⛄ Sync 2026-10-25 10:56:54

thermal-guard/Makefile

@@ -8,8 +8,8 @@
 include $(TOPDIR)/rules.mk
 
 PKG_NAME:=thermal-guard
-PKG_VERSION:=2.0.0
-PKG_RELEASE:=1
+PKG_VERSION:=2.1.1.0
+PKG_RELEASE:=2
 
 PKG_LICENSE:=GPL-2.0-only
 PKG_LICENSE_FILES:=LICENSE
@@ -29,7 +29,9 @@ define Package/thermal-guard/description
   Watches SoC and modem temperature and escalates in stages when a router runs
   hot, typically after a fan failure: fan to full speed, modem radio off, Wi-Fi
   off. Stages do not roll back on their own, clearing them is an operator
-  decision, and they survive a reboot. Fan control works either through the
+  decision, and they survive a reboot. One optional exception, off by default:
+  the second stage clears itself when the modem alone caused it and has read
+  cool for a while. Fan control works either through the
   kernel thermal zone or through sysfs PWM, modem temperature through Quectel or
   Fibocom AT commands, a file or a hook script. Warnings are posted to a URL.
   Where the kernel drives the fan, it can lower the fan's trip points while the
@@ -59,6 +61,7 @@ define Package/thermal-guard/install
 	$(INSTALL_BIN) ./files/usr/sbin/thermal-guard $(1)/usr/sbin/thermal-guard
 	$(INSTALL_DATA) ./files/usr/share/thermal-guard/backends.sh $(1)/usr/share/thermal-guard/backends.sh
 	$(INSTALL_DATA) ./files/usr/share/thermal-guard/trips.sh $(1)/usr/share/thermal-guard/trips.sh
+	$(INSTALL_DATA) ./files/usr/share/thermal-guard/modems $(1)/usr/share/thermal-guard/modems
 	$(INSTALL_CONF) ./files/etc/config/thermal-guard $(1)/etc/config/thermal-guard
 	$(INSTALL_BIN) ./files/etc/init.d/thermal-guard $(1)/etc/init.d/thermal-guard
 	$(INSTALL_BIN) ./files/etc/uci-defaults/90-thermal-guard-trip-base $(1)/etc/uci-defaults/

thermal-guard/files/etc/config/thermal-guard

@@ -1,20 +1,31 @@
 
 config thermal-guard 'main'
 	option enabled '1'
-	# seconds between readings, minimum 5
+	# seconds between readings, 5 to 300, values outside are set to the nearest limit
 	option interval '20'
 
 	# stage thresholds in degrees celsius, must be ordered warn < crit < emergency
 	option cpu_warn '70'
 	option cpu_crit '80'
 	option cpu_emergency '90'
-	option modem_warn '80'
-	option modem_crit '88'
+	# Unset takes the profile of the detected modem from
+	# /usr/share/thermal-guard/modems, otherwise 80 and 88. Set either to
+	# override it, warn below crit
+	#option modem_warn '80'
+	#option modem_crit '88'
 
 	# minutes a stage has to persist before the next one can trigger on time alone
 	option hold_minutes '3'
 	# degrees below the warn thresholds that count as cooled down
 	option hysteresis '10'
+	# Clear daemon stage 1 (stage 2 of 3 on the status page) on its own once
+	# the modem, which alone caused it, has read cool for recover_minutes
+	# (5 to 120), at most recover_max times in 24 hours (1 to 10). Stage 2, an
+	# incident where the processor was warm at any point and the 'command'
+	# action wait for a reset. Unset or 0 is off
+	#option auto_recover '1'
+	#option recover_minutes '10'
+	#option recover_max '3'
 
 	option cpu_temp_path '/sys/class/thermal/thermal_zone0/temp'
 	# also report the other temperature sensors of the board, Wi-Fi chips and so on
@@ -23,6 +34,10 @@ config thermal-guard 'main'
 	# none, cooling_device (kernel thermal zone), pwm (sysfs) or auto.
 	# none leaves the fan to whatever regulates it already. auto takes the cooling
 	# device if there is one and otherwise stays off the fan; it never picks pwm.
+	# The fan options below are read when a stage takes the fan over. Changed
+	# during a stage, they apply from the next takeover: the fan taken at stage
+	# entry stays in use (forced, checked, handed back) until it is released,
+	# also under none.
 	option fan_mode 'none'
 	# type of the fan's cooling device, for cooling_device, auto and trip
 	# management; empty means pwm-fan
@@ -72,7 +87,9 @@ config thermal-guard 'main'
 
 	# the message is posted here. Works with ntfy, Gotify, Matrix, Slack and
 	# anything else that takes a plain POST body. For a delivery HTTP cannot
-	# express, put an executable at /etc/thermal-guard/hooks/notify instead
+	# express, put an executable at /etc/thermal-guard/hooks/notify instead.
+	# Without curl the address is visible in the process list: it must not carry
+	# a secret in its path, and one with a query string or user info is refused
 	#option notify_url 'https://ntfy.sh/my-topic'
 	# only sent when curl is installed, uclient-fetch cannot set headers
 	#option notify_header 'Authorization: Bearer secret'

thermal-guard/files/etc/thermal-guard/hooks/README

@@ -22,12 +22,22 @@ delivery that way, and a missing modem-temp hook means no modem reading.
 
   modem-temp    Runs when modem_source is 'command'. Has to print the modem
                 temperature in whole degrees Celsius on stdout, nothing else.
+                A number of more than seven significant digits counts as no
+                reading (leading zeros are dropped first, a minus sign counts).
+
+  modem-model   Runs at start, in cycles with a modem reading, up to three times
+                until it prints a valid model, when modem_source is 'command'
+                or 'file'. Prints the modem model on stdout, for the modem
+                table in /usr/share/thermal-guard/modems. With ModemManager:
+                  mmcli -m any -K | sed -n 's/^modem\.generic\.model *: *//p'
+                Letters, digits, dot, underscore and dash, up to 32 characters.
 
 Every hook runs under a time limit: 15 seconds for notify, stage1 and stage2,
-5 seconds for modem-temp, which runs every cycle. A hook still running then
-gets TERM, and KILL 5 seconds later. It counts as failed: a stage action is
-reported as not carried out, a modem-temp hook as no reading. Keep hooks short;
-a mail or HTTP call inside one needs its own, shorter time limit.
+5 seconds for modem-temp, which runs every cycle, and for modem-model. A hook
+still running then gets TERM, and KILL 5 seconds later. It counts as failed: a
+stage action is reported as not carried out, a modem-temp hook as no reading.
+Keep hooks short; a mail or HTTP call inside one needs its own, shorter time
+limit.
 
 Example, a notify hook for a local mailer:
 

thermal-guard/files/usr/share/thermal-guard/modems

@@ -0,0 +1,17 @@
+# Modem profiles for thermal-guard: default modem limits per model.
+#
+# One line per profile: pattern, warn, crit, source, separated by blanks.
+#   pattern  shell glob, * and ? only, against the model the modem reports
+#            (AT+CGMM or the modem-model hook); case does not matter
+#   warn     modem_warn in whole degrees, 40 to 120
+#   crit     modem_crit, above warn, at most 120
+#   source   the rest of the line: the measurement or published limit the
+#            numbers come from. Required, a line without one is skipped.
+#
+# The first matching line wins, so specific patterns go above generic ones.
+# modem_warn and modem_crit set in /etc/config/thermal-guard override these
+# values one by one. thermal-guard reads the hottest sensor the modem
+# reports, which is a die or amplifier sensor, not the module case.
+#
+RM520N-GL*  85  95  GL-X3000 2026-10-19: 80 C after 90 min of 5G load, GL.iNet firmware sets no modem limit
+FM350-GL*   80  88  Banana Pi R3 Mini: plateau 73 to 74 C under sustained load (2026-09-19), 59 to 63 C with heatsink (2026-09-21)

thermal-guard/files/usr/share/thermal-guard/trips.sh

@@ -233,7 +233,7 @@ tg_trip_base_save() { # $1 whole degrees, space separated
 	mkdir -p "$PERSIST_DIR" 2>/dev/null &&
 		echo "$1" > "$f.tmp" 2>/dev/null && chmod 644 "$f.tmp" &&
 		mv "$f.tmp" "$f" 2>/dev/null || return 1
-	sync
+	tg_sync
 }
 
 tg_trip_base_read() {
@@ -336,7 +336,8 @@ tg_fan_levels() {
 	local f
 	for f in "$DT_ROOT"/*/cooling-levels "$DT_ROOT"/*/*/cooling-levels; do
 		[ -r "$f" ] || continue
-		tr '\0' '\n' < "${f%/*}/compatible" 2>/dev/null | grep -qx pwm-fan || continue
+		# The braces take the open error too, a node may have no compatible.
+		{ tr '\0' '\n' < "${f%/*}/compatible"; } 2>/dev/null | grep -qx pwm-fan || continue
 		hexdump -v -e '4/1 "%u " "\n"' "$f" 2>/dev/null |
 			awk 'NF == 4 { printf "%s%d", s, $1 * 16777216 + $2 * 65536 + $3 * 256 + $4; s = " " }'
 		return 0