]> pilppa.org Git - linux-2.6-omap-h63xx.git/blobdiff - arch/x86/kvm/x86.c
Merge branch 'linus' into x86/cleanups
[linux-2.6-omap-h63xx.git] / arch / x86 / kvm / x86.c
index 0ce556372a4d1df614499224a634f4967afc8eb3..63a77caa59f11bf55a8655b857aa9c541e4168a6 100644 (file)
@@ -492,8 +492,8 @@ static int do_set_msr(struct kvm_vcpu *vcpu, unsigned index, u64 *data)
 static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 {
        static int version;
-       struct kvm_wall_clock wc;
-       struct timespec wc_ts;
+       struct pvclock_wall_clock wc;
+       struct timespec now, sys, boot;
 
        if (!wall_clock)
                return;
@@ -502,10 +502,19 @@ static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
 
        kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 
-       wc_ts = current_kernel_time();
-       wc.wc_sec = wc_ts.tv_sec;
-       wc.wc_nsec = wc_ts.tv_nsec;
-       wc.wc_version = version;
+       /*
+        * The guest calculates current wall clock time by adding
+        * system time (updated by kvm_write_guest_time below) to the
+        * wall clock specified here.  guest system time equals host
+        * system time for us, thus we must fill in host boot time here.
+        */
+       now = current_kernel_time();
+       ktime_get_ts(&sys);
+       boot = ns_to_timespec(timespec_to_ns(&now) - timespec_to_ns(&sys));
+
+       wc.sec = boot.tv_sec;
+       wc.nsec = boot.tv_nsec;
+       wc.version = version;
 
        kvm_write_guest(kvm, wall_clock, &wc, sizeof(wc));
 
@@ -513,6 +522,45 @@ static void kvm_write_wall_clock(struct kvm *kvm, gpa_t wall_clock)
        kvm_write_guest(kvm, wall_clock, &version, sizeof(version));
 }
 
+static uint32_t div_frac(uint32_t dividend, uint32_t divisor)
+{
+       uint32_t quotient, remainder;
+
+       /* Don't try to replace with do_div(), this one calculates
+        * "(dividend << 32) / divisor" */
+       __asm__ ( "divl %4"
+                 : "=a" (quotient), "=d" (remainder)
+                 : "0" (0), "1" (dividend), "r" (divisor) );
+       return quotient;
+}
+
+static void kvm_set_time_scale(uint32_t tsc_khz, struct pvclock_vcpu_time_info *hv_clock)
+{
+       uint64_t nsecs = 1000000000LL;
+       int32_t  shift = 0;
+       uint64_t tps64;
+       uint32_t tps32;
+
+       tps64 = tsc_khz * 1000LL;
+       while (tps64 > nsecs*2) {
+               tps64 >>= 1;
+               shift--;
+       }
+
+       tps32 = (uint32_t)tps64;
+       while (tps32 <= (uint32_t)nsecs) {
+               tps32 <<= 1;
+               shift++;
+       }
+
+       hv_clock->tsc_shift = shift;
+       hv_clock->tsc_to_system_mul = div_frac(nsecs, tps32);
+
+       pr_debug("%s: tsc_khz %u, tsc_shift %d, tsc_mul %u\n",
+                __FUNCTION__, tsc_khz, hv_clock->tsc_shift,
+                hv_clock->tsc_to_system_mul);
+}
+
 static void kvm_write_guest_time(struct kvm_vcpu *v)
 {
        struct timespec ts;
@@ -523,6 +571,11 @@ static void kvm_write_guest_time(struct kvm_vcpu *v)
        if ((!vcpu->time_page))
                return;
 
+       if (unlikely(vcpu->hv_clock_tsc_khz != tsc_khz)) {
+               kvm_set_time_scale(tsc_khz, &vcpu->hv_clock);
+               vcpu->hv_clock_tsc_khz = tsc_khz;
+       }
+
        /* Keep irq disabled to prevent changes to the clock */
        local_irq_save(flags);
        kvm_get_msr(v, MSR_IA32_TIME_STAMP_COUNTER,
@@ -537,14 +590,14 @@ static void kvm_write_guest_time(struct kvm_vcpu *v)
        /*
         * The interface expects us to write an even number signaling that the
         * update is finished. Since the guest won't see the intermediate
-        * state, we just write "2" at the end
+        * state, we just increase by 2 at the end.
         */
-       vcpu->hv_clock.version = 2;
+       vcpu->hv_clock.version += 2;
 
        shared_kaddr = kmap_atomic(vcpu->time_page, KM_USER0);
 
        memcpy(shared_kaddr + vcpu->time_offset, &vcpu->hv_clock,
-               sizeof(vcpu->hv_clock));
+              sizeof(vcpu->hv_clock));
 
        kunmap_atomic(shared_kaddr, KM_USER0);
 
@@ -599,10 +652,6 @@ int kvm_set_msr_common(struct kvm_vcpu *vcpu, u32 msr, u64 data)
                /* ...but clean it before doing the actual write */
                vcpu->arch.time_offset = data & ~(PAGE_MASK | 1);
 
-               vcpu->arch.hv_clock.tsc_to_system_mul =
-                                       clocksource_khz2mult(tsc_khz, 22);
-               vcpu->arch.hv_clock.tsc_shift = 22;
-
                down_read(&current->mm->mmap_sem);
                vcpu->arch.time_page =
                                gfn_to_page(vcpu->kvm, data >> PAGE_SHIFT);
@@ -2417,6 +2466,9 @@ int kvm_arch_init(void *opaque)
 
        kvm_x86_ops = ops;
        kvm_mmu_set_nonpresent_ptes(0ull, 0ull);
+       kvm_mmu_set_base_ptes(PT_PRESENT_MASK);
+       kvm_mmu_set_mask_ptes(PT_USER_MASK, PT_ACCESSED_MASK,
+                       PT_DIRTY_MASK, PT64_NX_MASK, 0);
        return 0;
 
 out:
@@ -2755,7 +2807,9 @@ again:
 
        if (vcpu->requests) {
                if (test_and_clear_bit(KVM_REQ_MIGRATE_TIMER, &vcpu->requests))
-                       __kvm_migrate_apic_timer(vcpu);
+                       __kvm_migrate_timers(vcpu);
+               if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
+                       kvm_x86_ops->tlb_flush(vcpu);
                if (test_and_clear_bit(KVM_REQ_REPORT_TPR_ACCESS,
                                       &vcpu->requests)) {
                        kvm_run->exit_reason = KVM_EXIT_TPR_ACCESS;
@@ -2769,6 +2823,7 @@ again:
                }
        }
 
+       clear_bit(KVM_REQ_PENDING_TIMER, &vcpu->requests);
        kvm_inject_pending_timer_irqs(vcpu);
 
        preempt_disable();
@@ -2778,21 +2833,13 @@ again:
 
        local_irq_disable();
 
-       if (need_resched()) {
+       if (vcpu->requests || need_resched()) {
                local_irq_enable();
                preempt_enable();
                r = 1;
                goto out;
        }
 
-       if (vcpu->requests)
-               if (test_bit(KVM_REQ_MMU_RELOAD, &vcpu->requests)) {
-                       local_irq_enable();
-                       preempt_enable();
-                       r = 1;
-                       goto out;
-               }
-
        if (signal_pending(current)) {
                local_irq_enable();
                preempt_enable();
@@ -2822,9 +2869,6 @@ again:
 
        kvm_guest_enter();
 
-       if (vcpu->requests)
-               if (test_and_clear_bit(KVM_REQ_TLB_FLUSH, &vcpu->requests))
-                       kvm_x86_ops->tlb_flush(vcpu);
 
        KVMTRACE_0D(VMENTRY, vcpu, entryexit);
        kvm_x86_ops->run(vcpu, kvm_run);
@@ -3019,6 +3063,8 @@ int kvm_arch_vcpu_ioctl_set_regs(struct kvm_vcpu *vcpu, struct kvm_regs *regs)
 
        kvm_x86_ops->decache_regs(vcpu);
 
+       vcpu->arch.exception.pending = false;
+
        vcpu_put(vcpu);
 
        return 0;
@@ -3481,7 +3527,7 @@ int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
        }
 
        if (reason == TASK_SWITCH_IRET || reason == TASK_SWITCH_JMP) {
-               cseg_desc.type &= ~(1 << 8); //clear the B flag
+               cseg_desc.type &= ~(1 << 1); //clear the B flag
                save_guest_segment_descriptor(vcpu, tr_seg.selector,
                                              &cseg_desc);
        }
@@ -3507,7 +3553,7 @@ int kvm_task_switch(struct kvm_vcpu *vcpu, u16 tss_selector, int reason)
        }
 
        if (reason != TASK_SWITCH_IRET) {
-               nseg_desc.type |= (1 << 8);
+               nseg_desc.type |= (1 << 1);
                save_guest_segment_descriptor(vcpu, tss_selector,
                                              &nseg_desc);
        }
@@ -3698,10 +3744,19 @@ void fx_init(struct kvm_vcpu *vcpu)
 {
        unsigned after_mxcsr_mask;
 
+       /*
+        * Touch the fpu the first time in non atomic context as if
+        * this is the first fpu instruction the exception handler
+        * will fire before the instruction returns and it'll have to
+        * allocate ram with GFP_KERNEL.
+        */
+       if (!used_math())
+               fx_save(&vcpu->arch.host_fx_image);
+
        /* Initialize guest FPU by resetting ours and saving into guest's */
        preempt_disable();
        fx_save(&vcpu->arch.host_fx_image);
-       fpu_init();
+       fx_finit();
        fx_save(&vcpu->arch.guest_fx_image);
        fx_restore(&vcpu->arch.host_fx_image);
        preempt_enable();
@@ -3906,6 +3961,8 @@ void kvm_arch_destroy_vm(struct kvm *kvm)
        kvm_free_physmem(kvm);
        if (kvm->arch.apic_access_page)
                put_page(kvm->arch.apic_access_page);
+       if (kvm->arch.ept_identity_pagetable)
+               put_page(kvm->arch.ept_identity_pagetable);
        kfree(kvm);
 }