Update temp_metrics to look at skin temperatures

Updated temp_metrics to take the skin temperature into account instead
of PECI CPU temperature for controlling fan speed as well as Ivy Bridge
throttling.

BUG=chrome-os-partner:9193
TEST=Vary CPU and GPU load on the system. The fan speed and CPU/GPU
limits should change as the skin temperature responds to changes in the
load.
BRANCH=link

Change-Id: Ie3d85112de1043cf5b12a78ca1fc50f5eb6c0497
Signed-off-by: Sameer Nanda <snanda@chromium.org>
Reviewed-on: https://gerrit.chromium.org/gerrit/34221
Reviewed-by: Todd Broch <tbroch@chromium.org>
Reviewed-by: Vincent Palatin <vpalatin@chromium.org>
(cherry picked from commit bb66bec788c9447976d1be56f69a40b6c1ea36a3)
Reviewed-on: https://gerrit.chromium.org/gerrit/34388

1 files changed, 127 insertions, 30 deletions

diff --git a/util/temp_metrics.conf b/util/temp_metrics.conf
index 6a2aa3fc30..adca6850de 100644
--- a/util/temp_metrics.conf
+++ b/util/temp_metrics.conf
@@ -21,7 +21,6 @@ respawn
 
 script
   TEMP_OFFSET=273  # difference between K (reported by EC) and C (used in UMA)
-  PECI_TEMP_THRESHOLD=77  # CPU temperature threshold
 
   # Thermal loop fields
   CPU_MAX_FREQ_FIELD=1
@@ -56,7 +55,6 @@ script
   }
 
   get_field() {
-    #out=$(echo "$2" | awk -v z=$1 '{print $z}')
     out=$(echo "$2" | awk "{print \$$1}")
     echo $out
   }
@@ -67,6 +65,47 @@ script
     echo $tempc
   }
 
+  get_sensor_temp() {
+    s=$1
+    tempk=$(ectool temps $s | sed 's/[^0-9]//g')
+    tempc=$((tempk - $TEMP_OFFSET))
+    echo $tempc
+  }
+
+  get_sensor_list() {
+    # USB C-Object: 1 or 13
+    # PCH D-Object: 3
+    # Hinge C-Object: 5 or 15
+    # Charger D-Object: 7
+    if ectool tempsinfo 1 | grep -q "USB C-Object"; then
+      usb_c_object=1
+    else
+      usb_c_object=13
+    fi
+    pch_d_object=3
+    charger_d_object=7
+    echo $usb_c_object $pch_d_object $charger_d_object
+  }
+
+  max_skin_temp=0
+  sensor_temperatures=""
+  get_max_skin_temp() {
+    sensor_temperatures=""
+    max_skin_temp=0
+    for i in $*; do
+      t=$(get_sensor_temp $i)
+      sensor_temperatures=$sensor_temperatures$i:$t:
+      if [ $t -gt $max_skin_temp ]; then
+        max_skin_temp=$t
+      fi
+    done
+
+    # Record the PECI CPU temperature also.
+    i=9
+    t=$(get_sensor_temp $i)
+    sensor_temperatures=$sensor_temperatures$i:$t:
+  }
+
   set_cpu_freq() {
     max_freq=$1
     min_freq=$2
@@ -99,23 +138,66 @@ script
     wrmsr 0 0x1fc 0x000000000004005e
   }
 
+  disable_tcc_offset() {
+    wrmsr 0 0x1a2 0
+  }
+
+  log_message() {
+    logger -t temp_metrics "$*"
+  }
+
+  TEMP_THRESHOLD_1=40
+  TEMP_THRESHOLD_2=45
+  TEMP_THRESHOLD_3=50
+
+  TEMP_THRESHOLD_1_MIN_STEP=0
+  TEMP_THRESHOLD_1_MAX_STEP=4
+  TEMP_THRESHOLD_2_MIN_STEP=5
+  TEMP_THRESHOLD_2_MAX_STEP=9
+  TEMP_THRESHOLD_3_MIN_STEP=10
+  TEMP_THRESHOLD_3_MAX_STEP=13
+
   current_step=1
   new_step=0
 
   thermal_loop() {
-    peci_temp=$1
-    if [ $peci_temp -gt $PECI_TEMP_THRESHOLD ]; then
-      if [ $current_step -ne $max_steps ]; then
+    # Hack to reset turbo activation threshold since BIOS can change it
+    # underneath us.
+    wrmsr 0 0x64c 0x12
+
+    skin_temp=$1
+    if [ $skin_temp -gt $TEMP_THRESHOLD_3 ]; then
+      temp_threshold=$TEMP_THRESHOLD_3
+      min_step=$TEMP_THRESHOLD_3_MIN_STEP
+      max_step=$TEMP_THRESHOLD_3_MAX_STEP
+    elif [ $skin_temp -gt $TEMP_THRESHOLD_2 ]; then
+      temp_threshold=$TEMP_THRESHOLD_2
+      min_step=$TEMP_THRESHOLD_2_MIN_STEP
+      max_step=$TEMP_THRESHOLD_2_MAX_STEP
+    else
+      temp_threshold=$TEMP_THRESHOLD_1
+      min_step=$TEMP_THRESHOLD_1_MIN_STEP
+      max_step=$TEMP_THRESHOLD_1_MAX_STEP
+    fi
+
+    if [ $skin_temp -gt $temp_threshold ]; then
+      if [ $current_step -ne $max_step ]; then
         new_step=$(($current_step + 1))
       fi
-    elif [ $peci_temp -lt $PECI_TEMP_THRESHOLD ]; then
-      if [ $current_step -gt 0 ]; then
+    elif [ $skin_temp -lt $temp_threshold ]; then
+      if [ $current_step -gt $min_step ]; then
         new_step=$(($current_step - 1))
       fi
     else
       new_step=$current_step
     fi
 
+    if [ $new_step -gt $max_step ]; then
+        new_step=$max_step
+    elif [ $new_step -lt $min_step ]; then
+        new_step=$min_step
+    fi
+
     if  [ $new_step -eq $current_step ]; then
       return
     fi
@@ -123,6 +205,8 @@ script
     current_step=$new_step
     step=$(get_step $new_step)
 
+    log_message "Throttling (temps: $sensor_temperatures):" $step
+
     cpu_max_freq=$(get_field $CPU_MAX_FREQ_FIELD "$step")
     cpu_min_freq=$(get_field $CPU_MIN_FREQ_FIELD "$step")
     gpu_max_freq=$(get_field $GPU_MAX_FREQ_FIELD "$step")
@@ -151,28 +235,28 @@ script
   temp_low5=105
 
   fan_loop() {
-    peci_temp=$1
+    skin_temp=$1
 
-    if [ $peci_temp -gt 77 ] || [ $peci_temp -gt $temp_low1 ]; then
+    if [ $skin_temp -gt 48 ] || [ $skin_temp -gt $temp_low1 ]; then
       rpm=9300
       fan_reset_thresholds
-      temp_low1=68
-    elif [ $peci_temp -gt 68 ] || [ $peci_temp -gt $temp_low2 ]; then
-      rpm=7725
+      temp_low1=46
+    elif [ $skin_temp -gt 45 ] || [ $skin_temp -gt $temp_low2 ]; then
+      rpm=7800
       fan_reset_thresholds
-      temp_low2=61
-    elif [ $peci_temp -gt 61 ] || [ $peci_temp -gt $temp_low3 ]; then
-      rpm=6150
+      temp_low2=43
+    elif [ $skin_temp -gt 42 ] || [ $skin_temp -gt $temp_low3 ]; then
+      rpm=6000
       fan_reset_thresholds
-      temp_low3=54
-    elif [ $peci_temp -gt 54 ] || [ $peci_temp -gt $temp_low4 ]; then
-      rpm=4575
+      temp_low3=39
+    elif [ $skin_temp -gt 38 ] || [ $skin_temp -gt $temp_low4 ]; then
+      rpm=4000
       fan_reset_thresholds
-      temp_low4=47
-    elif [ $peci_temp -gt 47 ] || [ $peci_temp -gt $temp_low5 ]; then
+      temp_low4=34
+    elif [ $skin_temp -gt 33 ] || [ $skin_temp -gt $temp_low5 ]; then
       rpm=3000
       fan_reset_thresholds
-      temp_low5=41
+      temp_low5=30
     else
       rpm=0
       fan_reset_thresholds
@@ -182,6 +266,8 @@ script
       return
     fi
 
+    log_message "Setting fan RPM (temps: $sensor_temperatures): $last_rpm -> $rpm"
+
     last_rpm=$rpm
     ectool pwmsetfanrpm $rpm
   }
@@ -189,22 +275,33 @@ script
   # External prochot misfires occasionally. Disable it.
   disable_external_prochot
 
+  # Set TCC offset to 0.
+  disable_tcc_offset
+
+  # Get list of sensors to monitor.
+  sensor_list=$(get_sensor_list)
+
   loop_count=0
   while true; do
-    sleep 1
+    sleep 10
     loop_count=$(($loop_count + 1))
 
-    # read the CPU temperature
-    cpu_temp=$(get_peci_temp)
+    # Read the max skin temperature.
+    get_max_skin_temp $sensor_list
+
+    if [ $max_skin_temp -eq 0 ]; then
+      # TODO (snanda): use PECI temperature as a fallback.
+      log_message "Invalid max skin temp"
+    fi
 
-    # run the fan loop once a second
-    fan_loop $cpu_temp
+    # Run the fan loop every 10 seconds.
+    fan_loop $max_skin_temp
 
-    # run the thermal loop once a second
-    thermal_loop $cpu_temp
+    # Run the thermal loop every 10 seconds.
+    thermal_loop $max_skin_temp
 
-    # report the metrics once every 30 seconds
-    if [ $loop_count -lt 30 ]; then
+    # Report the metrics once every 30 seconds.
+    if [ $loop_count -lt 3 ]; then
       continue
     fi
     loop_count=0