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>
38 #include <palacios/vmm_barrier.h>
40 #ifdef V3_CONFIG_CHECKPOINT
41 #include <palacios/vmm_checkpoint.h>
44 #include <palacios/vmm_direct_paging.h>
46 #include <palacios/vmm_ctrl_regs.h>
47 #include <palacios/svm_io.h>
49 #include <palacios/vmm_sprintf.h>
52 #ifndef V3_CONFIG_DEBUG_SVM
54 #define PrintDebug(fmt, args...)
58 uint32_t v3_last_exit;
60 // This is a global pointer to the host's VMCB
61 static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
65 extern void v3_stgi();
66 extern void v3_clgi();
67 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
68 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
71 static vmcb_t * Allocate_VMCB() {
72 vmcb_t * vmcb_page = NULL;
73 addr_t vmcb_pa = (addr_t)V3_AllocPages(1);
75 if ((void *)vmcb_pa == NULL) {
76 PrintError("Error allocating VMCB\n");
80 vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
82 memset(vmcb_page, 0, 4096);
88 static int v3_svm_handle_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data)
92 // Call arch-independent handler
93 if ((status = v3_handle_efer_write(core, msr, src, priv_data)) != 0) {
99 // Ensure that hardware visible EFER.SVME bit is set (SVM Enable)
100 struct efer_64 * hw_efer = (struct efer_64 *)&(core->ctrl_regs.efer);
108 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
109 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
110 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
115 ctrl_area->svm_instrs.VMRUN = 1;
116 ctrl_area->svm_instrs.VMMCALL = 1;
117 ctrl_area->svm_instrs.VMLOAD = 1;
118 ctrl_area->svm_instrs.VMSAVE = 1;
119 ctrl_area->svm_instrs.STGI = 1;
120 ctrl_area->svm_instrs.CLGI = 1;
121 ctrl_area->svm_instrs.SKINIT = 1;
122 ctrl_area->svm_instrs.ICEBP = 1;
123 ctrl_area->svm_instrs.WBINVD = 1;
124 ctrl_area->svm_instrs.MONITOR = 1;
125 ctrl_area->svm_instrs.MWAIT_always = 1;
126 ctrl_area->svm_instrs.MWAIT_if_armed = 1;
127 ctrl_area->instrs.INVLPGA = 1;
128 ctrl_area->instrs.CPUID = 1;
130 ctrl_area->instrs.HLT = 1;
132 /* Set at VMM launch as needed */
133 ctrl_area->instrs.RDTSC = 0;
134 ctrl_area->svm_instrs.RDTSCP = 0;
136 // guest_state->cr0 = 0x00000001; // PE
139 ctrl_area->exceptions.de = 1;
140 ctrl_area->exceptions.df = 1;
142 ctrl_area->exceptions.ts = 1;
143 ctrl_area->exceptions.ss = 1;
144 ctrl_area->exceptions.ac = 1;
145 ctrl_area->exceptions.mc = 1;
146 ctrl_area->exceptions.gp = 1;
147 ctrl_area->exceptions.ud = 1;
148 ctrl_area->exceptions.np = 1;
149 ctrl_area->exceptions.of = 1;
151 ctrl_area->exceptions.nmi = 1;
155 ctrl_area->instrs.NMI = 1;
156 ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
157 ctrl_area->instrs.INIT = 1;
158 // ctrl_area->instrs.PAUSE = 1;
159 ctrl_area->instrs.shutdown_evts = 1;
162 /* DEBUG FOR RETURN CODE */
163 ctrl_area->exit_code = 1;
166 /* Setup Guest Machine state */
168 core->vm_regs.rsp = 0x00;
171 core->vm_regs.rdx = 0x00000f00;
176 core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
177 core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
178 core->ctrl_regs.efer |= EFER_MSR_svm_enable;
184 core->segments.cs.selector = 0xf000;
185 core->segments.cs.limit = 0xffff;
186 core->segments.cs.base = 0x0000000f0000LL;
188 // (raw attributes = 0xf3)
189 core->segments.cs.type = 0x3;
190 core->segments.cs.system = 0x1;
191 core->segments.cs.dpl = 0x3;
192 core->segments.cs.present = 1;
196 struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
197 &(core->segments.es), &(core->segments.fs),
198 &(core->segments.gs), NULL};
200 for ( i = 0; segregs[i] != NULL; i++) {
201 struct v3_segment * seg = segregs[i];
203 seg->selector = 0x0000;
204 // seg->base = seg->selector << 4;
205 seg->base = 0x00000000;
208 // (raw attributes = 0xf3)
215 core->segments.gdtr.limit = 0x0000ffff;
216 core->segments.gdtr.base = 0x0000000000000000LL;
217 core->segments.idtr.limit = 0x0000ffff;
218 core->segments.idtr.base = 0x0000000000000000LL;
220 core->segments.ldtr.selector = 0x0000;
221 core->segments.ldtr.limit = 0x0000ffff;
222 core->segments.ldtr.base = 0x0000000000000000LL;
223 core->segments.tr.selector = 0x0000;
224 core->segments.tr.limit = 0x0000ffff;
225 core->segments.tr.base = 0x0000000000000000LL;
228 core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
229 core->dbg_regs.dr7 = 0x0000000000000400LL;
232 ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
233 ctrl_area->instrs.IOIO_PROT = 1;
235 ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
236 ctrl_area->instrs.MSR_PROT = 1;
239 PrintDebug("Exiting on interrupts\n");
240 ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
241 ctrl_area->instrs.INTR = 1;
244 v3_hook_msr(core->vm_info, EFER_MSR,
245 &v3_handle_efer_read,
246 &v3_svm_handle_efer_write,
249 if (core->shdw_pg_mode == SHADOW_PAGING) {
250 PrintDebug("Creating initial shadow page table\n");
252 /* JRL: This is a performance killer, and a simplistic solution */
253 /* We need to fix this */
254 ctrl_area->TLB_CONTROL = 1;
255 ctrl_area->guest_ASID = 1;
258 if (v3_init_passthrough_pts(core) == -1) {
259 PrintError("Could not initialize passthrough page tables\n");
264 core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
265 PrintDebug("Created\n");
267 core->ctrl_regs.cr0 |= 0x80000000;
268 core->ctrl_regs.cr3 = core->direct_map_pt;
270 ctrl_area->cr_reads.cr0 = 1;
271 ctrl_area->cr_writes.cr0 = 1;
272 //ctrl_area->cr_reads.cr4 = 1;
273 ctrl_area->cr_writes.cr4 = 1;
274 ctrl_area->cr_reads.cr3 = 1;
275 ctrl_area->cr_writes.cr3 = 1;
279 ctrl_area->instrs.INVLPG = 1;
281 ctrl_area->exceptions.pf = 1;
283 guest_state->g_pat = 0x7040600070406ULL;
287 } else if (core->shdw_pg_mode == NESTED_PAGING) {
288 // Flush the TLB on entries/exits
289 ctrl_area->TLB_CONTROL = 1;
290 ctrl_area->guest_ASID = 1;
292 // Enable Nested Paging
293 ctrl_area->NP_ENABLE = 1;
295 PrintDebug("NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
297 // Set the Nested Page Table pointer
298 if (v3_init_passthrough_pts(core) == -1) {
299 PrintError("Could not initialize Nested page tables\n");
303 ctrl_area->N_CR3 = core->direct_map_pt;
305 guest_state->g_pat = 0x7040600070406ULL;
308 /* tell the guest that we don't support SVM */
309 v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
310 &v3_handle_vm_cr_read,
311 &v3_handle_vm_cr_write,
316 #define INT_PENDING_AMD_MSR 0xc0010055
318 v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
319 v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
320 v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
321 v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
322 v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
324 v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
325 v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
326 v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
329 v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
330 v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
332 // Passthrough read operations are ok.
333 v3_hook_msr(core->vm_info, INT_PENDING_AMD_MSR, NULL, v3_msr_unhandled_write, NULL);
338 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
340 PrintDebug("Allocating VMCB\n");
341 core->vmm_data = (void *)Allocate_VMCB();
343 if (core->vmm_data == NULL) {
344 PrintError("Could not allocate VMCB, Exiting...\n");
348 if (vm_class == V3_PC_VM) {
349 PrintDebug("Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
350 Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
352 PrintError("Invalid VM class\n");
356 core->core_run_state = CORE_STOPPED;
362 int v3_deinit_svm_vmcb(struct guest_info * core) {
363 V3_FreePages(V3_PAddr(core->vmm_data), 1);
368 #ifdef V3_CONFIG_CHECKPOINT
369 int v3_svm_save_core(struct guest_info * core, void * ctx){
371 if (v3_chkpt_save_8(ctx, "cpl", &(core->cpl)) == -1) {
372 PrintError("Could not save SVM cpl\n");
376 if (v3_chkpt_save(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data) == -1) {
377 PrintError("Could not save SVM vmcb\n");
384 int v3_svm_load_core(struct guest_info * core, void * ctx){
386 if (v3_chkpt_load_8(ctx, "cpl", &(core->cpl)) == -1) {
387 PrintError("Could not load SVM cpl\n");
391 if (v3_chkpt_load(ctx, "vmcb_data", PAGE_SIZE, core->vmm_data) == -1) {
399 static int update_irq_exit_state(struct guest_info * info) {
400 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
402 // Fix for QEMU bug using EVENTINJ as an internal cache
403 guest_ctrl->EVENTINJ.valid = 0;
405 if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
407 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
408 PrintDebug("INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
411 info->intr_core_state.irq_started = 1;
412 info->intr_core_state.irq_pending = 0;
414 v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
417 if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
418 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
419 PrintDebug("Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
422 // Interrupt was taken fully vectored
423 info->intr_core_state.irq_started = 0;
425 } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
426 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
427 PrintDebug("EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
435 static int update_irq_entry_state(struct guest_info * info) {
436 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
439 if (info->intr_core_state.irq_pending == 0) {
440 guest_ctrl->guest_ctrl.V_IRQ = 0;
441 guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
444 if (v3_excp_pending(info)) {
445 uint_t excp = v3_get_excp_number(info);
447 guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
449 if (info->excp_state.excp_error_code_valid) {
450 guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
451 guest_ctrl->EVENTINJ.ev = 1;
452 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
453 PrintDebug("Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
457 guest_ctrl->EVENTINJ.vector = excp;
459 guest_ctrl->EVENTINJ.valid = 1;
461 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
462 PrintDebug("<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
463 (int)info->num_exits,
464 guest_ctrl->EVENTINJ.vector,
465 (void *)(addr_t)info->ctrl_regs.cr2,
466 (void *)(addr_t)info->rip);
469 v3_injecting_excp(info, excp);
470 } else if (info->intr_core_state.irq_started == 1) {
471 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
472 PrintDebug("IRQ pending from previous injection\n");
474 guest_ctrl->guest_ctrl.V_IRQ = 1;
475 guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
476 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
477 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
480 switch (v3_intr_pending(info)) {
481 case V3_EXTERNAL_IRQ: {
482 uint32_t irq = v3_get_intr(info);
484 guest_ctrl->guest_ctrl.V_IRQ = 1;
485 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
486 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
487 guest_ctrl->guest_ctrl.V_INTR_PRIO = 0xf;
489 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
490 PrintDebug("Injecting Interrupt %d (EIP=%p)\n",
491 guest_ctrl->guest_ctrl.V_INTR_VECTOR,
492 (void *)(addr_t)info->rip);
495 info->intr_core_state.irq_pending = 1;
496 info->intr_core_state.irq_vector = irq;
501 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
503 case V3_SOFTWARE_INTR:
504 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
506 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
507 PrintDebug("Injecting software interrupt -- type: %d, vector: %d\n",
508 SVM_INJECTION_SOFT_INTR, info->intr_core_state.swintr_vector);
510 guest_ctrl->EVENTINJ.vector = info->intr_core_state.swintr_vector;
511 guest_ctrl->EVENTINJ.valid = 1;
513 /* reset swintr state */
514 info->intr_core_state.swintr_posted = 0;
515 info->intr_core_state.swintr_vector = 0;
519 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
522 case V3_INVALID_INTR:
533 v3_svm_config_tsc_virtualization(struct guest_info * info) {
534 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
536 if (info->time_state.flags & VM_TIME_TRAP_RDTSC) {
537 ctrl_area->instrs.RDTSC = 1;
538 ctrl_area->svm_instrs.RDTSCP = 1;
540 ctrl_area->instrs.RDTSC = 0;
541 ctrl_area->svm_instrs.RDTSCP = 0;
542 if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) {
543 ctrl_area->TSC_OFFSET = 0;
545 ctrl_area->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
552 * CAUTION and DANGER!!!
554 * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
555 * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
556 * on its contents will cause things to break. The contents at the time of the exit WILL
557 * change before the exit handler is executed.
559 int v3_svm_enter(struct guest_info * info) {
560 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
561 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
562 addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
563 uint64_t guest_cycles = 0;
565 // Conditionally yield the CPU if the timeslice has expired
568 // Update timer devices after being in the VM before doing
569 // IRQ updates, so that any interrupts they raise get seen
571 v3_advance_time(info, NULL);
572 v3_update_timers(info);
574 // disable global interrupts for vm state transition
577 // Synchronize the guest state to the VMCB
578 guest_state->cr0 = info->ctrl_regs.cr0;
579 guest_state->cr2 = info->ctrl_regs.cr2;
580 guest_state->cr3 = info->ctrl_regs.cr3;
581 guest_state->cr4 = info->ctrl_regs.cr4;
582 guest_state->dr6 = info->dbg_regs.dr6;
583 guest_state->dr7 = info->dbg_regs.dr7;
584 guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
585 guest_state->rflags = info->ctrl_regs.rflags;
586 guest_state->efer = info->ctrl_regs.efer;
588 /* Synchronize MSRs */
589 guest_state->star = info->msrs.star;
590 guest_state->lstar = info->msrs.lstar;
591 guest_state->sfmask = info->msrs.sfmask;
592 guest_state->KernelGsBase = info->msrs.kern_gs_base;
594 guest_state->cpl = info->cpl;
596 v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
598 guest_state->rax = info->vm_regs.rax;
599 guest_state->rip = info->rip;
600 guest_state->rsp = info->vm_regs.rsp;
602 #ifdef V3_CONFIG_SYMCALL
603 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
604 update_irq_entry_state(info);
607 update_irq_entry_state(info);
614 PrintDebug("SVM Entry to CS=%p rip=%p...\n",
615 (void *)(addr_t)info->segments.cs.base,
616 (void *)(addr_t)info->rip);
619 #ifdef V3_CONFIG_SYMCALL
620 if (info->sym_core_state.symcall_state.sym_call_active == 1) {
621 if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
622 V3_Print("!!! Injecting Interrupt during Sym call !!!\n");
627 v3_svm_config_tsc_virtualization(info);
629 //V3_Print("Calling v3_svm_launch\n");
631 uint64_t entry_tsc = 0;
632 uint64_t exit_tsc = 0;
636 v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
640 guest_cycles = exit_tsc - entry_tsc;
644 //V3_Print("SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
646 v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
648 v3_advance_time(info, &guest_cycles);
652 // Save Guest state from VMCB
653 info->rip = guest_state->rip;
654 info->vm_regs.rsp = guest_state->rsp;
655 info->vm_regs.rax = guest_state->rax;
657 info->cpl = guest_state->cpl;
659 info->ctrl_regs.cr0 = guest_state->cr0;
660 info->ctrl_regs.cr2 = guest_state->cr2;
661 info->ctrl_regs.cr3 = guest_state->cr3;
662 info->ctrl_regs.cr4 = guest_state->cr4;
663 info->dbg_regs.dr6 = guest_state->dr6;
664 info->dbg_regs.dr7 = guest_state->dr7;
665 info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
666 info->ctrl_regs.rflags = guest_state->rflags;
667 info->ctrl_regs.efer = guest_state->efer;
669 /* Synchronize MSRs */
670 info->msrs.star = guest_state->star;
671 info->msrs.lstar = guest_state->lstar;
672 info->msrs.sfmask = guest_state->sfmask;
673 info->msrs.kern_gs_base = guest_state->KernelGsBase;
675 v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
676 info->cpu_mode = v3_get_vm_cpu_mode(info);
677 info->mem_mode = v3_get_vm_mem_mode(info);
680 // save exit info here
681 exit_code = guest_ctrl->exit_code;
682 exit_info1 = guest_ctrl->exit_info1;
683 exit_info2 = guest_ctrl->exit_info2;
685 #ifdef V3_CONFIG_SYMCALL
686 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
687 update_irq_exit_state(info);
690 update_irq_exit_state(info);
693 // reenable global interrupts after vm exit
696 // Conditionally yield the CPU if the timeslice has expired
699 // This update timers is for time-dependent handlers
700 // if we're slaved to host time
701 v3_advance_time(info, NULL);
702 v3_update_timers(info);
705 int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2);
708 PrintError("Error in SVM exit handler (ret=%d)\n", ret);
709 PrintError(" last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code);
714 if (info->timeouts.timeout_active) {
715 /* Check to see if any timeouts have expired */
716 v3_handle_timeouts(info, guest_cycles);
724 int v3_start_svm_guest(struct guest_info * info) {
725 // vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
726 // vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
728 PrintDebug("Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
730 if (info->vcpu_id == 0) {
731 info->core_run_state = CORE_RUNNING;
733 PrintDebug("SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
735 while (info->core_run_state == CORE_STOPPED) {
737 if (info->vm_info->run_state == VM_STOPPED) {
738 // The VM was stopped before this core was initialized.
743 //PrintDebug("SVM core %u: still waiting for INIT\n", info->vcpu_id);
746 PrintDebug("SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id);
748 // We'll be paranoid about race conditions here
749 v3_wait_at_barrier(info);
752 PrintDebug("SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n",
753 info->vcpu_id, info->pcpu_id,
754 info->segments.cs.selector, (void *)(info->segments.cs.base),
755 info->segments.cs.limit, (void *)(info->rip));
759 PrintDebug("SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n",
760 info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
761 //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
767 if (info->vm_info->run_state == VM_STOPPED) {
768 info->core_run_state = CORE_STOPPED;
772 if (v3_svm_enter(info) == -1) {
773 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
775 addr_t linear_addr = 0;
777 info->vm_info->run_state = VM_ERROR;
779 V3_Print("SVM core %u: SVM ERROR!!\n", info->vcpu_id);
781 v3_print_guest_state(info);
783 V3_Print("SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code);
785 V3_Print("SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
786 V3_Print("SVM core %u: exit_info1 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
788 V3_Print("SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
789 V3_Print("SVM core %u: exit_info2 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
791 linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
793 if (info->mem_mode == PHYSICAL_MEM) {
794 v3_gpa_to_hva(info, linear_addr, &host_addr);
795 } else if (info->mem_mode == VIRTUAL_MEM) {
796 v3_gva_to_hva(info, linear_addr, &host_addr);
799 V3_Print("SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
801 V3_Print("SVM core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
802 v3_dump_mem((uint8_t *)host_addr, 15);
804 v3_print_stack(info);
809 v3_wait_at_barrier(info);
812 if (info->vm_info->run_state == VM_STOPPED) {
813 info->core_run_state = CORE_STOPPED;
820 if ((info->num_exits % 50000) == 0) {
821 V3_Print("SVM Exit number %d\n", (uint32_t)info->num_exits);
822 v3_print_guest_state(info);
828 // Need to take down the other cores on error...
836 int v3_reset_svm_vm_core(struct guest_info * core, addr_t rip) {
839 // Write the RIP, CS, and descriptor
840 // assume the rest is already good to go
842 // vector VV -> rip at 0
844 // This means we start executing at linear address VV000
846 // So the selector needs to be VV00
847 // and the base needs to be VV000
850 core->segments.cs.selector = rip << 8;
851 core->segments.cs.limit = 0xffff;
852 core->segments.cs.base = rip << 12;
862 /* Checks machine SVM capability */
863 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
864 int v3_is_svm_capable() {
865 uint_t vm_cr_low = 0, vm_cr_high = 0;
866 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
868 v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
870 PrintDebug("CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
872 if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
873 V3_Print("SVM Not Available\n");
876 v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
878 PrintDebug("SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
880 if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
881 V3_Print("SVM is available but is disabled.\n");
883 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
885 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
887 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
888 V3_Print("SVM BIOS Disabled, not unlockable\n");
890 V3_Print("SVM is locked with a key\n");
895 V3_Print("SVM is available and enabled.\n");
897 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
898 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
899 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
900 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
901 PrintDebug("CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
908 static int has_svm_nested_paging() {
909 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
911 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
913 //PrintDebug("CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
915 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
916 V3_Print("SVM Nested Paging not supported\n");
919 V3_Print("SVM Nested Paging supported\n");
926 void v3_init_svm_cpu(int cpu_id) {
928 extern v3_cpu_arch_t v3_cpu_types[];
930 // Enable SVM on the CPU
931 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
932 msr.e_reg.low |= EFER_MSR_svm_enable;
933 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
935 V3_Print("SVM Enabled\n");
937 // Setup the host state save area
938 host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4);
941 // msr.e_reg.high = 0;
942 //msr.e_reg.low = (uint_t)host_vmcb;
943 msr.r_reg = host_vmcbs[cpu_id];
945 PrintDebug("Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
946 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
949 if (has_svm_nested_paging() == 1) {
950 v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
952 v3_cpu_types[cpu_id] = V3_SVM_CPU;
958 void v3_deinit_svm_cpu(int cpu_id) {
960 extern v3_cpu_arch_t v3_cpu_types[];
962 // reset SVM_VM_HSAVE_PA_MSR
963 // Does setting it to NULL disable??
965 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
968 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
969 msr.e_reg.low &= ~EFER_MSR_svm_enable;
970 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
972 v3_cpu_types[cpu_id] = V3_INVALID_CPU;
974 V3_FreePages((void *)host_vmcbs[cpu_id], 4);
976 V3_Print("Host CPU %d host area freed, and SVM disabled\n", cpu_id);
1031 * Test VMSAVE/VMLOAD Latency
1033 #define vmsave ".byte 0x0F,0x01,0xDB ; "
1034 #define vmload ".byte 0x0F,0x01,0xDA ; "
1036 uint32_t start_lo, start_hi;
1037 uint32_t end_lo, end_hi;
1038 uint64_t start, end;
1040 __asm__ __volatile__ (
1042 "movl %%eax, %%esi ; "
1043 "movl %%edx, %%edi ; "
1044 "movq %%rcx, %%rax ; "
1047 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1048 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1059 PrintDebug("VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
1061 __asm__ __volatile__ (
1063 "movl %%eax, %%esi ; "
1064 "movl %%edx, %%edi ; "
1065 "movq %%rcx, %%rax ; "
1068 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1069 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1081 PrintDebug("VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
1083 /* End Latency Test */
1094 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
1095 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
1096 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
1100 guest_state->rsp = vm_info.vm_regs.rsp;
1101 guest_state->rip = vm_info.rip;
1104 /* I pretty much just gutted this from TVMM */
1105 /* Note: That means its probably wrong */
1107 // set the segment registers to mirror ours
1108 guest_state->cs.selector = 1<<3;
1109 guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
1110 guest_state->cs.attrib.fields.S = 1;
1111 guest_state->cs.attrib.fields.P = 1;
1112 guest_state->cs.attrib.fields.db = 1;
1113 guest_state->cs.attrib.fields.G = 1;
1114 guest_state->cs.limit = 0xfffff;
1115 guest_state->cs.base = 0;
1117 struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
1118 for ( i = 0; segregs[i] != NULL; i++) {
1119 struct vmcb_selector * seg = segregs[i];
1121 seg->selector = 2<<3;
1122 seg->attrib.fields.type = 0x2; // Data Segment+read/write
1123 seg->attrib.fields.S = 1;
1124 seg->attrib.fields.P = 1;
1125 seg->attrib.fields.db = 1;
1126 seg->attrib.fields.G = 1;
1127 seg->limit = 0xfffff;
1133 /* JRL THIS HAS TO GO */
1135 // guest_state->tr.selector = GetTR_Selector();
1136 guest_state->tr.attrib.fields.type = 0x9;
1137 guest_state->tr.attrib.fields.P = 1;
1138 // guest_state->tr.limit = GetTR_Limit();
1139 //guest_state->tr.base = GetTR_Base();// - 0x2000;
1147 guest_state->efer |= EFER_MSR_svm_enable;
1148 guest_state->rflags = 0x00000002; // The reserved bit is always 1
1149 ctrl_area->svm_instrs.VMRUN = 1;
1150 guest_state->cr0 = 0x00000001; // PE
1151 ctrl_area->guest_ASID = 1;
1154 // guest_state->cpl = 0;
1160 ctrl_area->cr_writes.cr4 = 1;
1162 ctrl_area->exceptions.de = 1;
1163 ctrl_area->exceptions.df = 1;
1164 ctrl_area->exceptions.pf = 1;
1165 ctrl_area->exceptions.ts = 1;
1166 ctrl_area->exceptions.ss = 1;
1167 ctrl_area->exceptions.ac = 1;
1168 ctrl_area->exceptions.mc = 1;
1169 ctrl_area->exceptions.gp = 1;
1170 ctrl_area->exceptions.ud = 1;
1171 ctrl_area->exceptions.np = 1;
1172 ctrl_area->exceptions.of = 1;
1173 ctrl_area->exceptions.nmi = 1;
1177 ctrl_area->instrs.IOIO_PROT = 1;
1178 ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3);
1182 tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
1183 memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
1186 ctrl_area->instrs.INTR = 1;
1193 memset(gdt_buf, 0, 6);
1194 memset(idt_buf, 0, 6);
1197 uint_t gdt_base, idt_base;
1198 ushort_t gdt_limit, idt_limit;
1201 gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
1202 gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
1203 PrintDebug("GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
1206 idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
1207 idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
1208 PrintDebug("IDT: base: %x, limit: %x\n",idt_base, idt_limit);
1211 // gdt_base -= 0x2000;
1212 //idt_base -= 0x2000;
1214 guest_state->gdtr.base = gdt_base;
1215 guest_state->gdtr.limit = gdt_limit;
1216 guest_state->idtr.base = idt_base;
1217 guest_state->idtr.limit = idt_limit;
1223 // also determine if CPU supports nested paging
1225 if (vm_info.page_tables) {
1227 // Flush the TLB on entries/exits
1228 ctrl_area->TLB_CONTROL = 1;
1230 // Enable Nested Paging
1231 ctrl_area->NP_ENABLE = 1;
1233 PrintDebug("NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
1235 // Set the Nested Page Table pointer
1236 ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1239 // ctrl_area->N_CR3 = Get_CR3();
1240 // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1242 guest_state->g_pat = 0x7040600070406ULL;
1244 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));
1245 PrintDebug("Set Guest CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4));
1247 // guest_state->cr0 |= 0x80000000;