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, Jack Lange <jarusl@cs.northwestern.edu>
11 * Copyright (c) 2008, 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/svm.h>
22 #include <palacios/vmm.h>
24 #include <palacios/vmcb.h>
25 #include <palacios/vmm_mem.h>
26 #include <palacios/vmm_paging.h>
27 #include <palacios/svm_handler.h>
29 #include <palacios/vmm_debug.h>
30 #include <palacios/vm_guest_mem.h>
32 #include <palacios/vmm_decoder.h>
33 #include <palacios/vmm_string.h>
34 #include <palacios/vmm_lowlevel.h>
35 #include <palacios/svm_msr.h>
37 #include <palacios/vmm_rbtree.h>
39 #include <palacios/vmm_direct_paging.h>
41 #include <palacios/vmm_ctrl_regs.h>
42 #include <palacios/svm_io.h>
44 #include <palacios/vmm_sprintf.h>
47 #ifndef CONFIG_DEBUG_SVM
49 #define PrintDebug(fmt, args...)
53 uint32_t v3_last_exit;
55 // This is a global pointer to the host's VMCB
56 static addr_t host_vmcbs[CONFIG_MAX_CPUS] = { [0 ... CONFIG_MAX_CPUS - 1] = 0};
60 extern void v3_stgi();
61 extern void v3_clgi();
62 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
63 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
66 static vmcb_t * Allocate_VMCB() {
67 vmcb_t * vmcb_page = NULL;
68 addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
70 if ((void *)vmcb_pa == NULL) {
71 PrintError("Error allocating VMCB\n");
75 vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
77 memset(vmcb_page, 0, 4096);
84 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
85 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
86 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
91 ctrl_area->svm_instrs.VMRUN = 1;
92 ctrl_area->svm_instrs.VMMCALL = 1;
93 ctrl_area->svm_instrs.VMLOAD = 1;
94 ctrl_area->svm_instrs.VMSAVE = 1;
95 ctrl_area->svm_instrs.STGI = 1;
96 ctrl_area->svm_instrs.CLGI = 1;
97 ctrl_area->svm_instrs.SKINIT = 1;
98 ctrl_area->svm_instrs.ICEBP = 1;
99 ctrl_area->svm_instrs.WBINVD = 1;
100 ctrl_area->svm_instrs.MONITOR = 1;
101 ctrl_area->svm_instrs.MWAIT_always = 1;
102 ctrl_area->svm_instrs.MWAIT_if_armed = 1;
103 ctrl_area->instrs.INVLPGA = 1;
104 ctrl_area->instrs.CPUID = 1;
106 ctrl_area->instrs.HLT = 1;
108 #ifdef CONFIG_TIME_VIRTUALIZE_TSC
109 ctrl_area->instrs.RDTSC = 1;
110 ctrl_area->svm_instrs.RDTSCP = 1;
113 // guest_state->cr0 = 0x00000001; // PE
116 ctrl_area->exceptions.de = 1;
117 ctrl_area->exceptions.df = 1;
119 ctrl_area->exceptions.ts = 1;
120 ctrl_area->exceptions.ss = 1;
121 ctrl_area->exceptions.ac = 1;
122 ctrl_area->exceptions.mc = 1;
123 ctrl_area->exceptions.gp = 1;
124 ctrl_area->exceptions.ud = 1;
125 ctrl_area->exceptions.np = 1;
126 ctrl_area->exceptions.of = 1;
128 ctrl_area->exceptions.nmi = 1;
132 ctrl_area->instrs.NMI = 1;
133 ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
134 ctrl_area->instrs.INIT = 1;
135 ctrl_area->instrs.PAUSE = 1;
136 ctrl_area->instrs.shutdown_evts = 1;
139 /* DEBUG FOR RETURN CODE */
140 ctrl_area->exit_code = 1;
143 /* Setup Guest Machine state */
145 core->vm_regs.rsp = 0x00;
148 core->vm_regs.rdx = 0x00000f00;
153 core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
154 core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
155 core->ctrl_regs.efer |= EFER_MSR_svm_enable;
161 core->segments.cs.selector = 0xf000;
162 core->segments.cs.limit = 0xffff;
163 core->segments.cs.base = 0x0000000f0000LL;
165 // (raw attributes = 0xf3)
166 core->segments.cs.type = 0x3;
167 core->segments.cs.system = 0x1;
168 core->segments.cs.dpl = 0x3;
169 core->segments.cs.present = 1;
173 struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
174 &(core->segments.es), &(core->segments.fs),
175 &(core->segments.gs), NULL};
177 for ( i = 0; segregs[i] != NULL; i++) {
178 struct v3_segment * seg = segregs[i];
180 seg->selector = 0x0000;
181 // seg->base = seg->selector << 4;
182 seg->base = 0x00000000;
185 // (raw attributes = 0xf3)
192 core->segments.gdtr.limit = 0x0000ffff;
193 core->segments.gdtr.base = 0x0000000000000000LL;
194 core->segments.idtr.limit = 0x0000ffff;
195 core->segments.idtr.base = 0x0000000000000000LL;
197 core->segments.ldtr.selector = 0x0000;
198 core->segments.ldtr.limit = 0x0000ffff;
199 core->segments.ldtr.base = 0x0000000000000000LL;
200 core->segments.tr.selector = 0x0000;
201 core->segments.tr.limit = 0x0000ffff;
202 core->segments.tr.base = 0x0000000000000000LL;
205 core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
206 core->dbg_regs.dr7 = 0x0000000000000400LL;
209 ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
210 ctrl_area->instrs.IOIO_PROT = 1;
212 ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
213 ctrl_area->instrs.MSR_PROT = 1;
216 PrintDebug("Exiting on interrupts\n");
217 ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
218 ctrl_area->instrs.INTR = 1;
221 v3_hook_msr(core->vm_info, EFER_MSR,
222 &v3_handle_efer_read,
223 &v3_handle_efer_write,
226 if (core->shdw_pg_mode == SHADOW_PAGING) {
227 PrintDebug("Creating initial shadow page table\n");
229 /* JRL: This is a performance killer, and a simplistic solution */
230 /* We need to fix this */
231 ctrl_area->TLB_CONTROL = 1;
232 ctrl_area->guest_ASID = 1;
235 if (v3_init_passthrough_pts(core) == -1) {
236 PrintError("Could not initialize passthrough page tables\n");
241 core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
242 PrintDebug("Created\n");
244 core->ctrl_regs.cr0 |= 0x80000000;
245 core->ctrl_regs.cr3 = core->direct_map_pt;
247 ctrl_area->cr_reads.cr0 = 1;
248 ctrl_area->cr_writes.cr0 = 1;
249 //ctrl_area->cr_reads.cr4 = 1;
250 ctrl_area->cr_writes.cr4 = 1;
251 ctrl_area->cr_reads.cr3 = 1;
252 ctrl_area->cr_writes.cr3 = 1;
256 ctrl_area->instrs.INVLPG = 1;
258 ctrl_area->exceptions.pf = 1;
260 guest_state->g_pat = 0x7040600070406ULL;
264 } else if (core->shdw_pg_mode == NESTED_PAGING) {
265 // Flush the TLB on entries/exits
266 ctrl_area->TLB_CONTROL = 1;
267 ctrl_area->guest_ASID = 1;
269 // Enable Nested Paging
270 ctrl_area->NP_ENABLE = 1;
272 PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
274 // Set the Nested Page Table pointer
275 if (v3_init_passthrough_pts(core) == -1) {
276 PrintError("Could not initialize Nested page tables\n");
280 ctrl_area->N_CR3 = core->direct_map_pt;
282 guest_state->g_pat = 0x7040600070406ULL;
285 /* tell the guest that we don't support SVM */
286 v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
287 &v3_handle_vm_cr_read,
288 &v3_handle_vm_cr_write,
293 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
295 PrintDebug("Allocating VMCB\n");
296 core->vmm_data = (void *)Allocate_VMCB();
298 if (core->vmm_data == NULL) {
299 PrintError("Could not allocate VMCB, Exiting...\n");
303 if (vm_class == V3_PC_VM) {
304 PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
305 Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
307 PrintError("Invalid VM class\n");
315 int v3_deinit_svm_vmcb(struct guest_info * core) {
316 V3_FreePages(V3_PAddr(core->vmm_data), 1);
321 static int update_irq_exit_state(struct guest_info * info) {
322 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
324 // Fix for QEMU bug using EVENTINJ as an internal cache
325 guest_ctrl->EVENTINJ.valid = 0;
327 if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
329 #ifdef CONFIG_DEBUG_INTERRUPTS
330 PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
333 info->intr_core_state.irq_started = 1;
334 info->intr_core_state.irq_pending = 0;
336 v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
339 if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
340 #ifdef CONFIG_DEBUG_INTERRUPTS
341 PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
344 // Interrupt was taken fully vectored
345 info->intr_core_state.irq_started = 0;
347 } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
348 #ifdef CONFIG_DEBUG_INTERRUPTS
349 PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
357 static int update_irq_entry_state(struct guest_info * info) {
358 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
361 if (info->intr_core_state.irq_pending == 0) {
362 guest_ctrl->guest_ctrl.V_IRQ = 0;
363 guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
366 if (v3_excp_pending(info)) {
367 uint_t excp = v3_get_excp_number(info);
369 guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
371 if (info->excp_state.excp_error_code_valid) {
372 guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
373 guest_ctrl->EVENTINJ.ev = 1;
374 #ifdef CONFIG_DEBUG_INTERRUPTS
375 PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
379 guest_ctrl->EVENTINJ.vector = excp;
381 guest_ctrl->EVENTINJ.valid = 1;
383 #ifdef CONFIG_DEBUG_INTERRUPTS
384 PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
385 (int)info->num_exits,
386 guest_ctrl->EVENTINJ.vector,
387 (void *)(addr_t)info->ctrl_regs.cr2,
388 (void *)(addr_t)info->rip);
391 v3_injecting_excp(info, excp);
392 } else if (info->intr_core_state.irq_started == 1) {
393 #ifdef CONFIG_DEBUG_INTERRUPTS
394 PrintDebug("IRQ pending from previous injection\n");
396 guest_ctrl->guest_ctrl.V_IRQ = 1;
397 guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
398 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
399 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
402 switch (v3_intr_pending(info)) {
403 case V3_EXTERNAL_IRQ: {
404 uint32_t irq = v3_get_intr(info);
406 guest_ctrl->guest_ctrl.V_IRQ = 1;
407 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
408 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
409 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
411 #ifdef CONFIG_DEBUG_INTERRUPTS
412 PrintDebug("Injecting Interrupt %d (EIP=%p)\n",
413 guest_ctrl->guest_ctrl.V_INTR_VECTOR,
414 (void *)(addr_t)info->rip);
417 info->intr_core_state.irq_pending = 1;
418 info->intr_core_state.irq_vector = irq;
423 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
425 case V3_SOFTWARE_INTR:
426 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
429 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
432 case V3_INVALID_INTR:
444 * CAUTION and DANGER!!!
446 * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
447 * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
448 * on its contents will cause things to break. The contents at the time of the exit WILL
449 * change before the exit handler is executed.
451 int v3_svm_enter(struct guest_info * info) {
452 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
453 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
454 addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
456 // Conditionally yield the CPU if the timeslice has expired
459 // Perform any additional yielding needed for time adjustment
460 v3_adjust_time(info);
462 // disable global interrupts for vm state transition
465 // Update timer devices prior to entering VM.
466 v3_update_timers(info);
468 // Synchronize the guest state to the VMCB
469 guest_state->cr0 = info->ctrl_regs.cr0;
470 guest_state->cr2 = info->ctrl_regs.cr2;
471 guest_state->cr3 = info->ctrl_regs.cr3;
472 guest_state->cr4 = info->ctrl_regs.cr4;
473 guest_state->dr6 = info->dbg_regs.dr6;
474 guest_state->dr7 = info->dbg_regs.dr7;
475 guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
476 guest_state->rflags = info->ctrl_regs.rflags;
477 guest_state->efer = info->ctrl_regs.efer;
479 guest_state->cpl = info->cpl;
481 v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
483 guest_state->rax = info->vm_regs.rax;
484 guest_state->rip = info->rip;
485 guest_state->rsp = info->vm_regs.rsp;
487 #ifdef CONFIG_SYMCALL
488 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
489 update_irq_entry_state(info);
492 update_irq_entry_state(info);
499 PrintDebug("SVM Entry to CS=%p rip=%p...\n",
500 (void *)(addr_t)info->segments.cs.base,
501 (void *)(addr_t)info->rip);
504 #ifdef CONFIG_SYMCALL
505 if (info->sym_core_state.symcall_state.sym_call_active == 1) {
506 if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
507 V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
512 v3_time_enter_vm(info);
513 guest_ctrl->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
515 //V3_Print("Calling v3_svm_launch\n");
517 v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
519 //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
521 v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
523 // Immediate exit from VM time bookkeeping
524 v3_time_exit_vm(info);
528 // Save Guest state from VMCB
529 info->rip = guest_state->rip;
530 info->vm_regs.rsp = guest_state->rsp;
531 info->vm_regs.rax = guest_state->rax;
533 info->cpl = guest_state->cpl;
535 info->ctrl_regs.cr0 = guest_state->cr0;
536 info->ctrl_regs.cr2 = guest_state->cr2;
537 info->ctrl_regs.cr3 = guest_state->cr3;
538 info->ctrl_regs.cr4 = guest_state->cr4;
539 info->dbg_regs.dr6 = guest_state->dr6;
540 info->dbg_regs.dr7 = guest_state->dr7;
541 info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
542 info->ctrl_regs.rflags = guest_state->rflags;
543 info->ctrl_regs.efer = guest_state->efer;
545 v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
546 info->cpu_mode = v3_get_vm_cpu_mode(info);
547 info->mem_mode = v3_get_vm_mem_mode(info);
551 // save exit info here
552 exit_code = guest_ctrl->exit_code;
553 exit_info1 = guest_ctrl->exit_info1;
554 exit_info2 = guest_ctrl->exit_info2;
557 #ifdef CONFIG_SYMCALL
558 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
559 update_irq_exit_state(info);
562 update_irq_exit_state(info);
566 // reenable global interrupts after vm exit
570 // Conditionally yield the CPU if the timeslice has expired
575 if (v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2) != 0) {
576 PrintError("Error in SVM exit handler\n");
577 PrintError(" last exit was %d\n", v3_last_exit);
586 int v3_start_svm_guest(struct guest_info * info) {
587 // vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
588 // vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
590 PrintDebug("Starting SVM core %u\n", info->cpu_id);
592 if (info->cpu_id == 0) {
593 info->core_run_state = CORE_RUNNING;
594 info->vm_info->run_state = VM_RUNNING;
596 PrintDebug("SVM core %u: Waiting for core initialization\n", info->cpu_id);
598 while (info->core_run_state == CORE_STOPPED) {
600 //PrintDebug("SVM core %u: still waiting for INIT\n",info->cpu_id);
603 PrintDebug("SVM core %u initialized\n", info->cpu_id);
606 PrintDebug("SVM core %u: I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n",
607 info->cpu_id, info->segments.cs.selector, (void *)(info->segments.cs.base),
608 info->segments.cs.limit, (void *)(info->rip));
612 PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p)\n", info->cpu_id, (void *)info->vmm_data);
613 //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
619 if (info->vm_info->run_state == VM_STOPPED) {
620 info->core_run_state = CORE_STOPPED;
624 if (v3_svm_enter(info) == -1) {
625 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
627 addr_t linear_addr = 0;
629 info->vm_info->run_state = VM_ERROR;
631 V3_Print("SVM core %u: SVM ERROR!!\n", info->cpu_id);
633 v3_print_guest_state(info);
635 V3_Print("SVM core %u: SVM Exit Code: %p\n", info->cpu_id, (void *)(addr_t)guest_ctrl->exit_code);
637 V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
638 V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
640 V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->cpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
641 V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->cpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
643 linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
645 if (info->mem_mode == PHYSICAL_MEM) {
646 v3_gpa_to_hva(info, linear_addr, &host_addr);
647 } else if (info->mem_mode == VIRTUAL_MEM) {
648 v3_gva_to_hva(info, linear_addr, &host_addr);
651 V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->cpu_id, (void *)host_addr);
653 V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->cpu_id, (void *)host_addr);
654 v3_dump_mem((uint8_t *)host_addr, 15);
656 v3_print_stack(info);
662 if (info->vm_info->run_state == VM_STOPPED) {
663 info->core_run_state = CORE_STOPPED;
669 if ((info->num_exits % 5000) == 0) {
670 V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
676 // Need to take down the other cores on error...
685 /* Checks machine SVM capability */
686 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
687 int v3_is_svm_capable() {
688 uint_t vm_cr_low = 0, vm_cr_high = 0;
689 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
691 v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
693 PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
695 if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
696 V3_Print("SVM Not Available\n");
699 v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
701 PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
703 if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
704 V3_Print("SVM is available but is disabled.\n");
706 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
708 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
710 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
711 V3_Print("SVM BIOS Disabled, not unlockable\n");
713 V3_Print("SVM is locked with a key\n");
718 V3_Print("SVM is available and enabled.\n");
720 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
721 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
722 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
723 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
724 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
731 static int has_svm_nested_paging() {
732 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
734 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
736 //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
738 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
739 V3_Print("SVM Nested Paging not supported\n");
742 V3_Print("SVM Nested Paging supported\n");
748 void v3_init_svm_cpu(int cpu_id) {
750 extern v3_cpu_arch_t v3_cpu_types[];
752 // Enable SVM on the CPU
753 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
754 msr.e_reg.low |= EFER_MSR_svm_enable;
755 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
757 V3_Print("SVM Enabled\n");
759 // Setup the host state save area
760 host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
763 // msr.e_reg.high = 0;
764 //msr.e_reg.low = (uint_t)host_vmcb;
765 msr.r_reg = host_vmcbs[cpu_id];
767 PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
768 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
771 if (has_svm_nested_paging() == 1) {
772 v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
774 v3_cpu_types[cpu_id] = V3_SVM_CPU;
780 void v3_deinit_svm_cpu(int cpu_id) {
782 extern v3_cpu_arch_t v3_cpu_types[];
784 // reset SVM_VM_HSAVE_PA_MSR
785 // Does setting it to NULL disable??
787 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
790 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
791 msr.e_reg.low &= ~EFER_MSR_svm_enable;
792 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
794 v3_cpu_types[cpu_id] = V3_INVALID_CPU;
796 V3_FreePages((void *)host_vmcbs[cpu_id], 4);
798 V3_Print("Host CPU %d host area freed, and SVM disabled\n", cpu_id);
853 * Test VMSAVE/VMLOAD Latency
855 #define vmsave ".byte 0x0F,0x01,0xDB ; "
856 #define vmload ".byte 0x0F,0x01,0xDA ; "
858 uint32_t start_lo, start_hi;
859 uint32_t end_lo, end_hi;
862 __asm__ __volatile__ (
864 "movl %%eax, %%esi ; "
865 "movl %%edx, %%edi ; "
866 "movq %%rcx, %%rax ; "
869 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
870 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
881 PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
883 __asm__ __volatile__ (
885 "movl %%eax, %%esi ; "
886 "movl %%edx, %%edi ; "
887 "movq %%rcx, %%rax ; "
890 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
891 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
903 PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
905 /* End Latency Test */
916 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
917 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
918 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
922 guest_state->rsp = vm_info.vm_regs.rsp;
923 guest_state->rip = vm_info.rip;
926 /* I pretty much just gutted this from TVMM */
927 /* Note: That means its probably wrong */
929 // set the segment registers to mirror ours
930 guest_state->cs.selector = 1<<3;
931 guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
932 guest_state->cs.attrib.fields.S = 1;
933 guest_state->cs.attrib.fields.P = 1;
934 guest_state->cs.attrib.fields.db = 1;
935 guest_state->cs.attrib.fields.G = 1;
936 guest_state->cs.limit = 0xfffff;
937 guest_state->cs.base = 0;
939 struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
940 for ( i = 0; segregs[i] != NULL; i++) {
941 struct vmcb_selector * seg = segregs[i];
943 seg->selector = 2<<3;
944 seg->attrib.fields.type = 0x2; // Data Segment+read/write
945 seg->attrib.fields.S = 1;
946 seg->attrib.fields.P = 1;
947 seg->attrib.fields.db = 1;
948 seg->attrib.fields.G = 1;
949 seg->limit = 0xfffff;
955 /* JRL THIS HAS TO GO */
957 // guest_state->tr.selector = GetTR_Selector();
958 guest_state->tr.attrib.fields.type = 0x9;
959 guest_state->tr.attrib.fields.P = 1;
960 // guest_state->tr.limit = GetTR_Limit();
961 //guest_state->tr.base = GetTR_Base();// - 0x2000;
969 guest_state->efer |= EFER_MSR_svm_enable;
970 guest_state->rflags = 0x00000002; // The reserved bit is always 1
971 ctrl_area->svm_instrs.VMRUN = 1;
972 guest_state->cr0 = 0x00000001; // PE
973 ctrl_area->guest_ASID = 1;
976 // guest_state->cpl = 0;
982 ctrl_area->cr_writes.cr4 = 1;
984 ctrl_area->exceptions.de = 1;
985 ctrl_area->exceptions.df = 1;
986 ctrl_area->exceptions.pf = 1;
987 ctrl_area->exceptions.ts = 1;
988 ctrl_area->exceptions.ss = 1;
989 ctrl_area->exceptions.ac = 1;
990 ctrl_area->exceptions.mc = 1;
991 ctrl_area->exceptions.gp = 1;
992 ctrl_area->exceptions.ud = 1;
993 ctrl_area->exceptions.np = 1;
994 ctrl_area->exceptions.of = 1;
995 ctrl_area->exceptions.nmi = 1;
999 ctrl_area->instrs.IOIO_PROT = 1;
1000 ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
1004 tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
1005 memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
1008 ctrl_area->instrs.INTR = 1;
1015 memset(gdt_buf, 0, 6);
1016 memset(idt_buf, 0, 6);
1019 uint_t gdt_base, idt_base;
1020 ushort_t gdt_limit, idt_limit;
1023 gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
1024 gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
1025 PrintDebug("GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
1028 idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
1029 idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
1030 PrintDebug("IDT: base: %x, limit: %x\n",idt_base, idt_limit);
1033 // gdt_base -= 0x2000;
1034 //idt_base -= 0x2000;
1036 guest_state->gdtr.base = gdt_base;
1037 guest_state->gdtr.limit = gdt_limit;
1038 guest_state->idtr.base = idt_base;
1039 guest_state->idtr.limit = idt_limit;
1045 // also determine if CPU supports nested paging
1047 if (vm_info.page_tables) {
1049 // Flush the TLB on entries/exits
1050 ctrl_area->TLB_CONTROL = 1;
1052 // Enable Nested Paging
1053 ctrl_area->NP_ENABLE = 1;
1055 PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
1057 // Set the Nested Page Table pointer
1058 ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1061 // ctrl_area->N_CR3 = Get_CR3();
1062 // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1064 guest_state->g_pat = 0x7040600070406ULL;
1066 PrintDebug("Set Nested CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(ctrl_area->N_CR3)), (uint_t)*((unsigned char *)&(ctrl_area->N_CR3) + 4));
1067 PrintDebug("Set Guest CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4));
1069 // guest_state->cr0 |= 0x80000000;