]> git.karo-electronics.de Git - mv-sheeva.git/commitdiff
KVM: SVM: Add VMEXIT handler and intercepts
authorAlexander Graf <agraf@suse.de>
Tue, 25 Nov 2008 19:17:08 +0000 (20:17 +0100)
committerAvi Kivity <avi@redhat.com>
Tue, 24 Mar 2009 09:02:47 +0000 (11:02 +0200)
This adds the #VMEXIT intercept, so we return to the level 1 guest
when something happens in the level 2 guest that should return to
the level 1 guest.

v2 implements HIF handling and cleans up exception interception
v3 adds support for V_INTR_MASKING_MASK
v4 uses the host page hsave
v5 removes IOPM merging code
v6 moves mmu code out of the atomic section

Acked-by: Joerg Roedel <joro@8bytes.org>
Signed-off-by: Alexander Graf <agraf@suse.de>
Signed-off-by: Avi Kivity <avi@redhat.com>
arch/x86/kvm/svm.c

index fad187cbfabe1c7c7df002c8070329de84e15e79..4cb2920b15278f70e96212c00ba7a08860cb85a6 100644 (file)
@@ -72,6 +72,13 @@ module_param(npt, int, S_IRUGO);
 static void kvm_reput_irq(struct vcpu_svm *svm);
 static void svm_flush_tlb(struct kvm_vcpu *vcpu);
 
+static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override);
+static int nested_svm_vmexit(struct vcpu_svm *svm);
+static int nested_svm_vmsave(struct vcpu_svm *svm, void *nested_vmcb,
+                            void *arg2, void *opaque);
+static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
+                                     bool has_error_code, u32 error_code);
+
 static inline struct vcpu_svm *to_svm(struct kvm_vcpu *vcpu)
 {
        return container_of(vcpu, struct vcpu_svm, vcpu);
@@ -221,6 +228,11 @@ static void svm_queue_exception(struct kvm_vcpu *vcpu, unsigned nr,
 {
        struct vcpu_svm *svm = to_svm(vcpu);
 
+       /* If we are within a nested VM we'd better #VMEXIT and let the
+          guest handle the exception */
+       if (nested_svm_check_exception(svm, nr, has_error_code, error_code))
+               return;
+
        svm->vmcb->control.event_inj = nr
                | SVM_EVTINJ_VALID
                | (has_error_code ? SVM_EVTINJ_VALID_ERR : 0)
@@ -1198,6 +1210,46 @@ static int nested_svm_check_permissions(struct vcpu_svm *svm)
        return 0;
 }
 
+static int nested_svm_check_exception(struct vcpu_svm *svm, unsigned nr,
+                                     bool has_error_code, u32 error_code)
+{
+       if (is_nested(svm)) {
+               svm->vmcb->control.exit_code = SVM_EXIT_EXCP_BASE + nr;
+               svm->vmcb->control.exit_code_hi = 0;
+               svm->vmcb->control.exit_info_1 = error_code;
+               svm->vmcb->control.exit_info_2 = svm->vcpu.arch.cr2;
+               if (nested_svm_exit_handled(svm, false)) {
+                       nsvm_printk("VMexit -> EXCP 0x%x\n", nr);
+
+                       nested_svm_vmexit(svm);
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+
+static inline int nested_svm_intr(struct vcpu_svm *svm)
+{
+       if (is_nested(svm)) {
+               if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
+                       return 0;
+
+               if (!(svm->vcpu.arch.hflags & HF_HIF_MASK))
+                       return 0;
+
+               svm->vmcb->control.exit_code = SVM_EXIT_INTR;
+
+               if (nested_svm_exit_handled(svm, false)) {
+                       nsvm_printk("VMexit -> INTR\n");
+                       nested_svm_vmexit(svm);
+                       return 1;
+               }
+       }
+
+       return 0;
+}
+
 static struct page *nested_svm_get_page(struct vcpu_svm *svm, u64 gpa)
 {
        struct page *page;
@@ -1258,6 +1310,228 @@ static int nested_svm_do(struct vcpu_svm *svm,
        return retval;
 }
 
+static int nested_svm_exit_handled_real(struct vcpu_svm *svm,
+                                       void *arg1,
+                                       void *arg2,
+                                       void *opaque)
+{
+       struct vmcb *nested_vmcb = (struct vmcb *)arg1;
+       bool kvm_overrides = *(bool *)opaque;
+       u32 exit_code = svm->vmcb->control.exit_code;
+
+       if (kvm_overrides) {
+               switch (exit_code) {
+               case SVM_EXIT_INTR:
+               case SVM_EXIT_NMI:
+                       return 0;
+               /* For now we are always handling NPFs when using them */
+               case SVM_EXIT_NPF:
+                       if (npt_enabled)
+                               return 0;
+                       break;
+               /* When we're shadowing, trap PFs */
+               case SVM_EXIT_EXCP_BASE + PF_VECTOR:
+                       if (!npt_enabled)
+                               return 0;
+                       break;
+               default:
+                       break;
+               }
+       }
+
+       switch (exit_code) {
+       case SVM_EXIT_READ_CR0 ... SVM_EXIT_READ_CR8: {
+               u32 cr_bits = 1 << (exit_code - SVM_EXIT_READ_CR0);
+               if (nested_vmcb->control.intercept_cr_read & cr_bits)
+                       return 1;
+               break;
+       }
+       case SVM_EXIT_WRITE_CR0 ... SVM_EXIT_WRITE_CR8: {
+               u32 cr_bits = 1 << (exit_code - SVM_EXIT_WRITE_CR0);
+               if (nested_vmcb->control.intercept_cr_write & cr_bits)
+                       return 1;
+               break;
+       }
+       case SVM_EXIT_READ_DR0 ... SVM_EXIT_READ_DR7: {
+               u32 dr_bits = 1 << (exit_code - SVM_EXIT_READ_DR0);
+               if (nested_vmcb->control.intercept_dr_read & dr_bits)
+                       return 1;
+               break;
+       }
+       case SVM_EXIT_WRITE_DR0 ... SVM_EXIT_WRITE_DR7: {
+               u32 dr_bits = 1 << (exit_code - SVM_EXIT_WRITE_DR0);
+               if (nested_vmcb->control.intercept_dr_write & dr_bits)
+                       return 1;
+               break;
+       }
+       case SVM_EXIT_EXCP_BASE ... SVM_EXIT_EXCP_BASE + 0x1f: {
+               u32 excp_bits = 1 << (exit_code - SVM_EXIT_EXCP_BASE);
+               if (nested_vmcb->control.intercept_exceptions & excp_bits)
+                       return 1;
+               break;
+       }
+       default: {
+               u64 exit_bits = 1ULL << (exit_code - SVM_EXIT_INTR);
+               nsvm_printk("exit code: 0x%x\n", exit_code);
+               if (nested_vmcb->control.intercept & exit_bits)
+                       return 1;
+       }
+       }
+
+       return 0;
+}
+
+static int nested_svm_exit_handled_msr(struct vcpu_svm *svm,
+                                      void *arg1, void *arg2,
+                                      void *opaque)
+{
+       struct vmcb *nested_vmcb = (struct vmcb *)arg1;
+       u8 *msrpm = (u8 *)arg2;
+        u32 t0, t1;
+       u32 msr = svm->vcpu.arch.regs[VCPU_REGS_RCX];
+       u32 param = svm->vmcb->control.exit_info_1 & 1;
+
+       if (!(nested_vmcb->control.intercept & (1ULL << INTERCEPT_MSR_PROT)))
+               return 0;
+
+       switch(msr) {
+       case 0 ... 0x1fff:
+               t0 = (msr * 2) % 8;
+               t1 = msr / 8;
+               break;
+       case 0xc0000000 ... 0xc0001fff:
+               t0 = (8192 + msr - 0xc0000000) * 2;
+               t1 = (t0 / 8);
+               t0 %= 8;
+               break;
+       case 0xc0010000 ... 0xc0011fff:
+               t0 = (16384 + msr - 0xc0010000) * 2;
+               t1 = (t0 / 8);
+               t0 %= 8;
+               break;
+       default:
+               return 1;
+               break;
+       }
+       if (msrpm[t1] & ((1 << param) << t0))
+               return 1;
+
+       return 0;
+}
+
+static int nested_svm_exit_handled(struct vcpu_svm *svm, bool kvm_override)
+{
+       bool k = kvm_override;
+
+       switch (svm->vmcb->control.exit_code) {
+       case SVM_EXIT_MSR:
+               return nested_svm_do(svm, svm->nested_vmcb,
+                                    svm->nested_vmcb_msrpm, NULL,
+                                    nested_svm_exit_handled_msr);
+       default: break;
+       }
+
+       return nested_svm_do(svm, svm->nested_vmcb, 0, &k,
+                            nested_svm_exit_handled_real);
+}
+
+static int nested_svm_vmexit_real(struct vcpu_svm *svm, void *arg1,
+                                 void *arg2, void *opaque)
+{
+       struct vmcb *nested_vmcb = (struct vmcb *)arg1;
+       struct vmcb *hsave = svm->hsave;
+       u64 nested_save[] = { nested_vmcb->save.cr0,
+                             nested_vmcb->save.cr3,
+                             nested_vmcb->save.cr4,
+                             nested_vmcb->save.efer,
+                             nested_vmcb->control.intercept_cr_read,
+                             nested_vmcb->control.intercept_cr_write,
+                             nested_vmcb->control.intercept_dr_read,
+                             nested_vmcb->control.intercept_dr_write,
+                             nested_vmcb->control.intercept_exceptions,
+                             nested_vmcb->control.intercept,
+                             nested_vmcb->control.msrpm_base_pa,
+                             nested_vmcb->control.iopm_base_pa,
+                             nested_vmcb->control.tsc_offset };
+
+       /* Give the current vmcb to the guest */
+       memcpy(nested_vmcb, svm->vmcb, sizeof(struct vmcb));
+       nested_vmcb->save.cr0 = nested_save[0];
+       if (!npt_enabled)
+               nested_vmcb->save.cr3 = nested_save[1];
+       nested_vmcb->save.cr4 = nested_save[2];
+       nested_vmcb->save.efer = nested_save[3];
+       nested_vmcb->control.intercept_cr_read = nested_save[4];
+       nested_vmcb->control.intercept_cr_write = nested_save[5];
+       nested_vmcb->control.intercept_dr_read = nested_save[6];
+       nested_vmcb->control.intercept_dr_write = nested_save[7];
+       nested_vmcb->control.intercept_exceptions = nested_save[8];
+       nested_vmcb->control.intercept = nested_save[9];
+       nested_vmcb->control.msrpm_base_pa = nested_save[10];
+       nested_vmcb->control.iopm_base_pa = nested_save[11];
+       nested_vmcb->control.tsc_offset = nested_save[12];
+
+       /* We always set V_INTR_MASKING and remember the old value in hflags */
+       if (!(svm->vcpu.arch.hflags & HF_VINTR_MASK))
+               nested_vmcb->control.int_ctl &= ~V_INTR_MASKING_MASK;
+
+       if ((nested_vmcb->control.int_ctl & V_IRQ_MASK) &&
+           (nested_vmcb->control.int_vector)) {
+               nsvm_printk("WARNING: IRQ 0x%x still enabled on #VMEXIT\n",
+                               nested_vmcb->control.int_vector);
+       }
+
+       /* Restore the original control entries */
+       svm->vmcb->control = hsave->control;
+
+       /* Kill any pending exceptions */
+       if (svm->vcpu.arch.exception.pending == true)
+               nsvm_printk("WARNING: Pending Exception\n");
+       svm->vcpu.arch.exception.pending = false;
+
+       /* Restore selected save entries */
+       svm->vmcb->save.es = hsave->save.es;
+       svm->vmcb->save.cs = hsave->save.cs;
+       svm->vmcb->save.ss = hsave->save.ss;
+       svm->vmcb->save.ds = hsave->save.ds;
+       svm->vmcb->save.gdtr = hsave->save.gdtr;
+       svm->vmcb->save.idtr = hsave->save.idtr;
+       svm->vmcb->save.rflags = hsave->save.rflags;
+       svm_set_efer(&svm->vcpu, hsave->save.efer);
+       svm_set_cr0(&svm->vcpu, hsave->save.cr0 | X86_CR0_PE);
+       svm_set_cr4(&svm->vcpu, hsave->save.cr4);
+       if (npt_enabled) {
+               svm->vmcb->save.cr3 = hsave->save.cr3;
+               svm->vcpu.arch.cr3 = hsave->save.cr3;
+       } else {
+               kvm_set_cr3(&svm->vcpu, hsave->save.cr3);
+       }
+       kvm_register_write(&svm->vcpu, VCPU_REGS_RAX, hsave->save.rax);
+       kvm_register_write(&svm->vcpu, VCPU_REGS_RSP, hsave->save.rsp);
+       kvm_register_write(&svm->vcpu, VCPU_REGS_RIP, hsave->save.rip);
+       svm->vmcb->save.dr7 = 0;
+       svm->vmcb->save.cpl = 0;
+       svm->vmcb->control.exit_int_info = 0;
+
+       svm->vcpu.arch.hflags &= ~HF_GIF_MASK;
+       /* Exit nested SVM mode */
+       svm->nested_vmcb = 0;
+
+       return 0;
+}
+
+static int nested_svm_vmexit(struct vcpu_svm *svm)
+{
+       nsvm_printk("VMexit\n");
+       if (nested_svm_do(svm, svm->nested_vmcb, 0,
+                         NULL, nested_svm_vmexit_real))
+               return 1;
+
+       kvm_mmu_reset_context(&svm->vcpu);
+       kvm_mmu_load(&svm->vcpu);
+
+       return 0;
+}
 
 static int nested_svm_vmrun_msrpm(struct vcpu_svm *svm, void *arg1,
                                  void *arg2, void *opaque)
@@ -1805,6 +2079,17 @@ static int handle_exit(struct kvm_run *kvm_run, struct kvm_vcpu *vcpu)
        KVMTRACE_3D(VMEXIT, vcpu, exit_code, (u32)svm->vmcb->save.rip,
                    (u32)((u64)svm->vmcb->save.rip >> 32), entryexit);
 
+       if (is_nested(svm)) {
+               nsvm_printk("nested handle_exit: 0x%x | 0x%lx | 0x%lx | 0x%lx\n",
+                           exit_code, svm->vmcb->control.exit_info_1,
+                           svm->vmcb->control.exit_info_2, svm->vmcb->save.rip);
+               if (nested_svm_exit_handled(svm, true)) {
+                       nested_svm_vmexit(svm);
+                       nsvm_printk("-> #VMEXIT\n");
+                       return 1;
+               }
+       }
+
        if (npt_enabled) {
                int mmu_reload = 0;
                if ((vcpu->arch.cr0 ^ svm->vmcb->save.cr0) & X86_CR0_PG) {
@@ -1892,6 +2177,8 @@ static void svm_set_irq(struct kvm_vcpu *vcpu, int irq)
 {
        struct vcpu_svm *svm = to_svm(vcpu);
 
+       nested_svm_intr(svm);
+
        svm_inject_irq(svm, irq);
 }
 
@@ -1937,6 +2224,9 @@ static void svm_intr_assist(struct kvm_vcpu *vcpu)
        if (!kvm_cpu_has_interrupt(vcpu))
                goto out;
 
+       if (nested_svm_intr(svm))
+               goto out;
+
        if (!(svm->vcpu.arch.hflags & HF_GIF_MASK))
                goto out;
 
@@ -1989,6 +2279,9 @@ static void do_interrupt_requests(struct kvm_vcpu *vcpu,
        struct vcpu_svm *svm = to_svm(vcpu);
        struct vmcb_control_area *control = &svm->vmcb->control;
 
+       if (nested_svm_intr(svm))
+               return;
+
        svm->vcpu.arch.interrupt_window_open =
                (!(control->int_state & SVM_INTERRUPT_SHADOW_MASK) &&
                 (svm->vmcb->save.rflags & X86_EFLAGS_IF) &&