per powernow spec, then provide something for people who
> unfortunately do have borken bios (those who put the processor not in
> max fid before booting OS).
> 
> Dave, could you please apply this patch?  I will provide the workaround
> for the unfortunate tomorow, after searching who do have the borken
> bios, and probably yet another powernow boot option :(
> 
> --- linux-2.6/arch/i386/kernel/cpu/cpufreq/powernow-k7.c	2004/07/01 15:01:46	1.1
> +++ linux-2.6/arch/i386/kernel/cpu/cpufreq/powernow-k7.c	2004/07/01 17:56:12
> @@ -565,7 +565,7 @@ static int __init powernow_cpu_init (str
>  	rdmsrl (MSR_K7_FID_VID_STATUS, fidvidstatus.val);
>  
>  	/* A K7 with powernow technology is set to max frequency by BIOS */
> -	fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.CFID];
> +	fsb = (10 * cpu_khz) / fid_codes[fidvidstatus.bits.MFID];
>  	if (!fsb) {
>  		printk(KERN_WARNING PFX "can not determine bus frequency\n");
>  		return -EINVAL;
> 
> -- 

With this patch the results are much better:

powernow: PowerNOW! Technology present. Can scale: frequency and voltage.
powernow: FSB: 133.374 MHz
powernow: Found PSB header at c00f1960
powernow: Table version: 0x12
powernow: Flags: 0x0 (Mobile voltage regulator)
powernow: Settling Time: 100 microseconds.
powernow: Has 1 PST tables. (Only dumping ones relevant to this CPU).
powernow: No PST tables match this cpuid (0x7a0)
powernow: This is indicative of a broken BIOS.
powernow: Trying ACPI perflib
powernow: acpi:  P0: 1867 MHz 24000 mW 125 uS control 009c4176 SGTC 10000
powernow:    FID: 0x16 (14.0x [1867MHz])        VID: 0xb (1.450V)
powernow: acpi:  P1: 1400 MHz 14486 mW 125 uS control 009c41cf SGTC 10000
powernow:    FID: 0xf (10.5x [1400MHz]) VID: 0xe (1.300V)
powernow: acpi:  P2: 1200 MHz 14486 mW 125 uS control 009c41cc SGTC 10000
powernow:    FID: 0xc (9.0x [1200MHz])  VID: 0xe (1.300V)
powernow: acpi:  P3: 1067 MHz 14486 mW 125 uS control 009c426a SGTC 10000
powernow:    FID: 0xa (8.0x [1066MHz])  VID: 0x13 (1.200V)
powernow: acpi:  P4: 933 MHz 14486 mW 125 uS control 009c4268 SGTC 10000
powernow:    FID: 0x8 (7.0x [933MHz])   VID: 0x13 (1.200V)
powernow: acpi:  P5: 668 MHz 14486 mW 125 uS control 009c42e4 SGTC 10000
powernow:    FID: 0x4 (5.0x [666MHz])   VID: 0x17 (1.100V)
powernow: Minimum speed 666 MHz. Maximum speed 1867 MHz.

# cat /sys/devices/system/cpu/cpu0/cpufreq/scaling_available_frequencies 
1867236 1400427 1200366 1066992 933618 666870 

[-- Attachment #1.2: Type: application/pgp-signature, Size: 189 bytes --]

[-- Attachment #2: Type: text/plain, Size: 143 bytes --]

_______________________________________________
Cpufreq mailing list
Cpufreq@www.linux.org.uk
http://www.linux.org.uk/mailman/listinfo/cpufreq