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) 2011, Jack Lange <jarusl@cs.northwestern.edu>
11 * Copyright (c) 2011, The V3VEE Project <http://www.v3vee.org>
12 * All rights reserved.
14 * Author: Jack Lange <jarusl@cs.northwestern.edu>
16 * This is free software. You are permitted to use,
17 * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
21 #include <palacios/vmx.h>
22 #include <palacios/vmm.h>
23 #include <palacios/vmx_handler.h>
24 #include <palacios/vmcs.h>
25 #include <palacios/vmx_lowlevel.h>
26 #include <palacios/vmm_lowlevel.h>
27 #include <palacios/vmm_ctrl_regs.h>
28 #include <palacios/vmm_config.h>
29 #include <palacios/vmm_time.h>
30 #include <palacios/vm_guest_mem.h>
31 #include <palacios/vmm_direct_paging.h>
32 #include <palacios/vmx_io.h>
33 #include <palacios/vmx_msr.h>
34 #include <palacios/vmm_decoder.h>
35 #include <palacios/vmm_barrier.h>
36 #include <palacios/vmm_timeout.h>
37 #include <palacios/vmm_debug.h>
39 #ifdef V3_CONFIG_CHECKPOINT
40 #include <palacios/vmm_checkpoint.h>
43 #include <palacios/vmx_ept.h>
44 #include <palacios/vmx_assist.h>
45 #include <palacios/vmx_hw_info.h>
47 #ifndef V3_CONFIG_DEBUG_VMX
49 #define PrintDebug(fmt, args...)
53 /* These fields contain the hardware feature sets supported by the local CPU */
54 static struct vmx_hw_info hw_info;
56 extern v3_cpu_arch_t v3_mach_type;
58 static addr_t host_vmcs_ptrs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
60 extern int v3_vmx_launch(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
61 extern int v3_vmx_resume(struct v3_gprs * vm_regs, struct guest_info * info, struct v3_ctrl_regs * ctrl_regs);
63 static int inline check_vmcs_write(vmcs_field_t field, addr_t val) {
66 ret = vmcs_write(field, val);
68 if (ret != VMX_SUCCESS) {
69 PrintError("VMWRITE error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
79 static int inline check_vmcs_read(vmcs_field_t field, void * val) {
82 ret = vmcs_read(field, val);
84 if (ret != VMX_SUCCESS) {
85 PrintError("VMREAD error on %s!: %d\n", v3_vmcs_field_to_str(field), ret);
94 static addr_t allocate_vmcs() {
96 struct vmcs_data * vmcs_page = NULL;
98 PrintDebug("Allocating page\n");
100 temp = V3_AllocPages(1);
102 PrintError("Cannot allocate VMCS\n");
105 vmcs_page = (struct vmcs_data *)V3_VAddr(temp);
106 memset(vmcs_page, 0, 4096);
108 vmcs_page->revision = hw_info.basic_info.revision;
109 PrintDebug("VMX Revision: 0x%x\n", vmcs_page->revision);
111 return (addr_t)V3_PAddr((void *)vmcs_page);
116 static int debug_efer_read(struct guest_info * core, uint_t msr, struct v3_msr * src, void * priv_data) {
117 struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer);
118 V3_Print("\n\nEFER READ (val = %p)\n", (void *)efer->value);
120 v3_print_guest_state(core);
124 src->value = efer->value;
128 static int debug_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data) {
129 struct v3_msr * efer = (struct v3_msr *)&(core->ctrl_regs.efer);
130 V3_Print("\n\nEFER WRITE (old_val = %p) (new_val = %p)\n", (void *)efer->value, (void *)src.value);
132 v3_print_guest_state(core);
135 efer->value = src.value;
142 static int init_vmcs_bios(struct guest_info * core, struct vmx_data * vmx_state) {
145 /* Get Available features */
146 struct vmx_pin_ctrls avail_pin_ctrls;
147 avail_pin_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.pin_ctrls));
151 // disable global interrupts for vm state initialization
154 PrintDebug("Loading VMCS\n");
155 vmx_ret = vmcs_load(vmx_state->vmcs_ptr_phys);
156 vmx_state->state = VMX_UNLAUNCHED;
158 if (vmx_ret != VMX_SUCCESS) {
159 PrintError("VMPTRLD failed\n");
164 /*** Setup default state from HW ***/
166 vmx_state->pin_ctrls.value = hw_info.pin_ctrls.def_val;
167 vmx_state->pri_proc_ctrls.value = hw_info.proc_ctrls.def_val;
168 vmx_state->exit_ctrls.value = hw_info.exit_ctrls.def_val;
169 vmx_state->entry_ctrls.value = hw_info.entry_ctrls.def_val;
170 vmx_state->sec_proc_ctrls.value = hw_info.sec_proc_ctrls.def_val;
172 /* Print Control MSRs */
173 V3_Print("CR0 MSR: req_val=%p, req_mask=%p\n", (void *)(addr_t)hw_info.cr0.req_val, (void *)(addr_t)hw_info.cr0.req_mask);
174 V3_Print("CR4 MSR: req_val=%p, req_mask=%p\n", (void *)(addr_t)hw_info.cr4.req_val, (void *)(addr_t)hw_info.cr4.req_mask);
178 /******* Setup Host State **********/
180 /* Cache GDTR, IDTR, and TR in host struct */
183 /********** Setup VMX Control Fields ***********/
185 /* Add external interrupts, NMI exiting, and virtual NMI */
186 vmx_state->pin_ctrls.nmi_exit = 1;
187 vmx_state->pin_ctrls.ext_int_exit = 1;
191 /* We enable the preemption timer by default to measure accurate guest time */
192 if (avail_pin_ctrls.active_preempt_timer) {
193 V3_Print("VMX Preemption Timer is available\n");
194 vmx_state->pin_ctrls.active_preempt_timer = 1;
195 vmx_state->exit_ctrls.save_preempt_timer = 1;
198 vmx_state->pri_proc_ctrls.hlt_exit = 1;
201 vmx_state->pri_proc_ctrls.pause_exit = 0;
202 vmx_state->pri_proc_ctrls.tsc_offset = 1;
203 #ifdef V3_CONFIG_TIME_VIRTUALIZE_TSC
204 vmx_state->pri_proc_ctrls.rdtsc_exit = 1;
208 vmx_state->pri_proc_ctrls.use_io_bitmap = 1;
209 vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_A_ADDR, (addr_t)V3_PAddr(core->vm_info->io_map.arch_data));
210 vmx_ret |= check_vmcs_write(VMCS_IO_BITMAP_B_ADDR,
211 (addr_t)V3_PAddr(core->vm_info->io_map.arch_data) + PAGE_SIZE_4KB);
214 vmx_state->pri_proc_ctrls.use_msr_bitmap = 1;
215 vmx_ret |= check_vmcs_write(VMCS_MSR_BITMAP, (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data));
220 // Ensure host runs in 64-bit mode at each VM EXIT
221 vmx_state->exit_ctrls.host_64_on = 1;
226 // Restore host's EFER register on each VM EXIT
227 vmx_state->exit_ctrls.ld_efer = 1;
229 // Save/restore guest's EFER register to/from VMCS on VM EXIT/ENTRY
230 vmx_state->exit_ctrls.save_efer = 1;
231 vmx_state->entry_ctrls.ld_efer = 1;
233 vmx_state->exit_ctrls.save_pat = 1;
234 vmx_state->exit_ctrls.ld_pat = 1;
235 vmx_state->entry_ctrls.ld_pat = 1;
237 /* Temporary GPF trap */
238 // vmx_state->excp_bmap.gp = 1;
240 // Setup Guests initial PAT field
241 vmx_ret |= check_vmcs_write(VMCS_GUEST_PAT, 0x0007040600070406LL);
243 // Capture CR8 mods so that we can keep the apic_tpr correct
244 vmx_state->pri_proc_ctrls.cr8_ld_exit = 1;
245 vmx_state->pri_proc_ctrls.cr8_str_exit = 1;
249 if (core->shdw_pg_mode == SHADOW_PAGING) {
250 PrintDebug("Creating initial shadow page table\n");
252 if (v3_init_passthrough_pts(core) == -1) {
253 PrintError("Could not initialize passthrough page tables\n");
257 #define CR0_PE 0x00000001
258 #define CR0_PG 0x80000000
259 #define CR0_WP 0x00010000 // To ensure mem hooks work
260 #define CR0_NE 0x00000020
261 vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP | CR0_NE));
264 // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
265 vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
267 core->ctrl_regs.cr3 = core->direct_map_pt;
269 // vmx_state->pinbased_ctrls |= NMI_EXIT;
272 vmx_state->pri_proc_ctrls.cr3_ld_exit = 1;
273 vmx_state->pri_proc_ctrls.cr3_str_exit = 1;
275 vmx_state->pri_proc_ctrls.invlpg_exit = 1;
277 /* Add page fault exits */
278 vmx_state->excp_bmap.pf = 1;
281 v3_vmxassist_init(core, vmx_state);
283 // Hook all accesses to EFER register
284 v3_hook_msr(core->vm_info, EFER_MSR,
285 &v3_handle_efer_read,
286 &v3_handle_efer_write,
289 } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
290 (v3_mach_type == V3_VMX_EPT_CPU)) {
292 #define CR0_PE 0x00000001
293 #define CR0_PG 0x80000000
294 #define CR0_WP 0x00010000 // To ensure mem hooks work
295 #define CR0_NE 0x00000020
296 vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, (CR0_PE | CR0_PG | CR0_WP | CR0_NE));
298 // vmx_state->pinbased_ctrls |= NMI_EXIT;
300 // Cause VM_EXIT whenever CR4.VMXE or CR4.PAE bits are written
301 vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE | CR4_PAE);
303 /* Disable CR exits */
304 vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
305 vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
307 vmx_state->pri_proc_ctrls.invlpg_exit = 0;
309 /* Add page fault exits */
310 // vmx_state->excp_bmap.pf = 1; // This should never happen..., enabled to catch bugs
313 v3_vmxassist_init(core, vmx_state);
316 vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
317 vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
321 if (v3_init_ept(core, &hw_info) == -1) {
322 PrintError("Error initializing EPT\n");
326 // Hook all accesses to EFER register
327 v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL);
329 } else if ((core->shdw_pg_mode == NESTED_PAGING) &&
330 (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
332 // For now we will assume that unrestricted guest mode is assured w/ EPT
335 core->vm_regs.rsp = 0x00;
337 core->vm_regs.rdx = 0x00000f00;
338 core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
339 core->ctrl_regs.cr0 = 0x60010030;
340 core->ctrl_regs.cr4 = 0x00002010; // Enable VMX and PSE flag
343 core->segments.cs.selector = 0xf000;
344 core->segments.cs.limit = 0xffff;
345 core->segments.cs.base = 0x0000000f0000LL;
347 // (raw attributes = 0xf3)
348 core->segments.cs.type = 0xb;
349 core->segments.cs.system = 0x1;
350 core->segments.cs.dpl = 0x0;
351 core->segments.cs.present = 1;
355 struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
356 &(core->segments.es), &(core->segments.fs),
357 &(core->segments.gs), NULL};
359 for ( i = 0; segregs[i] != NULL; i++) {
360 struct v3_segment * seg = segregs[i];
362 seg->selector = 0x0000;
363 // seg->base = seg->selector << 4;
364 seg->base = 0x00000000;
372 // seg->granularity = 1;
377 core->segments.gdtr.limit = 0x0000ffff;
378 core->segments.gdtr.base = 0x0000000000000000LL;
380 core->segments.idtr.limit = 0x0000ffff;
381 core->segments.idtr.base = 0x0000000000000000LL;
383 core->segments.ldtr.selector = 0x0000;
384 core->segments.ldtr.limit = 0x0000ffff;
385 core->segments.ldtr.base = 0x0000000000000000LL;
386 core->segments.ldtr.type = 0x2;
387 core->segments.ldtr.present = 1;
389 core->segments.tr.selector = 0x0000;
390 core->segments.tr.limit = 0x0000ffff;
391 core->segments.tr.base = 0x0000000000000000LL;
392 core->segments.tr.type = 0xb;
393 core->segments.tr.present = 1;
395 // core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
396 core->dbg_regs.dr7 = 0x0000000000000400LL;
399 vmx_state->pri_proc_ctrls.sec_ctrls = 1; // Enable secondary proc controls
400 vmx_state->sec_proc_ctrls.enable_ept = 1; // enable EPT paging
401 vmx_state->sec_proc_ctrls.unrstrct_guest = 1; // enable unrestricted guest operation
404 /* Disable shadow paging stuff */
405 vmx_state->pri_proc_ctrls.cr3_ld_exit = 0;
406 vmx_state->pri_proc_ctrls.cr3_str_exit = 0;
408 vmx_state->pri_proc_ctrls.invlpg_exit = 0;
411 // Cause VM_EXIT whenever the CR4.VMXE bit is set
412 vmx_ret |= check_vmcs_write(VMCS_CR4_MASK, CR4_VMXE);
413 #define CR0_NE 0x00000020
414 vmx_ret |= check_vmcs_write(VMCS_CR0_MASK, CR0_NE);
415 ((struct cr0_32 *)&(core->shdw_pg_state.guest_cr0))->ne = 1;
417 if (v3_init_ept(core, &hw_info) == -1) {
418 PrintError("Error initializing EPT\n");
422 // Hook all accesses to EFER register
423 // v3_hook_msr(core->vm_info, EFER_MSR, &debug_efer_read, &debug_efer_write, core);
424 v3_hook_msr(core->vm_info, EFER_MSR, NULL, NULL, NULL);
426 PrintError("Invalid Virtual paging mode (pg_mode=%d) (mach_type=%d)\n", core->shdw_pg_mode, v3_mach_type);
433 // Setup SYSCALL/SYSENTER MSRs in load/store area
435 // save STAR, LSTAR, FMASK, KERNEL_GS_BASE MSRs in MSR load/store area
438 struct vmcs_msr_save_area * msr_entries = NULL;
439 int max_msrs = (hw_info.misc_info.max_msr_cache_size + 1) * 4;
442 V3_Print("Setting up MSR load/store areas (max_msr_count=%d)\n", max_msrs);
445 PrintError("Max MSR cache size is too small (%d)\n", max_msrs);
449 vmx_state->msr_area_paddr = (addr_t)V3_AllocPages(1);
451 if (vmx_state->msr_area_paddr == (addr_t)NULL) {
452 PrintError("could not allocate msr load/store area\n");
456 msr_entries = (struct vmcs_msr_save_area *)V3_VAddr((void *)(vmx_state->msr_area_paddr));
457 vmx_state->msr_area = msr_entries; // cache in vmx_info
459 memset(msr_entries, 0, PAGE_SIZE);
461 msr_entries->guest_star.index = IA32_STAR_MSR;
462 msr_entries->guest_lstar.index = IA32_LSTAR_MSR;
463 msr_entries->guest_fmask.index = IA32_FMASK_MSR;
464 msr_entries->guest_kern_gs.index = IA32_KERN_GS_BASE_MSR;
466 msr_entries->host_star.index = IA32_STAR_MSR;
467 msr_entries->host_lstar.index = IA32_LSTAR_MSR;
468 msr_entries->host_fmask.index = IA32_FMASK_MSR;
469 msr_entries->host_kern_gs.index = IA32_KERN_GS_BASE_MSR;
471 msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_CNT, 4);
472 msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_CNT, 4);
473 msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_CNT, 4);
475 msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_STORE_ADDR, (addr_t)V3_PAddr(msr_entries->guest_msrs));
476 msr_ret |= check_vmcs_write(VMCS_ENTRY_MSR_LOAD_ADDR, (addr_t)V3_PAddr(msr_entries->guest_msrs));
477 msr_ret |= check_vmcs_write(VMCS_EXIT_MSR_LOAD_ADDR, (addr_t)V3_PAddr(msr_entries->host_msrs));
480 msr_ret |= v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
481 msr_ret |= v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
482 msr_ret |= v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
483 msr_ret |= v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
486 // IMPORTANT: These MSRs appear to be cached by the hardware....
487 msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
488 msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
489 msr_ret |= v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
491 msr_ret |= v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
492 msr_ret |= v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
494 msr_ret |= v3_hook_msr(core->vm_info, IA32_PAT_MSR, NULL, NULL, NULL);
496 // Not sure what to do about this... Does not appear to be an explicit hardware cache version...
497 msr_ret |= v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
500 PrintError("Error configuring MSR save/restore area\n");
507 /* Sanity check ctrl/reg fields against hw_defaults */
512 /*** Write all the info to the VMCS ***/
516 // IS THIS NECESSARY???
517 #define DEBUGCTL_MSR 0x1d9
518 struct v3_msr tmp_msr;
519 v3_get_msr(DEBUGCTL_MSR, &(tmp_msr.hi), &(tmp_msr.lo));
520 vmx_ret |= check_vmcs_write(VMCS_GUEST_DBG_CTL, tmp_msr.value);
521 core->dbg_regs.dr7 = 0x400;
526 vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffffffffffULL);
528 vmx_ret |= check_vmcs_write(VMCS_LINK_PTR, (addr_t)0xffffffffUL);
529 vmx_ret |= check_vmcs_write(VMCS_LINK_PTR_HIGH, (addr_t)0xffffffffUL);
536 if (v3_update_vmcs_ctrl_fields(core)) {
537 PrintError("Could not write control fields!\n");
542 if (v3_update_vmcs_host_state(core)) {
543 PrintError("Could not write host state\n");
548 // reenable global interrupts for vm state initialization now
549 // that the vm state is initialized. If another VM kicks us off,
550 // it'll update our vmx state so that we know to reload ourself
557 static void __init_vmx_vmcs(void * arg) {
558 struct guest_info * core = arg;
559 struct vmx_data * vmx_state = NULL;
562 vmx_state = (struct vmx_data *)V3_Malloc(sizeof(struct vmx_data));
565 PrintError("Unable to allocate in initializing vmx vmcs\n");
569 memset(vmx_state, 0, sizeof(struct vmx_data));
571 PrintDebug("vmx_data pointer: %p\n", (void *)vmx_state);
573 PrintDebug("Allocating VMCS\n");
574 vmx_state->vmcs_ptr_phys = allocate_vmcs();
576 PrintDebug("VMCS pointer: %p\n", (void *)(vmx_state->vmcs_ptr_phys));
578 core->vmm_data = vmx_state;
579 vmx_state->state = VMX_UNLAUNCHED;
581 PrintDebug("Initializing VMCS (addr=%p)\n", core->vmm_data);
583 // TODO: Fix vmcs fields so they're 32-bit
585 PrintDebug("Clearing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
586 vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
588 if (vmx_ret != VMX_SUCCESS) {
589 PrintError("VMCLEAR failed\n");
593 if (core->vm_info->vm_class == V3_PC_VM) {
594 PrintDebug("Initializing VMCS\n");
595 if (init_vmcs_bios(core, vmx_state) == -1) {
596 PrintError("Error initializing VMCS to BIOS state\n");
600 PrintError("Invalid VM Class\n");
604 PrintDebug("Serializing VMCS: %p\n", (void *)vmx_state->vmcs_ptr_phys);
605 vmx_ret = vmcs_clear(vmx_state->vmcs_ptr_phys);
607 core->core_run_state = CORE_STOPPED;
613 int v3_init_vmx_vmcs(struct guest_info * core, v3_vm_class_t vm_class) {
614 extern v3_cpu_arch_t v3_cpu_types[];
616 if (v3_cpu_types[V3_Get_CPU()] == V3_INVALID_CPU) {
619 for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
620 if (v3_cpu_types[i] != V3_INVALID_CPU) {
625 if (i == V3_CONFIG_MAX_CPUS) {
626 PrintError("Could not find VALID CPU for VMX guest initialization\n");
630 V3_Call_On_CPU(i, __init_vmx_vmcs, core);
633 __init_vmx_vmcs(core);
636 if (core->core_run_state != CORE_STOPPED) {
637 PrintError("Error initializing VMX Core\n");
645 int v3_deinit_vmx_vmcs(struct guest_info * core) {
646 struct vmx_data * vmx_state = core->vmm_data;
648 V3_FreePages((void *)(vmx_state->vmcs_ptr_phys), 1);
649 V3_FreePages(V3_PAddr(vmx_state->msr_area), 1);
658 #ifdef V3_CONFIG_CHECKPOINT
660 * JRL: This is broken
662 int v3_vmx_save_core(struct guest_info * core, void * ctx){
663 struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
665 // note that the vmcs pointer is an HPA, but we need an HVA
666 if (v3_chkpt_save(ctx, "vmcs_data", PAGE_SIZE_4KB,
667 V3_VAddr((void*) (vmx_info->vmcs_ptr_phys))) ==-1) {
668 PrintError("Could not save vmcs data for VMX\n");
675 int v3_vmx_load_core(struct guest_info * core, void * ctx){
676 struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
677 struct cr0_32 * shadow_cr0;
678 addr_t vmcs_page_paddr; //HPA
680 vmcs_page_paddr = (addr_t) V3_AllocPages(1);
682 if (!vmcs_page_paddr) {
683 PrintError("Could not allocate space for a vmcs in VMX\n");
687 if (v3_chkpt_load(ctx, "vmcs_data", PAGE_SIZE_4KB,
688 V3_VAddr((void *)vmcs_page_paddr)) == -1) {
689 PrintError("Could not load vmcs data for VMX\n");
693 vmcs_clear(vmx_info->vmcs_ptr_phys);
695 // Probably need to delete the old one...
696 V3_FreePages((void*)(vmx_info->vmcs_ptr_phys),1);
698 vmcs_load(vmcs_page_paddr);
700 v3_vmx_save_vmcs(core);
702 shadow_cr0 = (struct cr0_32 *)&(core->ctrl_regs.cr0);
705 /* Get the CPU mode to set the guest_ia32e entry ctrl */
707 if (core->shdw_pg_mode == SHADOW_PAGING) {
708 if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) {
709 if (v3_activate_shadow_pt(core) == -1) {
710 PrintError("Failed to activate shadow page tables\n");
714 if (v3_activate_passthrough_pt(core) == -1) {
715 PrintError("Failed to activate passthrough page tables\n");
726 void v3_flush_vmx_vm_core(struct guest_info * core) {
727 struct vmx_data * vmx_info = (struct vmx_data *)(core->vmm_data);
728 vmcs_clear(vmx_info->vmcs_ptr_phys);
729 vmx_info->state = VMX_UNLAUNCHED;
734 static int update_irq_exit_state(struct guest_info * info) {
735 struct vmx_exit_idt_vec_info idt_vec_info;
737 check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
739 if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 0)) {
740 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
741 V3_Print("Calling v3_injecting_intr\n");
743 info->intr_core_state.irq_started = 0;
744 v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
750 static int update_irq_entry_state(struct guest_info * info) {
751 struct vmx_exit_idt_vec_info idt_vec_info;
752 struct vmcs_interrupt_state intr_core_state;
753 struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
755 check_vmcs_read(VMCS_IDT_VECTOR_INFO, &(idt_vec_info.value));
756 check_vmcs_read(VMCS_GUEST_INT_STATE, &(intr_core_state));
758 /* Check for pending exceptions to inject */
759 if (v3_excp_pending(info)) {
760 struct vmx_entry_int_info int_info;
763 // In VMX, almost every exception is hardware
764 // Software exceptions are pretty much only for breakpoint or overflow
766 int_info.vector = v3_get_excp_number(info);
768 if (info->excp_state.excp_error_code_valid) {
769 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, info->excp_state.excp_error_code);
770 int_info.error_code = 1;
772 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
773 V3_Print("Injecting exception %d with error code %x\n",
774 int_info.vector, info->excp_state.excp_error_code);
779 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
780 V3_Print("Injecting exception %d (EIP=%p)\n", int_info.vector, (void *)(addr_t)info->rip);
782 check_vmcs_write(VMCS_ENTRY_INT_INFO, int_info.value);
784 v3_injecting_excp(info, int_info.vector);
786 } else if ((((struct rflags *)&(info->ctrl_regs.rflags))->intr == 1) &&
787 (intr_core_state.val == 0)) {
789 if ((info->intr_core_state.irq_started == 1) && (idt_vec_info.valid == 1)) {
791 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
792 V3_Print("IRQ pending from previous injection\n");
795 // Copy the IDT vectoring info over to reinject the old interrupt
796 if (idt_vec_info.error_code == 1) {
797 uint32_t err_code = 0;
799 check_vmcs_read(VMCS_IDT_VECTOR_ERR, &err_code);
800 check_vmcs_write(VMCS_ENTRY_EXCP_ERR, err_code);
803 idt_vec_info.undef = 0;
804 check_vmcs_write(VMCS_ENTRY_INT_INFO, idt_vec_info.value);
807 struct vmx_entry_int_info ent_int;
810 switch (v3_intr_pending(info)) {
811 case V3_EXTERNAL_IRQ: {
812 info->intr_core_state.irq_vector = v3_get_intr(info);
813 ent_int.vector = info->intr_core_state.irq_vector;
815 ent_int.error_code = 0;
818 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
819 V3_Print("Injecting Interrupt %d at exit %u(EIP=%p)\n",
820 info->intr_core_state.irq_vector,
821 (uint32_t)info->num_exits,
822 (void *)(addr_t)info->rip);
825 check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
826 info->intr_core_state.irq_started = 1;
831 PrintDebug("Injecting NMI\n");
836 check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
839 case V3_SOFTWARE_INTR:
840 PrintDebug("Injecting software interrupt\n");
844 check_vmcs_write(VMCS_ENTRY_INT_INFO, ent_int.value);
848 // Not sure what to do here, Intel doesn't have virtual IRQs
849 // May be the same as external interrupts/IRQs
852 case V3_INVALID_INTR:
857 } else if ((v3_intr_pending(info)) && (vmx_info->pri_proc_ctrls.int_wndw_exit == 0)) {
858 // Enable INTR window exiting so we know when IF=1
861 check_vmcs_read(VMCS_EXIT_INSTR_LEN, &instr_len);
863 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
864 V3_Print("Enabling Interrupt-Window exiting: %d\n", instr_len);
867 vmx_info->pri_proc_ctrls.int_wndw_exit = 1;
868 check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
877 static struct vmx_exit_info exit_log[10];
878 static uint64_t rip_log[10];
882 static void print_exit_log(struct guest_info * info) {
883 int cnt = info->num_exits % 10;
887 V3_Print("\nExit Log (%d total exits):\n", (uint32_t)info->num_exits);
889 for (i = 0; i < 10; i++) {
890 struct vmx_exit_info * tmp = &exit_log[cnt];
892 V3_Print("%d:\texit_reason = %p\n", i, (void *)(addr_t)tmp->exit_reason);
893 V3_Print("\texit_qual = %p\n", (void *)tmp->exit_qual);
894 V3_Print("\tint_info = %p\n", (void *)(addr_t)tmp->int_info);
895 V3_Print("\tint_err = %p\n", (void *)(addr_t)tmp->int_err);
896 V3_Print("\tinstr_info = %p\n", (void *)(addr_t)tmp->instr_info);
897 V3_Print("\tguest_linear_addr= %p\n", (void *)(addr_t)tmp->guest_linear_addr);
898 V3_Print("\tRIP = %p\n", (void *)rip_log[cnt]);
912 v3_vmx_config_tsc_virtualization(struct guest_info * info) {
913 struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
915 if (info->time_state.flags & VM_TIME_TRAP_RDTSC) {
916 if (!vmx_info->pri_proc_ctrls.rdtsc_exit) {
917 vmx_info->pri_proc_ctrls.rdtsc_exit = 1;
918 check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
922 uint32_t tsc_offset_low, tsc_offset_high;
924 if (vmx_info->pri_proc_ctrls.rdtsc_exit) {
925 vmx_info->pri_proc_ctrls.rdtsc_exit = 0;
926 check_vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
929 if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) {
932 tsc_offset = v3_tsc_host_offset(&info->time_state);
934 tsc_offset_high = (uint32_t)(( tsc_offset >> 32) & 0xffffffff);
935 tsc_offset_low = (uint32_t)(tsc_offset & 0xffffffff);
937 check_vmcs_write(VMCS_TSC_OFFSET_HIGH, tsc_offset_high);
938 check_vmcs_write(VMCS_TSC_OFFSET, tsc_offset_low);
944 * CAUTION and DANGER!!!
946 * The VMCS CANNOT(!!) be accessed outside of the cli/sti calls inside this function
947 * When exectuing a symbiotic call, the VMCS WILL be overwritten, so any dependencies
948 * on its contents will cause things to break. The contents at the time of the exit WILL
949 * change before the exit handler is executed.
951 int v3_vmx_enter(struct guest_info * info) {
953 struct vmx_exit_info exit_info;
954 struct vmx_data * vmx_info = (struct vmx_data *)(info->vmm_data);
955 uint64_t guest_cycles = 0;
957 // Conditionally yield the CPU if the timeslice has expired
958 v3_yield_cond(info,-1);
960 // Update timer devices late after being in the VM so that as much
961 // of the time in the VM is accounted for as possible. Also do it before
962 // updating IRQ entry state so that any interrupts the timers raise get
963 // handled on the next VM entry.
964 v3_advance_time(info, NULL);
965 v3_update_timers(info);
967 // disable global interrupts for vm state transition
970 if (vmcs_store() != vmx_info->vmcs_ptr_phys) {
971 vmcs_clear(vmx_info->vmcs_ptr_phys);
972 vmcs_load(vmx_info->vmcs_ptr_phys);
973 vmx_info->state = VMX_UNLAUNCHED;
976 v3_vmx_restore_vmcs(info);
979 #ifdef V3_CONFIG_SYMCALL
980 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
981 update_irq_entry_state(info);
984 update_irq_entry_state(info);
989 vmcs_read(VMCS_GUEST_CR3, &guest_cr3);
990 vmcs_write(VMCS_GUEST_CR3, guest_cr3);
994 // Perform last-minute time setup prior to entering the VM
995 v3_vmx_config_tsc_virtualization(info);
997 if (v3_update_vmcs_host_state(info)) {
999 PrintError("Could not write host state\n");
1003 if (vmx_info->pin_ctrls.active_preempt_timer) {
1004 /* Preemption timer is active */
1005 uint32_t preempt_window = 0xffffffff;
1007 if (info->timeouts.timeout_active) {
1008 preempt_window = info->timeouts.next_timeout;
1011 check_vmcs_write(VMCS_PREEMPT_TIMER, preempt_window);
1016 uint64_t entry_tsc = 0;
1017 uint64_t exit_tsc = 0;
1019 if (vmx_info->state == VMX_UNLAUNCHED) {
1020 vmx_info->state = VMX_LAUNCHED;
1022 ret = v3_vmx_launch(&(info->vm_regs), info, &(info->ctrl_regs));
1026 V3_ASSERT(vmx_info->state != VMX_UNLAUNCHED);
1028 ret = v3_vmx_resume(&(info->vm_regs), info, &(info->ctrl_regs));
1032 guest_cycles = exit_tsc - entry_tsc;
1035 // PrintDebug("VMX Exit: ret=%d\n", ret);
1037 if (ret != VMX_SUCCESS) {
1039 vmcs_read(VMCS_INSTR_ERR, &error);
1043 PrintError("VMENTRY Error: %d (launch_ret = %d)\n", error, ret);
1050 /* If we have the preemption time, then use it to get more accurate guest time */
1051 if (vmx_info->pin_ctrls.active_preempt_timer) {
1052 uint32_t cycles_left = 0;
1053 check_vmcs_read(VMCS_PREEMPT_TIMER, &(cycles_left));
1055 if (info->timeouts.timeout_active) {
1056 guest_cycles = info->timeouts.next_timeout - cycles_left;
1058 guest_cycles = 0xffffffff - cycles_left;
1062 // Immediate exit from VM time bookkeeping
1063 v3_advance_time(info, &guest_cycles);
1065 /* Update guest state */
1066 v3_vmx_save_vmcs(info);
1068 // info->cpl = info->segments.cs.selector & 0x3;
1070 info->mem_mode = v3_get_vm_mem_mode(info);
1071 info->cpu_mode = v3_get_vm_cpu_mode(info);
1075 check_vmcs_read(VMCS_EXIT_INSTR_LEN, &(exit_info.instr_len));
1076 check_vmcs_read(VMCS_EXIT_INSTR_INFO, &(exit_info.instr_info));
1077 check_vmcs_read(VMCS_EXIT_REASON, &(exit_info.exit_reason));
1078 check_vmcs_read(VMCS_EXIT_QUAL, &(exit_info.exit_qual));
1079 check_vmcs_read(VMCS_EXIT_INT_INFO, &(exit_info.int_info));
1080 check_vmcs_read(VMCS_EXIT_INT_ERR, &(exit_info.int_err));
1081 check_vmcs_read(VMCS_GUEST_LINEAR_ADDR, &(exit_info.guest_linear_addr));
1083 if (info->shdw_pg_mode == NESTED_PAGING) {
1084 check_vmcs_read(VMCS_GUEST_PHYS_ADDR, &(exit_info.ept_fault_addr));
1087 //PrintDebug("VMX Exit taken, id-qual: %u-%lu\n", exit_info.exit_reason, exit_info.exit_qual);
1089 exit_log[info->num_exits % 10] = exit_info;
1090 rip_log[info->num_exits % 10] = get_addr_linear(info, info->rip, &(info->segments.cs));
1092 #ifdef V3_CONFIG_SYMCALL
1093 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
1094 update_irq_exit_state(info);
1097 update_irq_exit_state(info);
1100 if (exit_info.exit_reason == VMX_EXIT_INTR_WINDOW) {
1101 // This is a special case whose only job is to inject an interrupt
1102 vmcs_read(VMCS_PROC_CTRLS, &(vmx_info->pri_proc_ctrls.value));
1103 vmx_info->pri_proc_ctrls.int_wndw_exit = 0;
1104 vmcs_write(VMCS_PROC_CTRLS, vmx_info->pri_proc_ctrls.value);
1106 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
1107 V3_Print("Interrupts available again! (RIP=%llx)\n", info->rip);
1111 // reenable global interrupts after vm exit
1114 // Conditionally yield the CPU if the timeslice has expired
1115 v3_yield_cond(info,-1);
1116 v3_advance_time(info, NULL);
1117 v3_update_timers(info);
1119 if (v3_handle_vmx_exit(info, &exit_info) == -1) {
1120 PrintError("Error in VMX exit handler (Exit reason=%x)\n", exit_info.exit_reason);
1124 if (info->timeouts.timeout_active) {
1125 /* Check to see if any timeouts have expired */
1126 v3_handle_timeouts(info, guest_cycles);
1133 int v3_start_vmx_guest(struct guest_info * info) {
1135 PrintDebug("Starting VMX core %u\n", info->vcpu_id);
1137 if (info->vcpu_id == 0) {
1138 info->core_run_state = CORE_RUNNING;
1141 PrintDebug("VMX core %u: Waiting for core initialization\n", info->vcpu_id);
1143 while (info->core_run_state == CORE_STOPPED) {
1145 if (info->vm_info->run_state == VM_STOPPED) {
1146 // The VM was stopped before this core was initialized.
1151 //PrintDebug("VMX core %u: still waiting for INIT\n",info->vcpu_id);
1154 PrintDebug("VMX core %u initialized\n", info->vcpu_id);
1156 // We'll be paranoid about race conditions here
1157 v3_wait_at_barrier(info);
1161 PrintDebug("VMX core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n",
1162 info->vcpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
1163 info->segments.cs.limit, (void *)(info->rip));
1166 PrintDebug("VMX core %u: Launching VMX VM on logical core %u\n", info->vcpu_id, info->pcpu_id);
1168 v3_start_time(info);
1172 if (info->vm_info->run_state == VM_STOPPED) {
1173 info->core_run_state = CORE_STOPPED;
1177 if (v3_vmx_enter(info) == -1) {
1180 addr_t linear_addr = 0;
1182 info->vm_info->run_state = VM_ERROR;
1184 V3_Print("VMX core %u: VMX ERROR!!\n", info->vcpu_id);
1186 v3_print_guest_state(info);
1188 V3_Print("VMX core %u\n", info->vcpu_id);
1190 linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
1192 if (info->mem_mode == PHYSICAL_MEM) {
1193 v3_gpa_to_hva(info, linear_addr, &host_addr);
1194 } else if (info->mem_mode == VIRTUAL_MEM) {
1195 v3_gva_to_hva(info, linear_addr, &host_addr);
1198 V3_Print("VMX core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
1200 V3_Print("VMX core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
1201 v3_dump_mem((uint8_t *)host_addr, 15);
1203 v3_print_stack(info);
1207 print_exit_log(info);
1211 v3_wait_at_barrier(info);
1214 if (info->vm_info->run_state == VM_STOPPED) {
1215 info->core_run_state = CORE_STOPPED;
1219 if ((info->num_exits % 5000) == 0) {
1220 V3_Print("VMX Exit number %d\n", (uint32_t)info->num_exits);
1232 #define VMX_FEATURE_CONTROL_MSR 0x0000003a
1233 #define CPUID_VMX_FEATURES 0x00000005 /* LOCK and VMXON */
1234 #define CPUID_1_ECX_VTXFLAG 0x00000020
1236 int v3_is_vmx_capable() {
1237 v3_msr_t feature_msr;
1238 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1240 v3_cpuid(0x1, &eax, &ebx, &ecx, &edx);
1242 PrintDebug("ECX: 0x%x\n", ecx);
1244 if (ecx & CPUID_1_ECX_VTXFLAG) {
1245 v3_get_msr(VMX_FEATURE_CONTROL_MSR, &(feature_msr.hi), &(feature_msr.lo));
1247 PrintDebug("MSRREGlow: 0x%.8x\n", feature_msr.lo);
1249 if ((feature_msr.lo & CPUID_VMX_FEATURES) != CPUID_VMX_FEATURES) {
1250 PrintDebug("VMX is locked -- enable in the BIOS\n");
1255 PrintDebug("VMX not supported on this cpu\n");
1263 int v3_reset_vmx_vm_core(struct guest_info * core, addr_t rip) {
1266 if ((core->shdw_pg_mode == NESTED_PAGING) &&
1267 (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
1270 core->segments.cs.selector = rip << 8;
1271 core->segments.cs.limit = 0xffff;
1272 core->segments.cs.base = rip << 12;
1274 core->vm_regs.rdx = core->vcpu_id;
1275 core->vm_regs.rbx = rip;
1283 void v3_init_vmx_cpu(int cpu_id) {
1284 addr_t vmx_on_region = 0;
1285 extern v3_cpu_arch_t v3_mach_type;
1286 extern v3_cpu_arch_t v3_cpu_types[];
1288 if (v3_mach_type == V3_INVALID_CPU) {
1289 if (v3_init_vmx_hw(&hw_info) == -1) {
1290 PrintError("Could not initialize VMX hardware features on cpu %d\n", cpu_id);
1298 // Setup VMXON Region
1299 vmx_on_region = allocate_vmcs();
1302 if (vmx_on(vmx_on_region) == VMX_SUCCESS) {
1303 V3_Print("VMX Enabled\n");
1304 host_vmcs_ptrs[cpu_id] = vmx_on_region;
1306 V3_Print("VMX already enabled\n");
1307 V3_FreePages((void *)vmx_on_region, 1);
1310 PrintDebug("VMXON pointer: 0x%p\n", (void *)host_vmcs_ptrs[cpu_id]);
1313 struct vmx_sec_proc_ctrls sec_proc_ctrls;
1314 sec_proc_ctrls.value = v3_vmx_get_ctrl_features(&(hw_info.sec_proc_ctrls));
1316 if (sec_proc_ctrls.enable_ept == 0) {
1317 V3_Print("VMX EPT (Nested) Paging not supported\n");
1318 v3_cpu_types[cpu_id] = V3_VMX_CPU;
1319 } else if (sec_proc_ctrls.unrstrct_guest == 0) {
1320 V3_Print("VMX EPT (Nested) Paging supported\n");
1321 v3_cpu_types[cpu_id] = V3_VMX_EPT_CPU;
1323 V3_Print("VMX EPT (Nested) Paging + Unrestricted guest supported\n");
1324 v3_cpu_types[cpu_id] = V3_VMX_EPT_UG_CPU;
1331 void v3_deinit_vmx_cpu(int cpu_id) {
1332 extern v3_cpu_arch_t v3_cpu_types[];
1333 v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1335 if (host_vmcs_ptrs[cpu_id] != 0) {
1336 V3_Print("Disabling VMX\n");
1338 if (vmx_off() != VMX_SUCCESS) {
1339 PrintError("Error executing VMXOFF\n");
1342 V3_FreePages((void *)host_vmcs_ptrs[cpu_id], 1);
1344 host_vmcs_ptrs[cpu_id] = 0;