ACPI reads wrong temperature

Hi,
I installed Debian Lenny (2.6.26-1-686) on my notebook (Samsung P40 with the latest Bios 09HK), but the notebook continued to shutting down with the message "Critical temperature reached (65535 C). Shutting down.". To stop this I patched the thermal module, so that it ignores every temperature above 1000 C.

Datum: Fri, 31 Oct 2008 14:30:30 +0000
Von: Matthew Garrett <[email protected]>
An: Peter Mahlknecht <[email protected]>
CC: [email protected]
Betreff: Re: ACPI reads wrong temperature

On Fri, Oct 31, 2008 at 03:02:47PM +0100, Peter Mahlknecht wrote:

Quoted 2 lines

Hi,
I installed Debian Lenny (2.6.26-1-686) on my notebook (Samsung P40 with

the latest Bios 09HK), but the notebook continued to shutting down with the message "Critical temperature reached (65535 C). Shutting down.".

Quoted 1 line

To stop this I patched the thermal module, so that it ignores every

temperature above 1000 C.

Do you get wrong temperatures if you never load the max6657 driver?

Thanks for the hint. I get the wrong temperatures after loading the lm90 module (hwmon driver), but also if i load em28xx_cx25843, which is a part of the driver for my tv-card. Both modules depend on i2c-core, so probably the bug is related to the i2c bus?

Quoted 4 lines

Thanks for the hint. I get the wrong temperatures after loading the lm90 module (hwmon driver), but also if i load em28xx_cx25843, which is a part of the driver for my tv-card. Both modules depend on i2c-core, so probably the bug is related to the i2c bus?

In the lm90 case it's almost certainly because you have the firmware and Linux trying to access the chip at the same time. Since it has indexed registers, this fails very miserably. I'm surprised that the TV card driver has any effect - it should be on an entirely spearate i2c bus. Are you sure you don't have lm90 loaded in that case?

Quoted 5 lines

Quoted 4 lines

Thanks for the hint. I get the wrong temperatures after loading the lm90 module (hwmon driver), but also if i load em28xx_cx25843, which is a part of the driver for my tv-card. Both modules depend on i2c-core, so probably the bug is related to the i2c bus?

Quoted 5 lines

In the lm90 case it's almost certainly because you have the firmware and Linux trying to access the chip at the same time. Since it has indexed registers, this fails very miserably. I'm surprised that the TV card driver has any effect - it should be on an entirely spearate i2c bus. Are you sure you don't have lm90 loaded in that case?

Yes, I am. I had also a talk with the author of the Tv-card driver and he suggested to but the problem on the lm-sensors ml: http://lists.lm-sensors.org/pipermail/lm-sensors/2008-November/024610.html
We know now that the problem is related to the i2c-i801 driver.

Peter

Datum: Mon, 03 Nov 2008 10:05:36 +0800
Von: Zhao Yakui <[email protected]>
An: Peter Mahlknecht <[email protected]>
CC: Matthew Garrett <[email protected]>, "[email protected]" <[email protected]>
Betreff: Re: ACPI reads wrong temperature

Quoted 6 lines · Peter Mahlknecht

Quoted 6 lines

Quoted 4 lines

Thanks for the hint. I get the wrong temperatures after loading the lm90 module (hwmon driver), but also if i load em28xx_cx25843, which is a part of the driver for my tv-card. Both modules depend on i2c-core, so probably the bug is related to the i2c bus?

In the lm90 case it's almost certainly because you have the firmware

and

Quoted 5 lines

Quoted 4 lines

Linux trying to access the chip at the same time. Since it has indexed registers, this fails very miserably. I'm surprised that the TV card driver has any effect - it should be on an entirely spearate i2c bus. Are you sure you don't have lm90 loaded in that case?

Yes, I am. I had also a talk with the author of the Tv-card driver and

he suggested to but the problem on the lm-sensors ml:
http://lists.lm-sensors.org/pipermail/lm-sensors/2008-November/024610.html

Quoted 1 line

We know now that the problem is related to the i2c-i801 driver.

Yes. This is related with i2c-i801 driver.
We have a similar bug.
http://bugzilla.kernel.org/show_bug.cgi?id=10245

In this bug after the driver of lm_sensor is loaded, the system read the incorrect temperature. The root cause is that the SMBus is used by ACPI for thermal management. After the lm_sensor is loaded, Linux will use the same bus. There is the conflict. So if the Linux doesn't attach the driver to the SMbus, it will be OK.

Will you please attach the output of acpidump, lspci -vxxx?

Datum: Mon, 03 Nov 2008 10:05:36 +0800
Von: Zhao Yakui <[email protected]>
An: Peter Mahlknecht <[email protected]>
CC: Matthew Garrett <[email protected]>, "[email protected]" <[email protected]>
Betreff: Re: ACPI reads wrong temperature

Quoted 6 lines · Peter Mahlknecht

Quoted 6 lines

Quoted 4 lines

Thanks for the hint. I get the wrong temperatures after loading the lm90 module (hwmon driver), but also if i load em28xx_cx25843, which is a part of the driver for my tv-card. Both modules depend on i2c-core, so probably the bug is related to the i2c bus?

In the lm90 case it's almost certainly because you have the firmware

and

Quoted 5 lines

Quoted 4 lines

Linux trying to access the chip at the same time. Since it has indexed registers, this fails very miserably. I'm surprised that the TV card driver has any effect - it should be on an entirely spearate i2c bus. Are you sure you don't have lm90 loaded in that case?

Yes, I am. I had also a talk with the author of the Tv-card driver and

he suggested to but the problem on the lm-sensors ml:
http://lists.lm-sensors.org/pipermail/lm-sensors/2008-November/024610.html

Quoted 1 line

We know now that the problem is related to the i2c-i801 driver.

Yes. This is related with i2c-i801 driver.
We have a similar bug.
http://bugzilla.kernel.org/show_bug.cgi?id=10245

In this bug after the driver of lm_sensor is loaded, the system read the incorrect temperature. The root cause is that the SMBus is used by ACPI for thermal management. After the lm_sensor is loaded, Linux will use the same bus. There is the conflict. So if the Linux doesn't attach the driver to the SMbus, it will be OK.

Will you please attach the output of acpidump, lspci -vxxx?

Thanks for the info.
From the attached info we can know that the I/O start address for SMbus is 0x18E0. And SMbus is internally used in AML code.

If the I2C-i801 driver is loaded, there is the conflict between BIOS and OS.

So I think that it is a BIOS bug. And this can't be fixed by Linux-kernel.

I don't think the BIOS is doing anything wrong here. It's entitled to assume that the operating system won't load drivers for hardware it's not declaring.

Quoted 4 lines

I don't think the BIOS is doing anything wrong here. It's entitled to assume that the operating system won't load drivers for hardware it's not declaring.

agreed, we need a linux solution.

Quoted 6 lines · Len Brown

Quoted 4 lines

I don't think the BIOS is doing anything wrong here. It's entitled to assume that the operating system won't load drivers for hardware it's not declaring.

agreed, we need a linux solution.

There's a kernel warning in this kind of situation, but IIRC we were still at the point where there were false positives.

On your box after the i2c-i801 driver is loaded, the smbus controller will be used by the AML code and i2c-i801 driver. Unfortunately there is no synchronization between them. IMO this is the BIOS fault. The SMbus will be accessed by the two different modules. But it is exported by BIOS.

So the better solution is that:
a. SMBus is hidden in BIOS. In such case the Linux OS can't detect
the SMbus controller . Of course it won't try to load the device driver for it.
b. Linux OS won't load the device driver for it.

Quoted 5 lines · Zhao Yakui

On your box after the i2c-i801 driver is loaded, the smbus controller will be used by the AML code and i2c-i801 driver. Unfortunately there is no synchronization between them. IMO this is the BIOS fault. The SMbus will be accessed by the two different modules. But it is exported by BIOS.

No, loading i2c-i801 is harmless. The breakage occurs when you attempt to drive the smbus.

Quoted 5 lines

So the better solution is that:
a. SMBus is hidden in BIOS. In such case the Linux OS can't detect
the SMbus controller . Of course it won't try to load the device driver for it.
b. Linux OS won't load the device driver for it.

The solution is for users not to load smbus drivers unless they know for certain that it's safe to do so, or alternatively for us to refuse to load i2c drivers if they overlap with ACPI resource allocations.

or alternatively for us to refuse to load i2c drivers

Quoted 16 lines · Matthew Garrett

Quoted 5 lines · Zhao Yakui

On your box after the i2c-i801 driver is loaded, the smbus controller will be used by the AML code and i2c-i801 driver. Unfortunately there is no synchronization between them. IMO this is the BIOS fault. The SMbus will be accessed by the two different modules. But it is exported by BIOS.

No, loading i2c-i801 is harmless. The breakage occurs when you attempt to drive the smbus.

Quoted 5 lines

So the better solution is that:
a. SMBus is hidden in BIOS. In such case the Linux OS can't detect
the SMbus controller . Of course it won't try to load the device driver for it.
b. Linux OS won't load the device driver for it.

Quoted 3 lines

The solution is for users not to load smbus drivers unless they know for certain that it's safe to do so, or alternatively for us to refuse to load i2c drivers if they overlap with ACPI resource allocations.

We already have that:

Quoted 1 line

or alternatively for us to refuse to load i2c drivers

in drivers/acpi/osl.c:

#define ENFORCE_RESOURCES_STRICT 2
#define ENFORCE_RESOURCES_LAX 1
#define ENFORCE_RESOURCES_NO 0

static unsigned int acpi_enforce_resources = ENFORCE_RESOURCES_LAX;

lax means print out warning which i2c driver conflicts with which ACPI Operation Region.
Did you get such a message:

Quoted 10 lines · Matthew Garrett

Quoted 5 lines · Zhao Yakui

On your box after the i2c-i801 driver is loaded, the smbus controller will be used by the AML code and i2c-i801 driver. Unfortunately there is no synchronization between them. IMO this is the BIOS fault. The SMbus will be accessed by the two different modules. But it is exported by BIOS.

No, loading i2c-i801 is harmless. The breakage occurs when you attempt to drive the smbus.

Quoted 8 lines

Quoted 5 lines

So the better solution is that:
a. SMBus is hidden in BIOS. In such case the Linux OS can't detect
the SMbus controller . Of course it won't try to load the device driver for it.
b. Linux OS won't load the device driver for it.

The solution is for users not to load smbus drivers unless they know for certain that it's safe to do so, or alternatively for us to refuse to

Quoted 1 line

load i2c drivers if they overlap with ACPI resource allocations.

We already have that:

Quoted 1 line

or alternatively for us to refuse to load i2c drivers

in drivers/acpi/osl.c:

#define ENFORCE_RESOURCES_STRICT 2
#define ENFORCE_RESOURCES_LAX 1
#define ENFORCE_RESOURCES_NO 0

static unsigned int acpi_enforce_resources = ENFORCE_RESOURCES_LAX;

lax means print out warning which i2c driver conflicts with which ACPI Operation Region.
Did you get such a message:
printk("%sACPI: %s resource %s [0x%llx-0x%llx]"

                               " conflicts with ACPI region %s"
                               " [0x%llx-0x%llx]\n",
                               acpi_enforce_resources == ENFORCE_RESOURCES_LAX
                               ? KERN_WARNING : KERN_ERR,
                               ioport ? "I/O" : "Memory", res->name,
                               (long long) res->start, (long long) res->end,
                               res_list_elem->name,
                               (long long) res_list_elem->start,
                               (long long) res_list_elem->end);

If not, probably this specific i2c driver has to be added to check it's resources against ACPI used Operation Regions.

Jean: Could you add that if necessary, pls.
This is the whole thread:
http://marc.info/?t=122546192400002&r=1&w=2