2 * This file is part of the Palacios Virtual Machine Monitor developed
3 * by the V3VEE Project with funding from the United States National
4 * Science Foundation and the Department of Energy.
6 * The V3VEE Project is a joint project between Northwestern University
7 * and the University of New Mexico. You can find out more at
10 * Copyright (c) 2008, Peter Dinda <pdinda@northwestern.edu>
11 * Copyright (c) 2008, Jack Lange <jarusl@cs.northwestern.edu>
12 * Copyright (c) 2008, The V3VEE Project <http://www.v3vee.org>
13 * All rights reserved.
15 * Author: Peter Dinda <pdinda@northwestern.edu>
16 * Jack Lange <jarusl@cs.northwestern.edu>
18 * This is free software. You are permitted to use,
19 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
23 #include <palacios/vmx.h>
24 #include <palacios/vmm.h>
25 #include <palacios/vmx_handler.h>
26 #include <palacios/vmcs.h>
27 #include <palacios/vmx_lowlevel.h>
28 #include <palacios/vmm_lowlevel.h>
29 #include <palacios/vmm_ctrl_regs.h>
30 #include <palacios/vmm_config.h>
31 #include <palacios/vmm_time.h>
32 #include <palacios/vm_guest_mem.h>
33 #include <palacios/vmm_direct_paging.h>
34 #include <palacios/vmx_io.h>
35 #include <palacios/vmx_msr.h>
38 #ifndef CONFIG_DEBUG_VMX
40 #define PrintDebug(fmt, args...)
44 static addr_t host_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
45 static addr_t active_vmcs_ptrs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
47 extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
48 extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
50 static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
53 ret = vmcs_write(field,val);
55 if (ret != VMX_SUCCESS) {
56 PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
63 static int inline check_vmcs_read(vmcs_field_t field, void * val) {
66 ret = vmcs_read(field, val);
68 if (ret != VMX_SUCCESS) {
69 PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
78 static addr_t allocate_vmcs() {
80 struct vmcs_data * vmcs_page = NULL;
82 PrintDebug("Allocating page\n");
84 vmcs_page = (struct vmcs_data *)V3_VAddr(V3_AllocPages(1));
85 memset(vmcs_page, 0, 4096);
87 v3_get_msr(VMX_BASIC_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
89 vmcs_page->revision = ((struct vmx_basic_msr*)&msr)->revision;
90 PrintDebug("VMX Revision: 0x%x\n",vmcs_page->revision);
92 return (addr_t)V3_PAddr((void *)vmcs_page);
98 static int init_vmcs_bios(struct guest_info * info, struct vmx_data * vmx_state) {
100 struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
102 // disable global interrupts for vm state initialization
105 PrintDebug("Loading VMCS\n");
106 vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
107 active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys;
108 vmx_state->state = VMX_UNLAUNCHED;
110 if (vmx_ret != VMX_SUCCESS) {
111 PrintError("VMPTRLD failed\n");
117 /******* Setup Host State **********/
119 /* Cache GDTR, IDTR, and TR in host struct */
124 } __attribute__((packed)) tmp_seg;
127 __asm__ __volatile__(
133 gdtr_base = tmp_seg.base;
134 vmx_state->host_state.gdtr.base = gdtr_base;
136 __asm__ __volatile__(
142 vmx_state->host_state.idtr.base = tmp_seg.base;
144 __asm__ __volatile__(
150 vmx_state->host_state.tr.selector = tmp_seg.selector;
152 /* The GDTR *index* is bits 3-15 of the selector. */
153 struct tss_descriptor * desc = NULL;
154 desc = (struct tss_descriptor *)(gdtr_base + (8 * (tmp_seg.selector >> 3)));
156 tmp_seg.base = ((desc->base1) |
157 (desc->base2 << 16) |
158 (desc->base3 << 24) |
160 ((uint64_t)desc->base4 << 32)
166 vmx_state->host_state.tr.base = tmp_seg.base;
170 /********** Setup and VMX Control Fields from MSR ***********/
174 struct v3_msr tmp_msr;
176 v3_get_msr(VMX_PINBASED_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
178 /* Add external interrupts, NMI exiting, and virtual NMI */
179 vmx_state->pin_ctrls.value = tmp_msr.lo;
180 vmx_state->pin_ctrls.nmi_exit = 1;
181 vmx_state->pin_ctrls.ext_int_exit = 1;
183 v3_get_msr(VMX_PROCBASED_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
185 vmx_state->pri_proc_ctrls.value = tmp_msr.lo;
186 vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
187 vmx_state->pri_proc_ctrls.hlt_exit = 1;
188 vmx_state->pri_proc_ctrls.invlpg_exit = 1;
189 vmx_state->pri_proc_ctrls.use_msr_bitmap = 1;
190 vmx_state->pri_proc_ctrls.pause_exit = 1;
191 vmx_state->pri_proc_ctrls.tsc_offset = 1;
192 #ifdef CONFIG_TIME_VIRTUALIZE_TSC
193 vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
196 vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(info->vm_info->io_map.arch_data));
197 vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
198 (addr_t)V3_PAddr(info->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
201 vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(info->vm_info->msr_map.arch_data));
203 v3_get_msr(VMX_EXIT_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
204 vmx_state->exit_ctrls.value = tmp_msr.lo;
205 vmx_state->exit_ctrls.host_64_on = 1;
207 if ((vmx_state->exit_ctrls.save_efer == 1) || (vmx_state->exit_ctrls.ld_efer == 1)) {
208 vmx_state->ia32e_avail = 1;
211 v3_get_msr(VMX_ENTRY_CTLS_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
212 vmx_state->entry_ctrls.value = tmp_msr.lo;
215 struct vmx_exception_bitmap excp_bmap;
220 vmx_ret |= check_vmcs_write(VMCS_EXCP_BITMAP, excp_bmap.value);
222 /******* Setup VMXAssist guest state ***********/
225 info->vm_regs.rsp = 0x80000;
227 struct rflags * flags = (struct rflags *)&(info->ctrl_regs.rflags);
230 /* Print Control MSRs */
231 v3_get_msr(VMX_CR0_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
232 PrintDebug("CR0 MSR: %p\n", (void *)(addr_t)tmp_msr.value);
234 v3_get_msr(VMX_CR4_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
235 PrintDebug("CR4 MSR: %p\n", (void *)(addr_t)tmp_msr.value);
238 #define GUEST_CR0 0x80000031
239 #define GUEST_CR4 0x00002000
240 info->ctrl_regs.cr0 = GUEST_CR0;
241 info->ctrl_regs.cr4 = GUEST_CR4;
243 ((struct cr0_32 *)&(info->shdw_pg_state.guest_cr0))->pe = 1;
246 if (info->shdw_pg_mode == SHADOW_PAGING) {
247 PrintDebug("Creating initial shadow page table\n");
249 if (v3_init_passthrough_pts(info) == -1) {
250 PrintError("Could not initialize passthrough page tables\n");
254 #define CR0_PE 0x00000001
255 #define CR0_PG 0x80000000
258 vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG) );
259 vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE);
261 info->ctrl_regs.cr3 = info->direct_map_pt;
263 // vmx_state->pinbased_ctrls |= NMI_EXIT;
266 vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
267 vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
270 // Setup segment registers
272 struct v3_segment * seg_reg = (struct v3_segment *)&(info->segments);
276 for (i = 0; i < 10; i++) {
277 seg_reg[i].selector = 3 << 3;
278 seg_reg[i].limit = 0xffff;
279 seg_reg[i].base = 0x0;
282 info->segments.cs.selector = 2<<3;
284 /* Set only the segment registers */
285 for (i = 0; i < 6; i++) {
286 seg_reg[i].limit = 0xfffff;
287 seg_reg[i].granularity = 1;
289 seg_reg[i].system = 1;
291 seg_reg[i].present = 1;
295 info->segments.cs.type = 0xb;
297 info->segments.ldtr.selector = 0x20;
298 info->segments.ldtr.type = 2;
299 info->segments.ldtr.system = 0;
300 info->segments.ldtr.present = 1;
301 info->segments.ldtr.granularity = 0;
304 /************* Map in GDT and vmxassist *************/
306 uint64_t gdt[] __attribute__ ((aligned(32))) = {
307 0x0000000000000000ULL, /* 0x00: reserved */
308 0x0000830000000000ULL, /* 0x08: 32-bit TSS */
309 //0x0000890000000000ULL, /* 0x08: 32-bit TSS */
310 0x00CF9b000000FFFFULL, /* 0x10: CS 32-bit */
311 0x00CF93000000FFFFULL, /* 0x18: DS 32-bit */
312 0x000082000000FFFFULL, /* 0x20: LDTR 32-bit */
315 #define VMXASSIST_GDT 0x10000
316 addr_t vmxassist_gdt = 0;
318 if (v3_gpa_to_hva(info, VMXASSIST_GDT, &vmxassist_gdt) == -1) {
319 PrintError("Could not find VMXASSIST GDT destination\n");
323 memcpy((void *)vmxassist_gdt, gdt, sizeof(uint64_t) * 5);
325 info->segments.gdtr.base = VMXASSIST_GDT;
327 #define VMXASSIST_TSS 0x40000
328 uint64_t vmxassist_tss = VMXASSIST_TSS;
329 gdt[0x08 / sizeof(gdt[0])] |=
330 ((vmxassist_tss & 0xFF000000) << (56 - 24)) |
331 ((vmxassist_tss & 0x00FF0000) << (32 - 16)) |
332 ((vmxassist_tss & 0x0000FFFF) << (16)) |
335 info->segments.tr.selector = 0x08;
336 info->segments.tr.base = vmxassist_tss;
338 //info->segments.tr.type = 0x9;
339 info->segments.tr.type = 0x3;
340 info->segments.tr.system = 0;
341 info->segments.tr.present = 1;
342 info->segments.tr.granularity = 0;
347 #define VMXASSIST_START 0x000d0000
348 extern uint8_t v3_vmxassist_start[];
349 extern uint8_t v3_vmxassist_end[];
350 addr_t vmxassist_dst = 0;
352 if (v3_gpa_to_hva(info, VMXASSIST_START, &vmxassist_dst) == -1) {
353 PrintError("Could not find VMXASSIST destination\n");
357 memcpy((void *)vmxassist_dst, v3_vmxassist_start, v3_vmxassist_end - v3_vmxassist_start);
360 /*** Write all the info to the VMCS ***/
362 #define DEBUGCTL_MSR 0x1d9
363 v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
364 vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
366 info->dbg_regs.dr7 = 0x400;
369 vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
371 vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
372 vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
375 if (v3_update_vmcs_ctrl_fields(info)) {
376 PrintError("Could not write control fields!\n");
380 if (v3_update_vmcs_host_state(info)) {
381 PrintError("Could not write host state\n");
386 vmx_state->assist_state = VMXASSIST_DISABLED;
388 // reenable global interrupts for vm state initialization now
389 // that the vm state is initialized. If another VM kicks us off,
390 // it'll update our vmx state so that we know to reload ourself
396 int v3_init_vmx_vmcs(struct guest_info * info, v3_vm_class_t vm_class) {
397 struct vmx_data * vmx_state = NULL;
400 vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
402 PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
404 PrintDebug("Allocating VMCS\n");
405 vmx_state->vmcs_ptr_phys = allocate_vmcs();
407 PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
409 info->vmm_data = vmx_state;
410 vmx_state->state = VMX_UNLAUNCHED;
412 PrintDebug("Initializing VMCS (addr=%p)\n", info->vmm_data);
414 // TODO: Fix vmcs fields so they're 32-bit
416 PrintDebug("Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
417 vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
419 if (vmx_ret != VMX_SUCCESS) {
420 PrintError("VMCLEAR failed\n");
424 if (vm_class == V3_PC_VM) {
425 PrintDebug("Initializing VMCS\n");
426 init_vmcs_bios(info, vmx_state);
428 PrintError("Invalid VM Class\n");
436 int v3_deinit_vmx_vmcs(struct guest_info * core) {
437 struct vmx_data * vmx_state = core->vmm_data;
439 V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
447 static int update_irq_exit_state(struct guest_info * info) {
448 struct vmx_exit_idt_vec_info idt_vec_info;
450 check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
452 if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
453 #ifdef CONFIG_DEBUG_INTERRUPTS
454 PrintDebug("Calling v3_injecting_intr\n");
456 info->intr_core_state.irq_started = 0;
457 v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
463 static int update_irq_entry_state(struct guest_info * info) {
464 struct vmx_exit_idt_vec_info idt_vec_info;
465 struct vmcs_interrupt_state intr_core_state;
466 struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
468 check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
469 check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
471 /* Check for pending exceptions to inject */
472 if (v3_excp_pending(info)) {
473 struct vmx_entry_int_info int_info;
476 // In VMX, almost every exception is hardware
477 // Software exceptions are pretty much only for breakpoint or overflow
479 int_info.vector = v3_get_excp_number(info);
481 if (info->excp_state.excp_error_code_valid) {
482 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
483 int_info.error_code = 1;
485 #ifdef CONFIG_DEBUG_INTERRUPTS
486 PrintDebug("Injecting exception %d with error code %x\n",
487 int_info.vector, info->excp_state.excp_error_code);
492 #ifdef CONFIG_DEBUG_INTERRUPTS
493 PrintDebug("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
495 check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
497 v3_injecting_excp(info, int_info.vector);
499 } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
500 (intr_core_state.val == 0)) {
502 if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
504 #ifdef CONFIG_DEBUG_INTERRUPTS
505 PrintDebug("IRQ pending from previous injection\n");
508 // Copy the IDT vectoring info over to reinject the old interrupt
509 if (idt_vec_info.error_code == 1) {
510 uint32_t err_code = 0;
512 check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
513 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
516 idt_vec_info.undef = 0;
517 check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
520 struct vmx_entry_int_info ent_int;
523 switch (v3_intr_pending(info)) {
524 case V3_EXTERNAL_IRQ: {
525 info->intr_core_state.irq_vector = v3_get_intr(info);
526 ent_int.vector = info->intr_core_state.irq_vector;
528 ent_int.error_code = 0;
531 #ifdef CONFIG_DEBUG_INTERRUPTS
532 PrintDebug("Injecting Interrupt %d at exit %u(EIP=%p)\n",
533 info->intr_core_state.irq_vector,
534 (uint32_t)info->num_exits,
535 (void *)(addr_t)info->rip);
538 check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
539 info->intr_core_state.irq_started = 1;
544 PrintDebug("Injecting NMI\n");
549 check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
552 case V3_SOFTWARE_INTR:
553 PrintDebug("Injecting software interrupt\n");
557 check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
561 // Not sure what to do here, Intel doesn't have virtual IRQs
562 // May be the same as external interrupts/IRQs
565 case V3_INVALID_INTR:
570 } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
571 // Enable INTR window exiting so we know when IF=1
574 check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
576 #ifdef CONFIG_DEBUG_INTERRUPTS
577 PrintDebug("Enabling Interrupt-Window exiting: %d\n", instr_len);
580 vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
581 check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
590 static struct vmx_exit_info exit_log[10];
592 static void print_exit_log(struct guest_info * info) {
593 int cnt = info->num_exits % 10;
597 V3_Print("\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
599 for (i = 0; i < 10; i++) {
600 struct vmx_exit_info * tmp = &exit_log[cnt];
602 V3_Print("%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
603 V3_Print("\texit_qual = %p\n", (void *)tmp->exit_qual);
604 V3_Print("\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
605 V3_Print("\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
606 V3_Print("\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
619 * CAUTION and DANGER!!!
621 * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
622 * When exectuing a symbiotic call, the VMCS WILL be overwritten, so any dependencies
623 * on its contents will cause things to break. The contents at the time of the exit WILL
624 * change before the exit handler is executed.
626 int v3_vmx_enter(struct guest_info * info) {
628 uint32_t tsc_offset_low, tsc_offset_high;
629 struct vmx_exit_info exit_info;
630 struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
632 // Conditionally yield the CPU if the timeslice has expired
635 // Perform any additional yielding needed for time adjustment
636 v3_adjust_time(info);
638 // Update timer devices prior to entering VM.
639 v3_update_timers(info);
641 // disable global interrupts for vm state transition
644 v3_vmx_restore_vmcs(info);
647 #ifdef CONFIG_SYMCALL
648 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
649 update_irq_entry_state(info);
652 update_irq_entry_state(info);
657 vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
658 vmcs_write(VMCS_GUEST_CR3, guest_cr3);
661 // Perform last-minute time bookkeeping prior to entering the VM
662 v3_time_enter_vm(info);
664 tsc_offset_high = (uint32_t)((v3_tsc_host_offset(&info->time_state) >> 32) & 0xffffffff);
665 tsc_offset_low = (uint32_t)(v3_tsc_host_offset(&info->time_state) & 0xffffffff);
666 check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
667 check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
669 if (active_vmcs_ptrs[V3_Get_CPU()] != vmx_info->vmcs_ptr_phys) {
670 vmcs_load(vmx_info->vmcs_ptr_phys);
671 active_vmcs_ptrs[V3_Get_CPU()] = vmx_info->vmcs_ptr_phys;
674 if (vmx_info->state == VMX_UNLAUNCHED) {
675 vmx_info->state = VMX_LAUNCHED;
676 info->vm_info->run_state = VM_RUNNING;
677 ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
679 V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
680 ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
683 // PrintDebug("VMX Exit: ret=%d\n", ret);
685 if (ret != VMX_SUCCESS) {
688 vmcs_read(VMCS_INSTR_ERR, &error);
689 PrintError("VMENTRY Error: %d\n", error);
694 // Immediate exit from VM time bookkeeping
695 v3_time_exit_vm(info);
699 /* Update guest state */
700 v3_vmx_save_vmcs(info);
702 // info->cpl = info->segments.cs.selector & 0x3;
704 info->mem_mode = v3_get_vm_mem_mode(info);
705 info->cpu_mode = v3_get_vm_cpu_mode(info);
708 check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
709 check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
710 check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
711 check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
712 check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
713 check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
714 check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
716 //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
718 exit_log[info->num_exits % 10] = exit_info;
721 #ifdef CONFIG_SYMCALL
722 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
723 update_irq_exit_state(info);
726 update_irq_exit_state(info);
729 // Handle any exits needed still in the atomic section
730 if (v3_handle_atomic_vmx_exit(info, &exit_info) == -1) {
731 PrintError("Error in atomic VMX exit handler\n");
735 // reenable global interrupts after vm exit
738 // Conditionally yield the CPU if the timeslice has expired
741 if (v3_handle_vmx_exit(info, &exit_info) == -1) {
742 PrintError("Error in VMX exit handler\n");
750 int v3_start_vmx_guest(struct guest_info * info) {
752 PrintDebug("Starting VMX core %u\n", info->cpu_id);
754 if (info->cpu_id == 0) {
755 info->core_run_state = CORE_RUNNING;
756 info->vm_info->run_state = VM_RUNNING;
759 PrintDebug("VMX core %u: Waiting for core initialization\n", info->cpu_id);
761 while (info->core_run_state == CORE_STOPPED) {
763 //PrintDebug("VMX core %u: still waiting for INIT\n",info->cpu_id);
766 PrintDebug("VMX core %u initialized\n", info->cpu_id);
770 PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n",
771 info->cpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
772 info->segments.cs.limit, (void *)(info->rip));
775 PrintDebug("VMX core %u: Launching VMX VM\n", info->cpu_id);
781 if (info->vm_info->run_state == VM_STOPPED) {
782 info->core_run_state = CORE_STOPPED;
786 if (v3_vmx_enter(info) == -1) {
788 print_exit_log(info);
794 if (info->vm_info->run_state == VM_STOPPED) {
795 info->core_run_state = CORE_STOPPED;
799 if ((info->num_exits % 5000) == 0) {
800 V3_Print("VMX Exit number %d\n", (uint32_t)info->num_exits);
810 int v3_is_vmx_capable() {
811 v3_msr_t feature_msr;
812 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
814 v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
816 PrintDebug("ECX: 0x%x\n", ecx);
818 if (ecx & CPUID_1_ECX_VTXFLAG) {
819 v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
821 PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo);
823 if ((feature_msr.lo & FEATURE_CONTROL_VALID) != FEATURE_CONTROL_VALID) {
824 PrintDebug("VMX is locked -- enable in the BIOS\n");
829 PrintDebug("VMX not supported on this cpu\n");
836 static int has_vmx_nested_paging() {
842 void v3_init_vmx_cpu(int cpu_id) {
843 extern v3_cpu_arch_t v3_cpu_types[];
844 struct v3_msr tmp_msr;
847 v3_get_msr(VMX_CR4_FIXED0_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
850 __asm__ __volatile__ (
852 "orq $0x00002000, %%rbx;"
859 if ((~ret & tmp_msr.value) == 0) {
860 __asm__ __volatile__ (
866 PrintError("Invalid CR4 Settings!\n");
870 __asm__ __volatile__ (
871 "movq %%cr0, %%rbx; "
872 "orq $0x00000020,%%rbx; "
879 __asm__ __volatile__ (
881 "orl $0x00002000, %%ecx;"
888 if ((~ret & tmp_msr.value) == 0) {
889 __asm__ __volatile__ (
895 PrintError("Invalid CR4 Settings!\n");
899 __asm__ __volatile__ (
900 "movl %%cr0, %%ecx; "
901 "orl $0x00000020,%%ecx; "
911 // Should check and return Error here....
914 // Setup VMXON Region
915 host_vmcs_ptrs[cpu_id] = allocate_vmcs();
917 PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
919 if (v3_enable_vmx(host_vmcs_ptrs[cpu_id]) == VMX_SUCCESS) {
920 PrintDebug("VMX Enabled\n");
922 PrintError("VMX initialization failure\n");
927 if (has_vmx_nested_paging() == 1) {
928 v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
930 v3_cpu_types[cpu_id] = V3_VMX_CPU;
936 void v3_deinit_vmx_cpu(int cpu_id) {
937 extern v3_cpu_arch_t v3_cpu_types[];
938 v3_cpu_types[cpu_id] = V3_INVALID_CPU;
939 V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);