vserver 1.9.5.x5
[linux-2.6.git] / arch / x86_64 / kernel / time.c
index ee94184..b0df27f 100644 (file)
@@ -34,6 +34,7 @@
 #include <asm/hpet.h>
 #include <asm/sections.h>
 #include <linux/cpufreq.h>
+#include <linux/hpet.h>
 #ifdef CONFIG_X86_LOCAL_APIC
 #include <asm/apic.h>
 #endif
@@ -48,14 +49,14 @@ static void cpufreq_delayed_get(void);
 
 extern int using_apic_timer;
 
-spinlock_t rtc_lock = SPIN_LOCK_UNLOCKED;
-spinlock_t i8253_lock = SPIN_LOCK_UNLOCKED;
+DEFINE_SPINLOCK(rtc_lock);
+DEFINE_SPINLOCK(i8253_lock);
 
 static int nohpet __initdata = 0;
+static int notsc __initdata = 0;
 
 #undef HPET_HACK_ENABLE_DANGEROUS
 
-
 unsigned int cpu_khz;                                  /* TSC clocks / usec, not used here */
 unsigned long hpet_period;                             /* fsecs / HPET clock */
 unsigned long hpet_tick;                               /* HPET clocks / interrupt */
@@ -330,7 +331,7 @@ static noinline void handle_lost_ticks(int lost, struct pt_regs *regs)
            print_symbol("rip %s)\n", regs->rip);
     }
 
-    if (lost_count == 100 && !warned) {
+    if (lost_count == 1000 && !warned) {
            printk(KERN_WARNING
                   "warning: many lost ticks.\n"
                   KERN_WARNING "Your time source seems to be instable or "
@@ -427,6 +428,9 @@ static irqreturn_t timer_interrupt(int irq, void *dev_id, struct pt_regs *regs)
  */
 
        do_timer(regs);
+#ifndef CONFIG_SMP
+       update_process_times(user_mode(regs));
+#endif
 
 /*
  * In the SMP case we use the local APIC timer interrupt to do the profiling,
@@ -723,31 +727,59 @@ static unsigned int __init pit_calibrate_tsc(void)
        return (end - start) / 50;
 }
 
-static int hpet_init(void)
+#ifdef CONFIG_HPET
+static __init int late_hpet_init(void)
 {
-       unsigned int cfg, id;
+       struct hpet_data        hd;
+       unsigned int            ntimer;
 
        if (!vxtime.hpet_address)
-               return -1;
-       set_fixmap_nocache(FIX_HPET_BASE, vxtime.hpet_address);
-       __set_fixmap(VSYSCALL_HPET, vxtime.hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
+          return -1;
 
-/*
- * Read the period, compute tick and quotient.
- */
+       memset(&hd, 0, sizeof (hd));
 
-       id = hpet_readl(HPET_ID);
+       ntimer = hpet_readl(HPET_ID);
+       ntimer = (ntimer & HPET_ID_NUMBER) >> HPET_ID_NUMBER_SHIFT;
+       ntimer++;
 
-       if (!(id & HPET_ID_VENDOR) || !(id & HPET_ID_NUMBER) ||
-           !(id & HPET_ID_LEGSUP))
-               return -1;
+       /*
+        * Register with driver.
+        * Timer0 and Timer1 is used by platform.
+        */
+       hd.hd_phys_address = vxtime.hpet_address;
+       hd.hd_address = (void *)fix_to_virt(FIX_HPET_BASE);
+       hd.hd_nirqs = ntimer;
+       hd.hd_flags = HPET_DATA_PLATFORM;
+       hpet_reserve_timer(&hd, 0);
+#ifdef CONFIG_HPET_EMULATE_RTC
+       hpet_reserve_timer(&hd, 1);
+#endif
+       hd.hd_irq[0] = HPET_LEGACY_8254;
+       hd.hd_irq[1] = HPET_LEGACY_RTC;
+       if (ntimer > 2) {
+               struct hpet             *hpet;
+               struct hpet_timer       *timer;
+               int                     i;
 
-       hpet_period = hpet_readl(HPET_PERIOD);
-       if (hpet_period < 100000 || hpet_period > 100000000)
-               return -1;
+               hpet = (struct hpet *) fix_to_virt(FIX_HPET_BASE);
 
-       hpet_tick = (1000000000L * (USEC_PER_SEC / HZ) + hpet_period / 2) /
-               hpet_period;
+               for (i = 2, timer = &hpet->hpet_timers[2]; i < ntimer;
+                    timer++, i++)
+                       hd.hd_irq[i] = (timer->hpet_config &
+                                       Tn_INT_ROUTE_CNF_MASK) >>
+                               Tn_INT_ROUTE_CNF_SHIFT;
+
+       }
+
+       hpet_alloc(&hd);
+       return 0;
+}
+fs_initcall(late_hpet_init);
+#endif
+
+static int hpet_timer_stop_set_go(unsigned long tick)
+{
+       unsigned int cfg;
 
 /*
  * Stop the timers and reset the main counter.
@@ -779,6 +811,40 @@ static int hpet_init(void)
        return 0;
 }
 
+static int hpet_init(void)
+{
+       unsigned int id;
+
+       if (!vxtime.hpet_address)
+               return -1;
+       set_fixmap_nocache(FIX_HPET_BASE, vxtime.hpet_address);
+       __set_fixmap(VSYSCALL_HPET, vxtime.hpet_address, PAGE_KERNEL_VSYSCALL_NOCACHE);
+
+/*
+ * Read the period, compute tick and quotient.
+ */
+
+       id = hpet_readl(HPET_ID);
+
+       if (!(id & HPET_ID_VENDOR) || !(id & HPET_ID_NUMBER) ||
+           !(id & HPET_ID_LEGSUP))
+               return -1;
+
+       hpet_period = hpet_readl(HPET_PERIOD);
+       if (hpet_period < 100000 || hpet_period > 100000000)
+               return -1;
+
+       hpet_tick = (1000000000L * (USEC_PER_SEC / HZ) + hpet_period / 2) /
+               hpet_period;
+
+       return hpet_timer_stop_set_go(hpet_tick);
+}
+
+static int hpet_reenable(void)
+{
+       return hpet_timer_stop_set_go(hpet_tick);
+}
+
 void __init pit_init(void)
 {
        unsigned long flags;
@@ -856,61 +922,93 @@ void __init time_init_smp(void)
 {
        char *timetype;
 
-       if (vxtime.hpet_address) {
+       /*
+        * AMD systems with more than one CPU don't have fully synchronized
+        * TSCs. Always use HPET gettimeofday for these, although it is slower.
+        * Intel SMP systems usually have synchronized TSCs, so use always
+        * the TSC.
+        *
+        * Exceptions:
+        * IBM Summit2 checked by oem_force_hpet_timer().
+        * AMD dual core may also not need HPET. Check me.
+        *
+        * Can be turned off with "notsc".
+        */
+       if (num_online_cpus() > 1 &&
+           boot_cpu_data.x86_vendor == X86_VENDOR_AMD)
+               notsc = 1;
+       /* Some systems will want to disable TSC and use HPET. */
+       if (oem_force_hpet_timer())
+               notsc = 1;
+       if (vxtime.hpet_address && notsc) {
                timetype = "HPET";
                vxtime.last = hpet_readl(HPET_T0_CMP) - hpet_tick;
                vxtime.mode = VXTIME_HPET;
                do_gettimeoffset = do_gettimeoffset_hpet;
        } else {
-               timetype = "PIT/TSC";
+               timetype = vxtime.hpet_address ? "HPET/TSC" : "PIT/TSC";
                vxtime.mode = VXTIME_TSC;
        }
+
        printk(KERN_INFO "time.c: Using %s based timekeeping.\n", timetype);
 }
 
 __setup("report_lost_ticks", time_setup);
 
 static long clock_cmos_diff;
+static unsigned long sleep_start;
 
-static int time_suspend(struct sys_device *dev, u32 state)
+static int timer_suspend(struct sys_device *dev, u32 state)
 {
        /*
         * Estimate time zone so that set_time can update the clock
         */
-       clock_cmos_diff = -get_cmos_time();
+       long cmos_time =  get_cmos_time();
+
+       clock_cmos_diff = -cmos_time;
        clock_cmos_diff += get_seconds();
+       sleep_start = cmos_time;
        return 0;
 }
 
-static int time_resume(struct sys_device *dev)
+static int timer_resume(struct sys_device *dev)
 {
        unsigned long flags;
-       unsigned long sec = get_cmos_time() + clock_cmos_diff;
+       unsigned long sec;
+       unsigned long ctime = get_cmos_time();
+       unsigned long sleep_length = (ctime - sleep_start) * HZ;
+
+       if (vxtime.hpet_address)
+               hpet_reenable();
+
+       sec = ctime + clock_cmos_diff;
        write_seqlock_irqsave(&xtime_lock,flags);
        xtime.tv_sec = sec;
        xtime.tv_nsec = 0;
        write_sequnlock_irqrestore(&xtime_lock,flags);
+       jiffies += sleep_length;
+       wall_jiffies += sleep_length;
        return 0;
 }
 
-static struct sysdev_class pit_sysclass = {
-       .resume = time_resume,
-       .suspend = time_suspend,
-       set_kset_name("pit"),
+static struct sysdev_class timer_sysclass = {
+       .resume = timer_resume,
+       .suspend = timer_suspend,
+       set_kset_name("timer"),
 };
 
 
 /* XXX this driverfs stuff should probably go elsewhere later -john */
-static struct sys_device device_i8253 = {
+static struct sys_device device_timer = {
        .id     = 0,
-       .cls    = &pit_sysclass,
+       .cls    = &timer_sysclass,
 };
 
 static int time_init_device(void)
 {
-       int error = sysdev_class_register(&pit_sysclass);
+       int error = sysdev_class_register(&timer_sysclass);
        if (!error)
-               error = sysdev_register(&device_i8253);
+               error = sysdev_register(&device_timer);
        return error;
 }
 
@@ -931,7 +1029,6 @@ device_initcall(time_init_device);
  * For (3), we use interrupts at 64Hz or user specified periodic
  * frequency, whichever is higher.
  */
-#include <linux/mc146818rtc.h>
 #include <linux/rtc.h>
 
 extern irqreturn_t rtc_interrupt(int irq, void *dev_id, struct pt_regs *regs);
@@ -1151,3 +1248,14 @@ static int __init nohpet_setup(char *s)
 } 
 
 __setup("nohpet", nohpet_setup);
+
+
+static int __init notsc_setup(char *s)
+{
+       notsc = 1;
+       return 0;
+}
+
+__setup("notsc", notsc_setup);
+
+