fix bug where unmonitored gpus contribute to fan speed

CMakeLists.txt

@@ -4,14 +4,27 @@ project(ipmifan LANGUAGES CXX)
 
 set(CMAKE_CXX_STANDARD 20)
 
-find_package(Doxygen)
+if(CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT)
+	set(CMAKE_INSTALL_PREFIX "/usr" CACHE PATH "..." FORCE)
+endif(CMAKE_INSTALL_PREFIX_INITIALIZED_TO_DEFAULT)
 
 find_package(PkgConfig REQUIRED)
-pkg_search_module(IPMI REQUIRED libipmimonitoring)
+pkg_search_module(IPMI_MONITORING REQUIRED libipmimonitoring)
+pkg_search_module(IPMI REQUIRED libfreeipmi)
+pkg_search_module(SYSTEMD systemd)
+
 
 add_executable(${PROJECT_NAME} main.cpp ipmi.cpp lm.cpp)
-target_link_libraries(${PROJECT_NAME} ${IPMI_LINK_LIBRARIES} ${IPMIPOSIX_LINK_LIBRARIES} sensors)
+target_link_libraries(${PROJECT_NAME} ${IPMI_LINK_LIBRARIES} ${IPMI_MONITORING_LINK_LIBRARIES} sensors)
-target_include_directories(${PROJECT_NAME} PRIVATE ${IPMI_INCLUDE_DIRS} ${IPMIPOSIX_INCLUDE_DIRS})
+target_include_directories(${PROJECT_NAME} PRIVATE ${IPMI_INCLUDE_DIRS} ${IPMI_MONITORING_INCLUDE_DIRS})
 target_compile_options(${PROJECT_NAME} PRIVATE "-Wall" "-O2" "-g" "-fno-strict-aliasing" "-Wfatal-errors" "-Wno-reorder")
 install(TARGETS ${PROJECT_NAME} RUNTIME DESTINATION bin)
 
+if(SYSTEMD_FOUND)
+	pkg_get_variable(SYSTEMD_UNIT_DIR_PKG systemd systemd_system_unit_path)
+	message(STATUS ${SYSTEMD_UNIT_DIR_PKG})
+	string(REPLACE ":" ";" SYSTEMD_UNIT_DIR_LIST ${SYSTEMD_UNIT_DIR_PKG})
+	list(GET SYSTEMD_UNIT_DIR_LIST 0 SYSTEMD_UNIT_DIR)
+
+	install(FILES ipmifan.service DESTINATION ${SYSTEMD_UNIT_DIR})
+endif(SYSTEMD_FOUND)

ipmi.cpp

@@ -2,6 +2,8 @@
 #include <algorithm>
 #include <iostream>
 
+static constexpr size_t IPMI_RAW_MAX_ARGS = 65536*2;
+
 static double ipmi_convert_sensor_reading(void *sensor_reading, int sensor_reading_type)
 {
 	if(sensor_reading_type == IPMI_MONITORING_SENSOR_READING_TYPE_UNSIGNED_INTEGER8_BOOL)
@@ -102,3 +104,37 @@ ipmi_monitoring_ctx_t init_ipmi_monitoring()
 
 	return ctx;
 }
+
+ipmi_ctx_t ipmi_open_context()
+{
+	ipmi_ctx_t ctx = ipmi_ctx_create();
+	if(!ctx)
+	{
+		std::cerr<<"Could not allocae raw context\n";
+		return nullptr;
+	}
+
+	ipmi_driver_type_t driver = IPMI_DEVICE_OPENIPMI;
+	int ret = ipmi_ctx_find_inband(ctx, &driver, false, 0, 0, nullptr, 0, 0);
+	if(ret < 0)
+	{
+		std::cerr<<"Could not create raw context "<<ipmi_ctx_errormsg(ctx)<<'\n';
+		ipmi_ctx_destroy(ctx);
+		return nullptr;
+	}
+	return ctx;
+}
+
+bool ipmi_set_fan_group(ipmi_ctx_t raw_ctx, uint8_t group, double speed)
+{
+	char converted_speed = std::max(std::min(static_cast<char>(100), static_cast<char>(speed*100)), static_cast<char>(0));
+	char command[] = {0x70, 0x66, 0x01, static_cast<char>(group), converted_speed};
+	char bytesrx[IPMI_RAW_MAX_ARGS] = {0};
+	int rxlen = ipmi_cmd_raw(raw_ctx, 0, 0x30, command, sizeof(command), bytesrx, IPMI_RAW_MAX_ARGS);
+	if(rxlen < 0)
+	{
+		std::cerr<<"Raw write to ipmi failed with: "<<ipmi_ctx_errormsg(raw_ctx);
+		return false;
+	}
+	return true;
+}

ipmi.h

@@ -1,6 +1,7 @@
 #include <vector>
 #include <string>
 #include <ipmi_monitoring.h>
+#include <freeipmi/freeipmi.h>
 
 #include "sensor.h"
 
@@ -11,3 +12,7 @@ bool ipmi_fill_sensor_ids(std::vector<Sensor>& sensors, ipmi_monitoring_ctx_t ct
 bool ipmi_update_sensors(std::vector<Sensor>& sensors, ipmi_monitoring_ctx_t ctx, struct ipmi_monitoring_ipmi_config* config);
 
 ipmi_monitoring_ctx_t init_ipmi_monitoring();
+
+ipmi_ctx_t ipmi_open_context();
+
+bool ipmi_set_fan_group(ipmi_ctx_t raw_ctx, uint8_t group, double speed);

ipmifan.service

@@ -0,0 +1,10 @@
+[Unit]
+Description=Start impi fan control
+After=lm_sensors.service systemd-modules-load.service
+
+[Service]
+Type=simple
+ExecStart=/usr/bin/ipmifan -q
+
+[Install]
+WantedBy=multi-user.target

main.cpp

@@ -1,3 +1,6 @@
+#include <cstddef>
+#include <cstdint>
+#include <freeipmi/api/ipmi-api.h>
 #include <vector>
 #include <string>
 #include <iostream>
@@ -7,6 +10,7 @@
 #include <sensors/error.h>
 #include <signal.h>
 #include <limits>
+#include <freeipmi/freeipmi.h>
 
 #include "ipmi.h"
 #include "lm.h"
@@ -19,6 +23,8 @@ void sig_handler(int sig)
 	running = false;
 }
 
+bool quiet;
+
 std::vector<Sensor> gather_sensors(std::vector<Sensor>& ipmi_sensors, ipmi_monitoring_ctx_t ctx, std::vector<const sensors_chip_name*>& lm_chips)
 {
 	std::vector<Sensor> out;
@@ -63,41 +69,71 @@ double fan_curve(double temperature, double min_fan, double max_fan, double low_
 
 double gpu_fan_zone(const std::vector<Sensor>& sensors)
 {
-	const char mi50Chip[] = "amdgpu-pci-2300";
+	std::vector<std::pair<std::string, bool>> gpus = {{"amdgpu-pci-0300", false}, {"amdgpu-pci-8300", false}, {"amdgpu-pci-8900", false}};
-	const char mi25Chip[] = "amdgpu-pci-4300";
 	const char monitored_sensor_name[] = "edge";
 
 	double max_temp = std::numeric_limits<double>::min();
 	for(const Sensor& sensor : sensors)
 	{
-		if((sensor.chip == mi50Chip || sensor.chip == mi25Chip) && sensor.name == monitored_sensor_name)
+		if(sensor.name == monitored_sensor_name)
 		{
-			if(max_temp < sensor.reading)
+			for(std::pair<std::string, bool>& gpu : gpus)
-				max_temp = sensor.reading;
+			{
+				if(sensor.chip == gpu.first)
+				{
+					if(max_temp < sensor.reading)
+						max_temp = sensor.reading;
+					gpu.second = true;
+				}
+			}
+		}
+	}
+	for(std::pair<std::string, bool>& gpu : gpus)
+	{
+		if(!gpu.second)
+		{
+			std::cerr<<"Could not get temperature from "<<gpu.first<<" ramping fans to maximum\n";
+			return 1.0;
 		}
 	}
 
-	return fan_curve(max_temp, 0.2, 1.0, 40, 75);
+	return fan_curve(max_temp, 0.05, 1.0, 45, 75);
 }
 
 double system_fan_zone(const std::vector<Sensor>& sensors)
 {
 	Sensor cpu("IPMI", "CPU Temp");
 	Sensor system("IPMI", "System Temp");
+	bool hitCpu = false;
+	bool hitSystem = false;
 	std::vector<double> out;
 
 	for(const Sensor& sensor : sensors)
 	{
 		if(cpu == sensor)
+		{
+			hitCpu = true;
 			cpu = sensor;
+		}
 		else if(sensor == system)
+		{
+			hitSystem = true;
 			system = sensor;
+		}
 	}
 
-	double fanSystem = fan_curve(system.reading, 0.2, 1.0, 35, 45);
+	if(hitCpu && hitSystem)
-	double fanCpu = fan_curve(cpu.reading, 0.2, 1.0, 40, 70);
+	{
+		double fanSystem = fan_curve(system.reading, 0.33, 1.0, 40, 65);
+		double fanCpu = fan_curve(cpu.reading, 0.33, 1.0, 40, 70);
 
-	return std::max(fanSystem, fanCpu);
+		return std::max(fanSystem, fanCpu);
+	}
+	else
+	{
+		std::cerr<<"Could not get temperature from System or Cpu! Ramping fans to maximum\n";
+		return 1;
+	}
 }
 
 std::vector<double> get_fan_zones(const std::vector<Sensor>& sensors)
@@ -108,17 +144,24 @@ std::vector<double> get_fan_zones(const std::vector<Sensor>& sensors)
 	return out;
 }
 
-int main (int argc, char **argv)
+int main_loop()
 {
-	signal(SIGABRT, sig_handler);
+	ipmi_ctx_t raw_ctx = ipmi_open_context();
-	signal(SIGTERM, sig_handler);
+	if(!raw_ctx)
-	signal(SIGHUP, sig_handler);
+	{
-	signal(SIGINT, sig_handler);
+		sensors_cleanup();
+		return 1;
+	}
 
 	int ret = sensors_init(nullptr);
 	if(ret < 0)
 	{
 		std::cerr<<"Could not init lm_sensors\n";
+		ipmi_set_fan_group(raw_ctx, 0, 1);
+		ipmi_set_fan_group(raw_ctx, 1, 1);
+		ipmi_ctx_close(raw_ctx);
+		ipmi_ctx_destroy(raw_ctx);
+		return 1;
 	}
 
 	std::vector<const sensors_chip_name*> lm_chips = lm_get_chips("amdgpu-*");
@@ -126,21 +169,80 @@ int main (int argc, char **argv)
 	ipmi_sensors.push_back(Sensor("IPMI", "CPU Temp"));
 	ipmi_sensors.push_back(Sensor("IPMI", "System Temp"));
 
-	ipmi_monitoring_ctx_t ctx = init_ipmi_monitoring();
+	if(lm_chips.size() < 2)
-	if(!ctx)
+	{
+		std::cerr<<"Could not get both monitored gpus!";
+		ipmi_set_fan_group(raw_ctx, 0, 1);
+		ipmi_set_fan_group(raw_ctx, 1, 1);
+		ipmi_ctx_close(raw_ctx);
+		ipmi_ctx_destroy(raw_ctx);
+		sensors_cleanup();
 		return 1;
+	}
+
+	ipmi_monitoring_ctx_t monitoring_ctx = init_ipmi_monitoring();
+	if(!monitoring_ctx)
+	{
+		ipmi_set_fan_group(raw_ctx, 0, 1);
+		ipmi_set_fan_group(raw_ctx, 1, 1);
+		ipmi_ctx_close(raw_ctx);
+		ipmi_ctx_destroy(raw_ctx);
+		sensors_cleanup();
+		return 1;
+	}
+
 
 	while(running)
 	{
-		std::vector<Sensor> sensors = gather_sensors(ipmi_sensors, ctx, lm_chips);
+		std::vector<Sensor> sensors = gather_sensors(ipmi_sensors, monitoring_ctx, lm_chips);
 		std::vector<double> fanzones = get_fan_zones(sensors);
-		for(const double fanzone : fanzones)
+
-			std::cout<<fanzone<<'\n';
+		if(!quiet)
-		sleep(1);
+		{
+			for(const Sensor& sensor : sensors)
+				std::cout<<"Sensor "<<sensor.chip<<':'<<sensor.name<<"\t= "<<sensor.reading<<'\n';
+			for(size_t i = 0; i < fanzones.size(); ++i)
+				std::cout<<"setting fan group "<<i<<" to "<<fanzones[i]*100<<"%\n";
+		}
+
+		ipmi_set_fan_group(raw_ctx, 0, fanzones[0]);
+		ipmi_set_fan_group(raw_ctx, 1, fanzones[1]);
+		sleep(10);
 	}
 
-	ipmi_monitoring_ctx_destroy(ctx);
+	ipmi_set_fan_group(raw_ctx, 0, 1);
+	ipmi_set_fan_group(raw_ctx, 1, 1);
+	ipmi_ctx_close(raw_ctx);
+	ipmi_ctx_destroy(raw_ctx);
+	ipmi_monitoring_ctx_destroy(monitoring_ctx);
 	sensors_cleanup();
+
 	return 0;
 }
 
+int main (int argc, char **argv)
+{
+	signal(SIGABRT, sig_handler);
+	signal(SIGTERM, sig_handler);
+	signal(SIGHUP, sig_handler);
+	signal(SIGINT, sig_handler);
+
+	if(argc > 1)
+		quiet = true;
+
+	int ret = 0;
+	for(size_t i = 0; i < 3; ++i)
+	{
+		ret = main_loop();
+		if(!running)
+			break;
+		std::cerr<<"Mainloop unable to start, retrying in 10 sec\n";
+		sleep(10);
+	}
+
+	if(ret != 0)
+		std::cerr<<"Error not clearing, giveing up\n";
+
+	return ret;
+}
+