MFcalloutng:

When CPU becomes idle, cpu_idleclock() calculates time to the next timer
event in order to reprogram hw timer. Return that time in sbintime_t to
the caller and pass it to acpi_cpu_idle(), where it can be used as one
more factor (quite precise) to extimate furter sleep time and choose
optimal sleep state. This is a preparatory change for further callout
improvements will be committed in the next days.

The commmit is not targeted for MFC.

11 files changed, 59 insertions, 48 deletions

diff --git a/sys/amd64/amd64/machdep.c b/sys/amd64/amd64/machdep.c
index abce826..1d7178f 100644
--- a/sys/amd64/amd64/machdep.c
+++ b/sys/amd64/amd64/machdep.c
@@ -658,7 +658,7 @@ cpu_halt(void)
 		halt();
 }
 
-void (*cpu_idle_hook)(void) = NULL;	/* ACPI idle hook. */
+void (*cpu_idle_hook)(sbintime_t) = NULL;	/* ACPI idle hook. */
 static int	cpu_ident_amdc1e = 0;	/* AMD C1E supported. */
 static int	idle_mwait = 1;		/* Use MONITOR/MWAIT for short idle. */
 TUNABLE_INT("machdep.idle_mwait", &idle_mwait);
@@ -670,7 +670,7 @@ SYSCTL_INT(_machdep, OID_AUTO, idle_mwait, CTLFLAG_RW, &idle_mwait,
 #define	STATE_SLEEPING	0x2
 
 static void
-cpu_idle_acpi(int busy)
+cpu_idle_acpi(sbintime_t sbt)
 {
 	int *state;
 
@@ -682,14 +682,14 @@ cpu_idle_acpi(int busy)
 	if (sched_runnable())
 		enable_intr();
 	else if (cpu_idle_hook)
-		cpu_idle_hook();
+		cpu_idle_hook(sbt);
 	else
 		__asm __volatile("sti; hlt");
 	*state = STATE_RUNNING;
 }
 
 static void
-cpu_idle_hlt(int busy)
+cpu_idle_hlt(sbintime_t sbt)
 {
 	int *state;
 
@@ -730,7 +730,7 @@ cpu_idle_hlt(int busy)
 #define	MWAIT_C4	0x30
 
 static void
-cpu_idle_mwait(int busy)
+cpu_idle_mwait(sbintime_t sbt)
 {
 	int *state;
 
@@ -753,7 +753,7 @@ cpu_idle_mwait(int busy)
 }
 
 static void
-cpu_idle_spin(int busy)
+cpu_idle_spin(sbintime_t sbt)
 {
 	int *state;
 	int i;
@@ -802,12 +802,13 @@ cpu_probe_amdc1e(void)
 	}
 }
 
-void (*cpu_idle_fn)(int) = cpu_idle_acpi;
+void (*cpu_idle_fn)(sbintime_t) = cpu_idle_acpi;
 
 void
 cpu_idle(int busy)
 {
 	uint64_t msr;
+	sbintime_t sbt = -1;
 
 	CTR2(KTR_SPARE2, "cpu_idle(%d) at %d",
 	    busy, curcpu);
@@ -825,7 +826,7 @@ cpu_idle(int busy)
 	/* If we have time - switch timers into idle mode. */
 	if (!busy) {
 		critical_enter();
-		cpu_idleclock();
+		sbt = cpu_idleclock();
 	}
 
 	/* Apply AMD APIC timer C1E workaround. */
@@ -836,7 +837,7 @@ cpu_idle(int busy)
 	}
 
 	/* Call main idle method. */
-	cpu_idle_fn(busy);
+	cpu_idle_fn(sbt);
 
 	/* Switch timers mack into active mode. */
 	if (!busy) {
diff --git a/sys/dev/acpica/acpi_cpu.c b/sys/dev/acpica/acpi_cpu.c
index 7f008f7..df4c4c2 100644
--- a/sys/dev/acpica/acpi_cpu.c
+++ b/sys/dev/acpica/acpi_cpu.c
@@ -168,7 +168,7 @@ static int	acpi_cpu_cx_cst(struct acpi_cpu_softc *sc);
 static void	acpi_cpu_startup(void *arg);
 static void	acpi_cpu_startup_cx(struct acpi_cpu_softc *sc);
 static void	acpi_cpu_cx_list(struct acpi_cpu_softc *sc);
-static void	acpi_cpu_idle(void);
+static void	acpi_cpu_idle(sbintime_t sbt);
 static void	acpi_cpu_notify(ACPI_HANDLE h, UINT32 notify, void *context);
 static int	acpi_cpu_quirks(void);
 static int	acpi_cpu_usage_sysctl(SYSCTL_HANDLER_ARGS);
@@ -954,13 +954,13 @@ acpi_cpu_startup_cx(struct acpi_cpu_softc *sc)
  * interrupts are re-enabled.
  */
 static void
-acpi_cpu_idle()
+acpi_cpu_idle(sbintime_t sbt)
 {
     struct	acpi_cpu_softc *sc;
     struct	acpi_cx *cx_next;
     uint64_t	cputicks;
     uint32_t	start_time, end_time;
-    int		bm_active, cx_next_idx, i;
+    int		bm_active, cx_next_idx, i, us;
 
     /*
      * Look up our CPU id to get our softc.  If it's NULL, we'll use C1
@@ -980,13 +980,16 @@ acpi_cpu_idle()
     }
 
     /* Find the lowest state that has small enough latency. */
+    us = sc->cpu_prev_sleep;
+    if (sbt >= 0 && us > sbt / SBT_1US)
+	us = sbt / SBT_1US;
     cx_next_idx = 0;
     if (cpu_disable_deep_sleep)
 	i = min(sc->cpu_cx_lowest, sc->cpu_non_c3);
     else
 	i = sc->cpu_cx_lowest;
     for (; i >= 0; i--) {
-	if (sc->cpu_cx_states[i].trans_lat * 3 <= sc->cpu_prev_sleep) {
+	if (sc->cpu_cx_states[i].trans_lat * 3 <= us) {
 	    cx_next_idx = i;
 	    break;
 	}
diff --git a/sys/i386/i386/machdep.c b/sys/i386/i386/machdep.c
index 744c3cb..3f15f08 100644
--- a/sys/i386/i386/machdep.c
+++ b/sys/i386/i386/machdep.c
@@ -1220,7 +1220,7 @@ cpu_halt(void)
 int scheduler_running;
 
 static void
-cpu_idle_hlt(int busy)
+cpu_idle_hlt(sbintime_t sbt)
 {
 
 	scheduler_running = 1;
@@ -1241,7 +1241,7 @@ cpu_halt(void)
 
 #endif
 
-void (*cpu_idle_hook)(void) = NULL;	/* ACPI idle hook. */
+void (*cpu_idle_hook)(sbintime_t) = NULL;	/* ACPI idle hook. */
 static int	cpu_ident_amdc1e = 0;	/* AMD C1E supported. */
 static int	idle_mwait = 1;		/* Use MONITOR/MWAIT for short idle. */
 TUNABLE_INT("machdep.idle_mwait", &idle_mwait);
@@ -1253,7 +1253,7 @@ SYSCTL_INT(_machdep, OID_AUTO, idle_mwait, CTLFLAG_RW, &idle_mwait,
 #define	STATE_SLEEPING	0x2
 
 static void
-cpu_idle_acpi(int busy)
+cpu_idle_acpi(sbintime_t sbt)
 {
 	int *state;
 
@@ -1265,7 +1265,7 @@ cpu_idle_acpi(int busy)
 	if (sched_runnable())
 		enable_intr();
 	else if (cpu_idle_hook)
-		cpu_idle_hook();
+		cpu_idle_hook(sbt);
 	else
 		__asm __volatile("sti; hlt");
 	*state = STATE_RUNNING;
@@ -1273,7 +1273,7 @@ cpu_idle_acpi(int busy)
 
 #ifndef XEN
 static void
-cpu_idle_hlt(int busy)
+cpu_idle_hlt(sbintime_t sbt)
 {
 	int *state;
 
@@ -1315,7 +1315,7 @@ cpu_idle_hlt(int busy)
 #define	MWAIT_C4	0x30
 
 static void
-cpu_idle_mwait(int busy)
+cpu_idle_mwait(sbintime_t sbt)
 {
 	int *state;
 
@@ -1338,7 +1338,7 @@ cpu_idle_mwait(int busy)
 }
 
 static void
-cpu_idle_spin(int busy)
+cpu_idle_spin(sbintime_t sbt)
 {
 	int *state;
 	int i;
@@ -1388,9 +1388,9 @@ cpu_probe_amdc1e(void)
 }
 
 #ifdef XEN
-void (*cpu_idle_fn)(int) = cpu_idle_hlt;
+void (*cpu_idle_fn)(sbintime_t) = cpu_idle_hlt;
 #else
-void (*cpu_idle_fn)(int) = cpu_idle_acpi;
+void (*cpu_idle_fn)(sbintime_t) = cpu_idle_acpi;
 #endif
 
 void
@@ -1399,6 +1399,7 @@ cpu_idle(int busy)
 #ifndef XEN
 	uint64_t msr;
 #endif
+	sbintime_t sbt = -1;
 
 	CTR2(KTR_SPARE2, "cpu_idle(%d) at %d",
 	    busy, curcpu);
@@ -1418,7 +1419,7 @@ cpu_idle(int busy)
 	/* If we have time - switch timers into idle mode. */
 	if (!busy) {
 		critical_enter();
-		cpu_idleclock();
+		sbt = cpu_idleclock();
 	}
 
 #ifndef XEN
@@ -1431,7 +1432,7 @@ cpu_idle(int busy)
 #endif
 
 	/* Call main idle method. */
-	cpu_idle_fn(busy);
+	cpu_idle_fn(sbt);
 
 	/* Switch timers mack into active mode. */
 	if (!busy) {
diff --git a/sys/ia64/ia64/machdep.c b/sys/ia64/ia64/machdep.c
index 98bcf44..ac6a829 100644
--- a/sys/ia64/ia64/machdep.c
+++ b/sys/ia64/ia64/machdep.c
@@ -171,7 +171,7 @@ extern vm_offset_t ksym_start, ksym_end;
 struct msgbuf *msgbufp = NULL;
 
 /* Other subsystems (e.g., ACPI) can hook this later. */
-void (*cpu_idle_hook)(void) = NULL;
+void (*cpu_idle_hook)(sbintime_t) = NULL;
 
 struct kva_md_info kmi;
 
@@ -408,10 +408,11 @@ void
 cpu_idle(int busy)
 {
 	register_t ie;
+	sbintime_t sbt = -1;
 
 	if (!busy) {
 		critical_enter();
-		cpu_idleclock();
+		sbt = cpu_idleclock();
 	}
 
 	ie = intr_disable();
@@ -420,7 +421,7 @@ cpu_idle(int busy)
 	if (sched_runnable())
 		ia64_enable_intr();
 	else if (cpu_idle_hook != NULL) {
-		(*cpu_idle_hook)();
+		(*cpu_idle_hook)(sbt);
 		/* The hook must enable interrupts! */
 	} else {
 		ia64_call_pal_static(PAL_HALT_LIGHT, 0, 0, 0);
diff --git a/sys/kern/kern_clocksource.c b/sys/kern/kern_clocksource.c
index 33f21e9..ac7e2de 100644
--- a/sys/kern/kern_clocksource.c
+++ b/sys/kern/kern_clocksource.c
@@ -767,7 +767,7 @@ cpu_stopprofclock(void)
 /*
  * Switch to idle mode (all ticks handled).
  */
-void
+sbintime_t
 cpu_idleclock(void)
 {
 	struct bintime now, t;
@@ -779,7 +779,7 @@ cpu_idleclock(void)
 	    || curcpu == CPU_FIRST()
 #endif
 	    )
-		return;
+		return (-1);
 	state = DPCPU_PTR(timerstate);
 	if (periodic)
 		now = state->now;
@@ -795,6 +795,8 @@ cpu_idleclock(void)
 	if (!periodic)
 		loadtimer(&now, 0);
 	ET_HW_UNLOCK(state);
+	bintime_sub(&t, &now);
+	return (MAX(bttosbt(t), 0));
 }
 
 /*
diff --git a/sys/pc98/pc98/machdep.c b/sys/pc98/pc98/machdep.c
index 3133d1b..b0d5acd 100644
--- a/sys/pc98/pc98/machdep.c
+++ b/sys/pc98/pc98/machdep.c
@@ -1145,7 +1145,7 @@ SYSCTL_INT(_machdep, OID_AUTO, idle_mwait, CTLFLAG_RW, &idle_mwait,
 #define	STATE_SLEEPING	0x2
 
 static void
-cpu_idle_hlt(int busy)
+cpu_idle_hlt(sbintime_t sbt)
 {
 	int *state;
 
@@ -1186,7 +1186,7 @@ cpu_idle_hlt(int busy)
 #define	MWAIT_C4	0x30
 
 static void
-cpu_idle_mwait(int busy)
+cpu_idle_mwait(sbintime_t sbt)
 {
 	int *state;
 
@@ -1209,7 +1209,7 @@ cpu_idle_mwait(int busy)
 }
 
 static void
-cpu_idle_spin(int busy)
+cpu_idle_spin(sbintime_t sbt)
 {
 	int *state;
 	int i;
@@ -1229,11 +1229,12 @@ cpu_idle_spin(int busy)
 	}
 }
 
-void (*cpu_idle_fn)(int) = cpu_idle_hlt;
+void (*cpu_idle_fn)(sbintime_t) = cpu_idle_hlt;
 
 void
 cpu_idle(int busy)
 {
+	sbintime_t sbt = -1;
 
 	CTR2(KTR_SPARE2, "cpu_idle(%d) at %d",
 	    busy, curcpu);
@@ -1251,11 +1252,11 @@ cpu_idle(int busy)
 	/* If we have time - switch timers into idle mode. */
 	if (!busy) {
 		critical_enter();
-		cpu_idleclock();
+		sbt = cpu_idleclock();
 	}
 
 	/* Call main idle method. */
-	cpu_idle_fn(busy);
+	cpu_idle_fn(sbt);
 
 	/* Switch timers mack into active mode. */
 	if (!busy) {
diff --git a/sys/powerpc/powerpc/cpu.c b/sys/powerpc/powerpc/cpu.c
index d177c8b..d67f359 100644
--- a/sys/powerpc/powerpc/cpu.c
+++ b/sys/powerpc/powerpc/cpu.c
@@ -79,9 +79,9 @@ static void	cpu_970_setup(int cpuid, uint16_t vers);
 static void	cpu_booke_setup(int cpuid, uint16_t vers);
 
 int powerpc_pow_enabled;
-void (*cpu_idle_hook)(void) = NULL;
-static void	cpu_idle_60x(void);
-static void	cpu_idle_booke(void);
+void (*cpu_idle_hook)(sbintime_t) = NULL;
+static void	cpu_idle_60x(sbintime_t);
+static void	cpu_idle_booke(sbintime_t);
 
 struct cputab {
 	const char	*name;
@@ -516,6 +516,7 @@ cpu_feature_bit(SYSCTL_HANDLER_ARGS)
 void
 cpu_idle(int busy)
 {
+	sbintime_t sbt = -1;
 
 #ifdef INVARIANTS
 	if ((mfmsr() & PSL_EE) != PSL_EE) {
@@ -531,9 +532,9 @@ cpu_idle(int busy)
 	if (cpu_idle_hook != NULL) {
 		if (!busy) {
 			critical_enter();
-			cpu_idleclock();
+			sbt = cpu_idleclock();
 		}
-		cpu_idle_hook();
+		cpu_idle_hook(sbt);
 		if (!busy) {
 			cpu_activeclock();
 			critical_exit();
@@ -551,7 +552,7 @@ cpu_idle_wakeup(int cpu)
 }
 
 static void
-cpu_idle_60x(void)
+cpu_idle_60x(sbintime_t sbt)
 {
 	register_t msr;
 	uint16_t vers;
@@ -586,7 +587,7 @@ cpu_idle_60x(void)
 }
 
 static void
-cpu_idle_booke(void)
+cpu_idle_booke(sbintime_t sbt)
 {
 	register_t msr;
 
diff --git a/sys/powerpc/ps3/platform_ps3.c b/sys/powerpc/ps3/platform_ps3.c
index 61ce873..207382d 100644
--- a/sys/powerpc/ps3/platform_ps3.c
+++ b/sys/powerpc/ps3/platform_ps3.c
@@ -70,7 +70,7 @@ static int ps3_smp_start_cpu(platform_t, struct pcpu *cpu);
 static struct cpu_group *ps3_smp_topo(platform_t);
 #endif
 static void ps3_reset(platform_t);
-static void ps3_cpu_idle(void);
+static void ps3_cpu_idle(sbintime_t);
 
 static platform_method_t ps3_methods[] = {
 	PLATFORMMETHOD(platform_probe, 		ps3_probe),
@@ -245,7 +245,7 @@ ps3_real_maxaddr(platform_t plat)
 }
 
 static void
-ps3_cpu_idle(void)
+ps3_cpu_idle(sbintime_t sbt)
 {
 	lv1_pause(0);
 }
diff --git a/sys/powerpc/wii/platform_wii.c b/sys/powerpc/wii/platform_wii.c
index 2bb6022..bc35105 100644
--- a/sys/powerpc/wii/platform_wii.c
+++ b/sys/powerpc/wii/platform_wii.c
@@ -60,7 +60,7 @@ static void		wii_mem_regions(platform_t, struct mem_region **,
 			    int *, struct mem_region **, int *);
 static unsigned long	wii_timebase_freq(platform_t, struct cpuref *cpuref);
 static void		wii_reset(platform_t);
-static void		wii_cpu_idle(void);
+static void		wii_cpu_idle(sbintime_t sbt);
 
 static platform_method_t wii_methods[] = {
 	PLATFORMMETHOD(platform_probe,		wii_probe),
@@ -155,6 +155,6 @@ wii_reset(platform_t plat)
 }
 
 static void
-wii_cpu_idle(void)
+wii_cpu_idle(sbintime_t sbt)
 {
 }
diff --git a/sys/sys/proc.h b/sys/sys/proc.h
index ade2864..46f5820 100644
--- a/sys/sys/proc.h
+++ b/sys/sys/proc.h
@@ -908,7 +908,7 @@ void	tidhash_add(struct thread *);
 void	tidhash_remove(struct thread *);
 void	cpu_idle(int);
 int	cpu_idle_wakeup(int);
-extern	void (*cpu_idle_hook)(void);	/* Hook to machdep CPU idler. */
+extern	void (*cpu_idle_hook)(sbintime_t);	/* Hook to machdep CPU idler. */
 void	cpu_switch(struct thread *, struct thread *, struct mtx *);
 void	cpu_throw(struct thread *, struct thread *) __dead2;
 void	unsleep(struct thread *);
diff --git a/sys/sys/systm.h b/sys/sys/systm.h
index 611d014..2bf8056 100644
--- a/sys/sys/systm.h
+++ b/sys/sys/systm.h
@@ -44,6 +44,7 @@
 #include <sys/cdefs.h>
 #include <sys/queue.h>
 #include <sys/stdint.h>		/* for people using printf mainly */
+#include <sys/time.h>
 
 extern int cold;		/* nonzero if we are doing a cold boot */
 extern int rebooting;		/* kern_reboot() has been called. */
@@ -267,7 +268,7 @@ void	startprofclock(struct proc *);
 void	stopprofclock(struct proc *);
 void	cpu_startprofclock(void);
 void	cpu_stopprofclock(void);
-void	cpu_idleclock(void);
+sbintime_t 	cpu_idleclock(void);
 void	cpu_activeclock(void);
 extern int	cpu_can_deep_sleep;
 extern int	cpu_disable_deep_sleep;