Merge tag 'iwlwifi-next-for-kalle-2014-12-30' of https://git.kernel.org/pub/scm/linux...
[cascardo/linux.git] / arch / x86 / kvm / lapic.c
index b8345dd..4f0c0b9 100644 (file)
@@ -68,6 +68,9 @@
 #define MAX_APIC_VECTOR                        256
 #define APIC_VECTORS_PER_REG           32
 
+#define APIC_BROADCAST                 0xFF
+#define X2APIC_BROADCAST               0xFFFFFFFFul
+
 #define VEC_POS(v) ((v) & (32 - 1))
 #define REG_POS(v) (((v) >> 5) << 4)
 
@@ -129,8 +132,6 @@ static inline int kvm_apic_id(struct kvm_lapic *apic)
        return (kvm_apic_get_reg(apic, APIC_ID) >> 24) & 0xff;
 }
 
-#define KVM_X2APIC_CID_BITS 0
-
 static void recalculate_apic_map(struct kvm *kvm)
 {
        struct kvm_apic_map *new, *old = NULL;
@@ -149,42 +150,56 @@ static void recalculate_apic_map(struct kvm *kvm)
        new->cid_shift = 8;
        new->cid_mask = 0;
        new->lid_mask = 0xff;
+       new->broadcast = APIC_BROADCAST;
 
        kvm_for_each_vcpu(i, vcpu, kvm) {
                struct kvm_lapic *apic = vcpu->arch.apic;
-               u16 cid, lid;
-               u32 ldr;
 
                if (!kvm_apic_present(vcpu))
                        continue;
 
+               if (apic_x2apic_mode(apic)) {
+                       new->ldr_bits = 32;
+                       new->cid_shift = 16;
+                       new->cid_mask = new->lid_mask = 0xffff;
+                       new->broadcast = X2APIC_BROADCAST;
+               } else if (kvm_apic_get_reg(apic, APIC_LDR)) {
+                       if (kvm_apic_get_reg(apic, APIC_DFR) ==
+                                                       APIC_DFR_CLUSTER) {
+                               new->cid_shift = 4;
+                               new->cid_mask = 0xf;
+                               new->lid_mask = 0xf;
+                       } else {
+                               new->cid_shift = 8;
+                               new->cid_mask = 0;
+                               new->lid_mask = 0xff;
+                       }
+               }
+
                /*
                 * All APICs have to be configured in the same mode by an OS.
                 * We take advatage of this while building logical id loockup
-                * table. After reset APICs are in xapic/flat mode, so if we
-                * find apic with different setting we assume this is the mode
+                * table. After reset APICs are in software disabled mode, so if
+                * we find apic with different setting we assume this is the mode
                 * OS wants all apics to be in; build lookup table accordingly.
                 */
-               if (apic_x2apic_mode(apic)) {
-                       new->ldr_bits = 32;
-                       new->cid_shift = 16;
-                       new->cid_mask = (1 << KVM_X2APIC_CID_BITS) - 1;
-                       new->lid_mask = 0xffff;
-               } else if (kvm_apic_sw_enabled(apic) &&
-                               !new->cid_mask /* flat mode */ &&
-                               kvm_apic_get_reg(apic, APIC_DFR) == APIC_DFR_CLUSTER) {
-                       new->cid_shift = 4;
-                       new->cid_mask = 0xf;
-                       new->lid_mask = 0xf;
-               }
+               if (kvm_apic_sw_enabled(apic))
+                       break;
+       }
 
-               new->phys_map[kvm_apic_id(apic)] = apic;
+       kvm_for_each_vcpu(i, vcpu, kvm) {
+               struct kvm_lapic *apic = vcpu->arch.apic;
+               u16 cid, lid;
+               u32 ldr, aid;
 
+               aid = kvm_apic_id(apic);
                ldr = kvm_apic_get_reg(apic, APIC_LDR);
                cid = apic_cluster_id(new, ldr);
                lid = apic_logical_id(new, ldr);
 
-               if (lid)
+               if (aid < ARRAY_SIZE(new->phys_map))
+                       new->phys_map[aid] = apic;
+               if (lid && cid < ARRAY_SIZE(new->logical_map))
                        new->logical_map[cid][ffs(lid) - 1] = apic;
        }
 out:
@@ -201,11 +216,13 @@ out:
 
 static inline void apic_set_spiv(struct kvm_lapic *apic, u32 val)
 {
-       u32 prev = kvm_apic_get_reg(apic, APIC_SPIV);
+       bool enabled = val & APIC_SPIV_APIC_ENABLED;
 
        apic_set_reg(apic, APIC_SPIV, val);
-       if ((prev ^ val) & APIC_SPIV_APIC_ENABLED) {
-               if (val & APIC_SPIV_APIC_ENABLED) {
+
+       if (enabled != apic->sw_enabled) {
+               apic->sw_enabled = enabled;
+               if (enabled) {
                        static_key_slow_dec_deferred(&apic_sw_disabled);
                        recalculate_apic_map(apic->vcpu->kvm);
                } else
@@ -237,21 +254,17 @@ static inline int apic_lvt_vector(struct kvm_lapic *apic, int lvt_type)
 
 static inline int apic_lvtt_oneshot(struct kvm_lapic *apic)
 {
-       return ((kvm_apic_get_reg(apic, APIC_LVTT) &
-               apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_ONESHOT);
+       return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_ONESHOT;
 }
 
 static inline int apic_lvtt_period(struct kvm_lapic *apic)
 {
-       return ((kvm_apic_get_reg(apic, APIC_LVTT) &
-               apic->lapic_timer.timer_mode_mask) == APIC_LVT_TIMER_PERIODIC);
+       return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_PERIODIC;
 }
 
 static inline int apic_lvtt_tscdeadline(struct kvm_lapic *apic)
 {
-       return ((kvm_apic_get_reg(apic, APIC_LVTT) &
-               apic->lapic_timer.timer_mode_mask) ==
-                       APIC_LVT_TIMER_TSCDEADLINE);
+       return apic->lapic_timer.timer_mode == APIC_LVT_TIMER_TSCDEADLINE;
 }
 
 static inline int apic_lvt_nmi_mode(u32 lvt_val)
@@ -326,8 +339,12 @@ EXPORT_SYMBOL_GPL(kvm_apic_update_irr);
 
 static inline void apic_set_irr(int vec, struct kvm_lapic *apic)
 {
-       apic->irr_pending = true;
        apic_set_vector(vec, apic->regs + APIC_IRR);
+       /*
+        * irr_pending must be true if any interrupt is pending; set it after
+        * APIC_IRR to avoid race with apic_clear_irr
+        */
+       apic->irr_pending = true;
 }
 
 static inline int apic_search_irr(struct kvm_lapic *apic)
@@ -359,13 +376,15 @@ static inline void apic_clear_irr(int vec, struct kvm_lapic *apic)
 
        vcpu = apic->vcpu;
 
-       apic_clear_vector(vec, apic->regs + APIC_IRR);
-       if (unlikely(kvm_apic_vid_enabled(vcpu->kvm)))
+       if (unlikely(kvm_apic_vid_enabled(vcpu->kvm))) {
                /* try to update RVI */
+               apic_clear_vector(vec, apic->regs + APIC_IRR);
                kvm_make_request(KVM_REQ_EVENT, vcpu);
-       else {
-               vec = apic_search_irr(apic);
-               apic->irr_pending = (vec != -1);
+       } else {
+               apic->irr_pending = false;
+               apic_clear_vector(vec, apic->regs + APIC_IRR);
+               if (apic_search_irr(apic) != -1)
+                       apic->irr_pending = true;
        }
 }
 
@@ -558,16 +577,25 @@ static void apic_set_tpr(struct kvm_lapic *apic, u32 tpr)
        apic_update_ppr(apic);
 }
 
-int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u16 dest)
+static int kvm_apic_broadcast(struct kvm_lapic *apic, u32 dest)
+{
+       return dest == (apic_x2apic_mode(apic) ?
+                       X2APIC_BROADCAST : APIC_BROADCAST);
+}
+
+int kvm_apic_match_physical_addr(struct kvm_lapic *apic, u32 dest)
 {
-       return dest == 0xff || kvm_apic_id(apic) == dest;
+       return kvm_apic_id(apic) == dest || kvm_apic_broadcast(apic, dest);
 }
 
-int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
+int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u32 mda)
 {
        int result = 0;
        u32 logical_id;
 
+       if (kvm_apic_broadcast(apic, mda))
+               return 1;
+
        if (apic_x2apic_mode(apic)) {
                logical_id = kvm_apic_get_reg(apic, APIC_LDR);
                return logical_id & mda;
@@ -595,7 +623,7 @@ int kvm_apic_match_logical_addr(struct kvm_lapic *apic, u8 mda)
 }
 
 int kvm_apic_match_dest(struct kvm_vcpu *vcpu, struct kvm_lapic *source,
-                          int short_hand, int dest, int dest_mode)
+                          int short_hand, unsigned int dest, int dest_mode)
 {
        int result = 0;
        struct kvm_lapic *target = vcpu->arch.apic;
@@ -657,15 +685,24 @@ bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
        if (!map)
                goto out;
 
+       if (irq->dest_id == map->broadcast)
+               goto out;
+
+       ret = true;
+
        if (irq->dest_mode == 0) { /* physical mode */
-               if (irq->delivery_mode == APIC_DM_LOWEST ||
-                               irq->dest_id == 0xff)
+               if (irq->dest_id >= ARRAY_SIZE(map->phys_map))
                        goto out;
-               dst = &map->phys_map[irq->dest_id & 0xff];
+
+               dst = &map->phys_map[irq->dest_id];
        } else {
                u32 mda = irq->dest_id << (32 - map->ldr_bits);
+               u16 cid = apic_cluster_id(map, mda);
+
+               if (cid >= ARRAY_SIZE(map->logical_map))
+                       goto out;
 
-               dst = map->logical_map[apic_cluster_id(map, mda)];
+               dst = map->logical_map[cid];
 
                bitmap = apic_logical_id(map, mda);
 
@@ -691,8 +728,6 @@ bool kvm_irq_delivery_to_apic_fast(struct kvm *kvm, struct kvm_lapic *src,
                        *r = 0;
                *r += kvm_apic_set_irq(dst[i]->vcpu, irq, dest_map);
        }
-
-       ret = true;
 out:
        rcu_read_unlock();
        return ret;
@@ -1034,6 +1069,26 @@ static void update_divide_count(struct kvm_lapic *apic)
                                   apic->divide_count);
 }
 
+static void apic_timer_expired(struct kvm_lapic *apic)
+{
+       struct kvm_vcpu *vcpu = apic->vcpu;
+       wait_queue_head_t *q = &vcpu->wq;
+
+       /*
+        * Note: KVM_REQ_PENDING_TIMER is implicitly checked in
+        * vcpu_enter_guest.
+        */
+       if (atomic_read(&apic->lapic_timer.pending))
+               return;
+
+       atomic_inc(&apic->lapic_timer.pending);
+       /* FIXME: this code should not know anything about vcpus */
+       kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
+
+       if (waitqueue_active(q))
+               wake_up_interruptible(q);
+}
+
 static void start_apic_timer(struct kvm_lapic *apic)
 {
        ktime_t now;
@@ -1096,9 +1151,10 @@ static void start_apic_timer(struct kvm_lapic *apic)
                if (likely(tscdeadline > guest_tsc)) {
                        ns = (tscdeadline - guest_tsc) * 1000000ULL;
                        do_div(ns, this_tsc_khz);
-               }
-               hrtimer_start(&apic->lapic_timer.timer,
-                       ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
+                       hrtimer_start(&apic->lapic_timer.timer,
+                               ktime_add_ns(now, ns), HRTIMER_MODE_ABS);
+               } else
+                       apic_timer_expired(apic);
 
                local_irq_restore(flags);
        }
@@ -1203,17 +1259,20 @@ static int apic_reg_write(struct kvm_lapic *apic, u32 reg, u32 val)
 
                break;
 
-       case APIC_LVTT:
-               if ((kvm_apic_get_reg(apic, APIC_LVTT) &
-                   apic->lapic_timer.timer_mode_mask) !=
-                  (val & apic->lapic_timer.timer_mode_mask))
+       case APIC_LVTT: {
+               u32 timer_mode = val & apic->lapic_timer.timer_mode_mask;
+
+               if (apic->lapic_timer.timer_mode != timer_mode) {
+                       apic->lapic_timer.timer_mode = timer_mode;
                        hrtimer_cancel(&apic->lapic_timer.timer);
+               }
 
                if (!kvm_apic_sw_enabled(apic))
                        val |= APIC_LVT_MASKED;
                val &= (apic_lvt_mask[0] | apic->lapic_timer.timer_mode_mask);
                apic_set_reg(apic, APIC_LVTT, val);
                break;
+       }
 
        case APIC_TMICT:
                if (apic_lvtt_tscdeadline(apic))
@@ -1320,7 +1379,7 @@ void kvm_free_lapic(struct kvm_vcpu *vcpu)
        if (!(vcpu->arch.apic_base & MSR_IA32_APICBASE_ENABLE))
                static_key_slow_dec_deferred(&apic_hw_disabled);
 
-       if (!(kvm_apic_get_reg(apic, APIC_SPIV) & APIC_SPIV_APIC_ENABLED))
+       if (!apic->sw_enabled)
                static_key_slow_dec_deferred(&apic_sw_disabled);
 
        if (apic->regs)
@@ -1355,9 +1414,6 @@ void kvm_set_lapic_tscdeadline_msr(struct kvm_vcpu *vcpu, u64 data)
                return;
 
        hrtimer_cancel(&apic->lapic_timer.timer);
-       /* Inject here so clearing tscdeadline won't override new value */
-       if (apic_has_pending_timer(vcpu))
-               kvm_inject_apic_timer_irqs(vcpu);
        apic->lapic_timer.tscdeadline = data;
        start_apic_timer(apic);
 }
@@ -1422,6 +1478,10 @@ void kvm_lapic_set_base(struct kvm_vcpu *vcpu, u64 value)
        apic->base_address = apic->vcpu->arch.apic_base &
                             MSR_IA32_APICBASE_BASE;
 
+       if ((value & MSR_IA32_APICBASE_ENABLE) &&
+            apic->base_address != APIC_DEFAULT_PHYS_BASE)
+               pr_warn_once("APIC base relocation is unsupported by KVM");
+
        /* with FSB delivery interrupt, we can restart APIC functionality */
        apic_debug("apic base msr is 0x%016" PRIx64 ", and base address is "
                   "0x%lx.\n", apic->vcpu->arch.apic_base, apic->base_address);
@@ -1447,6 +1507,7 @@ void kvm_lapic_reset(struct kvm_vcpu *vcpu)
 
        for (i = 0; i < APIC_LVT_NUM; i++)
                apic_set_reg(apic, APIC_LVTT + 0x10 * i, APIC_LVT_MASKED);
+       apic->lapic_timer.timer_mode = 0;
        apic_set_reg(apic, APIC_LVT0,
                     SET_APIC_DELIVERY_MODE(0, APIC_MODE_EXTINT));
 
@@ -1538,23 +1599,8 @@ static enum hrtimer_restart apic_timer_fn(struct hrtimer *data)
 {
        struct kvm_timer *ktimer = container_of(data, struct kvm_timer, timer);
        struct kvm_lapic *apic = container_of(ktimer, struct kvm_lapic, lapic_timer);
-       struct kvm_vcpu *vcpu = apic->vcpu;
-       wait_queue_head_t *q = &vcpu->wq;
-
-       /*
-        * There is a race window between reading and incrementing, but we do
-        * not care about potentially losing timer events in the !reinject
-        * case anyway. Note: KVM_REQ_PENDING_TIMER is implicitly checked
-        * in vcpu_enter_guest.
-        */
-       if (!atomic_read(&ktimer->pending)) {
-               atomic_inc(&ktimer->pending);
-               /* FIXME: this code should not know anything about vcpus */
-               kvm_make_request(KVM_REQ_PENDING_TIMER, vcpu);
-       }
 
-       if (waitqueue_active(q))
-               wake_up_interruptible(q);
+       apic_timer_expired(apic);
 
        if (lapic_is_periodic(apic)) {
                hrtimer_add_expires_ns(&ktimer->timer, ktimer->period);
@@ -1693,6 +1739,9 @@ void kvm_apic_post_state_restore(struct kvm_vcpu *vcpu,
        apic->isr_count = kvm_apic_vid_enabled(vcpu->kvm) ?
                                1 : count_vectors(apic->regs + APIC_ISR);
        apic->highest_isr_cache = -1;
+       if (kvm_x86_ops->hwapic_irr_update)
+               kvm_x86_ops->hwapic_irr_update(vcpu,
+                               apic_find_highest_irr(apic));
        kvm_x86_ops->hwapic_isr_update(vcpu->kvm, apic_find_highest_isr(apic));
        kvm_make_request(KVM_REQ_EVENT, vcpu);
        kvm_rtc_eoi_tracking_restore_one(vcpu);
@@ -1837,8 +1886,11 @@ int kvm_x2apic_msr_write(struct kvm_vcpu *vcpu, u32 msr, u64 data)
        if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
                return 1;
 
+       if (reg == APIC_ICR2)
+               return 1;
+
        /* if this is ICR write vector before command */
-       if (msr == 0x830)
+       if (reg == APIC_ICR)
                apic_reg_write(apic, APIC_ICR2, (u32)(data >> 32));
        return apic_reg_write(apic, reg, (u32)data);
 }
@@ -1851,9 +1903,15 @@ int kvm_x2apic_msr_read(struct kvm_vcpu *vcpu, u32 msr, u64 *data)
        if (!irqchip_in_kernel(vcpu->kvm) || !apic_x2apic_mode(apic))
                return 1;
 
+       if (reg == APIC_DFR || reg == APIC_ICR2) {
+               apic_debug("KVM_APIC_READ: read x2apic reserved register %x\n",
+                          reg);
+               return 1;
+       }
+
        if (apic_reg_read(apic, reg, 4, &low))
                return 1;
-       if (msr == 0x830)
+       if (reg == APIC_ICR)
                apic_reg_read(apic, APIC_ICR2, 4, &high);
 
        *data = (((u64)high) << 32) | low;
@@ -1908,7 +1966,7 @@ int kvm_lapic_enable_pv_eoi(struct kvm_vcpu *vcpu, u64 data)
 void kvm_apic_accept_events(struct kvm_vcpu *vcpu)
 {
        struct kvm_lapic *apic = vcpu->arch.apic;
-       unsigned int sipi_vector;
+       u8 sipi_vector;
        unsigned long pe;
 
        if (!kvm_vcpu_has_lapic(vcpu) || !apic->pending_events)