wiki:ventana/thermal

Ventana Thermal Information

See also:

Gateworks System Controller (GSC)

The Gateworks System Controller has a temperature sensor that measures board temperature as well as a fan controller on some models.

Get the Board Temperature (current GSC temp in C/10):

# DIR=$(find /sys/bus/i2c/devices/0-0029/ -name in0_input)
# DIR=${DIR%/*}
# cat $DIR/temp0_input
442
  • board temperature is 44.2C

For more information:

IMX6 SoC

The IMX6 System On Chip has an internal die temperature sensor.

There are several temperature grades of IMX6 SoC's that are used across various Gateworks products:

  • IMX6Q Automotive: -40C to 125C
  • IMX6DL Industrial: -40C to 105C
  • IMX6S Industrial: -40C to 105C

Get the CPU Temperature (milicelcius):

# cat /sys/class/thermal/thermal_zone0/temp
60380
  • CPU tempis 60.38C

Linux Kernel Support

The Linux IMX6 cpufreq cooling driver (drivers/thermal/imx_thermal.c) provides several functions:

  • display the thermal grade of CPU and min/max temperature
  • thermal sensing of the IMX6 chip
  • automatic CPU frequency throttling based on temperature feedback

There is a Linux sysfs API for accessing this information:

  • Note in first line, hwmon1 may be hwmon2
    cat /sys/class/thermal/thermal_zone0/temp # eg 62520 current CPU temp in milicelcius, eg 62.5C
    cat /sys/class/thermal/thermal_zone0/trip_point_0_temp # current passive temp in millicelcius
    cat /sys/class/thermal/thermal_zone0/trip_point_1_temp # current critical temp in millicelcius
    # set passive/crit to 105C/125C for automotive CPU
    echo 105000 > /sys/class/thermal/thermal_zone0/trip_point_0_temp
    echo 125000 > /sys/class/thermal/thermal_zone0/trip_point_1_temp
    

Notes:

  • the imx_thermal_zone temp currently starts scaling ARM cpufreq at 85C (until it lowers to 75C) and has a critical temp of 100C. This is really too conservative for the typical IMX6 temperature grades that Gateworks uses
  • there is a 10C hysteresis such that the temperature needs to lower by 10C before the alarm is cancelled
  • the passive trip point is a preventative measure before reaching critical that does things to lower temperature such as reducing cpu/gpu frequencies
  • the critical trip point should not be exceeded to avoid damaging the CPU and as such when this trip point is reached the Linux shutdown function is called to attempt to gracefully shut down the OS. The default implementation will end up simply resetting the board but before that occurs a usermode helper will exec /sbin/poweroff which the user could implement to do whatever they choose (ie, sync filesystems, use GSC sleep for 'x' seconds, etc).
  • the thresholds for the driver can be adjusted in userspace (see above) or passed in through the bootloader (see below).
  • Gateworks has an upstream Linux patch pending approval that will set the initial setpoints according to the temperature grade of the CPU determined via eFUSE registers. This patch will eventually be rolled into our various kernel support.
  • the passive trip point gpu scaling is controlled via the writable sysfs nodes: /sys/bus/platform/drivers/galcore/gpu3DMinClock and /sys/module/galcore/parameters/initgpu3DMinClock which contain a value that sets the gpu scaling to value/64. The Android BSP sets these values at runtime in the init.i.MX6Q.rc and init.i.MX6DL.rc files located under the device/gateworks/ventana/ directory in the source tree.
  • in mainline kernels, the critical temperature can not
  • mainline kernels, such as the 4.4 kernel in our OpenWrt 16.02 BSP, do not allow critical temperature edits using the method described above but our downstream vendor kernel contains a patch to revert this restriction.

U-Boot Support

The Gateworks Ventana Bootloader driver (drivers/thermal/imx_thermal.c) provides several functions:

  • display the thermal grade of CPU and min/max temperature
  • display current temperature of IMX6
  • busyloop if over IMX6 temperature is greater or equal to max temp - 5C

Additionally, the critical and passive thermal trip points mentioned above can be configured via kernel command line options passed in through the extra env variable:

setenv extra "${extra} imx_thermal.crit_delta=X imx_thermal.passive_delta=Y"

Where X and Y are the deltas from the max CPU temp (in milli-degrees Celcius).

Notes:

  • mainline kernels require a patch to allow kernel command line arguments to set the crit and passive trip points using this method.

Vivante GPU thermal management

The Vivante GPU has some frequency scaling based on temperature which enabled by default in the [drivers/mxc/gpu-viv/hal/os/linux/kernel/platform/freescale/gc_hal_kernel_platform_imx6q14.c Vivante driver]. If enabled the driver will scale the GPU frequency down by a specified divisor. By default this is 3/64 for the Gateworks Yocto and Android BSP's.

The GPU frequency scaling is triggered by the IMX thermal zone passive trip point (see above). Note that because the temperature is triggered by the IMX thermal zone, the ARM cpu frequency will be reduced at the same time.

The minimum devisor used to reduce GPU frequency by default is 1 (meaning 1/64th frequency scaling which is the maximum amount of frequency reduction possible) you can use the drivers sysfs interface:

  • On Yocto v1.8+ (3.14.x):
    echo 3 > /sys/bus/platform/drivers/galcore/gpu3DMinClock # set GPU freq reduction to 3/64
    
  • On Yocto v1.6/v1.7 (3.10.17):
    echo 3 > /sys/module/galcore/parameters/gpu3DMinClock # set GPU freq reduction to 3/64
    
  • On Linux 3.10.53 (Yocto master / Android kk)
    echo 3 > /sys/bus/platform/drivers/galcore/gpu3DMinClock # set GPU freq reduction to 3/64
    

You can change this value from 1 to 64, where 1 is 1/64 frequency (maximum reduction) and 64 is no reduction. Note that depending on what the GPU is being used for, you may want to increase this value. We have found that a value of 3 for example offers adequate cooling without sacrificing significant performance.

PCIe Switch

Besides the CPU, another item that generates significant heat is the PCIe switch used on some of the Ventana models (GW520x, GW53xx, GW54xx, GW55xx). The typical power usage is appx 1.35W up to a max of appx 2.6W (max being 85% traffic). The switch does support PCIExpress Active State Power Management (ASPM)and also will power down unused SerDes? lanes automatically to reduce power when possible.

The part is rated as follows:

  • Max Operating Junction Temp for Reliability = 125C
  • ThetaJC = 6.12C/W

On the GW53xx and GW54xx boards there are topside Mini-PCIe sockets which when used will be directly over the switch. For cases that require the use of these sites, it is recommended to use additional heatsinking or air cooling to keep the switch within the operating temperature range.

For low profile heatsinking, Panasonic makes graphite thermal sheets that can be attached to the device and then can extend out and over to the CPU or another area where a larger heatsink can be located. See the following link for information on this material: https://industrial.panasonic.com/ww/pgs2/mlpgs (available at Digikey as well).

Last modified 5 weeks ago Last modified on 05/25/17 10:58:45