[HVM] Save/restore: save HVM GPRs and DRs with the rest of the CPU state

and remove the use of xc_vcpu_getcontext() from HVM save/restore.
Also fixes loss of CR2, DR6 and DR7 contents across HVM save/restore.
Signed-off-by: Tim Deegan <Tim.Deegan@xensource.com>

diff --git a/tools/libxc/xc_domain_restore.c b/tools/libxc/xc_domain_restore.c
index 1159776f4fb2d0f730dd24f49345ac89fde00575..1b52c90e5b6caf0bea86a8877c884f418606ae2c 100644 (file)
--- a/tools/libxc/xc_domain_restore.c
+++ b/tools/libxc/xc_domain_restore.c
@@ -696,26 +696,6 @@ int xc_domain_restore(int xc_handle, int io_fd, uint32_t dom,
         xc_set_hvm_param(xc_handle, dom, HVM_PARAM_STORE_EVTCHN, store_evtchn);
         *store_mfn = magic_pfns[2];
 
-        /* Read vcpu contexts */
-        for ( i = 0; i <= max_vcpu_id; i++ )
-        {
-            if ( !(vcpumap & (1ULL << i)) )
-                continue;
-
-            if ( !read_exact(io_fd, &(ctxt), sizeof(ctxt)) )
-            {
-                ERROR("error read vcpu context.\n");
-                goto out;
-            }
-            
-            if ( (rc = xc_vcpu_setcontext(xc_handle, dom, i, &ctxt)) )
-            {
-                ERROR("Could not set vcpu context, rc=%d", rc);
-                goto out;
-            }
-            rc = 1;
-        }
-
         /* Read HVM context */
         if ( !read_exact(io_fd, &rec_len, sizeof(uint32_t)) )
         {

diff --git a/tools/libxc/xc_domain_save.c b/tools/libxc/xc_domain_save.c
index f8b51c2713789db54296db7956563b0a7486b3cc..ba7e005819af003d104fb6805cc06bd63e5434df 100644 (file)
--- a/tools/libxc/xc_domain_save.c
+++ b/tools/libxc/xc_domain_save.c
@@ -378,8 +378,7 @@ static int analysis_phase(int xc_handle, uint32_t domid, int p2m_size,
 
 
 static int suspend_and_state(int (*suspend)(int), int xc_handle, int io_fd,
-                             int dom, xc_dominfo_t *info,
-                             vcpu_guest_context_t *ctxt)
+                             int dom, xc_dominfo_t *info)
 {
     int i = 0;
 
@@ -397,10 +396,6 @@ static int suspend_and_state(int (*suspend)(int), int xc_handle, int io_fd,
         return -1;
     }
 
-    if ( xc_vcpu_getcontext(xc_handle, dom, 0, ctxt) )
-        ERROR("Could not get vcpu context");
-
-
     if ( info->dying )
     {
         ERROR("domain is dying");
@@ -663,10 +658,11 @@ static xen_pfn_t *xc_map_m2p(int xc_handle,
 static xen_pfn_t *map_and_save_p2m_table(int xc_handle, 
                                          int io_fd, 
                                          uint32_t dom,
-                                         vcpu_guest_context_t *ctxt,
                                          unsigned long p2m_size,
                                          shared_info_t *live_shinfo)
 {
+    vcpu_guest_context_t ctxt;
+
     /* Double and single indirect references to the live P2M table */
     xen_pfn_t *live_p2m_frame_list_list = NULL;
     xen_pfn_t *live_p2m_frame_list = NULL;
@@ -730,13 +726,19 @@ static xen_pfn_t *map_and_save_p2m_table(int xc_handle,
         }
     }
 
+    if ( xc_vcpu_getcontext(xc_handle, dom, 0, &ctxt) )
+    {
+        ERROR("Could not get vcpu context");
+        goto out;
+    }
+
     /*
      * Write an extended-info structure to inform the restore code that
      * a PAE guest understands extended CR3 (PDPTs above 4GB). Turns off
      * slow paths in the restore code.
      */
     if ( (pt_levels == 3) &&
-         (ctxt->vm_assist & (1UL << VMASST_TYPE_pae_extended_cr3)) )
+         (ctxt.vm_assist & (1UL << VMASST_TYPE_pae_extended_cr3)) )
     {
         unsigned long signature = ~0UL;
         uint32_t tot_sz   = sizeof(struct vcpu_guest_context) + 8;
@@ -746,7 +748,7 @@ static xen_pfn_t *map_and_save_p2m_table(int xc_handle,
              !write_exact(io_fd, &tot_sz,    sizeof(tot_sz)) ||
              !write_exact(io_fd, &chunk_sig, 4) ||
              !write_exact(io_fd, &chunk_sz,  sizeof(chunk_sz)) ||
-             !write_exact(io_fd, ctxt,       sizeof(*ctxt)) )
+             !write_exact(io_fd, &ctxt,      sizeof(ctxt)) )
         {
             ERROR("write: extended info");
             goto out;
@@ -853,11 +855,6 @@ int xc_domain_save(int xc_handle, int io_fd, uint32_t dom, uint32_t max_iters,
         return 1;
     }
 
-    if ( xc_vcpu_getcontext(xc_handle, dom, 0, &ctxt) )
-    {
-        ERROR("Could not get vcpu context");
-        goto out;
-    }
     shared_info_frame = info.shared_info_frame;
 
     /* Map the shared info frame */
@@ -900,7 +897,7 @@ int xc_domain_save(int xc_handle, int io_fd, uint32_t dom, uint32_t max_iters,
     else
     {
         /* This is a non-live suspend. Suspend the domain .*/
-        if ( suspend_and_state(suspend, xc_handle, io_fd, dom, &info, &ctxt) )
+        if ( suspend_and_state(suspend, xc_handle, io_fd, dom, &info) )
         {
             ERROR("Domain appears not to have suspended");
             goto out;
@@ -999,7 +996,7 @@ int xc_domain_save(int xc_handle, int io_fd, uint32_t dom, uint32_t max_iters,
 
         /* Map the P2M table, and write the list of P2M frames */
         live_p2m = map_and_save_p2m_table(xc_handle, io_fd, dom, 
-                                          &ctxt, p2m_size, live_shinfo);
+                                          p2m_size, live_shinfo);
         if ( live_p2m == NULL )
         {
             ERROR("Failed to map/save the p2m frame list");
@@ -1304,17 +1301,13 @@ int xc_domain_save(int xc_handle, int io_fd, uint32_t dom, uint32_t max_iters,
                 DPRINTF("Start last iteration\n");
                 last_iter = 1;
 
-                if ( suspend_and_state(suspend, xc_handle, io_fd, dom, &info,
-                                       &ctxt) )
+                if ( suspend_and_state(suspend, xc_handle, io_fd, dom, &info) )
                 {
                     ERROR("Domain appears not to have suspended");
                     goto out;
                 }
 
-                DPRINTF("SUSPEND shinfo %08lx eip %08lx edx %08lx\n",
-                        info.shared_info_frame,
-                        (unsigned long)ctxt.user_regs.eip,
-                        (unsigned long)ctxt.user_regs.edx);
+                DPRINTF("SUSPEND shinfo %08lx\n", info.shared_info_frame);
             }
 
             if ( xc_shadow_control(xc_handle, dom, 
@@ -1410,27 +1403,6 @@ int xc_domain_save(int xc_handle, int io_fd, uint32_t dom, uint32_t max_iters,
             goto out;
         }
 
-        /* Save vcpu contexts */
-
-        for ( i = 0; i <= info.max_vcpu_id; i++ )
-        {
-            if ( !(vcpumap & (1ULL << i)) )
-                continue;
-            
-            if ( xc_vcpu_getcontext(xc_handle, dom, i, &ctxt) )
-            {
-                ERROR("HVM:Could not get vcpu context");
-                goto out;
-            }
-            
-            DPRINTF("write vcpu %d context.\n", i); 
-            if ( !write_exact(io_fd, &(ctxt), sizeof(ctxt)) )
-            {
-                ERROR("write vcpu context failed!\n");
-                goto out;
-            }
-        }
-
         /* Get HVM context from Xen and save it too */
         if ( (rec_size = xc_domain_hvm_getcontext(xc_handle, dom, hvm_buf, 
                                                   hvm_buf_size)) == -1 )
@@ -1494,6 +1466,12 @@ int xc_domain_save(int xc_handle, int io_fd, uint32_t dom, uint32_t max_iters,
         }
     }
 
+    if ( xc_vcpu_getcontext(xc_handle, dom, 0, &ctxt) )
+    {
+        ERROR("Could not get vcpu context");
+        goto out;
+    }
+
     /* Canonicalise the suspend-record frame number. */
     if ( !translate_mfn_to_pfn(&ctxt.user_regs.edx) )
     {

diff --git a/xen/arch/x86/hvm/hvm.c b/xen/arch/x86/hvm/hvm.c
index 5cb283bf2ede6f8869cb137522b8000fa643b40e..7636da5614239a38dac86f2d1b9ef798fd1dbb20 100644 (file)
--- a/xen/arch/x86/hvm/hvm.c
+++ b/xen/arch/x86/hvm/hvm.c
@@ -191,6 +191,7 @@ static int hvm_save_cpu_ctxt(struct domain *d, hvm_domain_context_t *h)
 {
     struct vcpu *v;
     struct hvm_hw_cpu ctxt;
+    struct vcpu_guest_context *vc;
 
     for_each_vcpu(d, v)
     {
@@ -199,7 +200,40 @@ static int hvm_save_cpu_ctxt(struct domain *d, hvm_domain_context_t *h)
         if ( test_bit(_VPF_down, &v->pause_flags) ) 
             continue;
 
+        /* Architecture-specific vmcs/vmcb bits */
         hvm_funcs.save_cpu_ctxt(v, &ctxt);
+
+        /* Other vcpu register state */
+        vc = &v->arch.guest_context;
+        if ( vc->flags & VGCF_i387_valid )
+            memcpy(ctxt.fpu_regs, &vc->fpu_ctxt, sizeof(ctxt.fpu_regs));
+        else 
+            memset(ctxt.fpu_regs, 0, sizeof(ctxt.fpu_regs));
+        ctxt.rax = vc->user_regs.eax;
+        ctxt.rbx = vc->user_regs.ebx;
+        ctxt.rcx = vc->user_regs.ecx;
+        ctxt.rdx = vc->user_regs.edx;
+        ctxt.rbp = vc->user_regs.ebp;
+        ctxt.rsi = vc->user_regs.esi;
+        ctxt.rdi = vc->user_regs.edi;
+        /* %rsp handled by arch-specific call above */
+#ifdef __x86_64__        
+        ctxt.r8  = vc->user_regs.r8;
+        ctxt.r9  = vc->user_regs.r9;
+        ctxt.r10 = vc->user_regs.r10;
+        ctxt.r11 = vc->user_regs.r11;
+        ctxt.r12 = vc->user_regs.r12;
+        ctxt.r13 = vc->user_regs.r13;
+        ctxt.r14 = vc->user_regs.r14;
+        ctxt.r15 = vc->user_regs.r15;
+#endif
+        ctxt.dr0 = vc->debugreg[0];
+        ctxt.dr1 = vc->debugreg[1];
+        ctxt.dr2 = vc->debugreg[2];
+        ctxt.dr3 = vc->debugreg[3];
+        ctxt.dr6 = vc->debugreg[6];
+        ctxt.dr7 = vc->debugreg[7];
+
         if ( hvm_save_entry(CPU, v->vcpu_id, h, &ctxt) != 0 )
             return 1; 
     }
@@ -208,9 +242,10 @@ static int hvm_save_cpu_ctxt(struct domain *d, hvm_domain_context_t *h)
 
 static int hvm_load_cpu_ctxt(struct domain *d, hvm_domain_context_t *h)
 {
-    int vcpuid;
+    int vcpuid, rc;
     struct vcpu *v;
     struct hvm_hw_cpu ctxt;
+    struct vcpu_guest_context *vc;
 
     /* Which vcpu is this? */
     vcpuid = hvm_load_instance(h);
@@ -219,13 +254,52 @@ static int hvm_load_cpu_ctxt(struct domain *d, hvm_domain_context_t *h)
         gdprintk(XENLOG_ERR, "HVM restore: domain has no vcpu %u\n", vcpuid);
         return -EINVAL;
     }
+    vc = &v->arch.guest_context;
+
+    /* Need to init this vcpu before loading its contents */
+    LOCK_BIGLOCK(d);
+    if ( !v->is_initialised )
+        if ( (rc = boot_vcpu(d, vcpuid, vc)) != 0 )
+            return rc;
+    UNLOCK_BIGLOCK(d);
 
     if ( hvm_load_entry(CPU, h, &ctxt) != 0 ) 
         return -EINVAL;
 
+    /* Architecture-specific vmcs/vmcb bits */
     if ( hvm_funcs.load_cpu_ctxt(v, &ctxt) < 0 )
         return -EINVAL;
 
+    /* Other vcpu register state */
+    memcpy(&vc->fpu_ctxt, ctxt.fpu_regs, sizeof(ctxt.fpu_regs));
+    vc->user_regs.eax = ctxt.rax;
+    vc->user_regs.ebx = ctxt.rbx;
+    vc->user_regs.ecx = ctxt.rcx;
+    vc->user_regs.edx = ctxt.rdx;
+    vc->user_regs.ebp = ctxt.rbp;
+    vc->user_regs.esi = ctxt.rsi;
+    vc->user_regs.edi = ctxt.rdi;
+    vc->user_regs.esp = ctxt.rsp;
+#ifdef __x86_64__
+    vc->user_regs.r8; = ctxt.r8; 
+    vc->user_regs.r9; = ctxt.r9; 
+    vc->user_regs.r10 = ctxt.r10;
+    vc->user_regs.r11 = ctxt.r11;
+    vc->user_regs.r12 = ctxt.r12;
+    vc->user_regs.r13 = ctxt.r13;
+    vc->user_regs.r14 = ctxt.r14;
+    vc->user_regs.r15 = ctxt.r15;
+#endif
+    vc->debugreg[0] = ctxt.dr0;
+    vc->debugreg[1] = ctxt.dr1;
+    vc->debugreg[2] = ctxt.dr2;
+    vc->debugreg[3] = ctxt.dr3;
+    vc->debugreg[6] = ctxt.dr6;
+    vc->debugreg[7] = ctxt.dr7;
+
+    vc->flags = VGCF_i387_valid | VGCF_online;
+    v->fpu_initialised = 1;
+
     /* Auxiliary processors should be woken immediately. */
     if ( test_and_clear_bit(_VPF_down, &v->pause_flags) )
         vcpu_wake(v);

diff --git a/xen/arch/x86/hvm/svm/svm.c b/xen/arch/x86/hvm/svm/svm.c
index ae0b42ae66c48a8176dac2ed283707c4a2563467..cf3f9c647fd5e89d0da70514eba2f953d9320e4a 100644 (file)
--- a/xen/arch/x86/hvm/svm/svm.c
+++ b/xen/arch/x86/hvm/svm/svm.c
@@ -233,7 +233,7 @@ int svm_vmcb_save(struct vcpu *v, struct hvm_hw_cpu *c)
 {
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
 
-    c->eip = vmcb->rip;
+    c->rip = vmcb->rip;
 
 #ifdef HVM_DEBUG_SUSPEND
     printk("%s: eip=0x%"PRIx64".\n", 
@@ -241,10 +241,11 @@ int svm_vmcb_save(struct vcpu *v, struct hvm_hw_cpu *c)
            inst_len, c->eip);
 #endif
 
-    c->esp = vmcb->rsp;
-    c->eflags = vmcb->rflags;
+    c->rsp = vmcb->rsp;
+    c->rflags = vmcb->rflags;
 
     c->cr0 = v->arch.hvm_svm.cpu_shadow_cr0;
+    c->cr2 = v->arch.hvm_svm.cpu_cr2;
     c->cr3 = v->arch.hvm_svm.cpu_cr3;
     c->cr4 = v->arch.hvm_svm.cpu_shadow_cr4;
 
@@ -315,15 +316,17 @@ int svm_vmcb_restore(struct vcpu *v, struct hvm_hw_cpu *c)
     unsigned long mfn, old_base_mfn;
     struct vmcb_struct *vmcb = v->arch.hvm_svm.vmcb;
 
-    vmcb->rip    = c->eip;
-    vmcb->rsp    = c->esp;
-    vmcb->rflags = c->eflags;
+    vmcb->rip    = c->rip;
+    vmcb->rsp    = c->rsp;
+    vmcb->rflags = c->rflags;
 
     v->arch.hvm_svm.cpu_shadow_cr0 = c->cr0;
     vmcb->cr0 = c->cr0 | X86_CR0_WP | X86_CR0_ET;
     if ( !paging_mode_hap(v->domain) ) 
         vmcb->cr0 |= X86_CR0_PG;
 
+    v->arch.hvm_svm.cpu_cr2 = c->cr2;
+
 #ifdef HVM_DEBUG_SUSPEND
     printk("%s: cr3=0x%"PRIx64", cr0=0x%"PRIx64", cr4=0x%"PRIx64".\n",
            __func__,
@@ -421,6 +424,9 @@ int svm_vmcb_restore(struct vcpu *v, struct hvm_hw_cpu *c)
     vmcb->sysenter_esp = c->sysenter_esp;
     vmcb->sysenter_eip = c->sysenter_eip;
 
+    vmcb->dr6 = c->dr6;
+    vmcb->dr7 = c->dr7;
+
     paging_update_paging_modes(v);
     return 0;
  
@@ -440,6 +446,7 @@ void svm_save_cpu_state(struct vcpu *v, struct hvm_hw_cpu *data)
     data->msr_cstar        = vmcb->cstar;
     data->msr_syscall_mask = vmcb->sfmask;
     data->msr_efer         = v->arch.hvm_svm.cpu_shadow_efer;
+    data->msr_flags        = -1ULL;
 
     data->tsc = hvm_get_guest_time(v);
 }

diff --git a/xen/arch/x86/hvm/vmx/vmx.c b/xen/arch/x86/hvm/vmx/vmx.c
index b4cfe17b294c61afc6a7e5f7205319c98292c5be..1c47aadebcb5b6868c0b7b8fedcead269155b642 100644 (file)
--- a/xen/arch/x86/hvm/vmx/vmx.c
+++ b/xen/arch/x86/hvm/vmx/vmx.c
@@ -370,11 +370,12 @@ static inline void __restore_debug_registers(struct vcpu *v)
 
 int vmx_vmcs_save(struct vcpu *v, struct hvm_hw_cpu *c)
 {    
-    c->eip = __vmread(GUEST_RIP);
-    c->esp = __vmread(GUEST_RSP);
-    c->eflags = __vmread(GUEST_RFLAGS);
+    c->rip = __vmread(GUEST_RIP);
+    c->rsp = __vmread(GUEST_RSP);
+    c->rflags = __vmread(GUEST_RFLAGS);
 
     c->cr0 = v->arch.hvm_vmx.cpu_shadow_cr0;
+    c->cr2 = v->arch.hvm_vmx.cpu_cr2;
     c->cr3 = v->arch.hvm_vmx.cpu_cr3;
     c->cr4 = v->arch.hvm_vmx.cpu_shadow_cr4;
 
@@ -444,13 +445,15 @@ int vmx_vmcs_restore(struct vcpu *v, struct hvm_hw_cpu *c)
 
     vmx_vmcs_enter(v);
 
-    __vmwrite(GUEST_RIP, c->eip);
-    __vmwrite(GUEST_RSP, c->esp);
-    __vmwrite(GUEST_RFLAGS, c->eflags);
+    __vmwrite(GUEST_RIP, c->rip);
+    __vmwrite(GUEST_RSP, c->rsp);
+    __vmwrite(GUEST_RFLAGS, c->rflags);
 
     v->arch.hvm_vmx.cpu_shadow_cr0 = c->cr0;
     __vmwrite(CR0_READ_SHADOW, v->arch.hvm_vmx.cpu_shadow_cr0);
 
+    v->arch.hvm_vmx.cpu_cr2 = c->cr2;
+
 #ifdef HVM_DEBUG_SUSPEND
     printk("vmx_vmcs_restore: cr3=0x%"PRIx64", cr0=0x%"PRIx64", cr4=0x%"PRIx64".\n",
             c->cr3,
@@ -555,6 +558,8 @@ int vmx_vmcs_restore(struct vcpu *v, struct hvm_hw_cpu *c)
     __vmwrite(GUEST_SYSENTER_ESP, c->sysenter_esp);
     __vmwrite(GUEST_SYSENTER_EIP, c->sysenter_eip);
 
+    __vmwrite(GUEST_DR7, c->dr7);
+
     vmx_vmcs_exit(v);
 
     paging_update_paging_modes(v);
@@ -590,7 +595,7 @@ void vmx_save_cpu_state(struct vcpu *v, struct hvm_hw_cpu *data)
     data->shadow_gs = guest_state->shadow_gs;
 
     /* save msrs */
-    data->flags = guest_flags;
+    data->msr_flags        = guest_flags;
     data->msr_lstar        = guest_state->msrs[VMX_INDEX_MSR_LSTAR];
     data->msr_star         = guest_state->msrs[VMX_INDEX_MSR_STAR];
     data->msr_cstar        = guest_state->msrs[VMX_INDEX_MSR_CSTAR];
@@ -607,7 +612,7 @@ void vmx_load_cpu_state(struct vcpu *v, struct hvm_hw_cpu *data)
     struct vmx_msr_state *guest_state = &v->arch.hvm_vmx.msr_state;
 
     /* restore msrs */
-    guest_state->flags = data->flags;
+    guest_state->flags = data->msr_flags;
     guest_state->msrs[VMX_INDEX_MSR_LSTAR]        = data->msr_lstar;
     guest_state->msrs[VMX_INDEX_MSR_STAR]         = data->msr_star;
     guest_state->msrs[VMX_INDEX_MSR_CSTAR]        = data->msr_cstar;

diff --git a/xen/include/public/hvm/save.h b/xen/include/public/hvm/save.h
index c921b50a7864643d7893839e4c379a0827e07d1a..14f120f9a6bacd1612c58446045d8939d000f9df 100644 (file)
--- a/xen/include/public/hvm/save.h
+++ b/xen/include/public/hvm/save.h
@@ -87,13 +87,40 @@ DECLARE_HVM_SAVE_TYPE(HEADER, 1, struct hvm_save_header);
  */
 
 struct hvm_hw_cpu {
-    uint64_t eip;
-    uint64_t esp;
-    uint64_t eflags;
+    uint8_t  fpu_regs[512];
+
+    uint64_t rax;
+    uint64_t rbx;
+    uint64_t rcx;
+    uint64_t rdx;
+    uint64_t rbp;
+    uint64_t rsi;
+    uint64_t rdi;
+    uint64_t rsp;
+    uint64_t r8;
+    uint64_t r9;
+    uint64_t r10;
+    uint64_t r11;
+    uint64_t r12;
+    uint64_t r13;
+    uint64_t r14;
+    uint64_t r15;
+
+    uint64_t rip;
+    uint64_t rflags;
+
     uint64_t cr0;
+    uint64_t cr2;
     uint64_t cr3;
     uint64_t cr4;
 
+    uint64_t dr0;
+    uint64_t dr1;
+    uint64_t dr2;
+    uint64_t dr3;
+    uint64_t dr6;
+    uint64_t dr7;    
+
     uint32_t cs_sel;
     uint32_t ds_sel;
     uint32_t es_sel;
@@ -142,9 +169,9 @@ struct hvm_hw_cpu {
 
     /* msr for em64t */
     uint64_t shadow_gs;
-    uint64_t flags;
 
     /* msr content saved/restored. */
+    uint64_t msr_flags;
     uint64_t msr_lstar;
     uint64_t msr_star;
     uint64_t msr_cstar;