Prepare to handle thermal event

Intel's new hardware supports Hardware Feedback Interface to provide CPU
performance and energy efficiency information. Every update on this is
delivered via thermal event interrupt. The thermal framework in the Linux
kernel relays these notifications to userspace via a Netlink interface.

When a CPU's performance and efficiency are zero, irqbalance needs to mask
the CPU from interrupts. Introduce a new CPU mask to indicate banned CPUs
for this.

Before supporting this event handling, define functions. Their
implementation will be on the following patches.

This event is available only on x86-64 systems. And it can be subscribed
with help of Netlink libraries. So check them before building it.

Also add a new build option so users may opt out this support. Setting this
option on other systems will result in a build failure.

Signed-off-by: Chang S. Bae <chang.seok.bae@intel.com>

Makefile.am

@@ -27,7 +27,7 @@ EXTRA_DIST = COPYING autogen.sh misc/irqbalance.service misc/irqbalance.env
 SUBDIRS = tests
 
 UI_DIR = ui
-AM_CFLAGS = $(LIBCAP_NG_CFLAGS) $(GLIB2_CFLAGS) $(NUMA_CFLAGS)
+AM_CFLAGS = $(LIBCAP_NG_CFLAGS) $(GLIB2_CFLAGS) $(NUMA_CFLAGS) $(LIBNL3_CFLAGS)
 AM_CPPFLAGS = -I${top_srcdir} -W -Wall -Wshadow -Wformat -Wundef -D_GNU_SOURCE
 noinst_HEADERS = bitmap.h constants.h cpumask.h irqbalance.h non-atomic.h \
 	types.h $(UI_DIR)/helpers.h $(UI_DIR)/irqbalance-ui.h $(UI_DIR)/ui.h
@@ -39,7 +39,10 @@ endif
 
 irqbalance_SOURCES = activate.c bitmap.c classify.c cputree.c irqbalance.c \
 	irqlist.c numa.c placement.c procinterrupts.c
-irqbalance_LDADD = $(LIBCAP_NG_LIBS) $(GLIB2_LIBS) $(NUMA_LIBS)
+if THERMAL
+irqbalance_SOURCES += thermal.c
+endif
+irqbalance_LDADD = $(LIBCAP_NG_LIBS) $(GLIB2_LIBS) $(NUMA_LIBS) $(LIBNL3_LIBS)
 if IRQBALANCEUI
 irqbalance_ui_SOURCES = $(UI_DIR)/helpers.c $(UI_DIR)/irqbalance-ui.c \
 	$(UI_DIR)/ui.c

configure.ac

@@ -36,6 +36,28 @@ AS_IF([test "x$has_ncursesw" = "xyes"], [
   AC_SUBST([LIBS])
 ])
 
+AC_CANONICAL_HOST
+
+AC_ARG_ENABLE(thermal,
+  AS_HELP_STRING([--enable-thermal], [enable thermal event support [default=auto]]),,
+  AS_IF([test x"$host_cpu" = x"x86_64"], [enable_thermal=yes], [enable_thermal=no])
+)
+
+AS_IF([test x"$enable_thermal" = x"yes" && test x"$host_cpu" != x"x86_64"],
+  AC_MSG_ERROR([no thermal events support on $host_cpu systems.]),
+)
+
+AS_IF([test x"$enable_thermal" = x"yes"],
+  [PKG_CHECK_MODULES([LIBNL3], [libnl-3.0 libnl-genl-3.0], [have_thermal=yes],
+    AC_MSG_NOTICE([no thermal event support as libnl-3.0 is unavailable.])
+  )]
+)
+
+AS_IF([test "x$have_thermal" = xyes],
+  AC_DEFINE([HAVE_THERMAL], 1, [Build irqbalance to support thermal events])
+)
+AM_CONDITIONAL([THERMAL], [test "x$have_thermal" = xyes])
+
 AC_C_CONST
 AC_C_INLINE
 AM_PROG_CC_C_O

cputree.c

@@ -38,6 +38,7 @@
 #include <glib.h>
 
 #include "irqbalance.h"
+#include "thermal.h"
 
 #ifdef HAVE_IRQBALANCEUI
 extern char *banned_cpumask_from_ui;
@@ -162,6 +163,11 @@ static void setup_banned_cpus(void)
 	cpumask_scnprintf(buffer, 4096, nohz_full);
 	log(TO_CONSOLE, LOG_INFO, "Adaptive-ticks CPUs: %s\n", buffer);
 out:
+#ifdef HAVE_THERMAL
+	cpus_or(banned_cpus, banned_cpus, thermal_banned_cpus);
+	cpumask_scnprintf(buffer, 4096, thermal_banned_cpus);
+	log(TO_CONSOLE, LOG_INFO, "Thermal-banned CPUs: %s\n", buffer);
+#endif
 	cpumask_scnprintf(buffer, 4096, banned_cpus);
 	log(TO_CONSOLE, LOG_INFO, "Banned CPUs: %s\n", buffer);
 }

irqbalance.c

@@ -44,6 +44,7 @@
 #include <sys/socket.h>
 #endif
 #include "irqbalance.h"
+#include "thermal.h"
 
 volatile int keep_going = 1;
 int one_shot_mode;
@@ -703,6 +704,8 @@ int main(int argc, char** argv)
 		goto out;
 	}
 #endif
+	if (init_thermal())
+		log(TO_ALL, LOG_WARNING, "Failed to initialize thermal events.\n");
 	main_loop = g_main_loop_new(NULL, FALSE);
 	last_interval = sleep_interval;
 	g_timeout_add_seconds(sleep_interval, scan, NULL);
@@ -711,6 +714,7 @@ int main(int argc, char** argv)
 	g_main_loop_quit(main_loop);
 
 out:
+	deinit_thermal();
 	free_object_tree();
 	free_cl_opts();
 	free(polscript);

thermal.c

@@ -0,0 +1,83 @@
+/*
+ * This program is free software; you can redistribute it and/or modify
+ * it under the terms of the GNU General Public License version 2 as
+ * published by the Free Software Foundation.
+ */
+
+#include <stdio.h>
+
+#include <netlink/genl/genl.h>
+#include <netlink/genl/family.h>
+#include <netlink/genl/ctrl.h>
+
+#include "irqbalance.h"
+
+cpumask_t thermal_banned_cpus;
+
+static gboolean prepare_netlink(void)
+{
+	gboolean error = TRUE;
+
+	log(TO_ALL, LOG_ERR, "thermal: not yet implement to alloc memory for netlink.\n");
+	return error;
+}
+
+#define NL_FAMILY_NAME	"nlctrl"
+
+static gboolean establish_netlink(void)
+{
+	gboolean error = TRUE;
+
+	log(TO_ALL, LOG_ERR, "thermal: not yet implemented to establish netlink.\n");
+	return error;
+}
+
+static gboolean register_netlink_handler(nl_recvmsg_msg_cb_t handler __attribute__((unused)))
+{
+	gboolean error = TRUE;
+
+	log(TO_ALL, LOG_ERR, "thermal: not yet implemented to register thermal handler.\n");
+	return error;
+}
+
+static gboolean set_netlink_nonblocking(void)
+{
+	gboolean error = TRUE;
+
+	log(TO_ALL, LOG_ERR, "thermal: not yet implemented to set nonblocking socket.\n");
+	return error;
+}
+
+void deinit_thermal(void)
+{
+	return;
+}
+
+/*
+ * return value: TRUE with an error; otherwise, FALSE
+ */
+gboolean init_thermal(void)
+{
+	gboolean error;
+
+	error = prepare_netlink();
+	if (error)
+		goto err_out;
+
+	error = establish_netlink();
+	if (error)
+		goto err_out;
+
+	error = register_netlink_handler(NULL);
+	if (error)
+		goto err_out;
+
+	error = set_netlink_nonblocking();
+	if (error)
+		goto err_out;
+
+	return FALSE;
+err_out:
+	deinit_thermal();
+	return TRUE;
+}

thermal.h

@@ -0,0 +1,15 @@
+#ifndef __LINUX_THERMAL_H_
+#define __LINUX_THERMAL_H_
+
+#include <glib.h>
+
+#ifdef HAVE_THERMAL
+gboolean init_thermal(void);
+void deinit_thermal(void);
+extern cpumask_t thermal_banned_cpus;
+#else
+static inline gboolean init_thermal(void) { return FALSE; }
+#define deinit_thermal() do { } while (0)
+#endif
+
+#endif /* __LINUX_THERMAL_H_ */