hvm: Clean up AP initialisation. This allows AP bringup into emulated

real mode when running on VMX, as well as removing 100 LOC.
Signed-off-by: Keir Fraser <keir.fraser@citrix.com>

4 files changed, 88 insertions, 185 deletions

diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index c42ff62aa5..ff49865b99 100644
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -1590,57 +1590,90 @@ void hvm_hypercall_page_initialise(struct domain *d,
     hvm_funcs.init_hypercall_page(d, hypercall_page);
 }
 
-
-/*
- * only called in HVM domain BSP context
- * when booting, vcpuid is always equal to apic_id
- */
 int hvm_bringup_ap(int vcpuid, int trampoline_vector)
 {
-    struct vcpu *v;
     struct domain *d = current->domain;
+    struct vcpu *v;
     struct vcpu_guest_context *ctxt;
-    int rc = 0;
+    struct segment_register reg;
 
-    BUG_ON(!is_hvm_domain(d));
+    ASSERT(is_hvm_domain(d));
 
     if ( (v = d->vcpu[vcpuid]) == NULL )
         return -ENOENT;
 
-    if ( (ctxt = xmalloc(struct vcpu_guest_context)) == NULL )
+    v->fpu_initialised = 0;
+    v->arch.flags |= TF_kernel_mode;
+    v->is_initialised = 1;
+
+    ctxt = &v->arch.guest_context;
+    memset(ctxt, 0, sizeof(*ctxt));
+    ctxt->flags = VGCF_online;
+    ctxt->user_regs.eflags = 2;
+
+#ifdef VMXASSIST
+    if ( boot_cpu_data.x86_vendor == X86_VENDOR_INTEL )
     {
-        gdprintk(XENLOG_ERR,
-                "Failed to allocate memory in hvm_bringup_ap.\n");
-        return -ENOMEM;
+        ctxt->user_regs.eip = VMXASSIST_BASE;
+        ctxt->user_regs.edx = vcpuid;
+        ctxt->user_regs.ebx = trampoline_vector;
+        goto done;
     }
+#endif
+
+    v->arch.hvm_vcpu.guest_cr[0] = X86_CR0_ET;
+    hvm_update_guest_cr(v, 0);
+
+    v->arch.hvm_vcpu.guest_cr[2] = 0;
+    hvm_update_guest_cr(v, 2);
+
+    v->arch.hvm_vcpu.guest_cr[3] = 0;
+    hvm_update_guest_cr(v, 3);
+
+    v->arch.hvm_vcpu.guest_cr[4] = 0;
+    hvm_update_guest_cr(v, 4);
 
-    hvm_init_ap_context(ctxt, vcpuid, trampoline_vector);
+    v->arch.hvm_vcpu.guest_efer = 0;
+    hvm_update_guest_efer(v);
 
+    reg.sel = trampoline_vector << 8;
+    reg.base = (uint32_t)reg.sel << 4;
+    reg.limit = 0xffff;
+    reg.attr.bytes = 0x89b;
+    hvm_set_segment_register(v, x86_seg_cs, &reg);
+
+    reg.sel = reg.base = 0;
+    reg.limit = 0xffff;
+    reg.attr.bytes = 0x893;
+    hvm_set_segment_register(v, x86_seg_ds, &reg);
+    hvm_set_segment_register(v, x86_seg_es, &reg);
+    hvm_set_segment_register(v, x86_seg_fs, &reg);
+    hvm_set_segment_register(v, x86_seg_gs, &reg);
+    hvm_set_segment_register(v, x86_seg_ss, &reg);
+
+    reg.attr.bytes = 0x82; /* LDT */
+    hvm_set_segment_register(v, x86_seg_ldtr, &reg);
+
+    reg.attr.bytes = 0x8b; /* 32-bit TSS (busy) */
+    hvm_set_segment_register(v, x86_seg_tr, &reg);
+
+    reg.attr.bytes = 0;
+    hvm_set_segment_register(v, x86_seg_gdtr, &reg);
+    hvm_set_segment_register(v, x86_seg_idtr, &reg);
+
+#ifdef VMXASSIST
+ done:
+#endif
     /* Sync AP's TSC with BSP's. */
     v->arch.hvm_vcpu.cache_tsc_offset =
         v->domain->vcpu[0]->arch.hvm_vcpu.cache_tsc_offset;
     hvm_funcs.set_tsc_offset(v, v->arch.hvm_vcpu.cache_tsc_offset);
 
-    LOCK_BIGLOCK(d);
-    rc = -EEXIST;
-    if ( !v->is_initialised )
-        rc = boot_vcpu(d, vcpuid, ctxt);
-    UNLOCK_BIGLOCK(d);
-
-    if ( rc != 0 )
-    {
-        gdprintk(XENLOG_ERR,
-               "AP %d bringup failed in boot_vcpu %x.\n", vcpuid, rc);
-        goto out;
-    }
-
     if ( test_and_clear_bit(_VPF_down, &v->pause_flags) )
         vcpu_wake(v);
-    gdprintk(XENLOG_INFO, "AP %d bringup suceeded.\n", vcpuid);
 
- out:
-    xfree(ctxt);
-    return rc;
+    gdprintk(XENLOG_INFO, "AP %d bringup succeeded.\n", vcpuid);
+    return 0;
 }
 
 static int hvmop_set_pci_intx_level(
diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
index b33b20af77..f471474b55 100644
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -62,8 +62,6 @@ int inst_copy_from_guest(unsigned char *buf, unsigned long guest_eip,
                          int inst_len);
 asmlinkage void do_IRQ(struct cpu_user_regs *);
 
-static int svm_reset_to_realmode(
-    struct vcpu *v, struct cpu_user_regs *regs);
 static void svm_update_guest_cr(struct vcpu *v, unsigned int cr);
 static void svm_update_guest_efer(struct vcpu *v);
 static void svm_inject_exception(
@@ -617,8 +615,24 @@ static void svm_set_segment_register(struct vcpu *v, enum x86_segment seg,
                                      struct segment_register *reg)
 {
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
+    int sync = 0;
 
-    ASSERT(v == current);
+    ASSERT((v == current) || !vcpu_runnable(v));
+
+    switch ( seg )
+    {
+    case x86_seg_fs:
+    case x86_seg_gs:
+    case x86_seg_tr:
+    case x86_seg_ldtr:
+        sync = (v == current);
+        break;
+    default:
+        break;
+    }
+
+    if ( sync )
+        svm_sync_vmcb(v);
 
     switch ( seg )
     {
@@ -632,23 +646,17 @@ static void svm_set_segment_register(struct vcpu *v, enum x86_segment seg,
         memcpy(&vmcb->es, reg, sizeof(*reg));
         break;
     case x86_seg_fs:
-        svm_sync_vmcb(v);
         memcpy(&vmcb->fs, reg, sizeof(*reg));
-        svm_vmload(vmcb);
         break;
     case x86_seg_gs:
-        svm_sync_vmcb(v);
         memcpy(&vmcb->gs, reg, sizeof(*reg));
-        svm_vmload(vmcb);
         break;
     case x86_seg_ss:
         memcpy(&vmcb->ss, reg, sizeof(*reg));
         vmcb->cpl = vmcb->ss.attr.fields.dpl;
         break;
     case x86_seg_tr:
-        svm_sync_vmcb(v);
         memcpy(&vmcb->tr, reg, sizeof(*reg));
-        svm_vmload(vmcb);
         break;
     case x86_seg_gdtr:
         memcpy(&vmcb->gdtr, reg, sizeof(*reg));
@@ -657,13 +665,14 @@ static void svm_set_segment_register(struct vcpu *v, enum x86_segment seg,
         memcpy(&vmcb->idtr, reg, sizeof(*reg));
         break;
     case x86_seg_ldtr:
-        svm_sync_vmcb(v);
         memcpy(&vmcb->ldtr, reg, sizeof(*reg));
-        svm_vmload(vmcb);
         break;
     default:
         BUG();
     }
+
+    if ( sync )
+        svm_vmload(vmcb);
 }
 
 /* Make sure that xen intercepts any FP accesses from current */
@@ -684,47 +693,11 @@ static void svm_stts(struct vcpu *v)
     }
 }
 
-
 static void svm_set_tsc_offset(struct vcpu *v, u64 offset)
 {
     v->arch.hvm_svm.vmcb->tsc_offset = offset;
 }
 
-
-static void svm_init_ap_context(
-    struct vcpu_guest_context *ctxt, int vcpuid, int trampoline_vector)
-{
-    struct vcpu *v;
-    struct vmcb_struct *vmcb;
-    cpu_user_regs_t *regs;
-    u16 cs_sel;
-
-    /* We know this is safe because hvm_bringup_ap() does it */
-    v = current->domain->vcpu[vcpuid];
-    vmcb = v->arch.hvm_svm.vmcb;
-    regs = &v->arch.guest_context.user_regs;
-
-    memset(ctxt, 0, sizeof(*ctxt));
-
-    /*
-     * We execute the trampoline code in real mode. The trampoline vector
-     * passed to us is page alligned and is the physical frame number for
-     * the code. We will execute this code in real mode.
-     */
-    cs_sel = trampoline_vector << 8;
-    ctxt->user_regs.eip = 0x0;
-    ctxt->user_regs.cs = cs_sel;
-
-    /*
-     * This is the launch of an AP; set state so that we begin executing
-     * the trampoline code in real-mode.
-     */
-    svm_reset_to_realmode(v, regs);  
-    /* Adjust the vmcb's hidden register state. */
-    vmcb->cs.sel = cs_sel;
-    vmcb->cs.base = (cs_sel << 4);
-}
-
 static void svm_init_hypercall_page(struct domain *d, void *hypercall_page)
 {
     char *p;
@@ -916,7 +889,6 @@ static struct hvm_function_table svm_function_table = {
     .stts                 = svm_stts,
     .set_tsc_offset       = svm_set_tsc_offset,
     .inject_exception     = svm_inject_exception,
-    .init_ap_context      = svm_init_ap_context,
     .init_hypercall_page  = svm_init_hypercall_page,
     .event_pending        = svm_event_pending
 };
@@ -2037,90 +2009,6 @@ void svm_handle_invlpg(const short invlpga, struct cpu_user_regs *regs)
     domain_crash(v->domain);
 }
 
-
-/*
- * Reset to realmode causes execution to start at 0xF000:0xFFF0 in
- * 16-bit realmode.  Basically, this mimics a processor reset.
- *
- * returns 0 on success, non-zero otherwise
- */
-static int svm_reset_to_realmode(struct vcpu *v, 
-                                 struct cpu_user_regs *regs)
-{
-    struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
-
-    memset(regs, 0, sizeof(struct cpu_user_regs));
-
-    regs->eflags = 2;
-
-    v->arch.hvm_vcpu.guest_cr[0] = X86_CR0_ET;
-    svm_update_guest_cr(v, 0);
-
-    v->arch.hvm_vcpu.guest_cr[2] = 0;
-    svm_update_guest_cr(v, 2);
-
-    v->arch.hvm_vcpu.guest_cr[4] = 0;
-    svm_update_guest_cr(v, 4);
-
-    vmcb->efer = EFER_SVME;
-
-    /* This will jump to ROMBIOS */
-    regs->eip = 0xFFF0;
-
-    /* Set up the segment registers and all their hidden states. */
-    vmcb->cs.sel = 0xF000;
-    vmcb->cs.attr.bytes = 0x089b;
-    vmcb->cs.limit = 0xffff;
-    vmcb->cs.base = 0x000F0000;
-
-    vmcb->ss.sel = 0x00;
-    vmcb->ss.attr.bytes = 0x0893;
-    vmcb->ss.limit = 0xffff;
-    vmcb->ss.base = 0x00;
-
-    vmcb->ds.sel = 0x00;
-    vmcb->ds.attr.bytes = 0x0893;
-    vmcb->ds.limit = 0xffff;
-    vmcb->ds.base = 0x00;
-    
-    vmcb->es.sel = 0x00;
-    vmcb->es.attr.bytes = 0x0893;
-    vmcb->es.limit = 0xffff;
-    vmcb->es.base = 0x00;
-    
-    vmcb->fs.sel = 0x00;
-    vmcb->fs.attr.bytes = 0x0893;
-    vmcb->fs.limit = 0xffff;
-    vmcb->fs.base = 0x00;
-    
-    vmcb->gs.sel = 0x00;
-    vmcb->gs.attr.bytes = 0x0893;
-    vmcb->gs.limit = 0xffff;
-    vmcb->gs.base = 0x00;
-
-    vmcb->ldtr.sel = 0x00;
-    vmcb->ldtr.attr.bytes = 0x0000;
-    vmcb->ldtr.limit = 0x0;
-    vmcb->ldtr.base = 0x00;
-
-    vmcb->gdtr.sel = 0x00;
-    vmcb->gdtr.attr.bytes = 0x0000;
-    vmcb->gdtr.limit = 0x0;
-    vmcb->gdtr.base = 0x00;
-    
-    vmcb->tr.sel = 0;
-    vmcb->tr.attr.bytes = 0;
-    vmcb->tr.limit = 0x0;
-    vmcb->tr.base = 0;
-
-    vmcb->idtr.sel = 0x00;
-    vmcb->idtr.attr.bytes = 0x0000;
-    vmcb->idtr.limit = 0x3ff;
-    vmcb->idtr.base = 0x00;
-
-    return 0;
-}
-
 asmlinkage void svm_vmexit_handler(struct cpu_user_regs *regs)
 {
     unsigned int exit_reason;
diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c
index 667b59f964..ff2d8f3cbc 100644
--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -863,7 +863,7 @@ static void vmx_set_segment_register(struct vcpu *v, enum x86_segment seg,
 {
     uint32_t attr;
 
-    ASSERT(v == current);
+    ASSERT((v == current) || !vcpu_runnable(v));
 
     attr = reg->attr.bytes;
     attr = ((attr & 0xf00) << 4) | (attr & 0xff);
@@ -872,6 +872,8 @@ static void vmx_set_segment_register(struct vcpu *v, enum x86_segment seg,
     if ( !reg->attr.fields.p )
         attr |= (1u << 16);
 
+    vmx_vmcs_enter(v);
+
     switch ( seg )
     {
     case x86_seg_cs:
@@ -933,6 +935,8 @@ static void vmx_set_segment_register(struct vcpu *v, enum x86_segment seg,
     default:
         BUG();
     }
+
+    vmx_vmcs_exit(v);
 }
 
 /* Make sure that xen intercepts any FP accesses from current */
@@ -965,17 +969,6 @@ static void vmx_set_tsc_offset(struct vcpu *v, u64 offset)
     vmx_vmcs_exit(v);
 }
 
-static void vmx_init_ap_context(
-    struct vcpu_guest_context *ctxt, int vcpuid, int trampoline_vector)
-{
-    memset(ctxt, 0, sizeof(*ctxt));
-#ifdef VMXASSIST
-    ctxt->user_regs.eip = VMXASSIST_BASE;
-    ctxt->user_regs.edx = vcpuid;
-    ctxt->user_regs.ebx = trampoline_vector;
-#endif
-}
-
 void do_nmi(struct cpu_user_regs *);
 
 static void vmx_init_hypercall_page(struct domain *d, void *hypercall_page)
@@ -1159,7 +1152,6 @@ static struct hvm_function_table vmx_function_table = {
     .stts                 = vmx_stts,
     .set_tsc_offset       = vmx_set_tsc_offset,
     .inject_exception     = vmx_inject_exception,
-    .init_ap_context      = vmx_init_ap_context,
     .init_hypercall_page  = vmx_init_hypercall_page,
     .event_pending        = vmx_event_pending,
     .cpu_up               = vmx_cpu_up,
diff --git a/xen/include/asm-x86/hvm/hvm.h b/xen/include/asm-x86/hvm/hvm.h
index 96ad350eb6..079e3c5da2 100644
--- a/xen/include/asm-x86/hvm/hvm.h
+++ b/xen/include/asm-x86/hvm/hvm.h
@@ -121,9 +121,6 @@ struct hvm_function_table {
     void (*inject_exception)(unsigned int trapnr, int errcode,
                              unsigned long cr2);
 
-    void (*init_ap_context)(struct vcpu_guest_context *ctxt,
-                            int vcpuid, int trampoline_vector);
-
     void (*init_hypercall_page)(struct domain *d, void *hypercall_page);
 
     int  (*event_pending)(struct vcpu *v);
@@ -239,13 +236,6 @@ void hvm_migrate_timers(struct vcpu *v);
 void hvm_do_resume(struct vcpu *v);
 
 static inline void
-hvm_init_ap_context(struct vcpu_guest_context *ctxt,
-                    int vcpuid, int trampoline_vector)
-{
-    return hvm_funcs.init_ap_context(ctxt, vcpuid, trampoline_vector);
-}
-
-static inline void
 hvm_inject_exception(unsigned int trapnr, int errcode, unsigned long cr2)
 {
     hvm_funcs.inject_exception(trapnr, errcode, cr2);