[Nouveau] [TESTING CALL] Fan management

[Martin Peres]

Hi Nouveau users,

Just saw the bitching on Phoronix about lack of fan management in 
nouveau (no offence Michael, it was justified ;) ).
Since it has been working flawlessly for more than a week on my desktop, 
I decided to let you guys know about it and ask for testing.

Here is the code: 
http://gitorious.org/linux-nouveau-pm/linux-nouveau-pm/commits/thermal
Please read the following paragraphs before pulling this branch and fry 
your GPU.

The hardware and current limitations
---------------------------------------------------

This branch introduces code to drive the most common fan type (and the 
most problematic one, I guess/hope). We call them PWM fan.
The other types are TOGGLE (can only be ON or OFF) and I2C (driven using 
an I2C chip).

The TOGGLE fans aren't supported because I think I lack the hw (to be 
confirmed).

The PWM fans are the most common ones. It supports a full range of speed 
(from 0% to 100%).

The I2C-accessible fans are usually limited to high-end cards. At the 
moment, the only thing we can do for them is to launch their hwmon driver.
This is due to a limitation in hwmon that doesn't let kernel drivers 
accessing the i2c chips' methods. We are working with the hwmon team and 
provided
a patch to support that: 
http://www.spinics.net/lists/lm-sensors/msg32906.html
If you have such a fan, you can manage it from the userspace.

The implemented fan management support
-----------------------------------------------------------

Current fan management should only work on what we call PWM fans.

This work should be usable up to (not including) nva3 chipset generation.
This is because nva3+ cards all have hw to drive the speed of the fan 
(pwm or i2c).

Reading the fan speed (power): cat /sys/class/drm/cardX/device/pwm0

Reading the fan speed (RPM): cat /sys/class/drm/cardX/device/fan0_input

Setting the fan management method: echo X > 
/sys/class/drm/cardX/device/pwm0_enable
- X = 0: Fan management disabled, fan power set to 100%
- X = 1: Manual fan management (you can set the speed by writing 
[0->100] to /sys/class/drm/cardX/device/pwm0)
- X = 2: The default automatic fan management (equivalent to X = 3 for 
the moment)
- X = 3: Linear fan management. The fan speed is linearly correlated to 
the temperature.
- X = 4: Target temperature management. The fan is driven to keep the 
GPU to a certain temperature

The thermal zone implementation
------------------------------------------------

Since I believe thermal zones are great but massively under-used. I used 
thermal zones to define some trip points.

Here are some noteworthy trip points:
- Default mode: Listen to the user commands
- Fan boost: Whenever the GPU reaches this temperature, the fan will 
blow at 100%
- Downclocking: Whenever the GPU reaches this temperature, your GPU will 
be downclocked to the lowest perflvl
- Critical: Whenever the GPU reaches this temperature, your computer 
will shut down to prevent damages.

BIG FAT WARNING
-----------------------------

For those willing to try this, please watch your GPU temperature closely 
at any time using "$ sensors" and your fingers.
If you think sensors doesn't report the right temperature, please send 
us a mail with your vbios and show up on IRC (ask mupuf).

Check this before having fun with fan management, it is very important!

Installation
------------------

cd my_package_directory
git clone git://gitorious.org/linux-nouveau-pm/linux-nouveau-pm.git
git branch --track thermal origin/thermal

Then it is just a matter of compiling and testing your kernel. Please 
follow your distribution's guide.

Report
------------

Please report your story on our IRC channel. Please also send your 
kernel logs along with your vbios to martin.peres <at> free(dot)fr.

Conclusion
------------

Hope you like it. Keep us informed of how well it works ;)

Martin (AKA mupuf)

PS: Another testing call about power management should also follow but 
there is something I want to implement before people try it.
-------------- next part --------------
An HTML attachment was scrubbed...
URL: <http://lists.freedesktop.org/archives/nouveau/attachments/20110829/d57e97bb/attachment.htm>