Merge branch 'master' into upstream
[pandora-kernel.git] / arch / i386 / kernel / apic.c
index 8c844d0..9655c23 100644 (file)
@@ -25,6 +25,8 @@
 #include <linux/kernel_stat.h>
 #include <linux/sysdev.h>
 #include <linux/cpu.h>
+#include <linux/clockchips.h>
+#include <linux/acpi_pmtmr.h>
 #include <linux/module.h>
 
 #include <asm/atomic.h>
@@ -36,6 +38,7 @@
 #include <asm/hpet.h>
 #include <asm/i8253.h>
 #include <asm/nmi.h>
+#include <asm/idle.h>
 
 #include <mach_apic.h>
 #include <mach_apicdef.h>
 #include "io_ports.h"
 
 /*
- * cpu_mask that denotes the CPUs that needs timer interrupt coming in as
- * IPIs in place of local APIC timers
+ * Sanity check
  */
-static cpumask_t timer_bcast_ipi;
+#if (SPURIOUS_APIC_VECTOR & 0x0F) != 0x0F
+# error SPURIOUS_APIC_VECTOR definition error
+#endif
 
 /*
  * Knob to control our willingness to enable the local APIC.
+ *
+ * -1=force-disable, +1=force-enable
  */
-int enable_local_apic __initdata = 0; /* -1=force-disable, +1=force-enable */
+static int enable_local_apic __initdata = 0;
+
+/* Local APIC timer verification ok */
+static int local_apic_timer_verify_ok;
 
 /*
- * Debug level
+ * Debug level, exported for io_apic.c
  */
 int apic_verbosity;
 
+static unsigned int calibration_result;
 
+static int lapic_next_event(unsigned long delta,
+                           struct clock_event_device *evt);
+static void lapic_timer_setup(enum clock_event_mode mode,
+                             struct clock_event_device *evt);
+static void lapic_timer_broadcast(cpumask_t mask);
 static void apic_pm_activate(void);
 
+/*
+ * The local apic timer can be used for any function which is CPU local.
+ */
+static struct clock_event_device lapic_clockevent = {
+       .name           = "lapic",
+       .features       = CLOCK_EVT_FEAT_PERIODIC | CLOCK_EVT_FEAT_ONESHOT
+                       | CLOCK_EVT_FEAT_C3STOP | CLOCK_EVT_FEAT_DUMMY,
+       .shift          = 32,
+       .set_mode       = lapic_timer_setup,
+       .set_next_event = lapic_next_event,
+       .broadcast      = lapic_timer_broadcast,
+       .rating         = 100,
+       .irq            = -1,
+};
+static DEFINE_PER_CPU(struct clock_event_device, lapic_events);
+
+/* Local APIC was disabled by the BIOS and enabled by the kernel */
+static int enabled_via_apicbase;
+
+/*
+ * Get the LAPIC version
+ */
+static inline int lapic_get_version(void)
+{
+       return GET_APIC_VERSION(apic_read(APIC_LVR));
+}
+
+/*
+ * Check, if the APIC is integrated or a seperate chip
+ */
+static inline int lapic_is_integrated(void)
+{
+       return APIC_INTEGRATED(lapic_get_version());
+}
+
+/*
+ * Check, whether this is a modern or a first generation APIC
+ */
 static int modern_apic(void)
 {
-       unsigned int lvr, version;
        /* AMD systems use old APIC versions, so check the CPU */
        if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
-               boot_cpu_data.x86 >= 0xf)
+           boot_cpu_data.x86 >= 0xf)
                return 1;
-       lvr = apic_read(APIC_LVR);
-       version = GET_APIC_VERSION(lvr);
-       return version >= 0x14;
+       return lapic_get_version() >= 0x14;
+}
+
+/**
+ * enable_NMI_through_LVT0 - enable NMI through local vector table 0
+ */
+void enable_NMI_through_LVT0 (void * dummy)
+{
+       unsigned int v = APIC_DM_NMI;
+
+       /* Level triggered for 82489DX */
+       if (!lapic_is_integrated())
+               v |= APIC_LVT_LEVEL_TRIGGER;
+       apic_write_around(APIC_LVT0, v);
+}
+
+/**
+ * get_physical_broadcast - Get number of physical broadcast IDs
+ */
+int get_physical_broadcast(void)
+{
+       return modern_apic() ? 0xff : 0xf;
+}
+
+/**
+ * lapic_get_maxlvt - get the maximum number of local vector table entries
+ */
+int lapic_get_maxlvt(void)
+{
+       unsigned int v = apic_read(APIC_LVR);
+
+       /* 82489DXs do not report # of LVT entries. */
+       return APIC_INTEGRATED(GET_APIC_VERSION(v)) ? GET_APIC_MAXLVT(v) : 2;
 }
 
 /*
- * 'what should we do if we get a hw irq event on an illegal vector'.
- * each architecture has to answer this themselves.
+ * Local APIC timer
+ */
+
+/* Clock divisor is set to 16 */
+#define APIC_DIVISOR 16
+
+/*
+ * This function sets up the local APIC timer, with a timeout of
+ * 'clocks' APIC bus clock. During calibration we actually call
+ * this function twice on the boot CPU, once with a bogus timeout
+ * value, second time for real. The other (noncalibrating) CPUs
+ * call this function only once, with the real, calibrated value.
+ *
+ * We do reads before writes even if unnecessary, to get around the
+ * P5 APIC double write bug.
  */
-void ack_bad_irq(unsigned int irq)
+static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
 {
-       printk("unexpected IRQ trap at vector %02x\n", irq);
+       unsigned int lvtt_value, tmp_value;
+
+       lvtt_value = LOCAL_TIMER_VECTOR;
+       if (!oneshot)
+               lvtt_value |= APIC_LVT_TIMER_PERIODIC;
+       if (!lapic_is_integrated())
+               lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
+
+       if (!irqen)
+               lvtt_value |= APIC_LVT_MASKED;
+
+       apic_write_around(APIC_LVTT, lvtt_value);
+
        /*
-        * Currently unexpected vectors happen only on SMP and APIC.
-        * We _must_ ack these because every local APIC has only N
-        * irq slots per priority level, and a 'hanging, unacked' IRQ
-        * holds up an irq slot - in excessive cases (when multiple
-        * unexpected vectors occur) that might lock up the APIC
-        * completely.
-        * But only ack when the APIC is enabled -AK
+        * Divide PICLK by 16
         */
-       if (cpu_has_apic)
-               ack_APIC_irq();
+       tmp_value = apic_read(APIC_TDCR);
+       apic_write_around(APIC_TDCR, (tmp_value
+                               & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE))
+                               | APIC_TDR_DIV_16);
+
+       if (!oneshot)
+               apic_write_around(APIC_TMICT, clocks/APIC_DIVISOR);
 }
 
-void __init apic_intr_init(void)
+/*
+ * Program the next event, relative to now
+ */
+static int lapic_next_event(unsigned long delta,
+                           struct clock_event_device *evt)
+{
+       apic_write_around(APIC_TMICT, delta);
+       return 0;
+}
+
+/*
+ * Setup the lapic timer in periodic or oneshot mode
+ */
+static void lapic_timer_setup(enum clock_event_mode mode,
+                             struct clock_event_device *evt)
+{
+       unsigned long flags;
+       unsigned int v;
+
+       /* Lapic used for broadcast ? */
+       if (!local_apic_timer_verify_ok)
+               return;
+
+       local_irq_save(flags);
+
+       switch (mode) {
+       case CLOCK_EVT_MODE_PERIODIC:
+       case CLOCK_EVT_MODE_ONESHOT:
+               __setup_APIC_LVTT(calibration_result,
+                                 mode != CLOCK_EVT_MODE_PERIODIC, 1);
+               break;
+       case CLOCK_EVT_MODE_UNUSED:
+       case CLOCK_EVT_MODE_SHUTDOWN:
+               v = apic_read(APIC_LVTT);
+               v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
+               apic_write_around(APIC_LVTT, v);
+               break;
+       }
+
+       local_irq_restore(flags);
+}
+
+/*
+ * Local APIC timer broadcast function
+ */
+static void lapic_timer_broadcast(cpumask_t mask)
 {
 #ifdef CONFIG_SMP
-       smp_intr_init();
+       send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
 #endif
-       /* self generated IPI for local APIC timer */
-       set_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
+}
 
-       /* IPI vectors for APIC spurious and error interrupts */
-       set_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
-       set_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
+/*
+ * Setup the local APIC timer for this CPU. Copy the initilized values
+ * of the boot CPU and register the clock event in the framework.
+ */
+static void __devinit setup_APIC_timer(void)
+{
+       struct clock_event_device *levt = &__get_cpu_var(lapic_events);
 
-       /* thermal monitor LVT interrupt */
-#ifdef CONFIG_X86_MCE_P4THERMAL
-       set_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
-#endif
+       memcpy(levt, &lapic_clockevent, sizeof(*levt));
+       levt->cpumask = cpumask_of_cpu(smp_processor_id());
+
+       clockevents_register_device(levt);
 }
 
-/* Using APIC to generate smp_local_timer_interrupt? */
-int using_apic_timer __read_mostly = 0;
+/*
+ * In this functions we calibrate APIC bus clocks to the external timer.
+ *
+ * We want to do the calibration only once since we want to have local timer
+ * irqs syncron. CPUs connected by the same APIC bus have the very same bus
+ * frequency.
+ *
+ * This was previously done by reading the PIT/HPET and waiting for a wrap
+ * around to find out, that a tick has elapsed. I have a box, where the PIT
+ * readout is broken, so it never gets out of the wait loop again. This was
+ * also reported by others.
+ *
+ * Monitoring the jiffies value is inaccurate and the clockevents
+ * infrastructure allows us to do a simple substitution of the interrupt
+ * handler.
+ *
+ * The calibration routine also uses the pm_timer when possible, as the PIT
+ * happens to run way too slow (factor 2.3 on my VAIO CoreDuo, which goes
+ * back to normal later in the boot process).
+ */
+
+#define LAPIC_CAL_LOOPS                (HZ/10)
 
-static int enabled_via_apicbase;
+static __initdata volatile int lapic_cal_loops = -1;
+static __initdata long lapic_cal_t1, lapic_cal_t2;
+static __initdata unsigned long long lapic_cal_tsc1, lapic_cal_tsc2;
+static __initdata unsigned long lapic_cal_pm1, lapic_cal_pm2;
+static __initdata unsigned long lapic_cal_j1, lapic_cal_j2;
 
-void enable_NMI_through_LVT0 (void * dummy)
+/*
+ * Temporary interrupt handler.
+ */
+static void __init lapic_cal_handler(struct clock_event_device *dev)
 {
-       unsigned int v, ver;
+       unsigned long long tsc = 0;
+       long tapic = apic_read(APIC_TMCCT);
+       unsigned long pm = acpi_pm_read_early();
 
-       ver = apic_read(APIC_LVR);
-       ver = GET_APIC_VERSION(ver);
-       v = APIC_DM_NMI;                        /* unmask and set to NMI */
-       if (!APIC_INTEGRATED(ver))              /* 82489DX */
-               v |= APIC_LVT_LEVEL_TRIGGER;
-       apic_write_around(APIC_LVT0, v);
+       if (cpu_has_tsc)
+               rdtscll(tsc);
+
+       switch (lapic_cal_loops++) {
+       case 0:
+               lapic_cal_t1 = tapic;
+               lapic_cal_tsc1 = tsc;
+               lapic_cal_pm1 = pm;
+               lapic_cal_j1 = jiffies;
+               break;
+
+       case LAPIC_CAL_LOOPS:
+               lapic_cal_t2 = tapic;
+               lapic_cal_tsc2 = tsc;
+               if (pm < lapic_cal_pm1)
+                       pm += ACPI_PM_OVRRUN;
+               lapic_cal_pm2 = pm;
+               lapic_cal_j2 = jiffies;
+               break;
+       }
 }
 
-int get_physical_broadcast(void)
+/*
+ * Setup the boot APIC
+ *
+ * Calibrate and verify the result.
+ */
+void __init setup_boot_APIC_clock(void)
 {
-       if (modern_apic())
-               return 0xff;
-       else
-               return 0xf;
+       struct clock_event_device *levt = &__get_cpu_var(lapic_events);
+       const long pm_100ms = PMTMR_TICKS_PER_SEC/10;
+       const long pm_thresh = pm_100ms/100;
+       void (*real_handler)(struct clock_event_device *dev);
+       unsigned long deltaj;
+       long delta, deltapm;
+
+       apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n"
+                   "calibrating APIC timer ...\n");
+
+       local_irq_disable();
+
+       /* Replace the global interrupt handler */
+       real_handler = global_clock_event->event_handler;
+       global_clock_event->event_handler = lapic_cal_handler;
+
+       /*
+        * Setup the APIC counter to 1e9. There is no way the lapic
+        * can underflow in the 100ms detection time frame
+        */
+       __setup_APIC_LVTT(1000000000, 0, 0);
+
+       /* Let the interrupts run */
+       local_irq_enable();
+
+       while(lapic_cal_loops <= LAPIC_CAL_LOOPS);
+
+       local_irq_disable();
+
+       /* Restore the real event handler */
+       global_clock_event->event_handler = real_handler;
+
+       /* Build delta t1-t2 as apic timer counts down */
+       delta = lapic_cal_t1 - lapic_cal_t2;
+       apic_printk(APIC_VERBOSE, "... lapic delta = %ld\n", delta);
+
+       /* Check, if the PM timer is available */
+       deltapm = lapic_cal_pm2 - lapic_cal_pm1;
+       apic_printk(APIC_VERBOSE, "... PM timer delta = %ld\n", deltapm);
+
+       if (deltapm) {
+               unsigned long mult;
+               u64 res;
+
+               mult = clocksource_hz2mult(PMTMR_TICKS_PER_SEC, 22);
+
+               if (deltapm > (pm_100ms - pm_thresh) &&
+                   deltapm < (pm_100ms + pm_thresh)) {
+                       apic_printk(APIC_VERBOSE, "... PM timer result ok\n");
+               } else {
+                       res = (((u64) deltapm) *  mult) >> 22;
+                       do_div(res, 1000000);
+                       printk(KERN_WARNING "APIC calibration not consistent "
+                              "with PM Timer: %ldms instead of 100ms\n",
+                              (long)res);
+                       /* Correct the lapic counter value */
+                       res = (((u64) delta ) * pm_100ms);
+                       do_div(res, deltapm);
+                       printk(KERN_INFO "APIC delta adjusted to PM-Timer: "
+                              "%lu (%ld)\n", (unsigned long) res, delta);
+                       delta = (long) res;
+               }
+       }
+
+       /* Calculate the scaled math multiplication factor */
+       lapic_clockevent.mult = div_sc(delta, TICK_NSEC * LAPIC_CAL_LOOPS, 32);
+       lapic_clockevent.max_delta_ns =
+               clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
+       lapic_clockevent.min_delta_ns =
+               clockevent_delta2ns(0xF, &lapic_clockevent);
+
+       calibration_result = (delta * APIC_DIVISOR) / LAPIC_CAL_LOOPS;
+
+       apic_printk(APIC_VERBOSE, "..... delta %ld\n", delta);
+       apic_printk(APIC_VERBOSE, "..... mult: %ld\n", lapic_clockevent.mult);
+       apic_printk(APIC_VERBOSE, "..... calibration result: %u\n",
+                   calibration_result);
+
+       if (cpu_has_tsc) {
+               delta = (long)(lapic_cal_tsc2 - lapic_cal_tsc1);
+               apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
+                           "%ld.%04ld MHz.\n",
+                           (delta / LAPIC_CAL_LOOPS) / (1000000 / HZ),
+                           (delta / LAPIC_CAL_LOOPS) % (1000000 / HZ));
+       }
+
+       apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
+                   "%u.%04u MHz.\n",
+                   calibration_result / (1000000 / HZ),
+                   calibration_result % (1000000 / HZ));
+
+
+       apic_printk(APIC_VERBOSE, "... verify APIC timer\n");
+
+       /*
+        * Setup the apic timer manually
+        */
+       local_apic_timer_verify_ok = 1;
+       levt->event_handler = lapic_cal_handler;
+       lapic_timer_setup(CLOCK_EVT_MODE_PERIODIC, levt);
+       lapic_cal_loops = -1;
+
+       /* Let the interrupts run */
+       local_irq_enable();
+
+       while(lapic_cal_loops <= LAPIC_CAL_LOOPS);
+
+       local_irq_disable();
+
+       /* Stop the lapic timer */
+       lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, levt);
+
+       local_irq_enable();
+
+       /* Jiffies delta */
+       deltaj = lapic_cal_j2 - lapic_cal_j1;
+       apic_printk(APIC_VERBOSE, "... jiffies delta = %lu\n", deltaj);
+
+       /* Check, if the PM timer is available */
+       deltapm = lapic_cal_pm2 - lapic_cal_pm1;
+       apic_printk(APIC_VERBOSE, "... PM timer delta = %ld\n", deltapm);
+
+       local_apic_timer_verify_ok = 0;
+
+       if (deltapm) {
+               if (deltapm > (pm_100ms - pm_thresh) &&
+                   deltapm < (pm_100ms + pm_thresh)) {
+                       apic_printk(APIC_VERBOSE, "... PM timer result ok\n");
+                       /* Check, if the jiffies result is consistent */
+                       if (deltaj < LAPIC_CAL_LOOPS-2 ||
+                           deltaj > LAPIC_CAL_LOOPS+2) {
+                               /*
+                                * Not sure, what we can do about this one.
+                                * When high resultion timers are active
+                                * and the lapic timer does not stop in C3
+                                * we are fine. Otherwise more trouble might
+                                * be waiting. -- tglx
+                                */
+                               printk(KERN_WARNING "Global event device %s "
+                                      "has wrong frequency "
+                                      "(%lu ticks instead of %d)\n",
+                                      global_clock_event->name, deltaj,
+                                      LAPIC_CAL_LOOPS);
+                       }
+                       local_apic_timer_verify_ok = 1;
+               }
+       } else {
+               /* Check, if the jiffies result is consistent */
+               if (deltaj >= LAPIC_CAL_LOOPS-2 &&
+                   deltaj <= LAPIC_CAL_LOOPS+2) {
+                       apic_printk(APIC_VERBOSE, "... jiffies result ok\n");
+                       local_apic_timer_verify_ok = 1;
+               }
+       }
+
+       if (!local_apic_timer_verify_ok) {
+               printk(KERN_WARNING
+                      "APIC timer disabled due to verification failure.\n");
+               /* No broadcast on UP ! */
+               if (num_possible_cpus() == 1)
+                       return;
+       } else
+               lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
+
+       /* Setup the lapic or request the broadcast */
+       setup_APIC_timer();
 }
 
-int get_maxlvt(void)
+void __devinit setup_secondary_APIC_clock(void)
 {
-       unsigned int v, ver, maxlvt;
+       setup_APIC_timer();
+}
 
-       v = apic_read(APIC_LVR);
-       ver = GET_APIC_VERSION(v);
-       /* 82489DXs do not report # of LVT entries. */
-       maxlvt = APIC_INTEGRATED(ver) ? GET_APIC_MAXLVT(v) : 2;
-       return maxlvt;
+/*
+ * The guts of the apic timer interrupt
+ */
+static void local_apic_timer_interrupt(void)
+{
+       int cpu = smp_processor_id();
+       struct clock_event_device *evt = &per_cpu(lapic_events, cpu);
+
+       /*
+        * Normally we should not be here till LAPIC has been initialized but
+        * in some cases like kdump, its possible that there is a pending LAPIC
+        * timer interrupt from previous kernel's context and is delivered in
+        * new kernel the moment interrupts are enabled.
+        *
+        * Interrupts are enabled early and LAPIC is setup much later, hence
+        * its possible that when we get here evt->event_handler is NULL.
+        * Check for event_handler being NULL and discard the interrupt as
+        * spurious.
+        */
+       if (!evt->event_handler) {
+               printk(KERN_WARNING
+                      "Spurious LAPIC timer interrupt on cpu %d\n", cpu);
+               /* Switch it off */
+               lapic_timer_setup(CLOCK_EVT_MODE_SHUTDOWN, evt);
+               return;
+       }
+
+       per_cpu(irq_stat, cpu).apic_timer_irqs++;
+
+       evt->event_handler(evt);
+}
+
+/*
+ * Local APIC timer interrupt. This is the most natural way for doing
+ * local interrupts, but local timer interrupts can be emulated by
+ * broadcast interrupts too. [in case the hw doesn't support APIC timers]
+ *
+ * [ if a single-CPU system runs an SMP kernel then we call the local
+ *   interrupt as well. Thus we cannot inline the local irq ... ]
+ */
+
+void fastcall smp_apic_timer_interrupt(struct pt_regs *regs)
+{
+       struct pt_regs *old_regs = set_irq_regs(regs);
+
+       /*
+        * NOTE! We'd better ACK the irq immediately,
+        * because timer handling can be slow.
+        */
+       ack_APIC_irq();
+       /*
+        * update_process_times() expects us to have done irq_enter().
+        * Besides, if we don't timer interrupts ignore the global
+        * interrupt lock, which is the WrongThing (tm) to do.
+        */
+       exit_idle();
+       irq_enter();
+       local_apic_timer_interrupt();
+       irq_exit();
+
+       set_irq_regs(old_regs);
+}
+
+int setup_profiling_timer(unsigned int multiplier)
+{
+       return -EINVAL;
 }
 
+/*
+ * Local APIC start and shutdown
+ */
+
+/**
+ * clear_local_APIC - shutdown the local APIC
+ *
+ * This is called, when a CPU is disabled and before rebooting, so the state of
+ * the local APIC has no dangling leftovers. Also used to cleanout any BIOS
+ * leftovers during boot.
+ */
 void clear_local_APIC(void)
 {
-       int maxlvt;
+       int maxlvt = lapic_get_maxlvt();
        unsigned long v;
 
-       maxlvt = get_maxlvt();
-
        /*
         * Masking an LVT entry can trigger a local APIC error
         * if the vector is zero. Mask LVTERR first to prevent this.
@@ -178,7 +613,7 @@ void clear_local_APIC(void)
                apic_write_around(APIC_LVTPC, v | APIC_LVT_MASKED);
        }
 
-/* lets not touch this if we didn't frob it */
+       /* lets not touch this if we didn't frob it */
 #ifdef CONFIG_X86_MCE_P4THERMAL
        if (maxlvt >= 5) {
                v = apic_read(APIC_LVTTHMR);
@@ -200,105 +635,65 @@ void clear_local_APIC(void)
        if (maxlvt >= 5)
                apic_write_around(APIC_LVTTHMR, APIC_LVT_MASKED);
 #endif
-       v = GET_APIC_VERSION(apic_read(APIC_LVR));
-       if (APIC_INTEGRATED(v)) {       /* !82489DX */
-               if (maxlvt > 3)         /* Due to Pentium errata 3AP and 11AP. */
+       /* Integrated APIC (!82489DX) ? */
+       if (lapic_is_integrated()) {
+               if (maxlvt > 3)
+                       /* Clear ESR due to Pentium errata 3AP and 11AP */
                        apic_write(APIC_ESR, 0);
                apic_read(APIC_ESR);
        }
 }
 
-void __init connect_bsp_APIC(void)
+/**
+ * disable_local_APIC - clear and disable the local APIC
+ */
+void disable_local_APIC(void)
 {
-       if (pic_mode) {
-               /*
-                * Do not trust the local APIC being empty at bootup.
-                */
-               clear_local_APIC();
-               /*
-                * PIC mode, enable APIC mode in the IMCR, i.e.
-                * connect BSP's local APIC to INT and NMI lines.
-                */
-               apic_printk(APIC_VERBOSE, "leaving PIC mode, "
-                               "enabling APIC mode.\n");
-               outb(0x70, 0x22);
-               outb(0x01, 0x23);
+       unsigned long value;
+
+       clear_local_APIC();
+
+       /*
+        * Disable APIC (implies clearing of registers
+        * for 82489DX!).
+        */
+       value = apic_read(APIC_SPIV);
+       value &= ~APIC_SPIV_APIC_ENABLED;
+       apic_write_around(APIC_SPIV, value);
+
+       /*
+        * When LAPIC was disabled by the BIOS and enabled by the kernel,
+        * restore the disabled state.
+        */
+       if (enabled_via_apicbase) {
+               unsigned int l, h;
+
+               rdmsr(MSR_IA32_APICBASE, l, h);
+               l &= ~MSR_IA32_APICBASE_ENABLE;
+               wrmsr(MSR_IA32_APICBASE, l, h);
        }
-       enable_apic_mode();
 }
 
-void disconnect_bsp_APIC(int virt_wire_setup)
+/*
+ * If Linux enabled the LAPIC against the BIOS default disable it down before
+ * re-entering the BIOS on shutdown.  Otherwise the BIOS may get confused and
+ * not power-off.  Additionally clear all LVT entries before disable_local_APIC
+ * for the case where Linux didn't enable the LAPIC.
+ */
+void lapic_shutdown(void)
 {
-       if (pic_mode) {
-               /*
-                * Put the board back into PIC mode (has an effect
-                * only on certain older boards).  Note that APIC
-                * interrupts, including IPIs, won't work beyond
-                * this point!  The only exception are INIT IPIs.
-                */
-               apic_printk(APIC_VERBOSE, "disabling APIC mode, "
-                               "entering PIC mode.\n");
-               outb(0x70, 0x22);
-               outb(0x00, 0x23);
-       }
-       else {
-               /* Go back to Virtual Wire compatibility mode */
-               unsigned long value;
-
-               /* For the spurious interrupt use vector F, and enable it */
-               value = apic_read(APIC_SPIV);
-               value &= ~APIC_VECTOR_MASK;
-               value |= APIC_SPIV_APIC_ENABLED;
-               value |= 0xf;
-               apic_write_around(APIC_SPIV, value);
-
-               if (!virt_wire_setup) {
-                       /* For LVT0 make it edge triggered, active high, external and enabled */
-                       value = apic_read(APIC_LVT0);
-                       value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
-                               APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
-                               APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED );
-                       value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
-                       value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
-                       apic_write_around(APIC_LVT0, value);
-               }
-               else {
-                       /* Disable LVT0 */
-                       apic_write_around(APIC_LVT0, APIC_LVT_MASKED);
-               }
-
-               /* For LVT1 make it edge triggered, active high, nmi and enabled */
-               value = apic_read(APIC_LVT1);
-               value &= ~(
-                       APIC_MODE_MASK | APIC_SEND_PENDING |
-                       APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
-                       APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
-               value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
-               value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
-               apic_write_around(APIC_LVT1, value);
-       }
-}
+       unsigned long flags;
 
-void disable_local_APIC(void)
-{
-       unsigned long value;
+       if (!cpu_has_apic)
+               return;
 
+       local_irq_save(flags);
        clear_local_APIC();
 
-       /*
-        * Disable APIC (implies clearing of registers
-        * for 82489DX!).
-        */
-       value = apic_read(APIC_SPIV);
-       value &= ~APIC_SPIV_APIC_ENABLED;
-       apic_write_around(APIC_SPIV, value);
+       if (enabled_via_apicbase)
+               disable_local_APIC();
 
-       if (enabled_via_apicbase) {
-               unsigned int l, h;
-               rdmsr(MSR_IA32_APICBASE, l, h);
-               l &= ~MSR_IA32_APICBASE_ENABLE;
-               wrmsr(MSR_IA32_APICBASE, l, h);
-       }
+       local_irq_restore(flags);
 }
 
 /*
@@ -333,7 +728,7 @@ int __init verify_local_APIC(void)
        reg1 = GET_APIC_VERSION(reg0);
        if (reg1 == 0x00 || reg1 == 0xff)
                return 0;
-       reg1 = get_maxlvt();
+       reg1 = lapic_get_maxlvt();
        if (reg1 < 0x02 || reg1 == 0xff)
                return 0;
 
@@ -356,10 +751,15 @@ int __init verify_local_APIC(void)
        return 1;
 }
 
+/**
+ * sync_Arb_IDs - synchronize APIC bus arbitration IDs
+ */
 void __init sync_Arb_IDs(void)
 {
-       /* Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1
-          And not needed on AMD */
+       /*
+        * Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 And not
+        * needed on AMD.
+        */
        if (modern_apic())
                return;
        /*
@@ -372,14 +772,12 @@ void __init sync_Arb_IDs(void)
                                | APIC_DM_INIT);
 }
 
-extern void __error_in_apic_c (void);
-
 /*
  * An initial setup of the virtual wire mode.
  */
 void __init init_bsp_APIC(void)
 {
-       unsigned long value, ver;
+       unsigned long value;
 
        /*
         * Don't do the setup now if we have a SMP BIOS as the
@@ -388,9 +786,6 @@ void __init init_bsp_APIC(void)
        if (smp_found_config || !cpu_has_apic)
                return;
 
-       value = apic_read(APIC_LVR);
-       ver = GET_APIC_VERSION(value);
-
        /*
         * Do not trust the local APIC being empty at bootup.
         */
@@ -402,9 +797,10 @@ void __init init_bsp_APIC(void)
        value = apic_read(APIC_SPIV);
        value &= ~APIC_VECTOR_MASK;
        value |= APIC_SPIV_APIC_ENABLED;
-       
+
        /* This bit is reserved on P4/Xeon and should be cleared */
-       if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) && (boot_cpu_data.x86 == 15))
+       if ((boot_cpu_data.x86_vendor == X86_VENDOR_INTEL) &&
+           (boot_cpu_data.x86 == 15))
                value &= ~APIC_SPIV_FOCUS_DISABLED;
        else
                value |= APIC_SPIV_FOCUS_DISABLED;
@@ -416,14 +812,17 @@ void __init init_bsp_APIC(void)
         */
        apic_write_around(APIC_LVT0, APIC_DM_EXTINT);
        value = APIC_DM_NMI;
-       if (!APIC_INTEGRATED(ver))              /* 82489DX */
+       if (!lapic_is_integrated())             /* 82489DX */
                value |= APIC_LVT_LEVEL_TRIGGER;
        apic_write_around(APIC_LVT1, value);
 }
 
+/**
+ * setup_local_APIC - setup the local APIC
+ */
 void __devinit setup_local_APIC(void)
 {
-       unsigned long oldvalue, value, ver, maxlvt;
+       unsigned long oldvalue, value, maxlvt, integrated;
        int i, j;
 
        /* Pound the ESR really hard over the head with a big hammer - mbligh */
@@ -434,11 +833,7 @@ void __devinit setup_local_APIC(void)
                apic_write(APIC_ESR, 0);
        }
 
-       value = apic_read(APIC_LVR);
-       ver = GET_APIC_VERSION(value);
-
-       if ((SPURIOUS_APIC_VECTOR & 0x0f) != 0x0f)
-               __error_in_apic_c();
+       integrated = lapic_is_integrated();
 
        /*
         * Double-check whether this APIC is really registered.
@@ -509,13 +904,10 @@ void __devinit setup_local_APIC(void)
         * like LRU than MRU (the short-term load is more even across CPUs).
         * See also the comment in end_level_ioapic_irq().  --macro
         */
-#if 1
+
        /* Enable focus processor (bit==0) */
        value &= ~APIC_SPIV_FOCUS_DISABLED;
-#else
-       /* Disable focus processor (bit==1) */
-       value |= APIC_SPIV_FOCUS_DISABLED;
-#endif
+
        /*
         * Set spurious IRQ vector
         */
@@ -551,17 +943,18 @@ void __devinit setup_local_APIC(void)
                value = APIC_DM_NMI;
        else
                value = APIC_DM_NMI | APIC_LVT_MASKED;
-       if (!APIC_INTEGRATED(ver))              /* 82489DX */
+       if (!integrated)                /* 82489DX */
                value |= APIC_LVT_LEVEL_TRIGGER;
        apic_write_around(APIC_LVT1, value);
 
-       if (APIC_INTEGRATED(ver) && !esr_disable) {             /* !82489DX */
-               maxlvt = get_maxlvt();
+       if (integrated && !esr_disable) {               /* !82489DX */
+               maxlvt = lapic_get_maxlvt();
                if (maxlvt > 3)         /* Due to the Pentium erratum 3AP. */
                        apic_write(APIC_ESR, 0);
                oldvalue = apic_read(APIC_ESR);
 
-               value = ERROR_APIC_VECTOR;      // enables sending errors
+               /* enables sending errors */
+               value = ERROR_APIC_VECTOR;
                apic_write_around(APIC_LVTERR, value);
                /*
                 * spec says clear errors after enabling vector.
@@ -574,194 +967,30 @@ void __devinit setup_local_APIC(void)
                                "vector: 0x%08lx  after: 0x%08lx\n",
                                oldvalue, value);
        } else {
-               if (esr_disable)        
-                       /* 
-                        * Something untraceble is creating bad interrupts on 
+               if (esr_disable)
+                       /*
+                        * Something untraceble is creating bad interrupts on
                         * secondary quads ... for the moment, just leave the
                         * ESR disabled - we can't do anything useful with the
                         * errors anyway - mbligh
                         */
-                       printk("Leaving ESR disabled.\n");
-               else 
-                       printk("No ESR for 82489DX.\n");
+                       printk(KERN_INFO "Leaving ESR disabled.\n");
+               else
+                       printk(KERN_INFO "No ESR for 82489DX.\n");
        }
 
-       if (nmi_watchdog == NMI_LOCAL_APIC)
-               setup_apic_nmi_watchdog();
-       apic_pm_activate();
-}
-
-/*
- * If Linux enabled the LAPIC against the BIOS default
- * disable it down before re-entering the BIOS on shutdown.
- * Otherwise the BIOS may get confused and not power-off.
- * Additionally clear all LVT entries before disable_local_APIC
- * for the case where Linux didn't enable the LAPIC.
- */
-void lapic_shutdown(void)
-{
-       unsigned long flags;
-
-       if (!cpu_has_apic)
-               return;
-
-       local_irq_save(flags);
-       clear_local_APIC();
-
-       if (enabled_via_apicbase)
-               disable_local_APIC();
-
-       local_irq_restore(flags);
-}
-
-#ifdef CONFIG_PM
-
-static struct {
-       int active;
-       /* r/w apic fields */
-       unsigned int apic_id;
-       unsigned int apic_taskpri;
-       unsigned int apic_ldr;
-       unsigned int apic_dfr;
-       unsigned int apic_spiv;
-       unsigned int apic_lvtt;
-       unsigned int apic_lvtpc;
-       unsigned int apic_lvt0;
-       unsigned int apic_lvt1;
-       unsigned int apic_lvterr;
-       unsigned int apic_tmict;
-       unsigned int apic_tdcr;
-       unsigned int apic_thmr;
-} apic_pm_state;
-
-static int lapic_suspend(struct sys_device *dev, pm_message_t state)
-{
-       unsigned long flags;
-
-       if (!apic_pm_state.active)
-               return 0;
-
-       apic_pm_state.apic_id = apic_read(APIC_ID);
-       apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
-       apic_pm_state.apic_ldr = apic_read(APIC_LDR);
-       apic_pm_state.apic_dfr = apic_read(APIC_DFR);
-       apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
-       apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
-       apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
-       apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
-       apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
-       apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
-       apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
-       apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
-       apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
-       
-       local_irq_save(flags);
-       disable_local_APIC();
-       local_irq_restore(flags);
-       return 0;
-}
-
-static int lapic_resume(struct sys_device *dev)
-{
-       unsigned int l, h;
-       unsigned long flags;
-
-       if (!apic_pm_state.active)
-               return 0;
-
-       local_irq_save(flags);
-
-       /*
-        * Make sure the APICBASE points to the right address
-        *
-        * FIXME! This will be wrong if we ever support suspend on
-        * SMP! We'll need to do this as part of the CPU restore!
-        */
-       rdmsr(MSR_IA32_APICBASE, l, h);
-       l &= ~MSR_IA32_APICBASE_BASE;
-       l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
-       wrmsr(MSR_IA32_APICBASE, l, h);
-
-       apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
-       apic_write(APIC_ID, apic_pm_state.apic_id);
-       apic_write(APIC_DFR, apic_pm_state.apic_dfr);
-       apic_write(APIC_LDR, apic_pm_state.apic_ldr);
-       apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
-       apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
-       apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
-       apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
-       apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
-       apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
-       apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
-       apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
-       apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
-       apic_write(APIC_ESR, 0);
-       apic_read(APIC_ESR);
-       apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
-       apic_write(APIC_ESR, 0);
-       apic_read(APIC_ESR);
-       local_irq_restore(flags);
-       return 0;
-}
-
-/*
- * This device has no shutdown method - fully functioning local APICs
- * are needed on every CPU up until machine_halt/restart/poweroff.
- */
-
-static struct sysdev_class lapic_sysclass = {
-       set_kset_name("lapic"),
-       .resume         = lapic_resume,
-       .suspend        = lapic_suspend,
-};
-
-static struct sys_device device_lapic = {
-       .id     = 0,
-       .cls    = &lapic_sysclass,
-};
-
-static void __devinit apic_pm_activate(void)
-{
-       apic_pm_state.active = 1;
-}
-
-static int __init init_lapic_sysfs(void)
-{
-       int error;
-
-       if (!cpu_has_apic)
-               return 0;
-       /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
+       /* Disable the local apic timer */
+       value = apic_read(APIC_LVTT);
+       value |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
+       apic_write_around(APIC_LVTT, value);
 
-       error = sysdev_class_register(&lapic_sysclass);
-       if (!error)
-               error = sysdev_register(&device_lapic);
-       return error;
+       setup_apic_nmi_watchdog(NULL);
+       apic_pm_activate();
 }
-device_initcall(init_lapic_sysfs);
-
-#else  /* CONFIG_PM */
-
-static void apic_pm_activate(void) { }
-
-#endif /* CONFIG_PM */
 
 /*
- * Detect and enable local APICs on non-SMP boards.
- * Original code written by Keir Fraser.
+ * Detect and initialize APIC
  */
-
-static int __init apic_set_verbosity(char *str)
-{
-       if (strcmp("debug", str) == 0)
-               apic_verbosity = APIC_DEBUG;
-       else if (strcmp("verbose", str) == 0)
-               apic_verbosity = APIC_VERBOSE;
-       return 1;
-}
-
-__setup("apic=", apic_set_verbosity);
-
 static int __init detect_init_APIC (void)
 {
        u32 h, l, features;
@@ -773,7 +1002,7 @@ static int __init detect_init_APIC (void)
        switch (boot_cpu_data.x86_vendor) {
        case X86_VENDOR_AMD:
                if ((boot_cpu_data.x86 == 6 && boot_cpu_data.x86_model > 1) ||
-                   (boot_cpu_data.x86 == 15))      
+                   (boot_cpu_data.x86 == 15))
                        break;
                goto no_apic;
        case X86_VENDOR_INTEL:
@@ -787,23 +1016,23 @@ static int __init detect_init_APIC (void)
 
        if (!cpu_has_apic) {
                /*
-                * Over-ride BIOS and try to enable the local
-                * APIC only if "lapic" specified.
+                * Over-ride BIOS and try to enable the local APIC only if
+                * "lapic" specified.
                 */
                if (enable_local_apic <= 0) {
-                       printk("Local APIC disabled by BIOS -- "
+                       printk(KERN_INFO "Local APIC disabled by BIOS -- "
                               "you can enable it with \"lapic\"\n");
                        return -1;
                }
                /*
-                * Some BIOSes disable the local APIC in the
-                * APIC_BASE MSR. This can only be done in
-                * software for Intel P6 or later and AMD K7
-                * (Model > 1) or later.
+                * Some BIOSes disable the local APIC in the APIC_BASE
+                * MSR. This can only be done in software for Intel P6 or later
+                * and AMD K7 (Model > 1) or later.
                 */
                rdmsr(MSR_IA32_APICBASE, l, h);
                if (!(l & MSR_IA32_APICBASE_ENABLE)) {
-                       printk("Local APIC disabled by BIOS -- reenabling.\n");
+                       printk(KERN_INFO
+                              "Local APIC disabled by BIOS -- reenabling.\n");
                        l &= ~MSR_IA32_APICBASE_BASE;
                        l |= MSR_IA32_APICBASE_ENABLE | APIC_DEFAULT_PHYS_BASE;
                        wrmsr(MSR_IA32_APICBASE, l, h);
@@ -816,7 +1045,7 @@ static int __init detect_init_APIC (void)
         */
        features = cpuid_edx(1);
        if (!(features & (1 << X86_FEATURE_APIC))) {
-               printk("Could not enable APIC!\n");
+               printk(KERN_WARNING "Could not enable APIC!\n");
                return -1;
        }
        set_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
@@ -830,17 +1059,20 @@ static int __init detect_init_APIC (void)
        if (nmi_watchdog != NMI_NONE)
                nmi_watchdog = NMI_LOCAL_APIC;
 
-       printk("Found and enabled local APIC!\n");
+       printk(KERN_INFO "Found and enabled local APIC!\n");
 
        apic_pm_activate();
 
        return 0;
 
 no_apic:
-       printk("No local APIC present or hardware disabled\n");
+       printk(KERN_INFO "No local APIC present or hardware disabled\n");
        return -1;
 }
 
+/**
+ * init_apic_mappings - initialize APIC mappings
+ */
 void __init init_apic_mappings(void)
 {
        unsigned long apic_phys;
@@ -900,385 +1132,96 @@ fake_ioapic_page:
 }
 
 /*
- * This part sets up the APIC 32 bit clock in LVTT1, with HZ interrupts
- * per second. We assume that the caller has already set up the local
- * APIC.
- *
- * The APIC timer is not exactly sync with the external timer chip, it
- * closely follows bus clocks.
- */
-
-/*
- * The timer chip is already set up at HZ interrupts per second here,
- * but we do not accept timer interrupts yet. We only allow the BP
- * to calibrate.
+ * This initializes the IO-APIC and APIC hardware if this is
+ * a UP kernel.
  */
-static unsigned int __devinit get_8254_timer_count(void)
+int __init APIC_init_uniprocessor (void)
 {
-       unsigned long flags;
+       if (enable_local_apic < 0)
+               clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
 
-       unsigned int count;
+       if (!smp_found_config && !cpu_has_apic)
+               return -1;
 
-       spin_lock_irqsave(&i8253_lock, flags);
+       /*
+        * Complain if the BIOS pretends there is one.
+        */
+       if (!cpu_has_apic &&
+           APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
+               printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
+                      boot_cpu_physical_apicid);
+               clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
+               return -1;
+       }
 
-       outb_p(0x00, PIT_MODE);
-       count = inb_p(PIT_CH0);
-       count |= inb_p(PIT_CH0) << 8;
+       verify_local_APIC();
 
-       spin_unlock_irqrestore(&i8253_lock, flags);
-
-       return count;
-}
-
-/* next tick in 8254 can be caught by catching timer wraparound */
-static void __devinit wait_8254_wraparound(void)
-{
-       unsigned int curr_count, prev_count;
-
-       curr_count = get_8254_timer_count();
-       do {
-               prev_count = curr_count;
-               curr_count = get_8254_timer_count();
-
-               /* workaround for broken Mercury/Neptune */
-               if (prev_count >= curr_count + 0x100)
-                       curr_count = get_8254_timer_count();
-
-       } while (prev_count >= curr_count);
-}
-
-/*
- * Default initialization for 8254 timers. If we use other timers like HPET,
- * we override this later
- */
-void (*wait_timer_tick)(void) __devinitdata = wait_8254_wraparound;
-
-/*
- * This function sets up the local APIC timer, with a timeout of
- * 'clocks' APIC bus clock. During calibration we actually call
- * this function twice on the boot CPU, once with a bogus timeout
- * value, second time for real. The other (noncalibrating) CPUs
- * call this function only once, with the real, calibrated value.
- *
- * We do reads before writes even if unnecessary, to get around the
- * P5 APIC double write bug.
- */
-
-#define APIC_DIVISOR 16
-
-static void __setup_APIC_LVTT(unsigned int clocks)
-{
-       unsigned int lvtt_value, tmp_value, ver;
-       int cpu = smp_processor_id();
-
-       ver = GET_APIC_VERSION(apic_read(APIC_LVR));
-       lvtt_value = APIC_LVT_TIMER_PERIODIC | LOCAL_TIMER_VECTOR;
-       if (!APIC_INTEGRATED(ver))
-               lvtt_value |= SET_APIC_TIMER_BASE(APIC_TIMER_BASE_DIV);
-
-       if (cpu_isset(cpu, timer_bcast_ipi))
-               lvtt_value |= APIC_LVT_MASKED;
-
-       apic_write_around(APIC_LVTT, lvtt_value);
+       connect_bsp_APIC();
 
        /*
-        * Divide PICLK by 16
+        * Hack: In case of kdump, after a crash, kernel might be booting
+        * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
+        * might be zero if read from MP tables. Get it from LAPIC.
         */
-       tmp_value = apic_read(APIC_TDCR);
-       apic_write_around(APIC_TDCR, (tmp_value
-                               & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE))
-                               | APIC_TDR_DIV_16);
-
-       apic_write_around(APIC_TMICT, clocks/APIC_DIVISOR);
-}
-
-static void __devinit setup_APIC_timer(unsigned int clocks)
-{
-       unsigned long flags;
-
-       local_irq_save(flags);
+#ifdef CONFIG_CRASH_DUMP
+       boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
+#endif
+       phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid);
 
-       /*
-        * Wait for IRQ0's slice:
-        */
-       wait_timer_tick();
+       setup_local_APIC();
 
-       __setup_APIC_LVTT(clocks);
+#ifdef CONFIG_X86_IO_APIC
+       if (smp_found_config)
+               if (!skip_ioapic_setup && nr_ioapics)
+                       setup_IO_APIC();
+#endif
+       setup_boot_clock();
 
-       local_irq_restore(flags);
+       return 0;
 }
 
 /*
- * In this function we calibrate APIC bus clocks to the external
- * timer. Unfortunately we cannot use jiffies and the timer irq
- * to calibrate, since some later bootup code depends on getting
- * the first irq? Ugh.
- *
- * We want to do the calibration only once since we
- * want to have local timer irqs syncron. CPUs connected
- * by the same APIC bus have the very same bus frequency.
- * And we want to have irqs off anyways, no accidental
- * APIC irq that way.
+ * APIC command line parameters
  */
-
-static int __init calibrate_APIC_clock(void)
-{
-       unsigned long long t1 = 0, t2 = 0;
-       long tt1, tt2;
-       long result;
-       int i;
-       const int LOOPS = HZ/10;
-
-       apic_printk(APIC_VERBOSE, "calibrating APIC timer ...\n");
-
-       /*
-        * Put whatever arbitrary (but long enough) timeout
-        * value into the APIC clock, we just want to get the
-        * counter running for calibration.
-        */
-       __setup_APIC_LVTT(1000000000);
-
-       /*
-        * The timer chip counts down to zero. Let's wait
-        * for a wraparound to start exact measurement:
-        * (the current tick might have been already half done)
-        */
-
-       wait_timer_tick();
-
-       /*
-        * We wrapped around just now. Let's start:
-        */
-       if (cpu_has_tsc)
-               rdtscll(t1);
-       tt1 = apic_read(APIC_TMCCT);
-
-       /*
-        * Let's wait LOOPS wraprounds:
-        */
-       for (i = 0; i < LOOPS; i++)
-               wait_timer_tick();
-
-       tt2 = apic_read(APIC_TMCCT);
-       if (cpu_has_tsc)
-               rdtscll(t2);
-
-       /*
-        * The APIC bus clock counter is 32 bits only, it
-        * might have overflown, but note that we use signed
-        * longs, thus no extra care needed.
-        *
-        * underflown to be exact, as the timer counts down ;)
-        */
-
-       result = (tt1-tt2)*APIC_DIVISOR/LOOPS;
-
-       if (cpu_has_tsc)
-               apic_printk(APIC_VERBOSE, "..... CPU clock speed is "
-                       "%ld.%04ld MHz.\n",
-                       ((long)(t2-t1)/LOOPS)/(1000000/HZ),
-                       ((long)(t2-t1)/LOOPS)%(1000000/HZ));
-
-       apic_printk(APIC_VERBOSE, "..... host bus clock speed is "
-               "%ld.%04ld MHz.\n",
-               result/(1000000/HZ),
-               result%(1000000/HZ));
-
-       return result;
-}
-
-static unsigned int calibration_result;
-
-void __init setup_boot_APIC_clock(void)
-{
-       unsigned long flags;
-       apic_printk(APIC_VERBOSE, "Using local APIC timer interrupts.\n");
-       using_apic_timer = 1;
-
-       local_irq_save(flags);
-
-       calibration_result = calibrate_APIC_clock();
-       /*
-        * Now set up the timer for real.
-        */
-       setup_APIC_timer(calibration_result);
-
-       local_irq_restore(flags);
-}
-
-void __devinit setup_secondary_APIC_clock(void)
-{
-       setup_APIC_timer(calibration_result);
-}
-
-void disable_APIC_timer(void)
-{
-       if (using_apic_timer) {
-               unsigned long v;
-
-               v = apic_read(APIC_LVTT);
-               /*
-                * When an illegal vector value (0-15) is written to an LVT
-                * entry and delivery mode is Fixed, the APIC may signal an
-                * illegal vector error, with out regard to whether the mask
-                * bit is set or whether an interrupt is actually seen on input.
-                *
-                * Boot sequence might call this function when the LVTT has
-                * '0' vector value. So make sure vector field is set to
-                * valid value.
-                */
-               v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
-               apic_write_around(APIC_LVTT, v);
-       }
-}
-
-void enable_APIC_timer(void)
+static int __init parse_lapic(char *arg)
 {
-       int cpu = smp_processor_id();
-
-       if (using_apic_timer &&
-           !cpu_isset(cpu, timer_bcast_ipi)) {
-               unsigned long v;
-
-               v = apic_read(APIC_LVTT);
-               apic_write_around(APIC_LVTT, v & ~APIC_LVT_MASKED);
-       }
+       enable_local_apic = 1;
+       return 0;
 }
+early_param("lapic", parse_lapic);
 
-void switch_APIC_timer_to_ipi(void *cpumask)
+static int __init parse_nolapic(char *arg)
 {
-       cpumask_t mask = *(cpumask_t *)cpumask;
-       int cpu = smp_processor_id();
-
-       if (cpu_isset(cpu, mask) &&
-           !cpu_isset(cpu, timer_bcast_ipi)) {
-               disable_APIC_timer();
-               cpu_set(cpu, timer_bcast_ipi);
-       }
+       enable_local_apic = -1;
+       clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
+       return 0;
 }
-EXPORT_SYMBOL(switch_APIC_timer_to_ipi);
+early_param("nolapic", parse_nolapic);
 
-void switch_ipi_to_APIC_timer(void *cpumask)
+static int __init apic_set_verbosity(char *str)
 {
-       cpumask_t mask = *(cpumask_t *)cpumask;
-       int cpu = smp_processor_id();
-
-       if (cpu_isset(cpu, mask) &&
-           cpu_isset(cpu, timer_bcast_ipi)) {
-               cpu_clear(cpu, timer_bcast_ipi);
-               enable_APIC_timer();
-       }
+       if (strcmp("debug", str) == 0)
+               apic_verbosity = APIC_DEBUG;
+       else if (strcmp("verbose", str) == 0)
+               apic_verbosity = APIC_VERBOSE;
+       return 1;
 }
-EXPORT_SYMBOL(switch_ipi_to_APIC_timer);
-
-#undef APIC_DIVISOR
-
-/*
- * Local timer interrupt handler. It does both profiling and
- * process statistics/rescheduling.
- *
- * We do profiling in every local tick, statistics/rescheduling
- * happen only every 'profiling multiplier' ticks. The default
- * multiplier is 1 and it can be changed by writing the new multiplier
- * value into /proc/profile.
- */
 
-inline void smp_local_timer_interrupt(struct pt_regs * regs)
-{
-       profile_tick(CPU_PROFILING, regs);
-#ifdef CONFIG_SMP
-       update_process_times(user_mode_vm(regs));
-#endif
+__setup("apic=", apic_set_verbosity);
 
-       /*
-        * We take the 'long' return path, and there every subsystem
-        * grabs the apropriate locks (kernel lock/ irq lock).
-        *
-        * we might want to decouple profiling from the 'long path',
-        * and do the profiling totally in assembly.
-        *
-        * Currently this isn't too much of an issue (performance wise),
-        * we can take more than 100K local irqs per second on a 100 MHz P5.
-        */
-}
 
 /*
- * Local APIC timer interrupt. This is the most natural way for doing
- * local interrupts, but local timer interrupts can be emulated by
- * broadcast interrupts too. [in case the hw doesn't support APIC timers]
- *
- * [ if a single-CPU system runs an SMP kernel then we call the local
- *   interrupt as well. Thus we cannot inline the local irq ... ]
+ * Local APIC interrupts
  */
 
-fastcall void smp_apic_timer_interrupt(struct pt_regs *regs)
-{
-       int cpu = smp_processor_id();
-
-       /*
-        * the NMI deadlock-detector uses this.
-        */
-       per_cpu(irq_stat, cpu).apic_timer_irqs++;
-
-       /*
-        * NOTE! We'd better ACK the irq immediately,
-        * because timer handling can be slow.
-        */
-       ack_APIC_irq();
-       /*
-        * update_process_times() expects us to have done irq_enter().
-        * Besides, if we don't timer interrupts ignore the global
-        * interrupt lock, which is the WrongThing (tm) to do.
-        */
-       irq_enter();
-       smp_local_timer_interrupt(regs);
-       irq_exit();
-}
-
-#ifndef CONFIG_SMP
-static void up_apic_timer_interrupt_call(struct pt_regs *regs)
-{
-       int cpu = smp_processor_id();
-
-       /*
-        * the NMI deadlock-detector uses this.
-        */
-       per_cpu(irq_stat, cpu).apic_timer_irqs++;
-
-       smp_local_timer_interrupt(regs);
-}
-#endif
-
-void smp_send_timer_broadcast_ipi(struct pt_regs *regs)
-{
-       cpumask_t mask;
-
-       cpus_and(mask, cpu_online_map, timer_bcast_ipi);
-       if (!cpus_empty(mask)) {
-#ifdef CONFIG_SMP
-               send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
-#else
-               /*
-                * We can directly call the apic timer interrupt handler
-                * in UP case. Minus all irq related functions
-                */
-               up_apic_timer_interrupt_call(regs);
-#endif
-       }
-}
-
-int setup_profiling_timer(unsigned int multiplier)
-{
-       return -EINVAL;
-}
-
 /*
  * This interrupt should _never_ happen with our APIC/SMP architecture
  */
-fastcall void smp_spurious_interrupt(struct pt_regs *regs)
+void smp_spurious_interrupt(struct pt_regs *regs)
 {
        unsigned long v;
 
+       exit_idle();
        irq_enter();
        /*
         * Check if this really is a spurious interrupt and ACK it
@@ -1290,19 +1233,19 @@ fastcall void smp_spurious_interrupt(struct pt_regs *regs)
                ack_APIC_irq();
 
        /* see sw-dev-man vol 3, chapter 7.4.13.5 */
-       printk(KERN_INFO "spurious APIC interrupt on CPU#%d, should never happen.\n",
-                       smp_processor_id());
+       printk(KERN_INFO "spurious APIC interrupt on CPU#%d, "
+              "should never happen.\n", smp_processor_id());
        irq_exit();
 }
 
 /*
  * This interrupt should never happen with our APIC/SMP architecture
  */
-
-fastcall void smp_error_interrupt(struct pt_regs *regs)
+void smp_error_interrupt(struct pt_regs *regs)
 {
        unsigned long v, v1;
 
+       exit_idle();
        irq_enter();
        /* First tickle the hardware, only then report what went on. -- REW */
        v = apic_read(APIC_ESR);
@@ -1322,54 +1265,261 @@ fastcall void smp_error_interrupt(struct pt_regs *regs)
           7: Illegal register address
        */
        printk (KERN_DEBUG "APIC error on CPU%d: %02lx(%02lx)\n",
-               smp_processor_id(), v , v1);
+               smp_processor_id(), v , v1);
        irq_exit();
 }
 
 /*
- * This initializes the IO-APIC and APIC hardware if this is
- * a UP kernel.
+ * Initialize APIC interrupts
  */
-int __init APIC_init_uniprocessor (void)
+void __init apic_intr_init(void)
 {
-       if (enable_local_apic < 0)
-               clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
+#ifdef CONFIG_SMP
+       smp_intr_init();
+#endif
+       /* self generated IPI for local APIC timer */
+       set_intr_gate(LOCAL_TIMER_VECTOR, apic_timer_interrupt);
 
-       if (!smp_found_config && !cpu_has_apic)
-               return -1;
+       /* IPI vectors for APIC spurious and error interrupts */
+       set_intr_gate(SPURIOUS_APIC_VECTOR, spurious_interrupt);
+       set_intr_gate(ERROR_APIC_VECTOR, error_interrupt);
 
-       /*
-        * Complain if the BIOS pretends there is one.
-        */
-       if (!cpu_has_apic && APIC_INTEGRATED(apic_version[boot_cpu_physical_apicid])) {
-               printk(KERN_ERR "BIOS bug, local APIC #%d not detected!...\n",
-                       boot_cpu_physical_apicid);
-               clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
-               return -1;
+       /* thermal monitor LVT interrupt */
+#ifdef CONFIG_X86_MCE_P4THERMAL
+       set_intr_gate(THERMAL_APIC_VECTOR, thermal_interrupt);
+#endif
+}
+
+/**
+ * connect_bsp_APIC - attach the APIC to the interrupt system
+ */
+void __init connect_bsp_APIC(void)
+{
+       if (pic_mode) {
+               /*
+                * Do not trust the local APIC being empty at bootup.
+                */
+               clear_local_APIC();
+               /*
+                * PIC mode, enable APIC mode in the IMCR, i.e.  connect BSP's
+                * local APIC to INT and NMI lines.
+                */
+               apic_printk(APIC_VERBOSE, "leaving PIC mode, "
+                               "enabling APIC mode.\n");
+               outb(0x70, 0x22);
+               outb(0x01, 0x23);
        }
+       enable_apic_mode();
+}
 
-       verify_local_APIC();
+/**
+ * disconnect_bsp_APIC - detach the APIC from the interrupt system
+ * @virt_wire_setup:   indicates, whether virtual wire mode is selected
+ *
+ * Virtual wire mode is necessary to deliver legacy interrupts even when the
+ * APIC is disabled.
+ */
+void disconnect_bsp_APIC(int virt_wire_setup)
+{
+       if (pic_mode) {
+               /*
+                * Put the board back into PIC mode (has an effect only on
+                * certain older boards).  Note that APIC interrupts, including
+                * IPIs, won't work beyond this point!  The only exception are
+                * INIT IPIs.
+                */
+               apic_printk(APIC_VERBOSE, "disabling APIC mode, "
+                               "entering PIC mode.\n");
+               outb(0x70, 0x22);
+               outb(0x00, 0x23);
+       } else {
+               /* Go back to Virtual Wire compatibility mode */
+               unsigned long value;
 
-       connect_bsp_APIC();
+               /* For the spurious interrupt use vector F, and enable it */
+               value = apic_read(APIC_SPIV);
+               value &= ~APIC_VECTOR_MASK;
+               value |= APIC_SPIV_APIC_ENABLED;
+               value |= 0xf;
+               apic_write_around(APIC_SPIV, value);
+
+               if (!virt_wire_setup) {
+                       /*
+                        * For LVT0 make it edge triggered, active high,
+                        * external and enabled
+                        */
+                       value = apic_read(APIC_LVT0);
+                       value &= ~(APIC_MODE_MASK | APIC_SEND_PENDING |
+                               APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
+                               APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED );
+                       value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
+                       value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_EXTINT);
+                       apic_write_around(APIC_LVT0, value);
+               } else {
+                       /* Disable LVT0 */
+                       apic_write_around(APIC_LVT0, APIC_LVT_MASKED);
+               }
+
+               /*
+                * For LVT1 make it edge triggered, active high, nmi and
+                * enabled
+                */
+               value = apic_read(APIC_LVT1);
+               value &= ~(
+                       APIC_MODE_MASK | APIC_SEND_PENDING |
+                       APIC_INPUT_POLARITY | APIC_LVT_REMOTE_IRR |
+                       APIC_LVT_LEVEL_TRIGGER | APIC_LVT_MASKED);
+               value |= APIC_LVT_REMOTE_IRR | APIC_SEND_PENDING;
+               value = SET_APIC_DELIVERY_MODE(value, APIC_MODE_NMI);
+               apic_write_around(APIC_LVT1, value);
+       }
+}
+
+/*
+ * Power management
+ */
+#ifdef CONFIG_PM
+
+static struct {
+       int active;
+       /* r/w apic fields */
+       unsigned int apic_id;
+       unsigned int apic_taskpri;
+       unsigned int apic_ldr;
+       unsigned int apic_dfr;
+       unsigned int apic_spiv;
+       unsigned int apic_lvtt;
+       unsigned int apic_lvtpc;
+       unsigned int apic_lvt0;
+       unsigned int apic_lvt1;
+       unsigned int apic_lvterr;
+       unsigned int apic_tmict;
+       unsigned int apic_tdcr;
+       unsigned int apic_thmr;
+} apic_pm_state;
+
+static int lapic_suspend(struct sys_device *dev, pm_message_t state)
+{
+       unsigned long flags;
+       int maxlvt;
+
+       if (!apic_pm_state.active)
+               return 0;
+
+       maxlvt = lapic_get_maxlvt();
+
+       apic_pm_state.apic_id = apic_read(APIC_ID);
+       apic_pm_state.apic_taskpri = apic_read(APIC_TASKPRI);
+       apic_pm_state.apic_ldr = apic_read(APIC_LDR);
+       apic_pm_state.apic_dfr = apic_read(APIC_DFR);
+       apic_pm_state.apic_spiv = apic_read(APIC_SPIV);
+       apic_pm_state.apic_lvtt = apic_read(APIC_LVTT);
+       if (maxlvt >= 4)
+               apic_pm_state.apic_lvtpc = apic_read(APIC_LVTPC);
+       apic_pm_state.apic_lvt0 = apic_read(APIC_LVT0);
+       apic_pm_state.apic_lvt1 = apic_read(APIC_LVT1);
+       apic_pm_state.apic_lvterr = apic_read(APIC_LVTERR);
+       apic_pm_state.apic_tmict = apic_read(APIC_TMICT);
+       apic_pm_state.apic_tdcr = apic_read(APIC_TDCR);
+#ifdef CONFIG_X86_MCE_P4THERMAL
+       if (maxlvt >= 5)
+               apic_pm_state.apic_thmr = apic_read(APIC_LVTTHMR);
+#endif
+
+       local_irq_save(flags);
+       disable_local_APIC();
+       local_irq_restore(flags);
+       return 0;
+}
+
+static int lapic_resume(struct sys_device *dev)
+{
+       unsigned int l, h;
+       unsigned long flags;
+       int maxlvt;
+
+       if (!apic_pm_state.active)
+               return 0;
+
+       maxlvt = lapic_get_maxlvt();
+
+       local_irq_save(flags);
 
        /*
-        * Hack: In case of kdump, after a crash, kernel might be booting
-        * on a cpu with non-zero lapic id. But boot_cpu_physical_apicid
-        * might be zero if read from MP tables. Get it from LAPIC.
+        * Make sure the APICBASE points to the right address
+        *
+        * FIXME! This will be wrong if we ever support suspend on
+        * SMP! We'll need to do this as part of the CPU restore!
         */
-#ifdef CONFIG_CRASH_DUMP
-       boot_cpu_physical_apicid = GET_APIC_ID(apic_read(APIC_ID));
+       rdmsr(MSR_IA32_APICBASE, l, h);
+       l &= ~MSR_IA32_APICBASE_BASE;
+       l |= MSR_IA32_APICBASE_ENABLE | mp_lapic_addr;
+       wrmsr(MSR_IA32_APICBASE, l, h);
+
+       apic_write(APIC_LVTERR, ERROR_APIC_VECTOR | APIC_LVT_MASKED);
+       apic_write(APIC_ID, apic_pm_state.apic_id);
+       apic_write(APIC_DFR, apic_pm_state.apic_dfr);
+       apic_write(APIC_LDR, apic_pm_state.apic_ldr);
+       apic_write(APIC_TASKPRI, apic_pm_state.apic_taskpri);
+       apic_write(APIC_SPIV, apic_pm_state.apic_spiv);
+       apic_write(APIC_LVT0, apic_pm_state.apic_lvt0);
+       apic_write(APIC_LVT1, apic_pm_state.apic_lvt1);
+#ifdef CONFIG_X86_MCE_P4THERMAL
+       if (maxlvt >= 5)
+               apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
 #endif
-       phys_cpu_present_map = physid_mask_of_physid(boot_cpu_physical_apicid);
+       if (maxlvt >= 4)
+               apic_write(APIC_LVTPC, apic_pm_state.apic_lvtpc);
+       apic_write(APIC_LVTT, apic_pm_state.apic_lvtt);
+       apic_write(APIC_TDCR, apic_pm_state.apic_tdcr);
+       apic_write(APIC_TMICT, apic_pm_state.apic_tmict);
+       apic_write(APIC_ESR, 0);
+       apic_read(APIC_ESR);
+       apic_write(APIC_LVTERR, apic_pm_state.apic_lvterr);
+       apic_write(APIC_ESR, 0);
+       apic_read(APIC_ESR);
+       local_irq_restore(flags);
+       return 0;
+}
 
-       setup_local_APIC();
+/*
+ * This device has no shutdown method - fully functioning local APICs
+ * are needed on every CPU up until machine_halt/restart/poweroff.
+ */
 
-#ifdef CONFIG_X86_IO_APIC
-       if (smp_found_config)
-               if (!skip_ioapic_setup && nr_ioapics)
-                       setup_IO_APIC();
-#endif
-       setup_boot_APIC_clock();
+static struct sysdev_class lapic_sysclass = {
+       set_kset_name("lapic"),
+       .resume         = lapic_resume,
+       .suspend        = lapic_suspend,
+};
 
-       return 0;
+static struct sys_device device_lapic = {
+       .id     = 0,
+       .cls    = &lapic_sysclass,
+};
+
+static void __devinit apic_pm_activate(void)
+{
+       apic_pm_state.active = 1;
 }
+
+static int __init init_lapic_sysfs(void)
+{
+       int error;
+
+       if (!cpu_has_apic)
+               return 0;
+       /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
+
+       error = sysdev_class_register(&lapic_sysclass);
+       if (!error)
+               error = sysdev_register(&device_lapic);
+       return error;
+}
+device_initcall(init_lapic_sysfs);
+
+#else  /* CONFIG_PM */
+
+static void apic_pm_activate(void) { }
+
+#endif /* CONFIG_PM */