[lm-sensors] lm_sensors and nvidia graphic card temperature

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Christoph Anton Mitterer]

Hi.

Modern Nvidia graphic cards support temperature monitoring of the GPU.
It would be nice to see lm_sensors support that. Is there anything going 
on in that field?

Greetings,
Chris.

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Christoph Anton Mitterer]

Rudolf Marek wrote:

>You can try:
>1) load i2c-rivatv (or whatever driver for the i2c nvidia bus)
>  
>
Is this going to be available for 2.6.x kernels?

Chris.

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Jean Delvare]

Hi Christoph,

> > You can try:
> > 1) load i2c-rivatv (or whatever driver for the i2c nvidia bus)
>
> Is this going to be available for 2.6.x kernels?

It is supposed to be already, except that the exact name is "rivatv",
not "i2c-rivatv".

http://rivatv.sourceforge.net/

Alternatively, you should be able to gain control of the nVidia I2C
busses using the in-kernel-tree rivafb or nvidiafb framebuffer drivers.
Make sure to enable "DDC support" in either case.

-- 
Jean Delvare

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Volker Kuhlmann]

> Alternatively, you should be able to gain control of the nVidia I2C
> busses using the in-kernel-tree rivafb or nvidiafb framebuffer drivers.

I have a rivafb module, which loads fine and produces no output on
stderr or syslog. What does "gain control of the nvidia i2c busses"
mean? That sensors-detect should find them? It doesn't for me:

modprobe -v rivafb
sensors-detect
...
Driver `eeprom' (should be inserted):
  Detects correctly:
  * Bus `SMBus nForce2 adapter at 1c00'
    Busdriver `i2c-nforce2', I2C address 0x50
    Chip `SPD EEPROM' (confidence: 8)
  * Bus `SMBus nForce2 adapter at 1c00'
    Busdriver `i2c-nforce2', I2C address 0x51
    Chip `SPD EEPROM' (confidence: 8)
Driver `it87' (should be inserted):
  Detects correctly:
  * ISA bus address 0x0290 (Busdriver `i2c-isa')
    Chip `ITE 8712F Super IO Sensors' (confidence: 9)
...

Same output as always.

Are we talking about the same thing?

Linux .. 2.6.13-15.7-default #1 Tue Nov 29 14:32:29 UTC 2005 x86_64
sensors-2.9.1-4
Gigabyte PCI-express graphics card with nvidia geforce 6600 chip

> Make sure to enable "DDC support" in either case.

Where? The kernel config doesn't seem to know about it:
zgrep -i ddc /proc/config.gz
gives no output. Grepping through the kernel source doesn't tell me
anything more, and none of the 5 rivafb module paramters have to do with
DDC.

Thanks,

Volker

-- 
Volker Kuhlmann			is possibly list0570 with the domain in header
http://volker.dnsalias.net/		Please do not CC list postings to me.

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Jean Delvare]

Hi Volker,

> > Alternatively, you should be able to gain control of the nVidia I2C
> > busses using the in-kernel-tree rivafb or nvidiafb framebuffer drivers.
> 
> I have a rivafb module, which loads fine and produces no output on
> stderr or syslog. What does "gain control of the nvidia i2c busses"
> mean? That sensors-detect should find them? It doesn't for me:

"Gain control" means that i2c/smbus chip drivers can "see" the bus and
try attaching to devices thereon. You can use i2cdetect to list and
probe these busses.

> modprobe -v rivafb
> sensors-detect
> ...
> Driver `eeprom' (should be inserted):
>   Detects correctly:
>   * Bus `SMBus nForce2 adapter at 1c00'
>     Busdriver `i2c-nforce2', I2C address 0x50
>     Chip `SPD EEPROM' (confidence: 8)
>   * Bus `SMBus nForce2 adapter at 1c00'
>     Busdriver `i2c-nforce2', I2C address 0x51
>     Chip `SPD EEPROM' (confidence: 8)
> Driver `it87' (should be inserted):
>   Detects correctly:
>   * ISA bus address 0x0290 (Busdriver `i2c-isa')
>     Chip `ITE 8712F Super IO Sensors' (confidence: 9)
> ...
> 
> Same output as always.
> 
> Are we talking about the same thing?

If you have a recent graphics adapter, you most likely need the
nvidiafb driver. Just because rivafb loads silently doesn't mean it did
anything useful. Actually, the fact that it was silent suggests very
strongly that it does NOT support your graphics adapter.

> Linux .. 2.6.13-15.7-default #1 Tue Nov 29 14:32:29 UTC 2005 x86_64
> sensors-2.9.1-4
> Gigabyte PCI-express graphics card with nvidia geforce 6600 chip
> 
> > Make sure to enable "DDC support" in either case.
> 
> Where? The kernel config doesn't seem to know about it:
> zgrep -i ddc /proc/config.gz
> gives no output. Grepping through the kernel source doesn't tell me
> anything more, and none of the 5 rivafb module paramters have to do with
> DDC.

For some reason the symbols have "I2C" rather than "DDC" in their names:

CONFIG_FB_RIVA_I2C:
CONFIG_FB_NVIDIA_I2C

-- 
Jean Delvare

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Volker Kuhlmann]

Hi Jean,

> If you have a recent graphics adapter, you most likely need the
> nvidiafb driver. Just because rivafb loads silently doesn't mean it did
> anything useful. Actually, the fact that it was silent suggests very
> strongly that it does NOT support your graphics adapter.

Thanks, that makes sense. Loading nvidiafb gives in syslog

kernel: nvidiafb: module not supported by Novell, setting U taint flag.
kernel: ACPI: PCI Interrupt 0000:05:00.0[A] -> Link [APC3] -> GSI 18 (level, low) -> IRQ 58
kernel: PCI: Unable to reserve mem region #2:10000000 at e0000000 for device 0000:05:00.0
kernel: nvidiafb: cannot request PCI regions
kernel: ACPI: PCI interrupt for device 0000:05:00.0 disabled
kernel: klogd 1.4.1, ---------- state change ----------

Which doesn't look too good to me, and sensors-detect doesn't find
anything more. From lspci

05:00.0 VGA compatible controller: nVidia Corporation NV43 [GeForce 6600] (rev a2)

There was something about this on this list in March, but that involved
kernel oopses.

At least the display doesn't crash and burn on me... Everything seems to
continue to operate normally.

Linux  2.6.13-15.7-default #1 Tue Nov 29 14:32:29 UTC 2005 x86_64 x86_64 x86_64 GNU/Linux

Booting with acpi=off at the grub command line leads to a stone dead
system before I've lifted the finger off the enter key.

# i2cdetect -l
i2c-2   dummy   ISA main adapter                ISA bus algorithm
i2c-1   unknown SMBus nForce2 adapter at 1c40   Algorithm unavailable
i2c-0   unknown SMBus nForce2 adapter at 1c00   Algorithm unavailable

These are already accounted for (mobo super io, 2 dimms).

I'm so far running with xorg's nv driver, though I could try nvidia's.

Thanks + greets,

Volker

-- 
Volker Kuhlmann			is possibly list0570 with the domain in header
http://volker.dnsalias.net/		Please do not CC list postings to me.

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Jean Delvare]

Hi Volker,

> Thanks, that makes sense. Loading nvidiafb gives in syslog
> 
> kernel: nvidiafb: module not supported by Novell, setting U taint flag.
> kernel: ACPI: PCI Interrupt 0000:05:00.0[A] -> Link [APC3] -> GSI 18 (level, low) -> IRQ 58
> kernel: PCI: Unable to reserve mem region #2:10000000 at e0000000 for device 0000:05:00.0
> kernel: nvidiafb: cannot request PCI regions
> kernel: ACPI: PCI interrupt for device 0000:05:00.0 disabled
> kernel: klogd 1.4.1, ---------- state change ----------
> 
> Which doesn't look too good to me, and sensors-detect doesn't find
> anything more. From lspci
> 
> 05:00.0 VGA compatible controller: nVidia Corporation NV43 [GeForce 6600] (rev a2)

Look for a conflicting device in /proc/iomem.

-- 
Jean Delvare

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Volker Kuhlmann]

Hi Jean,

> > kernel: nvidiafb: cannot request PCI regions

> Look for a conflicting device in /proc/iomem.

Thanks! Now nvidiafb loads without trouble, but I didn't do anything.
Perhaps it has something to do with switching to nvidia's binary-only
geforce driver, or updating the mobo bios to latest. No additional
anything though from either sensors-detect or i2cdetect -l.

For fun, I tried the rivafb module again too, it now also loads without
syslog message, but still doesn't show me any i2c anything on the
graphics card.

I conclude therefore that the card doesn't have any i2c facilities. For
anyone else trying the same, the card is a

Gigabyte GV-NX66256DP 6600 PCIE Video Card
05:00.0 VGA compatible controller: nVidia Corporation NV43 [GeForce 6600] (rev a2)
05:00.0 Class 0300: 10de:0141 (rev a2)

Thanks for all your help, it's really appreciated. And thanks for some
cool software!

Volker

-- 
Volker Kuhlmann			is possibly list0570 with the domain in header
http://volker.dnsalias.net/		Please do not CC list postings to me.

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Jean Delvare]

Hi Volker,

> [nvidia graphics card temp monitoring]
> 
> > You can try:
> > 1) load i2c-rivatv (or whatever driver for the i2c nvidia bus)
> 
> i2c-nvidia2 is loaded automatically, but sensors-detect only finds 2
> dimm eeproms and the mobo's it8712. However, it finds two adapters, but
> no success on the first: 
> 
> Next adapter: SMBus nForce2 adapter at 1c40
> Do you want to scan it? (YES/no/selectively): 
> Client found at address 0x08
> Client found at address 0x4e
> [all failed]
> 
> Next adapter: SMBus nForce2 adapter at 1c00
> Do you want to scan it? (YES/no/selectively): 
> Client found at address 0x08
> Client found at address 0x50
> [2 eeproms found]

i2c-nvidia2 doesn't exist as far as I know. You must refer to
i2c-nforce2, I presume? If so, this is totally unrelated with the
busses found on graphics adapters. i2c-nforce2 is meant for motherboard
SMBus.

> What kernel module should I load exactly? Google doesn't know anything
> about i2c-rivatv.

Search for rivatv instead - see my other post.

> > Run sensors detect it  should find he w83785
> 
> modprobe w83l785ts is fine. No change with sensors-detect though.

Rudolf's w83785 driver is meant for the Winbond W83L785R chips, which
is a full-featured hardware monitoring chip. My w83l785ts driver, OTOH,
is meant for the Winbond W83L785TS-S, which is a very simple
single-temperature sensor. I think I remember both devices are
compatible, in that you should get proper temperature readings if
forcing the w83l785ts driver on a W83L785R chip. Both drivers could be
merged to make things easier, but nobody had an interest in working on
it so far, it seems.

> Being able to get the temp of the geforce 6600 chip would be cool, as I
> have a temp problem with it. What else could I try?

rivafb, nvidiafb.

-- 
Jean Delvare

[lm-sensors] lm_sensors and nvidia graphic card temperature

[Volker Kuhlmann]

> i2c-nvidia2 doesn't exist as far as I know. You must refer to
> i2c-nforce2, I presume?

Yes, sorry.

> Search for rivatv instead - see my other post.

Yes, very informative, thanks!

The 0.86 release doesn't compile, but the cvs one does. The module loads
fine, syslog:

 kernel: Linux video capture interface: v1.00
 kernel: rivatv: module not supported by Novell, setting U taint flag.
 kernel: rivatv: Video4Linux driver for NVIDIA cards
 kernel: rivatv: MMX processor extension enabled

> Rudolf's w83785 driver is meant for the Winbond W83L785R chips, which
> is a full-featured hardware monitoring chip. My w83l785ts driver, OTOH,
> is meant for the Winbond W83L785TS-S, which is a very simple
> single-temperature sensor.

Thanks for the clarification. I can't get modprobe w83l785ts with
various probe= and force= options to do anything useful. Loading
i2c-nforce2 makes two i2c busesses avilable, numbered 0 and 1.
Presumably something similar would have to happen if the graphics card
has an i2c bus too? Dunno how though. :(
I've insmoded all /lib/modules/2.6.13-15.7-default/kernel/drivers/i2c/busses/*
ignoring the errors, and rivafb and rivatv, sensors-detect finds a lot
more but it's all bogus. sensors -s (with the original sensors.conf),
then sensors outputs only zeros for anything that wasn't there before.
Might have to give up. :((

Thanks again for all your insights.

Volker

-- 
Volker Kuhlmann			is possibly list0570 with the domain in header
http://volker.dnsalias.net/		Please do not CC list postings to me.