#include <palacios/vmm_sprintf.h>
+#ifndef CONFIG_DEBUG_SVM
+#undef PrintDebug
+#define PrintDebug(fmt, args...)
+#endif
+
+
uint32_t v3_last_exit;
// This is a global pointer to the host's VMCB
static vmcb_t * Allocate_VMCB() {
- vmcb_t * vmcb_page = (vmcb_t *)V3_VAddr(V3_AllocPages(1));
+ vmcb_t * vmcb_page = NULL;
+ addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
+
+ if ((void *)vmcb_pa == NULL) {
+ PrintError("Error allocating VMCB\n");
+ return NULL;
+ }
+
+ vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
memset(vmcb_page, 0, 4096);
-static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * vm_info) {
+static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
uint_t i;
//
-
-
ctrl_area->svm_instrs.VMRUN = 1;
ctrl_area->svm_instrs.VMMCALL = 1;
ctrl_area->svm_instrs.VMLOAD = 1;
ctrl_area->svm_instrs.STGI = 1;
ctrl_area->svm_instrs.CLGI = 1;
ctrl_area->svm_instrs.SKINIT = 1;
- ctrl_area->svm_instrs.RDTSCP = 1;
ctrl_area->svm_instrs.ICEBP = 1;
ctrl_area->svm_instrs.WBINVD = 1;
ctrl_area->svm_instrs.MONITOR = 1;
ctrl_area->instrs.CPUID = 1;
ctrl_area->instrs.HLT = 1;
+
+#ifdef CONFIG_TIME_VIRTUALIZE_TSC
+ ctrl_area->instrs.RDTSC = 1;
+ ctrl_area->svm_instrs.RDTSCP = 1;
+#endif
+
// guest_state->cr0 = 0x00000001; // PE
/*
ctrl_area->instrs.NMI = 1;
- ctrl_area->instrs.SMI = 1;
+ ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
ctrl_area->instrs.INIT = 1;
ctrl_area->instrs.PAUSE = 1;
ctrl_area->instrs.shutdown_evts = 1;
/* Setup Guest Machine state */
- vm_info->vm_regs.rsp = 0x00;
- vm_info->rip = 0xfff0;
+ core->vm_regs.rsp = 0x00;
+ core->rip = 0xfff0;
- vm_info->vm_regs.rdx = 0x00000f00;
+ core->vm_regs.rdx = 0x00000f00;
- vm_info->cpl = 0;
+ core->cpl = 0;
- vm_info->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
- vm_info->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
- vm_info->ctrl_regs.efer |= EFER_MSR_svm_enable;
+ core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
+ core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
+ core->ctrl_regs.efer |= EFER_MSR_svm_enable;
- vm_info->segments.cs.selector = 0xf000;
- vm_info->segments.cs.limit = 0xffff;
- vm_info->segments.cs.base = 0x0000000f0000LL;
+ core->segments.cs.selector = 0xf000;
+ core->segments.cs.limit = 0xffff;
+ core->segments.cs.base = 0x0000000f0000LL;
// (raw attributes = 0xf3)
- vm_info->segments.cs.type = 0x3;
- vm_info->segments.cs.system = 0x1;
- vm_info->segments.cs.dpl = 0x3;
- vm_info->segments.cs.present = 1;
+ core->segments.cs.type = 0x3;
+ core->segments.cs.system = 0x1;
+ core->segments.cs.dpl = 0x3;
+ core->segments.cs.present = 1;
- struct v3_segment * segregs [] = {&(vm_info->segments.ss), &(vm_info->segments.ds),
- &(vm_info->segments.es), &(vm_info->segments.fs),
- &(vm_info->segments.gs), NULL};
+ struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
+ &(core->segments.es), &(core->segments.fs),
+ &(core->segments.gs), NULL};
for ( i = 0; segregs[i] != NULL; i++) {
struct v3_segment * seg = segregs[i];
seg->present = 1;
}
- vm_info->segments.gdtr.limit = 0x0000ffff;
- vm_info->segments.gdtr.base = 0x0000000000000000LL;
- vm_info->segments.idtr.limit = 0x0000ffff;
- vm_info->segments.idtr.base = 0x0000000000000000LL;
+ core->segments.gdtr.limit = 0x0000ffff;
+ core->segments.gdtr.base = 0x0000000000000000LL;
+ core->segments.idtr.limit = 0x0000ffff;
+ core->segments.idtr.base = 0x0000000000000000LL;
- vm_info->segments.ldtr.selector = 0x0000;
- vm_info->segments.ldtr.limit = 0x0000ffff;
- vm_info->segments.ldtr.base = 0x0000000000000000LL;
- vm_info->segments.tr.selector = 0x0000;
- vm_info->segments.tr.limit = 0x0000ffff;
- vm_info->segments.tr.base = 0x0000000000000000LL;
+ core->segments.ldtr.selector = 0x0000;
+ core->segments.ldtr.limit = 0x0000ffff;
+ core->segments.ldtr.base = 0x0000000000000000LL;
+ core->segments.tr.selector = 0x0000;
+ core->segments.tr.limit = 0x0000ffff;
+ core->segments.tr.base = 0x0000000000000000LL;
- vm_info->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
- vm_info->dbg_regs.dr7 = 0x0000000000000400LL;
+ core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
+ core->dbg_regs.dr7 = 0x0000000000000400LL;
- v3_init_svm_io_map(vm_info);
- ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(vm_info->io_map.arch_data);
+ ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
ctrl_area->instrs.IOIO_PROT = 1;
-
-
- v3_init_svm_msr_map(vm_info);
- ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(vm_info->msr_map.arch_data);
- ctrl_area->instrs.MSR_PROT = 1;
+
+ ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
+ ctrl_area->instrs.MSR_PROT = 1;
PrintDebug("Exiting on interrupts\n");
ctrl_area->instrs.INTR = 1;
- if (vm_info->shdw_pg_mode == SHADOW_PAGING) {
+ v3_hook_msr(core->vm_info, EFER_MSR,
+ &v3_handle_efer_read,
+ &v3_handle_efer_write,
+ core);
+
+ if (core->shdw_pg_mode == SHADOW_PAGING) {
PrintDebug("Creating initial shadow page table\n");
/* JRL: This is a performance killer, and a simplistic solution */
ctrl_area->guest_ASID = 1;
- if (v3_init_passthrough_pts(vm_info) == -1) {
+ if (v3_init_passthrough_pts(core) == -1) {
PrintError("Could not initialize passthrough page tables\n");
return ;
}
- vm_info->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
+ core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
PrintDebug("Created\n");
- vm_info->ctrl_regs.cr0 |= 0x80000000;
- vm_info->ctrl_regs.cr3 = vm_info->direct_map_pt;
+ core->ctrl_regs.cr0 |= 0x80000000;
+ core->ctrl_regs.cr3 = core->direct_map_pt;
ctrl_area->cr_reads.cr0 = 1;
ctrl_area->cr_writes.cr0 = 1;
ctrl_area->cr_reads.cr3 = 1;
ctrl_area->cr_writes.cr3 = 1;
- v3_hook_msr(vm_info, EFER_MSR,
- &v3_handle_efer_read,
- &v3_handle_efer_write,
- vm_info);
+
ctrl_area->instrs.INVLPG = 1;
- } else if (vm_info->shdw_pg_mode == NESTED_PAGING) {
+ } else if (core->shdw_pg_mode == NESTED_PAGING) {
// Flush the TLB on entries/exits
ctrl_area->TLB_CONTROL = 1;
ctrl_area->guest_ASID = 1;
PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
// Set the Nested Page Table pointer
- if (v3_init_passthrough_pts(vm_info) == -1) {
+ if (v3_init_passthrough_pts(core) == -1) {
PrintError("Could not initialize Nested page tables\n");
return ;
}
- ctrl_area->N_CR3 = vm_info->direct_map_pt;
+ ctrl_area->N_CR3 = core->direct_map_pt;
guest_state->g_pat = 0x7040600070406ULL;
}
+
+ /* tell the guest that we don't support SVM */
+ v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
+ &v3_handle_vm_cr_read,
+ &v3_handle_vm_cr_write,
+ core);
}
-int v3_init_svm_vmcb(struct guest_info * info, v3_vm_class_t vm_class) {
+int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
PrintDebug("Allocating VMCB\n");
- info->vmm_data = (void*)Allocate_VMCB();
+ core->vmm_data = (void *)Allocate_VMCB();
+ if (core->vmm_data == NULL) {
+ PrintError("Could not allocate VMCB, Exiting...\n");
+ return -1;
+ }
+
if (vm_class == V3_PC_VM) {
- PrintDebug("Initializing VMCB (addr=%p)\n", (void *)info->vmm_data);
- Init_VMCB_BIOS((vmcb_t*)(info->vmm_data), info);
+ PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
+ Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
} else {
PrintError("Invalid VM class\n");
return -1;
}
+int v3_deinit_svm_vmcb(struct guest_info * core) {
+ V3_FreePages(V3_PAddr(core->vmm_data), 1);
+ return 0;
+}
+
static int update_irq_exit_state(struct guest_info * info) {
vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
- if ((info->intr_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
+ // Fix for QEMU bug using EVENTINJ as an internal cache
+ guest_ctrl->EVENTINJ.valid = 0;
+
+ if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
#ifdef CONFIG_DEBUG_INTERRUPTS
- PrintDebug("INTAK cycle completed for irq %d\n", info->intr_state.irq_vector);
+ PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
#endif
- info->intr_state.irq_started = 1;
- info->intr_state.irq_pending = 0;
+ info->intr_core_state.irq_started = 1;
+ info->intr_core_state.irq_pending = 0;
- v3_injecting_intr(info, info->intr_state.irq_vector, V3_EXTERNAL_IRQ);
+ v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
}
- if ((info->intr_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
+ if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
#ifdef CONFIG_DEBUG_INTERRUPTS
- PrintDebug("Interrupt %d taken by guest\n", info->intr_state.irq_vector);
+ PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
#endif
// Interrupt was taken fully vectored
- info->intr_state.irq_started = 0;
+ info->intr_core_state.irq_started = 0;
- } else {
+ } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
#ifdef CONFIG_DEBUG_INTERRUPTS
PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
#endif
static int update_irq_entry_state(struct guest_info * info) {
vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
+
+ if (info->intr_core_state.irq_pending == 0) {
+ guest_ctrl->guest_ctrl.V_IRQ = 0;
+ guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
+ }
+
if (v3_excp_pending(info)) {
uint_t excp = v3_get_excp_number(info);
#endif
v3_injecting_excp(info, excp);
- } else if (info->intr_state.irq_started == 1) {
+ } else if (info->intr_core_state.irq_started == 1) {
#ifdef CONFIG_DEBUG_INTERRUPTS
PrintDebug("IRQ pending from previous injection\n");
#endif
guest_ctrl->guest_ctrl.V_IRQ = 1;
- guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_state.irq_vector;
+ guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
(void *)(addr_t)info->rip);
#endif
- info->intr_state.irq_pending = 1;
- info->intr_state.irq_vector = irq;
+ info->intr_core_state.irq_pending = 1;
+ info->intr_core_state.irq_vector = irq;
break;
}
int v3_svm_enter(struct guest_info * info) {
vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
- ullong_t tmp_tsc;
addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
// Conditionally yield the CPU if the timeslice has expired
v3_yield_cond(info);
+ // Perform any additional yielding needed for time adjustment
+ v3_adjust_time(info);
+
// disable global interrupts for vm state transition
v3_clgi();
+ // Update timer devices prior to entering VM.
+ v3_update_timers(info);
+
// Synchronize the guest state to the VMCB
guest_state->cr0 = info->ctrl_regs.cr0;
guest_state->cr2 = info->ctrl_regs.cr2;
guest_state->rip = info->rip;
guest_state->rsp = info->vm_regs.rsp;
-#ifdef CONFIG_SYMBIOTIC
- if (info->sym_state.sym_call_active == 0) {
+#ifdef CONFIG_SYMCALL
+ if (info->sym_core_state.symcall_state.sym_call_active == 0) {
update_irq_entry_state(info);
}
#else
(void *)(addr_t)info->rip);
*/
-#ifdef CONFIG_SYMBIOTIC
- if (info->sym_state.sym_call_active == 1) {
+#ifdef CONFIG_SYMCALL
+ if (info->sym_core_state.symcall_state.sym_call_active == 1) {
if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
}
}
#endif
-
- rdtscll(info->time_state.cached_host_tsc);
- guest_ctrl->TSC_OFFSET = info->time_state.guest_tsc - info->time_state.cached_host_tsc;
+ v3_time_enter_vm(info);
+ guest_ctrl->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
//V3_Print("Calling v3_svm_launch\n");
-
- v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[info->cpu_id]);
-
- //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
+ v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
+
+ //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
- rdtscll(tmp_tsc);
+ // Immediate exit from VM time bookkeeping
+ v3_time_exit_vm(info);
- //PrintDebug("SVM Returned\n");
-
info->num_exits++;
- v3_update_time(info, tmp_tsc - info->time_state.cached_host_tsc);
-
-
// Save Guest state from VMCB
info->rip = guest_state->rip;
info->vm_regs.rsp = guest_state->rsp;
exit_info2 = guest_ctrl->exit_info2;
-#ifdef CONFIG_SYMBIOTIC
- if (info->sym_state.sym_call_active == 0) {
+#ifdef CONFIG_SYMCALL
+ if (info->sym_core_state.symcall_state.sym_call_active == 0) {
update_irq_exit_state(info);
}
#else
v3_yield_cond(info);
+
if (v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2) != 0) {
PrintError("Error in SVM exit handler\n");
+ PrintError(" last exit was %d\n", v3_last_exit);
return -1;
}
}
-int v3_start_svm_guest(struct guest_info *info) {
+int v3_start_svm_guest(struct guest_info * info) {
// vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
// vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
+ PrintDebug("Starting SVM core %u\n", info->cpu_id);
+
+ if (info->cpu_id == 0) {
+ info->core_run_state = CORE_RUNNING;
+ info->vm_info->run_state = VM_RUNNING;
+ } else {
+ PrintDebug("SVM core %u: Waiting for core initialization\n", info->cpu_id);
+
+ while (info->core_run_state == CORE_STOPPED) {
+ v3_yield(info);
+ //PrintDebug("SVM core %u: still waiting for INIT\n",info->cpu_id);
+ }
+
+ PrintDebug("SVM core %u initialized\n", info->cpu_id);
+ }
+
+ PrintDebug("SVM core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n",
+ info->cpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
+ info->segments.cs.limit, (void *)(info->rip));
- PrintDebug("Launching SVM VM (vmcb=%p)\n", (void *)info->vmm_data);
+
+ PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p)\n", info->cpu_id, (void *)info->vmm_data);
//PrintDebugVMCB((vmcb_t*)(info->vmm_data));
- info->run_state = VM_RUNNING;
- rdtscll(info->yield_start_cycle);
-
+ v3_start_time(info);
while (1) {
+
+ if (info->vm_info->run_state == VM_STOPPED) {
+ info->core_run_state = CORE_STOPPED;
+ break;
+ }
+
if (v3_svm_enter(info) == -1) {
vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
addr_t host_addr;
addr_t linear_addr = 0;
- info->run_state = VM_ERROR;
+ info->vm_info->run_state = VM_ERROR;
- V3_Print("SVM ERROR!!\n");
+ V3_Print("SVM core %u: SVM ERROR!!\n", info->cpu_id);
v3_print_guest_state(info);
- V3_Print("SVM Exit Code: %p\n", (void *)(addr_t)guest_ctrl->exit_code);
+ V3_Print("SVM core %u: SVM Exit Code: %p\n", info->cpu_id, (void *)(addr_t)guest_ctrl->exit_code);
- V3_Print("exit_info1 low = 0x%.8x\n", *(uint_t*)&(guest_ctrl->exit_info1));
- V3_Print("exit_info1 high = 0x%.8x\n", *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
+ V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
+ V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
- V3_Print("exit_info2 low = 0x%.8x\n", *(uint_t*)&(guest_ctrl->exit_info2));
- V3_Print("exit_info2 high = 0x%.8x\n", *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
+ V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
+ V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
if (info->mem_mode == PHYSICAL_MEM) {
- guest_pa_to_host_va(info, linear_addr, &host_addr);
+ v3_gpa_to_hva(info, linear_addr, &host_addr);
} else if (info->mem_mode == VIRTUAL_MEM) {
- guest_va_to_host_va(info, linear_addr, &host_addr);
+ v3_gva_to_hva(info, linear_addr, &host_addr);
}
- V3_Print("Host Address of rip = 0x%p\n", (void *)host_addr);
+ V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->cpu_id, (void *)host_addr);
- V3_Print("Instr (15 bytes) at %p:\n", (void *)host_addr);
+ V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->cpu_id, (void *)host_addr);
v3_dump_mem((uint8_t *)host_addr, 15);
v3_print_stack(info);
break;
}
+
+
+ if (info->vm_info->run_state == VM_STOPPED) {
+ info->core_run_state = CORE_STOPPED;
+ break;
+ }
+
/*
if ((info->num_exits % 5000) == 0) {
*/
}
+
+ // Need to take down the other cores on error...
+
return 0;
}
+void v3_deinit_svm_cpu(int cpu_id) {
+ reg_ex_t msr;
+ extern v3_cpu_arch_t v3_cpu_types[];
+ // reset SVM_VM_HSAVE_PA_MSR
+ // Does setting it to NULL disable??
+ msr.r_reg = 0;
+ v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
+
+ // Disable SVM?
+ v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
+ msr.e_reg.low &= ~EFER_MSR_svm_enable;
+ v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
+
+ v3_cpu_types[cpu_id] = V3_INVALID_CPU;
+
+ V3_FreePages((void *)host_vmcbs[cpu_id], 4);
+
+ V3_Print("Host CPU %d host area freed, and SVM disabled\n", cpu_id);
+ return;
+}
