[lm-sensors] Support for Family 10h CPUs - Not all sensors are

[lm-sensors] Support for Family 10h CPUs - Not all sensors are

[Jan Bessai]

Hi,

according to AMD (
http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/41322.pdf
) not all AMD K10 CPUs have defect sensors. In fact only DR-* revisions
are affected. All CPUs with Socket AM3 as well as all Phenom II Processors
and some Opterons are unaffected (DA-*, RB-*, HY-*).
Please reconsider the statement "Embedded sensors are known to be
unreliable, and won't be supported ever.". Linux should not lack
temperature sensor support for such popular Processors.

Kind regards,

 Jan


_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Support for Family 10h CPUs - Not all sensors are

[Jean Delvare]

On Sat, 21 Nov 2009 18:28:25 +0100, Jan Bessai wrote:
> according to AMD (
> http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/41322.pdf
> ) not all AMD K10 CPUs have defect sensors. In fact only DR-* revisions
> are affected. All CPUs with Socket AM3 as well as all Phenom II Processors
> and some Opterons are unaffected (DA-*, RB-*, HY-*).
> Please reconsider the statement "Embedded sensors are known to be
> unreliable, and won't be supported ever.".

We can do that. But how does this translate into CPUID numbers? That's
all we can check.

> Linux should not lack temperature sensor support for such popular
> Processors.

This is a pointless statement, I hope you realize it. There are a lot
of things that should not be, but are still, for various reasons. Want
an example? A big company such as AMD shouldn't have released millions
of CPUs with broken thermal sensors. As it stands, we do what we can to
cope with their mistake.

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Support for Family 10h CPUs - Not all sensors are

[Jan Bessai]

Jean Delvare wrote:
>> Linux should not lack temperature sensor support for such popular
>> Processors.
>>     
>
> This is a pointless statement, I hope you realize it. There are a lot
> of things that should not be, but are still, for various reasons. Want
> an example? A big company such as AMD shouldn't have released millions
> of CPUs with broken thermal sensors. As it stands, we do what we can to
> cope with their mistake.
Pardon, this was meant to be an argument for adding the drivers -
without any critical intention. I appreciate your work, I see that it
was AMD who released broken products and I just wanted to give a reason
for not dropping support for the complete k10 cpu family.



_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Support for Family 10h CPUs - Not all sensors are

[Clemens Ladisch]

Jean Delvare wrote:
> On Sat, 21 Nov 2009 18:28:25 +0100, Jan Bessai wrote:
> > according to AMD (
> > http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/41322.pdf
> > ) not all AMD K10 CPUs have defect sensors. In fact only DR-* revisions
> > are affected. All CPUs with Socket AM3 as well as all Phenom II Processors
> > and some Opterons are unaffected (DA-*, RB-*, HY-*).
> > Please reconsider the statement "Embedded sensors are known to be
> > unreliable, and won't be supported ever.".
>  
> We can do that. But how does this translate into CPUID numbers?

Family 0xF, extended family 1, model 2.

Or, to quote a certain driver that happend to be posted on this list
recently:

static bool __devinit has_erratum_319(void)
{
	/*
	 * Erratum 319: The thermal sensor of older Family 10h processors
	 *              (B steppings) may be unreliable.
	 */
	return boot_cpu_data.x86 = 0x10 && boot_cpu_data.x86_model <= 2;
}


HTH
Clemens

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Support for Family 10h CPUs - Not all sensors are

[Jean Delvare]

On Mon, 23 Nov 2009 08:58:37 +0100, Clemens Ladisch wrote:
> Jean Delvare wrote:
> > On Sat, 21 Nov 2009 18:28:25 +0100, Jan Bessai wrote:
> > > according to AMD (
> > > http://www.amd.com/us-en/assets/content_type/white_papers_and_tech_docs/41322.pdf
> > > ) not all AMD K10 CPUs have defect sensors. In fact only DR-* revisions
> > > are affected. All CPUs with Socket AM3 as well as all Phenom II Processors
> > > and some Opterons are unaffected (DA-*, RB-*, HY-*).
> > > Please reconsider the statement "Embedded sensors are known to be
> > > unreliable, and won't be supported ever.".
> >  
> > We can do that. But how does this translate into CPUID numbers?
> 
> Family 0xF, extended family 1, model 2.
> 
> Or, to quote a certain driver that happend to be posted on this list
> recently:
> 
> static bool __devinit has_erratum_319(void)
> {
> 	/*
> 	 * Erratum 319: The thermal sensor of older Family 10h processors
> 	 *              (B steppings) may be unreliable.
> 	 */
> 	return boot_cpu_data.x86 = 0x10 && boot_cpu_data.x86_model <= 2;
> }

OK, this would lead to the following patch for sensors-detect. Can you
please test it?

Index: CHANGES
=================================--- CHANGES	(révision 5799)
+++ CHANGES	(copie de travail)
@@ -25,7 +25,7 @@
                   The SMSC LPC47M233 isn't currently supported
                   Support upcoming sysfs path to i2c adapters
                   Rename the modprobe configuration file to lm_sensors.conf
-                  Drop AMD K10 detection (unreliable sensors)
+                  Only report AMD Family 10h CPU with model > 2
                   Add detection for AMD Family 11h thermal sensors
                   Add detection for Intel Atom thermal sensors
                   Add detection for National Semiconductor LM73
Index: prog/detect/sensors-detect
=================================--- prog/detect/sensors-detect	(révision 5799)
+++ prog/detect/sensors-detect	(copie de travail)
@@ -1878,6 +1878,10 @@
 		driver => "k8temp",
 		detect => \&k8temp_pci_detect,
 	}, {
+		name => "AMD Family 10h thermal sensors",
+		driver => "to-be-written",
+		detect => \&fam10h_pci_detect,
+	}, {
 		name => "AMD Family 11h thermal sensors",
 		driver => "to-be-written",
 		detect => \&fam11h_pci_detect,
@@ -5334,6 +5338,25 @@
 	return 9;
 }
 
+sub fam10h_pci_detect
+{
+	return unless exists $pci_list{'1022:1203'};
+
+	# Errata 319 (Inaccurate Temperature Measurement) affects
+	# revisions DR-BA, DR-B2 and DR-B3, all have model number = 2.
+	# So we only pick CPU with model number > 2.
+	my $probecpu;
+	foreach $probecpu (@cpu) {
+		if ($probecpu->{vendor_id} eq 'AuthenticAMC' &&
+				$probecpu->{'cpu family'} = 0x10 &&
+				$probecpu->{model} > 2) {
+			return 9;
+		}
+	}
+	
+	return;
+}
+
 sub fam11h_pci_detect
 {
 	return unless exists $pci_list{'1022:1303'};

I have also updated wiki/Devices.

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Support for Family 10h CPUs - Not all sensors are

[Clemens Ladisch]

Jean Delvare wrote:
> +		if ($probecpu->{vendor_id} eq 'AuthenticAMC' &&

  +		if ($probecpu->{vendor_id} eq 'AuthenticAMD' &&

-- 
Clemens

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors

Re: [lm-sensors] Support for Family 10h CPUs - Not all sensors are

[Jean Delvare]

On Mon, 23 Nov 2009 12:27:54 +0100, Clemens Ladisch wrote:
> Jean Delvare wrote:
> > +		if ($probecpu->{vendor_id} eq 'AuthenticAMC' &&
> 
>   +		if ($probecpu->{vendor_id} eq 'AuthenticAMD' &&

Oops, thanks. Fixed and committed.

-- 
Jean Delvare

_______________________________________________
lm-sensors mailing list
lm-sensors@lm-sensors.org
http://lists.lm-sensors.org/mailman/listinfo/lm-sensors