KVM: MMU: cleanup for function unaccount_shadowed()
[safe/jmp/linux-2.6] / arch / x86 / kvm / i8254.c
index e665d1c..0150aff 100644 (file)
  *   Based on QEMU and Xen.
  */
 
+#define pr_fmt(fmt) "pit: " fmt
+
 #include <linux/kvm_host.h>
+#include <linux/slab.h>
 
 #include "irq.h"
 #include "i8254.h"
@@ -98,6 +101,40 @@ static int pit_get_gate(struct kvm *kvm, int channel)
        return kvm->arch.vpit->pit_state.channels[channel].gate;
 }
 
+static s64 __kpit_elapsed(struct kvm *kvm)
+{
+       s64 elapsed;
+       ktime_t remaining;
+       struct kvm_kpit_state *ps = &kvm->arch.vpit->pit_state;
+
+       if (!ps->pit_timer.period)
+               return 0;
+
+       /*
+        * The Counter does not stop when it reaches zero. In
+        * Modes 0, 1, 4, and 5 the Counter ``wraps around'' to
+        * the highest count, either FFFF hex for binary counting
+        * or 9999 for BCD counting, and continues counting.
+        * Modes 2 and 3 are periodic; the Counter reloads
+        * itself with the initial count and continues counting
+        * from there.
+        */
+       remaining = hrtimer_get_remaining(&ps->pit_timer.timer);
+       elapsed = ps->pit_timer.period - ktime_to_ns(remaining);
+       elapsed = mod_64(elapsed, ps->pit_timer.period);
+
+       return elapsed;
+}
+
+static s64 kpit_elapsed(struct kvm *kvm, struct kvm_kpit_channel_state *c,
+                       int channel)
+{
+       if (channel == 0)
+               return __kpit_elapsed(kvm);
+
+       return ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
+}
+
 static int pit_get_count(struct kvm *kvm, int channel)
 {
        struct kvm_kpit_channel_state *c =
@@ -107,7 +144,7 @@ static int pit_get_count(struct kvm *kvm, int channel)
 
        WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
 
-       t = ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
+       t = kpit_elapsed(kvm, c, channel);
        d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
 
        switch (c->mode) {
@@ -137,7 +174,7 @@ static int pit_get_out(struct kvm *kvm, int channel)
 
        WARN_ON(!mutex_is_locked(&kvm->arch.vpit->pit_state.lock));
 
-       t = ktime_to_ns(ktime_sub(ktime_get(), c->count_load_time));
+       t = kpit_elapsed(kvm, c, channel);
        d = muldiv64(t, KVM_PIT_FREQ, NSEC_PER_SEC);
 
        switch (c->mode) {
@@ -193,30 +230,11 @@ static void pit_latch_status(struct kvm *kvm, int channel)
        }
 }
 
-static int __pit_timer_fn(struct kvm_kpit_state *ps)
-{
-       struct kvm_vcpu *vcpu0 = ps->pit->kvm->vcpus[0];
-       struct kvm_kpit_timer *pt = &ps->pit_timer;
-
-       if (!atomic_inc_and_test(&pt->pending))
-               set_bit(KVM_REQ_PENDING_TIMER, &vcpu0->requests);
-
-       if (vcpu0 && waitqueue_active(&vcpu0->wq))
-               wake_up_interruptible(&vcpu0->wq);
-
-       hrtimer_add_expires_ns(&pt->timer, pt->period);
-       pt->scheduled = hrtimer_get_expires_ns(&pt->timer);
-       if (pt->period)
-               ps->channels[0].count_load_time = hrtimer_get_expires(&pt->timer);
-
-       return (pt->period == 0 ? 0 : 1);
-}
-
 int pit_has_pending_timer(struct kvm_vcpu *vcpu)
 {
        struct kvm_pit *pit = vcpu->kvm->arch.vpit;
 
-       if (pit && vcpu->vcpu_id == 0 && pit->pit_state.irq_ack)
+       if (pit && kvm_vcpu_is_bsp(vcpu) && pit->pit_state.irq_ack)
                return atomic_read(&pit->pit_state.pit_timer.pending);
        return 0;
 }
@@ -225,26 +243,11 @@ static void kvm_pit_ack_irq(struct kvm_irq_ack_notifier *kian)
 {
        struct kvm_kpit_state *ps = container_of(kian, struct kvm_kpit_state,
                                                 irq_ack_notifier);
-       spin_lock(&ps->inject_lock);
+       raw_spin_lock(&ps->inject_lock);
        if (atomic_dec_return(&ps->pit_timer.pending) < 0)
                atomic_inc(&ps->pit_timer.pending);
        ps->irq_ack = 1;
-       spin_unlock(&ps->inject_lock);
-}
-
-static enum hrtimer_restart pit_timer_fn(struct hrtimer *data)
-{
-       struct kvm_kpit_state *ps;
-       int restart_timer = 0;
-
-       ps = container_of(data, struct kvm_kpit_state, pit_timer.timer);
-
-       restart_timer = __pit_timer_fn(ps);
-
-       if (restart_timer)
-               return HRTIMER_RESTART;
-       else
-               return HRTIMER_NORESTART;
+       raw_spin_unlock(&ps->inject_lock);
 }
 
 void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
@@ -252,7 +255,7 @@ void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
        struct kvm_pit *pit = vcpu->kvm->arch.vpit;
        struct hrtimer *timer;
 
-       if (vcpu->vcpu_id != 0 || !pit)
+       if (!kvm_vcpu_is_bsp(vcpu) || !pit)
                return;
 
        timer = &pit->pit_state.pit_timer.timer;
@@ -260,25 +263,42 @@ void __kvm_migrate_pit_timer(struct kvm_vcpu *vcpu)
                hrtimer_start_expires(timer, HRTIMER_MODE_ABS);
 }
 
-static void destroy_pit_timer(struct kvm_kpit_timer *pt)
+static void destroy_pit_timer(struct kvm_timer *pt)
 {
-       pr_debug("pit: execute del timer!\n");
+       pr_debug("execute del timer!\n");
        hrtimer_cancel(&pt->timer);
 }
 
+static bool kpit_is_periodic(struct kvm_timer *ktimer)
+{
+       struct kvm_kpit_state *ps = container_of(ktimer, struct kvm_kpit_state,
+                                                pit_timer);
+       return ps->is_periodic;
+}
+
+static struct kvm_timer_ops kpit_ops = {
+       .is_periodic = kpit_is_periodic,
+};
+
 static void create_pit_timer(struct kvm_kpit_state *ps, u32 val, int is_period)
 {
-       struct kvm_kpit_timer *pt = &ps->pit_timer;
+       struct kvm_timer *pt = &ps->pit_timer;
        s64 interval;
 
        interval = muldiv64(val, NSEC_PER_SEC, KVM_PIT_FREQ);
 
-       pr_debug("pit: create pit timer, interval is %llu nsec\n", interval);
+       pr_debug("create pit timer, interval is %llu nsec\n", interval);
 
        /* TODO The new value only affected after the retriggered */
        hrtimer_cancel(&pt->timer);
-       pt->period = (is_period == 0) ? 0 : interval;
-       pt->timer.function = pit_timer_fn;
+       pt->period = interval;
+       ps->is_periodic = is_period;
+
+       pt->timer.function = kvm_timer_fn;
+       pt->t_ops = &kpit_ops;
+       pt->kvm = ps->pit->kvm;
+       pt->vcpu = pt->kvm->bsp_vcpu;
+
        atomic_set(&pt->pending, 0);
        ps->irq_ack = 1;
 
@@ -292,56 +312,85 @@ static void pit_load_count(struct kvm *kvm, int channel, u32 val)
 
        WARN_ON(!mutex_is_locked(&ps->lock));
 
-       pr_debug("pit: load_count val is %d, channel is %d\n", val, channel);
+       pr_debug("load_count val is %d, channel is %d\n", val, channel);
 
        /*
-        * Though spec said the state of 8254 is undefined after power-up,
-        * seems some tricky OS like Windows XP depends on IRQ0 interrupt
-        * when booting up.
-        * So here setting initialize rate for it, and not a specific number
+        * The largest possible initial count is 0; this is equivalent
+        * to 216 for binary counting and 104 for BCD counting.
         */
        if (val == 0)
                val = 0x10000;
 
-       ps->channels[channel].count_load_time = ktime_get();
        ps->channels[channel].count = val;
 
-       if (channel != 0)
+       if (channel != 0) {
+               ps->channels[channel].count_load_time = ktime_get();
                return;
+       }
 
        /* Two types of timer
         * mode 1 is one shot, mode 2 is period, otherwise del timer */
        switch (ps->channels[0].mode) {
+       case 0:
        case 1:
         /* FIXME: enhance mode 4 precision */
        case 4:
-               create_pit_timer(ps, val, 0);
+               if (!(ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)) {
+                       create_pit_timer(ps, val, 0);
+               }
                break;
        case 2:
        case 3:
-               create_pit_timer(ps, val, 1);
+               if (!(ps->flags & KVM_PIT_FLAGS_HPET_LEGACY)){
+                       create_pit_timer(ps, val, 1);
+               }
                break;
        default:
                destroy_pit_timer(&ps->pit_timer);
        }
 }
 
-void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val)
+void kvm_pit_load_count(struct kvm *kvm, int channel, u32 val, int hpet_legacy_start)
+{
+       u8 saved_mode;
+       if (hpet_legacy_start) {
+               /* save existing mode for later reenablement */
+               saved_mode = kvm->arch.vpit->pit_state.channels[0].mode;
+               kvm->arch.vpit->pit_state.channels[0].mode = 0xff; /* disable timer */
+               pit_load_count(kvm, channel, val);
+               kvm->arch.vpit->pit_state.channels[0].mode = saved_mode;
+       } else {
+               pit_load_count(kvm, channel, val);
+       }
+}
+
+static inline struct kvm_pit *dev_to_pit(struct kvm_io_device *dev)
+{
+       return container_of(dev, struct kvm_pit, dev);
+}
+
+static inline struct kvm_pit *speaker_to_pit(struct kvm_io_device *dev)
 {
-       mutex_lock(&kvm->arch.vpit->pit_state.lock);
-       pit_load_count(kvm, channel, val);
-       mutex_unlock(&kvm->arch.vpit->pit_state.lock);
+       return container_of(dev, struct kvm_pit, speaker_dev);
 }
 
-static void pit_ioport_write(struct kvm_io_device *this,
-                            gpa_t addr, int len, const void *data)
+static inline int pit_in_range(gpa_t addr)
 {
-       struct kvm_pit *pit = (struct kvm_pit *)this->private;
+       return ((addr >= KVM_PIT_BASE_ADDRESS) &&
+               (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
+}
+
+static int pit_ioport_write(struct kvm_io_device *this,
+                           gpa_t addr, int len, const void *data)
+{
+       struct kvm_pit *pit = dev_to_pit(this);
        struct kvm_kpit_state *pit_state = &pit->pit_state;
        struct kvm *kvm = pit->kvm;
        int channel, access;
        struct kvm_kpit_channel_state *s;
        u32 val = *(u32 *) data;
+       if (!pit_in_range(addr))
+               return -EOPNOTSUPP;
 
        val  &= 0xff;
        addr &= KVM_PIT_CHANNEL_MASK;
@@ -349,8 +398,8 @@ static void pit_ioport_write(struct kvm_io_device *this,
        mutex_lock(&pit_state->lock);
 
        if (val != 0)
-               pr_debug("pit: write addr is 0x%x, len is %d, val is 0x%x\n",
-                         (unsigned int)addr, len, val);
+               pr_debug("write addr is 0x%x, len is %d, val is 0x%x\n",
+                        (unsigned int)addr, len, val);
 
        if (addr == 3) {
                channel = val >> 6;
@@ -404,18 +453,24 @@ static void pit_ioport_write(struct kvm_io_device *this,
        }
 
        mutex_unlock(&pit_state->lock);
+       return 0;
 }
 
-static void pit_ioport_read(struct kvm_io_device *this,
-                           gpa_t addr, int len, void *data)
+static int pit_ioport_read(struct kvm_io_device *this,
+                          gpa_t addr, int len, void *data)
 {
-       struct kvm_pit *pit = (struct kvm_pit *)this->private;
+       struct kvm_pit *pit = dev_to_pit(this);
        struct kvm_kpit_state *pit_state = &pit->pit_state;
        struct kvm *kvm = pit->kvm;
        int ret, count;
        struct kvm_kpit_channel_state *s;
+       if (!pit_in_range(addr))
+               return -EOPNOTSUPP;
 
        addr &= KVM_PIT_CHANNEL_MASK;
+       if (addr == 3)
+               return 0;
+
        s = &pit_state->channels[addr];
 
        mutex_lock(&pit_state->lock);
@@ -468,37 +523,36 @@ static void pit_ioport_read(struct kvm_io_device *this,
        memcpy(data, (char *)&ret, len);
 
        mutex_unlock(&pit_state->lock);
+       return 0;
 }
 
-static int pit_in_range(struct kvm_io_device *this, gpa_t addr,
-                       int len, int is_write)
-{
-       return ((addr >= KVM_PIT_BASE_ADDRESS) &&
-               (addr < KVM_PIT_BASE_ADDRESS + KVM_PIT_MEM_LENGTH));
-}
-
-static void speaker_ioport_write(struct kvm_io_device *this,
-                                gpa_t addr, int len, const void *data)
+static int speaker_ioport_write(struct kvm_io_device *this,
+                               gpa_t addr, int len, const void *data)
 {
-       struct kvm_pit *pit = (struct kvm_pit *)this->private;
+       struct kvm_pit *pit = speaker_to_pit(this);
        struct kvm_kpit_state *pit_state = &pit->pit_state;
        struct kvm *kvm = pit->kvm;
        u32 val = *(u32 *) data;
+       if (addr != KVM_SPEAKER_BASE_ADDRESS)
+               return -EOPNOTSUPP;
 
        mutex_lock(&pit_state->lock);
        pit_state->speaker_data_on = (val >> 1) & 1;
        pit_set_gate(kvm, 2, val & 1);
        mutex_unlock(&pit_state->lock);
+       return 0;
 }
 
-static void speaker_ioport_read(struct kvm_io_device *this,
-                               gpa_t addr, int len, void *data)
+static int speaker_ioport_read(struct kvm_io_device *this,
+                              gpa_t addr, int len, void *data)
 {
-       struct kvm_pit *pit = (struct kvm_pit *)this->private;
+       struct kvm_pit *pit = speaker_to_pit(this);
        struct kvm_kpit_state *pit_state = &pit->pit_state;
        struct kvm *kvm = pit->kvm;
        unsigned int refresh_clock;
        int ret;
+       if (addr != KVM_SPEAKER_BASE_ADDRESS)
+               return -EOPNOTSUPP;
 
        /* Refresh clock toggles at about 15us. We approximate as 2^14ns. */
        refresh_clock = ((unsigned int)ktime_to_ns(ktime_get()) >> 14) & 1;
@@ -510,12 +564,7 @@ static void speaker_ioport_read(struct kvm_io_device *this,
                len = sizeof(ret);
        memcpy(data, (char *)&ret, len);
        mutex_unlock(&pit_state->lock);
-}
-
-static int speaker_in_range(struct kvm_io_device *this, gpa_t addr,
-                           int len, int is_write)
-{
-       return (addr == KVM_SPEAKER_BASE_ADDRESS);
+       return 0;
 }
 
 void kvm_pit_reset(struct kvm_pit *pit)
@@ -524,6 +573,7 @@ void kvm_pit_reset(struct kvm_pit *pit)
        struct kvm_kpit_channel_state *c;
 
        mutex_lock(&pit->pit_state.lock);
+       pit->pit_state.flags = 0;
        for (i = 0; i < 3; i++) {
                c = &pit->pit_state.channels[i];
                c->mode = 0xff;
@@ -536,18 +586,38 @@ void kvm_pit_reset(struct kvm_pit *pit)
        pit->pit_state.irq_ack = 1;
 }
 
-struct kvm_pit *kvm_create_pit(struct kvm *kvm)
+static void pit_mask_notifer(struct kvm_irq_mask_notifier *kimn, bool mask)
+{
+       struct kvm_pit *pit = container_of(kimn, struct kvm_pit, mask_notifier);
+
+       if (!mask) {
+               atomic_set(&pit->pit_state.pit_timer.pending, 0);
+               pit->pit_state.irq_ack = 1;
+       }
+}
+
+static const struct kvm_io_device_ops pit_dev_ops = {
+       .read     = pit_ioport_read,
+       .write    = pit_ioport_write,
+};
+
+static const struct kvm_io_device_ops speaker_dev_ops = {
+       .read     = speaker_ioport_read,
+       .write    = speaker_ioport_write,
+};
+
+/* Caller must hold slots_lock */
+struct kvm_pit *kvm_create_pit(struct kvm *kvm, u32 flags)
 {
        struct kvm_pit *pit;
        struct kvm_kpit_state *pit_state;
+       int ret;
 
        pit = kzalloc(sizeof(struct kvm_pit), GFP_KERNEL);
        if (!pit)
                return NULL;
 
-       mutex_lock(&kvm->lock);
        pit->irq_source_id = kvm_request_irq_source_id(kvm);
-       mutex_unlock(&kvm->lock);
        if (pit->irq_source_id < 0) {
                kfree(pit);
                return NULL;
@@ -555,20 +625,7 @@ struct kvm_pit *kvm_create_pit(struct kvm *kvm)
 
        mutex_init(&pit->pit_state.lock);
        mutex_lock(&pit->pit_state.lock);
-       spin_lock_init(&pit->pit_state.inject_lock);
-
-       /* Initialize PIO device */
-       pit->dev.read = pit_ioport_read;
-       pit->dev.write = pit_ioport_write;
-       pit->dev.in_range = pit_in_range;
-       pit->dev.private = pit;
-       kvm_io_bus_register_dev(&kvm->pio_bus, &pit->dev);
-
-       pit->speaker_dev.read = speaker_ioport_read;
-       pit->speaker_dev.write = speaker_ioport_write;
-       pit->speaker_dev.in_range = speaker_in_range;
-       pit->speaker_dev.private = pit;
-       kvm_io_bus_register_dev(&kvm->pio_bus, &pit->speaker_dev);
+       raw_spin_lock_init(&pit->pit_state.inject_lock);
 
        kvm->arch.vpit = pit;
        pit->kvm = kvm;
@@ -580,11 +637,39 @@ struct kvm_pit *kvm_create_pit(struct kvm *kvm)
        pit_state->irq_ack_notifier.gsi = 0;
        pit_state->irq_ack_notifier.irq_acked = kvm_pit_ack_irq;
        kvm_register_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
+       pit_state->pit_timer.reinject = true;
        mutex_unlock(&pit->pit_state.lock);
 
        kvm_pit_reset(pit);
 
+       pit->mask_notifier.func = pit_mask_notifer;
+       kvm_register_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
+
+       kvm_iodevice_init(&pit->dev, &pit_dev_ops);
+       ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS, &pit->dev);
+       if (ret < 0)
+               goto fail;
+
+       if (flags & KVM_PIT_SPEAKER_DUMMY) {
+               kvm_iodevice_init(&pit->speaker_dev, &speaker_dev_ops);
+               ret = kvm_io_bus_register_dev(kvm, KVM_PIO_BUS,
+                                               &pit->speaker_dev);
+               if (ret < 0)
+                       goto fail_unregister;
+       }
+
        return pit;
+
+fail_unregister:
+       kvm_io_bus_unregister_dev(kvm, KVM_PIO_BUS, &pit->dev);
+
+fail:
+       kvm_unregister_irq_mask_notifier(kvm, 0, &pit->mask_notifier);
+       kvm_unregister_irq_ack_notifier(kvm, &pit_state->irq_ack_notifier);
+       kvm_free_irq_source_id(kvm, pit->irq_source_id);
+
+       kfree(pit);
+       return NULL;
 }
 
 void kvm_free_pit(struct kvm *kvm)
@@ -592,6 +677,10 @@ void kvm_free_pit(struct kvm *kvm)
        struct hrtimer *timer;
 
        if (kvm->arch.vpit) {
+               kvm_unregister_irq_mask_notifier(kvm, 0,
+                                              &kvm->arch.vpit->mask_notifier);
+               kvm_unregister_irq_ack_notifier(kvm,
+                               &kvm->arch.vpit->pit_state.irq_ack_notifier);
                mutex_lock(&kvm->arch.vpit->pit_state.lock);
                timer = &kvm->arch.vpit->pit_state.pit_timer.timer;
                hrtimer_cancel(timer);
@@ -606,10 +695,8 @@ static void __inject_pit_timer_intr(struct kvm *kvm)
        struct kvm_vcpu *vcpu;
        int i;
 
-       mutex_lock(&kvm->lock);
        kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 1);
        kvm_set_irq(kvm, kvm->arch.vpit->irq_source_id, 0, 0);
-       mutex_unlock(&kvm->lock);
 
        /*
         * Provides NMI watchdog support via Virtual Wire mode.
@@ -621,11 +708,8 @@ static void __inject_pit_timer_intr(struct kvm *kvm)
         * VCPU0, and only if its LVT0 is in EXTINT mode.
         */
        if (kvm->arch.vapics_in_nmi_mode > 0)
-               for (i = 0; i < KVM_MAX_VCPUS; ++i) {
-                       vcpu = kvm->vcpus[i];
-                       if (vcpu)
-                               kvm_apic_nmi_wd_deliver(vcpu);
-               }
+               kvm_for_each_vcpu(i, vcpu, kvm)
+                       kvm_apic_nmi_wd_deliver(vcpu);
 }
 
 void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu)
@@ -634,19 +718,19 @@ void kvm_inject_pit_timer_irqs(struct kvm_vcpu *vcpu)
        struct kvm *kvm = vcpu->kvm;
        struct kvm_kpit_state *ps;
 
-       if (vcpu && pit) {
+       if (pit) {
                int inject = 0;
                ps = &pit->pit_state;
 
                /* Try to inject pending interrupts when
                 * last one has been acked.
                 */
-               spin_lock(&ps->inject_lock);
+               raw_spin_lock(&ps->inject_lock);
                if (atomic_read(&ps->pit_timer.pending) && ps->irq_ack) {
                        ps->irq_ack = 0;
                        inject = 1;
                }
-               spin_unlock(&ps->inject_lock);
+               raw_spin_unlock(&ps->inject_lock);
                if (inject)
                        __inject_pit_timer_intr(kvm);
        }