Rename statclock_disable variable to atrtcclock_disable that it actually is,

and hide it inside of atrtc driver. Add new tunable hint.atrtc.0.clock
controlling it. Setting it to 0 disables using RTC clock as stat-/
profclock sources.

Teach i386 and amd64 SMP platforms to emulate stat-/profclocks using i8254
hardclock, when LAPIC and RTC clocks are disabled.

This allows to reduce global interrupt rate of idle system down to about
100 interrupts per core, permitting C3 and deeper C-states provide maximum
CPU power efficiency.

4 files changed, 17 insertions, 24 deletions

diff --git a/sys/i386/bios/apm.c b/sys/i386/bios/apm.c
index 2ac82c5..0d8532f 100644
--- a/sys/i386/bios/apm.c
+++ b/sys/i386/bios/apm.c
@@ -59,6 +59,7 @@ __FBSDID("$FreeBSD$");
 #include <vm/vm_param.h>
 
 #include <i386/bios/apm.h>
+#include <isa/rtc.h>
 
 /* Used by the apm_saver screen saver module */
 int apm_display(int newstate);
@@ -1155,7 +1156,7 @@ apm_attach(device_t dev)
 	cv_init(&sc->cv, "cbb cv");
 
 	if (device_get_flags(dev) & 0x20)
-		statclock_disable = 1;
+		atrtcclock_disable = 1;
 
 	sc->initialized = 0;
 
diff --git a/sys/i386/include/clock.h b/sys/i386/include/clock.h
index a1d4c15..c0c55ed 100644
--- a/sys/i386/include/clock.h
+++ b/sys/i386/include/clock.h
@@ -15,7 +15,6 @@
  * XXX large parts of the driver and its interface are misplaced.
  */
 extern int	clkintr_pending;
-extern int	statclock_disable;
 extern u_int	i8254_freq;
 extern int	i8254_max_count;
 extern uint64_t	tsc_freq;
diff --git a/sys/i386/isa/clock.c b/sys/i386/isa/clock.c
index 989a622..c97e9c8 100644
--- a/sys/i386/isa/clock.c
+++ b/sys/i386/isa/clock.c
@@ -91,7 +91,6 @@ __FBSDID("$FreeBSD$");
 int	clkintr_pending;
 static int pscnt = 1;
 static int psdiv = 1;
-int	statclock_disable;
 #ifndef TIMER_FREQ
 #define TIMER_FREQ   1193182
 #endif
@@ -106,6 +105,7 @@ static	u_int32_t i8254_lastcount;
 static	u_int32_t i8254_offset;
 static	int	(*i8254_pending)(struct intsrc *);
 static	int	i8254_ticked;
+static	int	using_atrtc_timer;
 static	int	using_lapic_timer;
 
 /* Values for timerX_state: */
@@ -137,6 +137,8 @@ hardclockintr(struct trapframe *frame)
 		hardclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
 	else
 		hardclock_cpu(TRAPF_USERMODE(frame));
+	if (!using_atrtc_timer)
+		statclockintr(frame);
 	return (FILTER_HANDLED);
 }
 
@@ -190,10 +192,7 @@ clkintr(struct trapframe *frame)
 	if (smp_started)
 		ipi_all_but_self(IPI_HARDCLOCK);
 #endif 
-	if (PCPU_GET(cpuid) == 0)
-		hardclock(TRAPF_USERMODE(frame), TRAPF_PC(frame));
-	else
-		hardclock_cpu(TRAPF_USERMODE(frame));
+	hardclockintr(frame);
 #ifdef DEV_MCA
 	/* Reset clock interrupt by asserting bit 7 of port 0x61 */
 	if (MCA_system)
@@ -508,7 +507,6 @@ startrtclock()
 void
 cpu_initclocks()
 {
-	int diag;
 
 #ifdef DEV_APIC
 	using_lapic_timer = lapic_setup_clock();
@@ -542,21 +540,17 @@ cpu_initclocks()
 	 * kernel clocks, then setup the RTC to periodically interrupt to
 	 * drive statclock() and profclock().
 	 */
-	if (!statclock_disable && !using_lapic_timer) {
-		diag = rtcin(RTC_DIAG);
-		if (diag != 0)
-			printf("RTC BIOS diagnostic error %b\n",
-			    diag, RTCDG_BITS);
-
-	        /* Setting stathz to nonzero early helps avoid races. */
-		stathz = RTC_NOPROFRATE;
-		profhz = RTC_PROFRATE;
-
-		/* Enable periodic interrupts from the RTC. */
-		intr_add_handler("rtc", 8,
-		    (driver_filter_t *)rtcintr, NULL, NULL,
-		    INTR_TYPE_CLK, NULL);
-		atrtc_enable_intr();
+	if (!using_lapic_timer) {
+		using_atrtc_timer = atrtc_setup_clock();
+		if (using_atrtc_timer) {
+			/* Enable periodic interrupts from the RTC. */
+			intr_add_handler("rtc", 8,
+			    (driver_filter_t *)rtcintr, NULL, NULL,
+			    INTR_TYPE_CLK, NULL);
+			atrtc_enable_intr();
+		} else {
+			profhz = stathz = hz;
+		}
 	}
 
 	init_TSC_tc();
diff --git a/sys/i386/xen/clock.c b/sys/i386/xen/clock.c
index ce1c615..383eff4 100644
--- a/sys/i386/xen/clock.c
+++ b/sys/i386/xen/clock.c
@@ -120,7 +120,6 @@ int adjkerntz;		/* local offset from GMT in seconds */
 int clkintr_pending;
 int pscnt = 1;
 int psdiv = 1;
-int statclock_disable;
 int wall_cmos_clock;
 u_int timer_freq = TIMER_FREQ;
 static int independent_wallclock;