x86: make mxcsr_feature_mask static again
[sfrench/cifs-2.6.git] / arch / x86 / kernel / apic_64.c
index 925758dbca0c6016b52e91ab90e7503c90ea08c0..d8d03e09dea24a45e94903ae1c61a646d65777d1 100644 (file)
 #include <linux/mc146818rtc.h>
 #include <linux/kernel_stat.h>
 #include <linux/sysdev.h>
-#include <linux/module.h>
 #include <linux/ioport.h>
+#include <linux/clockchips.h>
+#include <linux/acpi_pmtmr.h>
+#include <linux/module.h>
 
 #include <asm/atomic.h>
 #include <asm/smp.h>
 #include <asm/mtrr.h>
 #include <asm/mpspec.h>
+#include <asm/hpet.h>
 #include <asm/pgalloc.h>
 #include <asm/mach_apic.h>
 #include <asm/nmi.h>
 #include <asm/idle.h>
 #include <asm/proto.h>
 #include <asm/timex.h>
-#include <asm/hpet.h>
 #include <asm/apic.h>
 
-int apic_mapped;
-int apic_verbosity;
-int apic_runs_main_timer;
-int apic_calibrate_pmtmr __initdata;
-
-int disable_apic_timer __initdata;
+int disable_apic_timer __cpuinitdata;
+static int apic_calibrate_pmtmr __initdata;
+int disable_apic;
 
-/* Local APIC timer works in C2? */
+/* Local APIC timer works in C2 */
 int local_apic_timer_c2_ok;
 EXPORT_SYMBOL_GPL(local_apic_timer_c2_ok);
 
-static struct resource *ioapic_resources;
+/*
+ * Debug level, exported for io_apic.c
+ */
+int apic_verbosity;
+
 static struct resource lapic_resource = {
        .name = "Local APIC",
        .flags = IORESOURCE_MEM | IORESOURCE_BUSY,
 };
 
+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);
+
+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);
+
+static unsigned long apic_phys;
+
 /*
- * cpu_mask that denotes the CPUs that needs timer interrupt coming in as
- * IPIs in place of local APIC timers
+ * Get the LAPIC version
  */
-static cpumask_t timer_interrupt_broadcast_ipi_mask;
+static inline int lapic_get_version(void)
+{
+       return GET_APIC_VERSION(apic_read(APIC_LVR));
+}
 
-/* Using APIC to generate smp_local_timer_interrupt? */
-int using_apic_timer __read_mostly = 0;
+/*
+ * Check, if the APIC is integrated or a seperate chip
+ */
+static inline int lapic_is_integrated(void)
+{
+       return 1;
+}
 
-static void apic_pm_activate(void);
+/*
+ * Check, whether this is a modern or a first generation APIC
+ */
+static int modern_apic(void)
+{
+       /* AMD systems use old APIC versions, so check the CPU */
+       if (boot_cpu_data.x86_vendor == X86_VENDOR_AMD &&
+           boot_cpu_data.x86 >= 0xf)
+               return 1;
+       return lapic_get_version() >= 0x14;
+}
 
 void apic_wait_icr_idle(void)
 {
@@ -73,9 +117,9 @@ void apic_wait_icr_idle(void)
                cpu_relax();
 }
 
-unsigned int safe_apic_wait_icr_idle(void)
+u32 safe_apic_wait_icr_idle(void)
 {
-       unsigned int send_status;
+       u32 send_status;
        int timeout;
 
        timeout = 0;
@@ -89,7 +133,10 @@ unsigned int safe_apic_wait_icr_idle(void)
        return send_status;
 }
 
-void enable_NMI_through_LVT0 (void * dummy)
+/**
+ * enable_NMI_through_LVT0 - enable NMI through local vector table 0
+ */
+void __cpuinit enable_NMI_through_LVT0(void)
 {
        unsigned int v;
 
@@ -98,7 +145,10 @@ void enable_NMI_through_LVT0 (void * dummy)
        apic_write(APIC_LVT0, v);
 }
 
-int get_maxlvt(void)
+/**
+ * lapic_get_maxlvt - get the maximum number of local vector table entries
+ */
+int lapic_get_maxlvt(void)
 {
        unsigned int v, maxlvt;
 
@@ -108,225 +158,556 @@ int get_maxlvt(void)
 }
 
 /*
- * 'what should we do if we get a hw irq event on an illegal vector'.
- * each architecture has to answer this themselves.
+ * 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)
-{
-       printk("unexpected IRQ trap at vector %02x\n", irq);
-       /*
-        * 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 don't ack when the APIC is disabled. -AK
-        */
-       if (!disable_apic)
-               ack_APIC_irq();
-}
 
-void clear_local_APIC(void)
+static void __setup_APIC_LVTT(unsigned int clocks, int oneshot, int irqen)
 {
-       int maxlvt;
-       unsigned int v;
+       unsigned int lvtt_value, tmp_value;
 
-       maxlvt = get_maxlvt();
+       lvtt_value = LOCAL_TIMER_VECTOR;
+       if (!oneshot)
+               lvtt_value |= APIC_LVT_TIMER_PERIODIC;
+       if (!irqen)
+               lvtt_value |= APIC_LVT_MASKED;
 
-       /*
-        * Masking an LVT entry can trigger a local APIC error
-        * if the vector is zero. Mask LVTERR first to prevent this.
-        */
-       if (maxlvt >= 3) {
-               v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
-               apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
-       }
-       /*
-        * Careful: we have to set masks only first to deassert
-        * any level-triggered sources.
-        */
-       v = apic_read(APIC_LVTT);
-       apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
-       v = apic_read(APIC_LVT0);
-       apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
-       v = apic_read(APIC_LVT1);
-       apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
-       if (maxlvt >= 4) {
-               v = apic_read(APIC_LVTPC);
-               apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
-       }
+       apic_write(APIC_LVTT, lvtt_value);
 
        /*
-        * Clean APIC state for other OSs:
+        * Divide PICLK by 16
         */
-       apic_write(APIC_LVTT, APIC_LVT_MASKED);
-       apic_write(APIC_LVT0, APIC_LVT_MASKED);
-       apic_write(APIC_LVT1, APIC_LVT_MASKED);
-       if (maxlvt >= 3)
-               apic_write(APIC_LVTERR, APIC_LVT_MASKED);
-       if (maxlvt >= 4)
-               apic_write(APIC_LVTPC, APIC_LVT_MASKED);
-       apic_write(APIC_ESR, 0);
-       apic_read(APIC_ESR);
+       tmp_value = apic_read(APIC_TDCR);
+       apic_write(APIC_TDCR, (tmp_value
+                               & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE))
+                               | APIC_TDR_DIV_16);
+
+       if (!oneshot)
+               apic_write(APIC_TMICT, clocks);
 }
 
-void disconnect_bsp_APIC(int virt_wire_setup)
-{
-       /* Go back to Virtual Wire compatibility mode */
-       unsigned long value;
+/*
+ * Setup extended LVT, AMD specific (K8, family 10h)
+ *
+ * Vector mappings are hard coded. On K8 only offset 0 (APIC500) and
+ * MCE interrupts are supported. Thus MCE offset must be set to 0.
+ */
 
-       /* 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(APIC_SPIV, value);
+#define APIC_EILVT_LVTOFF_MCE 0
+#define APIC_EILVT_LVTOFF_IBS 1
 
-       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(APIC_LVT0, value);
-       } else {
-               /* Disable LVT0 */
-               apic_write(APIC_LVT0, APIC_LVT_MASKED);
-       }
+static void setup_APIC_eilvt(u8 lvt_off, u8 vector, u8 msg_type, u8 mask)
+{
+       unsigned long reg = (lvt_off << 4) + APIC_EILVT0;
+       unsigned int  v   = (mask << 16) | (msg_type << 8) | vector;
 
-       /* 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(APIC_LVT1, value);
+       apic_write(reg, v);
 }
 
-void disable_local_APIC(void)
+u8 setup_APIC_eilvt_mce(u8 vector, u8 msg_type, u8 mask)
 {
-       unsigned int value;
-
-       clear_local_APIC();
+       setup_APIC_eilvt(APIC_EILVT_LVTOFF_MCE, vector, msg_type, mask);
+       return APIC_EILVT_LVTOFF_MCE;
+}
 
-       /*
-        * Disable APIC (implies clearing of registers
-        * for 82489DX!).
-        */
-       value = apic_read(APIC_SPIV);
-       value &= ~APIC_SPIV_APIC_ENABLED;
-       apic_write(APIC_SPIV, value);
+u8 setup_APIC_eilvt_ibs(u8 vector, u8 msg_type, u8 mask)
+{
+       setup_APIC_eilvt(APIC_EILVT_LVTOFF_IBS, vector, msg_type, mask);
+       return APIC_EILVT_LVTOFF_IBS;
 }
 
 /*
- * This is to verify that we're looking at a real local APIC.
- * Check these against your board if the CPUs aren't getting
- * started for no apparent reason.
+ * Program the next event, relative to now
  */
-int __init verify_local_APIC(void)
+static int lapic_next_event(unsigned long delta,
+                           struct clock_event_device *evt)
 {
-       unsigned int reg0, reg1;
+       apic_write(APIC_TMICT, delta);
+       return 0;
+}
 
-       /*
-        * The version register is read-only in a real APIC.
       */
-       reg0 = apic_read(APIC_LVR);
-       apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
-       apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
-       reg1 = apic_read(APIC_LVR);
-       apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
+/*
+ * 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;
 
-       /*
-        * The two version reads above should print the same
-        * numbers.  If the second one is different, then we
-        * poke at a non-APIC.
-        */
-       if (reg1 != reg0)
-               return 0;
+       /* Lapic used as dummy for broadcast ? */
+       if (evt->features & CLOCK_EVT_FEAT_DUMMY)
+               return;
 
-       /*
-        * Check if the version looks reasonably.
-        */
-       reg1 = GET_APIC_VERSION(reg0);
-       if (reg1 == 0x00 || reg1 == 0xff)
-               return 0;
-       reg1 = get_maxlvt();
-       if (reg1 < 0x02 || reg1 == 0xff)
-               return 0;
+       local_irq_save(flags);
 
-       /*
-        * The ID register is read/write in a real APIC.
-        */
-       reg0 = apic_read(APIC_ID);
-       apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
-       apic_write(APIC_ID, reg0 ^ APIC_ID_MASK);
-       reg1 = apic_read(APIC_ID);
-       apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
-       apic_write(APIC_ID, reg0);
-       if (reg1 != (reg0 ^ APIC_ID_MASK))
-               return 0;
+       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(APIC_LVTT, v);
+               break;
+       case CLOCK_EVT_MODE_RESUME:
+               /* Nothing to do here */
+               break;
+       }
 
-       /*
-        * The next two are just to see if we have sane values.
-        * They're only really relevant if we're in Virtual Wire
-        * compatibility mode, but most boxes are anymore.
-        */
-       reg0 = apic_read(APIC_LVT0);
-       apic_printk(APIC_DEBUG,"Getting LVT0: %x\n", reg0);
-       reg1 = apic_read(APIC_LVT1);
-       apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
+       local_irq_restore(flags);
+}
 
-       return 1;
+/*
+ * Local APIC timer broadcast function
+ */
+static void lapic_timer_broadcast(cpumask_t mask)
+{
+#ifdef CONFIG_SMP
+       send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
+#endif
 }
 
-void __init sync_Arb_IDs(void)
+/*
+ * 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 setup_APIC_timer(void)
 {
-       /* Unsupported on P4 - see Intel Dev. Manual Vol. 3, Ch. 8.6.1 */
-       unsigned int ver = GET_APIC_VERSION(apic_read(APIC_LVR));
-       if (ver >= 0x14)        /* P4 or higher */
-               return;
+       struct clock_event_device *levt = &__get_cpu_var(lapic_events);
 
-       /*
-        * Wait for idle.
-        */
-       apic_wait_icr_idle();
+       memcpy(levt, &lapic_clockevent, sizeof(*levt));
+       levt->cpumask = cpumask_of_cpu(smp_processor_id());
 
-       apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
-       apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG
-                               | APIC_DM_INIT);
+       clockevents_register_device(levt);
 }
 
 /*
- * An initial setup of the virtual wire mode.
+ * 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.
  */
-void __init init_bsp_APIC(void)
-{
-       unsigned int value;
 
-       /*
-        * Don't do the setup now if we have a SMP BIOS as the
-        * through-I/O-APIC virtual wire mode might be active.
-        */
-       if (smp_found_config || !cpu_has_apic)
-               return;
+#define TICK_COUNT 100000000
 
-       value = apic_read(APIC_LVR);
+static void __init calibrate_APIC_clock(void)
+{
+       unsigned apic, apic_start;
+       unsigned long tsc, tsc_start;
+       int result;
 
-       /*
-        * Do not trust the local APIC being empty at bootup.
-        */
-       clear_local_APIC();
+       local_irq_disable();
 
        /*
-        * Enable APIC.
+        * Put whatever arbitrary (but long enough) timeout
+        * value into the APIC clock, we just want to get the
+        * counter running for calibration.
+        *
+        * No interrupt enable !
         */
-       value = apic_read(APIC_SPIV);
-       value &= ~APIC_VECTOR_MASK;
-       value |= APIC_SPIV_APIC_ENABLED;
+       __setup_APIC_LVTT(250000000, 0, 0);
+
+       apic_start = apic_read(APIC_TMCCT);
+#ifdef CONFIG_X86_PM_TIMER
+       if (apic_calibrate_pmtmr && pmtmr_ioport) {
+               pmtimer_wait(5000);  /* 5ms wait */
+               apic = apic_read(APIC_TMCCT);
+               result = (apic_start - apic) * 1000L / 5;
+       } else
+#endif
+       {
+               rdtscll(tsc_start);
+
+               do {
+                       apic = apic_read(APIC_TMCCT);
+                       rdtscll(tsc);
+               } while ((tsc - tsc_start) < TICK_COUNT &&
+                               (apic_start - apic) < TICK_COUNT);
+
+               result = (apic_start - apic) * 1000L * tsc_khz /
+                                       (tsc - tsc_start);
+       }
+
+       local_irq_enable();
+
+       printk(KERN_DEBUG "APIC timer calibration result %d\n", result);
+
+       printk(KERN_INFO "Detected %d.%03d MHz APIC timer.\n",
+               result / 1000 / 1000, result / 1000 % 1000);
+
+       /* Calculate the scaled math multiplication factor */
+       lapic_clockevent.mult = div_sc(result, NSEC_PER_SEC, 32);
+       lapic_clockevent.max_delta_ns =
+               clockevent_delta2ns(0x7FFFFF, &lapic_clockevent);
+       lapic_clockevent.min_delta_ns =
+               clockevent_delta2ns(0xF, &lapic_clockevent);
+
+       calibration_result = result / HZ;
+}
+
+/*
+ * Setup the boot APIC
+ *
+ * Calibrate and verify the result.
+ */
+void __init setup_boot_APIC_clock(void)
+{
+       /*
+        * The local apic timer can be disabled via the kernel commandline.
+        * Register the lapic timer as a dummy clock event source on SMP
+        * systems, so the broadcast mechanism is used. On UP systems simply
+        * ignore it.
+        */
+       if (disable_apic_timer) {
+               printk(KERN_INFO "Disabling APIC timer\n");
+               /* No broadcast on UP ! */
+               if (num_possible_cpus() > 1) {
+                       lapic_clockevent.mult = 1;
+                       setup_APIC_timer();
+               }
+               return;
+       }
+
+       printk(KERN_INFO "Using local APIC timer interrupts.\n");
+       calibrate_APIC_clock();
+
+       /*
+        * Do a sanity check on the APIC calibration result
+        */
+       if (calibration_result < (1000000 / HZ)) {
+               printk(KERN_WARNING
+                      "APIC frequency too slow, disabling apic timer\n");
+               /* No broadcast on UP ! */
+               if (num_possible_cpus() > 1)
+                       setup_APIC_timer();
+               return;
+       }
+
+       /*
+        * If nmi_watchdog is set to IO_APIC, we need the
+        * PIT/HPET going.  Otherwise register lapic as a dummy
+        * device.
+        */
+       if (nmi_watchdog != NMI_IO_APIC)
+               lapic_clockevent.features &= ~CLOCK_EVT_FEAT_DUMMY;
+       else
+               printk(KERN_WARNING "APIC timer registered as dummy,"
+                      " due to nmi_watchdog=1!\n");
+
+       setup_APIC_timer();
+}
+
+/*
+ * AMD C1E enabled CPUs have a real nasty problem: Some BIOSes set the
+ * C1E flag only in the secondary CPU, so when we detect the wreckage
+ * we already have enabled the boot CPU local apic timer. Check, if
+ * disable_apic_timer is set and the DUMMY flag is cleared. If yes,
+ * set the DUMMY flag again and force the broadcast mode in the
+ * clockevents layer.
+ */
+void __cpuinit check_boot_apic_timer_broadcast(void)
+{
+       if (!disable_apic_timer ||
+           (lapic_clockevent.features & CLOCK_EVT_FEAT_DUMMY))
+               return;
+
+       printk(KERN_INFO "AMD C1E detected late. Force timer broadcast.\n");
+       lapic_clockevent.features |= CLOCK_EVT_FEAT_DUMMY;
+
+       local_irq_enable();
+       clockevents_notify(CLOCK_EVT_NOTIFY_BROADCAST_FORCE, &boot_cpu_id);
+       local_irq_disable();
+}
+
+void __cpuinit setup_secondary_APIC_clock(void)
+{
+       check_boot_apic_timer_broadcast();
+       setup_APIC_timer();
+}
+
+/*
+ * 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;
+       }
+
+       /*
+        * the NMI deadlock-detector uses this.
+        */
+       add_pda(apic_timer_irqs, 1);
+
+       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 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 = lapic_get_maxlvt();
+       u32 v;
+
+       /* APIC hasn't been mapped yet */
+       if (!apic_phys)
+               return;
+
+       maxlvt = lapic_get_maxlvt();
+       /*
+        * Masking an LVT entry can trigger a local APIC error
+        * if the vector is zero. Mask LVTERR first to prevent this.
+        */
+       if (maxlvt >= 3) {
+               v = ERROR_APIC_VECTOR; /* any non-zero vector will do */
+               apic_write(APIC_LVTERR, v | APIC_LVT_MASKED);
+       }
+       /*
+        * Careful: we have to set masks only first to deassert
+        * any level-triggered sources.
+        */
+       v = apic_read(APIC_LVTT);
+       apic_write(APIC_LVTT, v | APIC_LVT_MASKED);
+       v = apic_read(APIC_LVT0);
+       apic_write(APIC_LVT0, v | APIC_LVT_MASKED);
+       v = apic_read(APIC_LVT1);
+       apic_write(APIC_LVT1, v | APIC_LVT_MASKED);
+       if (maxlvt >= 4) {
+               v = apic_read(APIC_LVTPC);
+               apic_write(APIC_LVTPC, v | APIC_LVT_MASKED);
+       }
+
+       /*
+        * Clean APIC state for other OSs:
+        */
+       apic_write(APIC_LVTT, APIC_LVT_MASKED);
+       apic_write(APIC_LVT0, APIC_LVT_MASKED);
+       apic_write(APIC_LVT1, APIC_LVT_MASKED);
+       if (maxlvt >= 3)
+               apic_write(APIC_LVTERR, APIC_LVT_MASKED);
+       if (maxlvt >= 4)
+               apic_write(APIC_LVTPC, APIC_LVT_MASKED);
+       apic_write(APIC_ESR, 0);
+       apic_read(APIC_ESR);
+}
+
+/**
+ * disable_local_APIC - clear and disable the local APIC
+ */
+void disable_local_APIC(void)
+{
+       unsigned int value;
+
+       clear_local_APIC();
+
+       /*
+        * Disable APIC (implies clearing of registers
+        * for 82489DX!).
+        */
+       value = apic_read(APIC_SPIV);
+       value &= ~APIC_SPIV_APIC_ENABLED;
+       apic_write(APIC_SPIV, value);
+}
+
+void lapic_shutdown(void)
+{
+       unsigned long flags;
+
+       if (!cpu_has_apic)
+               return;
+
+       local_irq_save(flags);
+
+       disable_local_APIC();
+
+       local_irq_restore(flags);
+}
+
+/*
+ * This is to verify that we're looking at a real local APIC.
+ * Check these against your board if the CPUs aren't getting
+ * started for no apparent reason.
+ */
+int __init verify_local_APIC(void)
+{
+       unsigned int reg0, reg1;
+
+       /*
+        * The version register is read-only in a real APIC.
+        */
+       reg0 = apic_read(APIC_LVR);
+       apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg0);
+       apic_write(APIC_LVR, reg0 ^ APIC_LVR_MASK);
+       reg1 = apic_read(APIC_LVR);
+       apic_printk(APIC_DEBUG, "Getting VERSION: %x\n", reg1);
+
+       /*
+        * The two version reads above should print the same
+        * numbers.  If the second one is different, then we
+        * poke at a non-APIC.
+        */
+       if (reg1 != reg0)
+               return 0;
+
+       /*
+        * Check if the version looks reasonably.
+        */
+       reg1 = GET_APIC_VERSION(reg0);
+       if (reg1 == 0x00 || reg1 == 0xff)
+               return 0;
+       reg1 = lapic_get_maxlvt();
+       if (reg1 < 0x02 || reg1 == 0xff)
+               return 0;
+
+       /*
+        * The ID register is read/write in a real APIC.
+        */
+       reg0 = apic_read(APIC_ID);
+       apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg0);
+       apic_write(APIC_ID, reg0 ^ APIC_ID_MASK);
+       reg1 = apic_read(APIC_ID);
+       apic_printk(APIC_DEBUG, "Getting ID: %x\n", reg1);
+       apic_write(APIC_ID, reg0);
+       if (reg1 != (reg0 ^ APIC_ID_MASK))
+               return 0;
+
+       /*
+        * The next two are just to see if we have sane values.
+        * They're only really relevant if we're in Virtual Wire
+        * compatibility mode, but most boxes are anymore.
+        */
+       reg0 = apic_read(APIC_LVT0);
+       apic_printk(APIC_DEBUG, "Getting LVT0: %x\n", reg0);
+       reg1 = apic_read(APIC_LVT1);
+       apic_printk(APIC_DEBUG, "Getting LVT1: %x\n", reg1);
+
+       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 */
+       if (modern_apic())
+               return;
+
+       /*
+        * Wait for idle.
+        */
+       apic_wait_icr_idle();
+
+       apic_printk(APIC_DEBUG, "Synchronizing Arb IDs.\n");
+       apic_write(APIC_ICR, APIC_DEST_ALLINC | APIC_INT_LEVELTRIG
+                               | APIC_DM_INIT);
+}
+
+/*
+ * An initial setup of the virtual wire mode.
+ */
+void __init init_bsp_APIC(void)
+{
+       unsigned int value;
+
+       /*
+        * Don't do the setup now if we have a SMP BIOS as the
+        * through-I/O-APIC virtual wire mode might be active.
+        */
+       if (smp_found_config || !cpu_has_apic)
+               return;
+
+       value = apic_read(APIC_LVR);
+
+       /*
+        * Do not trust the local APIC being empty at bootup.
+        */
+       clear_local_APIC();
+
+       /*
+        * Enable APIC.
+        */
+       value = apic_read(APIC_SPIV);
+       value &= ~APIC_VECTOR_MASK;
+       value |= APIC_SPIV_APIC_ENABLED;
        value |= APIC_SPIV_FOCUS_DISABLED;
        value |= SPURIOUS_APIC_VECTOR;
        apic_write(APIC_SPIV, value);
@@ -339,9 +720,12 @@ void __init init_bsp_APIC(void)
        apic_write(APIC_LVT1, value);
 }
 
-void __cpuinit setup_local_APIC (void)
+/**
+ * setup_local_APIC - setup the local APIC
+ */
+void __cpuinit setup_local_APIC(void)
 {
-       unsigned int value, maxlvt;
+       unsigned int value;
        int i, j;
 
        value = apic_read(APIC_LVR);
@@ -420,10 +804,12 @@ void __cpuinit setup_local_APIC (void)
        value = apic_read(APIC_LVT0) & APIC_LVT_MASKED;
        if (!smp_processor_id() && !value) {
                value = APIC_DM_EXTINT;
-               apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n", smp_processor_id());
+               apic_printk(APIC_VERBOSE, "enabled ExtINT on CPU#%d\n",
+                           smp_processor_id());
        } else {
                value = APIC_DM_EXTINT | APIC_LVT_MASKED;
-               apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n", smp_processor_id());
+               apic_printk(APIC_VERBOSE, "masked ExtINT on CPU#%d\n",
+                           smp_processor_id());
        }
        apic_write(APIC_LVT0, value);
 
@@ -435,183 +821,27 @@ void __cpuinit setup_local_APIC (void)
        else
                value = APIC_DM_NMI | APIC_LVT_MASKED;
        apic_write(APIC_LVT1, value);
-
-       {
-               unsigned oldvalue;
-               maxlvt = get_maxlvt();
-               oldvalue = apic_read(APIC_ESR);
-               value = ERROR_APIC_VECTOR;      // enables sending errors
-               apic_write(APIC_LVTERR, value);
-               /*
-                * spec says clear errors after enabling vector.
-                */
-               if (maxlvt > 3)
-                       apic_write(APIC_ESR, 0);
-               value = apic_read(APIC_ESR);
-               if (value != oldvalue)
-                       apic_printk(APIC_VERBOSE,
-                       "ESR value after enabling vector: %08x, after %08x\n",
-                       oldvalue, value);
-       }
-
-       nmi_watchdog_default();
-       setup_apic_nmi_watchdog(NULL);
-       apic_pm_activate();
-}
-
-#ifdef CONFIG_PM
-
-static struct {
-       /* 'active' is true if the local APIC was enabled by us and
-          not the BIOS; this signifies that we are also responsible
-          for disabling it before entering apm/acpi suspend */
-       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 = 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_INTEL
-       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 = get_maxlvt();
-
-       local_irq_save(flags);
-       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_INTEL
-       if (maxlvt >= 5)
-               apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
-#endif
-       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;
-}
-
-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 __cpuinit 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) { }
+void __cpuinit lapic_setup_esr(void)
+{
+       unsigned maxlvt = lapic_get_maxlvt();
 
-#endif /* CONFIG_PM */
+       apic_write(APIC_LVTERR, ERROR_APIC_VECTOR);
+       /*
+        * spec says clear errors after enabling vector.
+        */
+       if (maxlvt > 3)
+               apic_write(APIC_ESR, 0);
+}
 
-static int __init apic_set_verbosity(char *str)
+void __cpuinit end_local_APIC_setup(void)
 {
-       if (str == NULL)  {
-               skip_ioapic_setup = 0;
-               ioapic_force = 1;
-               return 0;
-       }
-       if (strcmp("debug", str) == 0)
-               apic_verbosity = APIC_DEBUG;
-       else if (strcmp("verbose", str) == 0)
-               apic_verbosity = APIC_VERBOSE;
-       else {
-               printk(KERN_WARNING "APIC Verbosity level %s not recognised"
-                               " use apic=verbose or apic=debug\n", str);
-               return -EINVAL;
-       }
-
-       return 0;
+       lapic_setup_esr();
+       nmi_watchdog_default();
+       setup_apic_nmi_watchdog(NULL);
+       apic_pm_activate();
 }
-early_param("apic", apic_set_verbosity);
 
 /*
  * Detect and enable local APICs on non-SMP boards.
@@ -619,8 +849,7 @@ early_param("apic", apic_set_verbosity);
  * On AMD64 we trust the BIOS - if it says no APIC it is likely
  * not correctly set up (usually the APIC timer won't work etc.)
  */
-
-static int __init detect_init_APIC (void)
+static int __init detect_init_APIC(void)
 {
        if (!cpu_has_apic) {
                printk(KERN_INFO "No local APIC present\n");
@@ -632,68 +861,11 @@ static int __init detect_init_APIC (void)
        return 0;
 }
 
-#ifdef CONFIG_X86_IO_APIC
-static struct resource * __init ioapic_setup_resources(void)
-{
-#define IOAPIC_RESOURCE_NAME_SIZE 11
-       unsigned long n;
-       struct resource *res;
-       char *mem;
-       int i;
-
-       if (nr_ioapics <= 0)
-               return NULL;
-
-       n = IOAPIC_RESOURCE_NAME_SIZE + sizeof(struct resource);
-       n *= nr_ioapics;
-
-       mem = alloc_bootmem(n);
-       res = (void *)mem;
-
-       if (mem != NULL) {
-               memset(mem, 0, n);
-               mem += sizeof(struct resource) * nr_ioapics;
-
-               for (i = 0; i < nr_ioapics; i++) {
-                       res[i].name = mem;
-                       res[i].flags = IORESOURCE_MEM | IORESOURCE_BUSY;
-                       sprintf(mem,  "IOAPIC %u", i);
-                       mem += IOAPIC_RESOURCE_NAME_SIZE;
-               }
-       }
-
-       ioapic_resources = res;
-
-       return res;
-}
-
-static int __init ioapic_insert_resources(void)
-{
-       int i;
-       struct resource *r = ioapic_resources;
-
-       if (!r) {
-               printk("IO APIC resources could be not be allocated.\n");
-               return -1;
-       }
-
-       for (i = 0; i < nr_ioapics; i++) {
-               insert_resource(&iomem_resource, r);
-               r++;
-       }
-
-       return 0;
-}
-
-/* Insert the IO APIC resources after PCI initialization has occured to handle
- * IO APICS that are mapped in on a BAR in PCI space. */
-late_initcall(ioapic_insert_resources);
-#endif
-
+/**
+ * init_apic_mappings - initialize APIC mappings
+ */
 void __init init_apic_mappings(void)
 {
-       unsigned long apic_phys;
-
        /*
         * If no local APIC can be found then set up a fake all
         * zeroes page to simulate the local APIC and another
@@ -706,8 +878,8 @@ void __init init_apic_mappings(void)
                apic_phys = mp_lapic_addr;
 
        set_fixmap_nocache(FIX_APIC_BASE, apic_phys);
-       apic_mapped = 1;
-       apic_printk(APIC_VERBOSE,"mapped APIC to %16lx (%16lx)\n", APIC_BASE, apic_phys);
+       apic_printk(APIC_VERBOSE, "mapped APIC to %16lx (%16lx)\n",
+                               APIC_BASE, apic_phys);
 
        /* Put local APIC into the resource map. */
        lapic_resource.start = apic_phys;
@@ -719,349 +891,281 @@ void __init init_apic_mappings(void)
         * default configuration (or the MP table is broken).
         */
        boot_cpu_id = GET_APIC_ID(apic_read(APIC_ID));
-
-       {
-               unsigned long ioapic_phys, idx = FIX_IO_APIC_BASE_0;
-               int i;
-               struct resource *ioapic_res;
-
-               ioapic_res = ioapic_setup_resources();
-               for (i = 0; i < nr_ioapics; i++) {
-                       if (smp_found_config) {
-                               ioapic_phys = mp_ioapics[i].mpc_apicaddr;
-                       } else {
-                               ioapic_phys = (unsigned long) alloc_bootmem_pages(PAGE_SIZE);
-                               ioapic_phys = __pa(ioapic_phys);
-                       }
-                       set_fixmap_nocache(idx, ioapic_phys);
-                       apic_printk(APIC_VERBOSE,"mapped IOAPIC to %016lx (%016lx)\n",
-                                       __fix_to_virt(idx), ioapic_phys);
-                       idx++;
-
-                       if (ioapic_res != NULL) {
-                               ioapic_res->start = ioapic_phys;
-                               ioapic_res->end = ioapic_phys + (4 * 1024) - 1;
-                               ioapic_res++;
-                       }
-               }
-       }
 }
 
 /*
- * 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.
+ * This initializes the IO-APIC and APIC hardware if this is
+ * a UP kernel.
  */
-
-#define APIC_DIVISOR 16
-
-static void __setup_APIC_LVTT(unsigned int clocks)
+int __init APIC_init_uniprocessor(void)
 {
-       unsigned int lvtt_value, tmp_value;
-       int cpu = smp_processor_id();
+       if (disable_apic) {
+               printk(KERN_INFO "Apic disabled\n");
+               return -1;
+       }
+       if (!cpu_has_apic) {
+               disable_apic = 1;
+               printk(KERN_INFO "Apic disabled by BIOS\n");
+               return -1;
+       }
 
-       lvtt_value = APIC_LVT_TIMER_PERIODIC | LOCAL_TIMER_VECTOR;
+       verify_local_APIC();
 
-       if (cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask))
-               lvtt_value |= APIC_LVT_MASKED;
+       phys_cpu_present_map = physid_mask_of_physid(boot_cpu_id);
+       apic_write(APIC_ID, SET_APIC_ID(boot_cpu_id));
 
-       apic_write(APIC_LVTT, lvtt_value);
+       setup_local_APIC();
 
        /*
-        * Divide PICLK by 16
+        * Now enable IO-APICs, actually call clear_IO_APIC
+        * We need clear_IO_APIC before enabling vector on BP
         */
-       tmp_value = apic_read(APIC_TDCR);
-       apic_write(APIC_TDCR, (tmp_value
-                               & ~(APIC_TDR_DIV_1 | APIC_TDR_DIV_TMBASE))
-                               | APIC_TDR_DIV_16);
-
-       apic_write(APIC_TMICT, clocks/APIC_DIVISOR);
-}
+       if (!skip_ioapic_setup && nr_ioapics)
+               enable_IO_APIC();
 
-static void setup_APIC_timer(unsigned int clocks)
-{
-       unsigned long flags;
-
-       local_irq_save(flags);
+       end_local_APIC_setup();
 
-       /* wait for irq slice */
-       if (hpet_address && hpet_use_timer) {
-               u32 trigger = hpet_readl(HPET_T0_CMP);
-               while (hpet_readl(HPET_T0_CMP) == trigger)
-                       /* do nothing */ ;
-       } else {
-               int c1, c2;
-               outb_p(0x00, 0x43);
-               c2 = inb_p(0x40);
-               c2 |= inb_p(0x40) << 8;
-               do {
-                       c1 = c2;
-                       outb_p(0x00, 0x43);
-                       c2 = inb_p(0x40);
-                       c2 |= inb_p(0x40) << 8;
-               } while (c2 - c1 < 300);
-       }
-       __setup_APIC_LVTT(clocks);
-       /* Turn off PIT interrupt if we use APIC timer as main timer.
-          Only works with the PM timer right now
-          TBD fix it for HPET too. */
-       if ((pmtmr_ioport != 0) &&
-               smp_processor_id() == boot_cpu_id &&
-               apic_runs_main_timer == 1 &&
-               !cpu_isset(boot_cpu_id, timer_interrupt_broadcast_ipi_mask)) {
-               stop_timer_interrupt();
-               apic_runs_main_timer++;
-       }
-       local_irq_restore(flags);
+       if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
+               setup_IO_APIC();
+       else
+               nr_ioapics = 0;
+       setup_boot_APIC_clock();
+       check_nmi_watchdog();
+       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.
+ * Local APIC interrupts
  */
 
-#define TICK_COUNT 100000000
-
-static int __init calibrate_APIC_clock(void)
+/*
+ * This interrupt should _never_ happen with our APIC/SMP architecture
+ */
+asmlinkage void smp_spurious_interrupt(void)
 {
-       unsigned apic, apic_start;
-       unsigned long tsc, tsc_start;
-       int result;
+       unsigned int v;
+       exit_idle();
+       irq_enter();
        /*
-        * Put whatever arbitrary (but long enough) timeout
-        * value into the APIC clock, we just want to get the
-        * counter running for calibration.
+        * Check if this really is a spurious interrupt and ACK it
+        * if it is a vectored one.  Just in case...
+        * Spurious interrupts should not be ACKed.
         */
-       __setup_APIC_LVTT(4000000000);
-
-       apic_start = apic_read(APIC_TMCCT);
-#ifdef CONFIG_X86_PM_TIMER
-       if (apic_calibrate_pmtmr && pmtmr_ioport) {
-               pmtimer_wait(5000);  /* 5ms wait */
-               apic = apic_read(APIC_TMCCT);
-               result = (apic_start - apic) * 1000L / 5;
-       } else
-#endif
-       {
-               rdtscll(tsc_start);
-
-               do {
-                       apic = apic_read(APIC_TMCCT);
-                       rdtscll(tsc);
-               } while ((tsc - tsc_start) < TICK_COUNT &&
-                               (apic_start - apic) < TICK_COUNT);
-
-               result = (apic_start - apic) * 1000L * tsc_khz /
-                                       (tsc - tsc_start);
-       }
-       printk("result %d\n", result);
-
-
-       printk(KERN_INFO "Detected %d.%03d MHz APIC timer.\n",
-               result / 1000 / 1000, result / 1000 % 1000);
+       v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
+       if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
+               ack_APIC_irq();
 
-       return result * APIC_DIVISOR / HZ;
+       add_pda(irq_spurious_count, 1);
+       irq_exit();
 }
 
-static unsigned int calibration_result;
-
-void __init setup_boot_APIC_clock (void)
+/*
+ * This interrupt should never happen with our APIC/SMP architecture
+ */
+asmlinkage void smp_error_interrupt(void)
 {
-       if (disable_apic_timer) {
-               printk(KERN_INFO "Disabling APIC timer\n");
-               return;
-       }
-
-       printk(KERN_INFO "Using local APIC timer interrupts.\n");
-       using_apic_timer = 1;
-
-       local_irq_disable();
+       unsigned int v, v1;
 
-       calibration_result = calibrate_APIC_clock();
-       /*
-        * Now set up the timer for real.
-        */
-       setup_APIC_timer(calibration_result);
+       exit_idle();
+       irq_enter();
+       /* First tickle the hardware, only then report what went on. -- REW */
+       v = apic_read(APIC_ESR);
+       apic_write(APIC_ESR, 0);
+       v1 = apic_read(APIC_ESR);
+       ack_APIC_irq();
+       atomic_inc(&irq_err_count);
 
-       local_irq_enable();
+       /* Here is what the APIC error bits mean:
+          0: Send CS error
+          1: Receive CS error
+          2: Send accept error
+          3: Receive accept error
+          4: Reserved
+          5: Send illegal vector
+          6: Received illegal vector
+          7: Illegal register address
+       */
+       printk(KERN_DEBUG "APIC error on CPU%d: %02x(%02x)\n",
+               smp_processor_id(), v , v1);
+       irq_exit();
 }
 
-void __cpuinit setup_secondary_APIC_clock(void)
+void disconnect_bsp_APIC(int virt_wire_setup)
 {
-       local_irq_disable(); /* FIXME: Do we need this? --RR */
-       setup_APIC_timer(calibration_result);
-       local_irq_enable();
-}
+       /* Go back to Virtual Wire compatibility mode */
+       unsigned long value;
 
-void disable_APIC_timer(void)
-{
-       if (using_apic_timer) {
-               unsigned long v;
+       /* 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(APIC_SPIV, value);
 
-               v = apic_read(APIC_LVTT);
+       if (!virt_wire_setup) {
                /*
-                * 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.
+                * For LVT0 make it edge triggered, active high,
+                * external and enabled
                 */
-               v |= (APIC_LVT_MASKED | LOCAL_TIMER_VECTOR);
-               apic_write(APIC_LVTT, v);
+               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(APIC_LVT0, value);
+       } else {
+               /* Disable LVT0 */
+               apic_write(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(APIC_LVT1, value);
 }
 
-void enable_APIC_timer(void)
+/*
+ * Power management
+ */
+#ifdef CONFIG_PM
+
+static struct {
+       /* 'active' is true if the local APIC was enabled by us and
+          not the BIOS; this signifies that we are also responsible
+          for disabling it before entering apm/acpi suspend */
+       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)
 {
-       int cpu = smp_processor_id();
-
-       if (using_apic_timer &&
-           !cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask)) {
-               unsigned long v;
+       unsigned long flags;
+       int maxlvt;
 
-               v = apic_read(APIC_LVTT);
-               apic_write(APIC_LVTT, v & ~APIC_LVT_MASKED);
-       }
-}
+       if (!apic_pm_state.active)
+               return 0;
 
-void switch_APIC_timer_to_ipi(void *cpumask)
-{
-       cpumask_t mask = *(cpumask_t *)cpumask;
-       int cpu = smp_processor_id();
+       maxlvt = lapic_get_maxlvt();
 
-       if (cpu_isset(cpu, mask) &&
-           !cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask)) {
-               disable_APIC_timer();
-               cpu_set(cpu, timer_interrupt_broadcast_ipi_mask);
-       }
+       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_INTEL
+       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;
 }
-EXPORT_SYMBOL(switch_APIC_timer_to_ipi);
 
-void smp_send_timer_broadcast_ipi(void)
+static int lapic_resume(struct sys_device *dev)
 {
-       int cpu = smp_processor_id();
-       cpumask_t mask;
+       unsigned int l, h;
+       unsigned long flags;
+       int maxlvt;
 
-       cpus_and(mask, cpu_online_map, timer_interrupt_broadcast_ipi_mask);
+       if (!apic_pm_state.active)
+               return 0;
 
-       if (cpu_isset(cpu, mask)) {
-               cpu_clear(cpu, mask);
-               add_pda(apic_timer_irqs, 1);
-               smp_local_timer_interrupt();
-       }
+       maxlvt = lapic_get_maxlvt();
 
-       if (!cpus_empty(mask)) {
-               send_IPI_mask(mask, LOCAL_TIMER_VECTOR);
-       }
+       local_irq_save(flags);
+       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_INTEL
+       if (maxlvt >= 5)
+               apic_write(APIC_LVTTHMR, apic_pm_state.apic_thmr);
+#endif
+       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;
 }
 
-void switch_ipi_to_APIC_timer(void *cpumask)
-{
-       cpumask_t mask = *(cpumask_t *)cpumask;
-       int cpu = smp_processor_id();
+static struct sysdev_class lapic_sysclass = {
+       .name           = "lapic",
+       .resume         = lapic_resume,
+       .suspend        = lapic_suspend,
+};
 
-       if (cpu_isset(cpu, mask) &&
-           cpu_isset(cpu, timer_interrupt_broadcast_ipi_mask)) {
-               cpu_clear(cpu, timer_interrupt_broadcast_ipi_mask);
-               enable_APIC_timer();
-       }
-}
-EXPORT_SYMBOL(switch_ipi_to_APIC_timer);
+static struct sys_device device_lapic = {
+       .id     = 0,
+       .cls    = &lapic_sysclass,
+};
 
-int setup_profiling_timer(unsigned int multiplier)
+static void __cpuinit apic_pm_activate(void)
 {
-       return -EINVAL;
+       apic_pm_state.active = 1;
 }
 
-void setup_APIC_extended_lvt(unsigned char lvt_off, unsigned char vector,
-                            unsigned char msg_type, unsigned char mask)
+static int __init init_lapic_sysfs(void)
 {
-       unsigned long reg = (lvt_off << 4) + K8_APIC_EXT_LVT_BASE;
-       unsigned int  v   = (mask << 16) | (msg_type << 8) | vector;
-       apic_write(reg, v);
-}
-
-#undef APIC_DIVISOR
+       int error;
 
-/*
- * 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.
- */
+       if (!cpu_has_apic)
+               return 0;
+       /* XXX: remove suspend/resume procs if !apic_pm_state.active? */
 
-void smp_local_timer_interrupt(void)
-{
-       profile_tick(CPU_PROFILING);
-#ifdef CONFIG_SMP
-       update_process_times(user_mode(get_irq_regs()));
-#endif
-       if (apic_runs_main_timer > 1 && smp_processor_id() == boot_cpu_id)
-               main_timer_handler();
-       /*
-        * We take the 'long' return path, and there every subsystem
-        * grabs the appropriate 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.
-        */
+       error = sysdev_class_register(&lapic_sysclass);
+       if (!error)
+               error = sysdev_register(&device_lapic);
+       return error;
 }
+device_initcall(init_lapic_sysfs);
 
-/*
- * 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 smp_apic_timer_interrupt(struct pt_regs *regs)
-{
-       struct pt_regs *old_regs = set_irq_regs(regs);
+#else  /* CONFIG_PM */
 
-       /*
-        * the NMI deadlock-detector uses this.
-        */
-       add_pda(apic_timer_irqs, 1);
+static void apic_pm_activate(void) { }
 
-       /*
-        * 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();
-       smp_local_timer_interrupt();
-       irq_exit();
-       set_irq_regs(old_regs);
-}
+#endif /* CONFIG_PM */
 
 /*
  * apic_is_clustered_box() -- Check if we can expect good TSC
@@ -1076,21 +1180,34 @@ __cpuinit int apic_is_clustered_box(void)
 {
        int i, clusters, zeros;
        unsigned id;
+       u16 *bios_cpu_apicid = x86_bios_cpu_apicid_early_ptr;
        DECLARE_BITMAP(clustermap, NUM_APIC_CLUSTERS);
 
        bitmap_zero(clustermap, NUM_APIC_CLUSTERS);
 
        for (i = 0; i < NR_CPUS; i++) {
-               id = bios_cpu_apicid[i];
+               /* are we being called early in kernel startup? */
+               if (bios_cpu_apicid) {
+                       id = bios_cpu_apicid[i];
+               }
+               else if (i < nr_cpu_ids) {
+                       if (cpu_present(i))
+                               id = per_cpu(x86_bios_cpu_apicid, i);
+                       else
+                               continue;
+               }
+               else
+                       break;
+
                if (id != BAD_APICID)
                        __set_bit(APIC_CLUSTERID(id), clustermap);
        }
 
        /* Problem:  Partially populated chassis may not have CPUs in some of
         * the APIC clusters they have been allocated.  Only present CPUs have
-        * bios_cpu_apicid entries, thus causing zeroes in the bitmap.  Since
-        * clusters are allocated sequentially, count zeros only if they are
-        * bounded by ones.
+        * x86_bios_cpu_apicid entries, thus causing zeroes in the bitmap.
+        * Since clusters are allocated sequentially, count zeros only if
+        * they are bounded by ones.
         */
        clusters = 0;
        zeros = 0;
@@ -1111,95 +1228,33 @@ __cpuinit int apic_is_clustered_box(void)
 }
 
 /*
- * This interrupt should _never_ happen with our APIC/SMP architecture
- */
-asmlinkage void smp_spurious_interrupt(void)
-{
-       unsigned int v;
-       exit_idle();
-       irq_enter();
-       /*
-        * Check if this really is a spurious interrupt and ACK it
-        * if it is a vectored one.  Just in case...
-        * Spurious interrupts should not be ACKed.
-        */
-       v = apic_read(APIC_ISR + ((SPURIOUS_APIC_VECTOR & ~0x1f) >> 1));
-       if (v & (1 << (SPURIOUS_APIC_VECTOR & 0x1f)))
-               ack_APIC_irq();
-
-       irq_exit();
-}
-
-/*
- * This interrupt should never happen with our APIC/SMP architecture
- */
-
-asmlinkage void smp_error_interrupt(void)
-{
-       unsigned int v, v1;
-
-       exit_idle();
-       irq_enter();
-       /* First tickle the hardware, only then report what went on. -- REW */
-       v = apic_read(APIC_ESR);
-       apic_write(APIC_ESR, 0);
-       v1 = apic_read(APIC_ESR);
-       ack_APIC_irq();
-       atomic_inc(&irq_err_count);
-
-       /* Here is what the APIC error bits mean:
-          0: Send CS error
-          1: Receive CS error
-          2: Send accept error
-          3: Receive accept error
-          4: Reserved
-          5: Send illegal vector
-          6: Received illegal vector
-          7: Illegal register address
-       */
-       printk (KERN_DEBUG "APIC error on CPU%d: %02x(%02x)\n",
-               smp_processor_id(), v , v1);
-       irq_exit();
-}
-
-int disable_apic;
-
-/*
- * This initializes the IO-APIC and APIC hardware if this is
- * a UP kernel.
+ * APIC command line parameters
  */
-int __init APIC_init_uniprocessor (void)
+static int __init apic_set_verbosity(char *str)
 {
-       if (disable_apic) {
-               printk(KERN_INFO "Apic disabled\n");
-               return -1;
+       if (str == NULL)  {
+               skip_ioapic_setup = 0;
+               ioapic_force = 1;
+               return 0;
        }
-       if (!cpu_has_apic) {
-               disable_apic = 1;
-               printk(KERN_INFO "Apic disabled by BIOS\n");
-               return -1;
+       if (strcmp("debug", str) == 0)
+               apic_verbosity = APIC_DEBUG;
+       else if (strcmp("verbose", str) == 0)
+               apic_verbosity = APIC_VERBOSE;
+       else {
+               printk(KERN_WARNING "APIC Verbosity level %s not recognised"
+                               " use apic=verbose or apic=debug\n", str);
+               return -EINVAL;
        }
 
-       verify_local_APIC();
-
-       phys_cpu_present_map = physid_mask_of_physid(boot_cpu_id);
-       apic_write(APIC_ID, SET_APIC_ID(boot_cpu_id));
-
-       setup_local_APIC();
-
-       if (smp_found_config && !skip_ioapic_setup && nr_ioapics)
-               setup_IO_APIC();
-       else
-               nr_ioapics = 0;
-       setup_boot_APIC_clock();
-       check_nmi_watchdog();
        return 0;
 }
+early_param("apic", apic_set_verbosity);
 
 static __init int setup_disableapic(char *str)
 {
        disable_apic = 1;
-       clear_bit(X86_FEATURE_APIC, boot_cpu_data.x86_capability);
+       clear_cpu_cap(&boot_cpu_data, X86_FEATURE_APIC);
        return 0;
 }
 early_param("disableapic", setup_disableapic);
@@ -1225,29 +1280,13 @@ static __init int setup_noapictimer(char *str)
        disable_apic_timer = 1;
        return 1;
 }
-
-static __init int setup_apicmaintimer(char *str)
-{
-       apic_runs_main_timer = 1;
-       nohpet = 1;
-       return 1;
-}
-__setup("apicmaintimer", setup_apicmaintimer);
-
-static __init int setup_noapicmaintimer(char *str)
-{
-       apic_runs_main_timer = -1;
-       return 1;
-}
-__setup("noapicmaintimer", setup_noapicmaintimer);
+__setup("noapictimer", setup_noapictimer);
 
 static __init int setup_apicpmtimer(char *s)
 {
        apic_calibrate_pmtmr = 1;
        notsc_setup(NULL);
-       return setup_apicmaintimer(NULL);
+       return 0;
 }
 __setup("apicpmtimer", setup_apicpmtimer);
 
-__setup("noapictimer", setup_noapictimer);
-