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".
22 #include <palacios/svm.h>
23 #include <palacios/vmm.h>
25 #include <palacios/vmcb.h>
26 #include <palacios/vmm_mem.h>
27 #include <palacios/vmm_paging.h>
28 #include <palacios/svm_handler.h>
30 #include <palacios/vmm_debug.h>
31 #include <palacios/vm_guest_mem.h>
33 #include <palacios/vmm_decoder.h>
34 #include <palacios/vmm_string.h>
35 #include <palacios/vmm_lowlevel.h>
36 #include <palacios/svm_msr.h>
38 #include <palacios/vmm_rbtree.h>
39 #include <palacios/vmm_barrier.h>
40 #include <palacios/vmm_debug.h>
44 #ifdef V3_CONFIG_CHECKPOINT
45 #include <palacios/vmm_checkpoint.h>
48 #include <palacios/vmm_direct_paging.h>
50 #include <palacios/vmm_ctrl_regs.h>
51 #include <palacios/svm_io.h>
53 #include <palacios/vmm_sprintf.h>
56 #ifndef V3_CONFIG_DEBUG_SVM
58 #define PrintDebug(fmt, args...)
62 uint32_t v3_last_exit;
64 // This is a global pointer to the host's VMCB
65 static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
69 extern void v3_stgi();
70 extern void v3_clgi();
71 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
72 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
75 static vmcb_t * Allocate_VMCB() {
76 vmcb_t * vmcb_page = NULL;
77 addr_t vmcb_pa = (addr_t)V3_AllocPages(1); // need not be shadow safe, not exposed to guest
79 if ((void *)vmcb_pa == NULL) {
80 PrintError(VM_NONE, VCORE_NONE, "Error allocating VMCB\n");
84 vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
86 memset(vmcb_page, 0, 4096);
92 static int v3_svm_handle_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data)
96 // Call arch-independent handler
97 if ((status = v3_handle_efer_write(core, msr, src, priv_data)) != 0) {
103 // Ensure that hardware visible EFER.SVME bit is set (SVM Enable)
104 struct efer_64 * hw_efer = (struct efer_64 *)&(core->ctrl_regs.efer);
112 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
113 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
114 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
119 ctrl_area->svm_instrs.VMRUN = 1;
120 ctrl_area->svm_instrs.VMMCALL = 1;
121 ctrl_area->svm_instrs.VMLOAD = 1;
122 ctrl_area->svm_instrs.VMSAVE = 1;
123 ctrl_area->svm_instrs.STGI = 1;
124 ctrl_area->svm_instrs.CLGI = 1;
125 ctrl_area->svm_instrs.SKINIT = 1;
126 ctrl_area->svm_instrs.ICEBP = 1;
127 ctrl_area->svm_instrs.WBINVD = 1;
128 ctrl_area->svm_instrs.MONITOR = 1;
129 ctrl_area->svm_instrs.MWAIT_always = 1;
130 ctrl_area->svm_instrs.MWAIT_if_armed = 1;
131 ctrl_area->instrs.INVLPGA = 1;
132 ctrl_area->instrs.CPUID = 1;
134 ctrl_area->instrs.HLT = 1;
136 /* Set at VMM launch as needed */
137 ctrl_area->instrs.RDTSC = 0;
138 ctrl_area->svm_instrs.RDTSCP = 0;
140 // guest_state->cr0 = 0x00000001; // PE
143 ctrl_area->exceptions.de = 1;
144 ctrl_area->exceptions.df = 1;
146 ctrl_area->exceptions.ts = 1;
147 ctrl_area->exceptions.ss = 1;
148 ctrl_area->exceptions.ac = 1;
149 ctrl_area->exceptions.mc = 1;
150 ctrl_area->exceptions.gp = 1;
151 ctrl_area->exceptions.ud = 1;
152 ctrl_area->exceptions.np = 1;
153 ctrl_area->exceptions.of = 1;
155 ctrl_area->exceptions.nmi = 1;
159 ctrl_area->instrs.NMI = 1;
160 ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
161 ctrl_area->instrs.INIT = 1;
162 // ctrl_area->instrs.PAUSE = 1;
163 ctrl_area->instrs.shutdown_evts = 1;
166 /* DEBUG FOR RETURN CODE */
167 ctrl_area->exit_code = 1;
170 /* Setup Guest Machine state */
172 core->vm_regs.rsp = 0x00;
175 core->vm_regs.rdx = 0x00000f00;
180 core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
181 core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
182 core->ctrl_regs.efer |= EFER_MSR_svm_enable;
188 core->segments.cs.selector = 0xf000;
189 core->segments.cs.limit = 0xffff;
190 core->segments.cs.base = 0x0000000f0000LL;
192 // (raw attributes = 0xf3)
193 core->segments.cs.type = 0x3;
194 core->segments.cs.system = 0x1;
195 core->segments.cs.dpl = 0x3;
196 core->segments.cs.present = 1;
200 struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
201 &(core->segments.es), &(core->segments.fs),
202 &(core->segments.gs), NULL};
204 for ( i = 0; segregs[i] != NULL; i++) {
205 struct v3_segment * seg = segregs[i];
207 seg->selector = 0x0000;
208 // seg->base = seg->selector << 4;
209 seg->base = 0x00000000;
212 // (raw attributes = 0xf3)
219 core->segments.gdtr.limit = 0x0000ffff;
220 core->segments.gdtr.base = 0x0000000000000000LL;
221 core->segments.idtr.limit = 0x0000ffff;
222 core->segments.idtr.base = 0x0000000000000000LL;
224 core->segments.ldtr.selector = 0x0000;
225 core->segments.ldtr.limit = 0x0000ffff;
226 core->segments.ldtr.base = 0x0000000000000000LL;
227 core->segments.tr.selector = 0x0000;
228 core->segments.tr.limit = 0x0000ffff;
229 core->segments.tr.base = 0x0000000000000000LL;
232 core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
233 core->dbg_regs.dr7 = 0x0000000000000400LL;
236 ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
237 ctrl_area->instrs.IOIO_PROT = 1;
239 ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
240 ctrl_area->instrs.MSR_PROT = 1;
243 PrintDebug(core->vm_info, core, "Exiting on interrupts\n");
244 ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
245 ctrl_area->instrs.INTR = 1;
246 // The above also assures the TPR changes (CR8) are only virtual
249 // However, we need to see TPR writes since they will
250 // affect the virtual apic
251 // we reflect out cr8 to ctrl_regs->apic_tpr
252 ctrl_area->cr_reads.cr8 = 1;
253 ctrl_area->cr_writes.cr8 = 1;
254 // We will do all TPR comparisons in the virtual apic
255 // We also do not want the V_TPR to be able to mask the PIC
256 ctrl_area->guest_ctrl.V_IGN_TPR = 1;
260 v3_hook_msr(core->vm_info, EFER_MSR,
261 &v3_handle_efer_read,
262 &v3_svm_handle_efer_write,
265 if (core->shdw_pg_mode == SHADOW_PAGING) {
266 PrintDebug(core->vm_info, core, "Creating initial shadow page table\n");
268 /* JRL: This is a performance killer, and a simplistic solution */
269 /* We need to fix this */
270 ctrl_area->TLB_CONTROL = 1;
271 ctrl_area->guest_ASID = 1;
274 if (v3_init_passthrough_pts(core) == -1) {
275 PrintError(core->vm_info, core, "Could not initialize passthrough page tables\n");
280 core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
281 PrintDebug(core->vm_info, core, "Created\n");
283 core->ctrl_regs.cr0 |= 0x80000000;
284 core->ctrl_regs.cr3 = core->direct_map_pt;
286 ctrl_area->cr_reads.cr0 = 1;
287 ctrl_area->cr_writes.cr0 = 1;
288 //ctrl_area->cr_reads.cr4 = 1;
289 ctrl_area->cr_writes.cr4 = 1;
290 ctrl_area->cr_reads.cr3 = 1;
291 ctrl_area->cr_writes.cr3 = 1;
294 ctrl_area->instrs.INVLPG = 1;
296 ctrl_area->exceptions.pf = 1;
298 guest_state->g_pat = 0x7040600070406ULL;
301 } else if (core->shdw_pg_mode == NESTED_PAGING) {
302 // Flush the TLB on entries/exits
303 ctrl_area->TLB_CONTROL = 1;
304 ctrl_area->guest_ASID = 1;
306 // Enable Nested Paging
307 ctrl_area->NP_ENABLE = 1;
309 PrintDebug(core->vm_info, core, "NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
311 // Set the Nested Page Table pointer
312 if (v3_init_passthrough_pts(core) == -1) {
313 PrintError(core->vm_info, core, "Could not initialize Nested page tables\n");
317 ctrl_area->N_CR3 = core->direct_map_pt;
319 guest_state->g_pat = 0x7040600070406ULL;
322 /* tell the guest that we don't support SVM */
323 v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
324 &v3_handle_vm_cr_read,
325 &v3_handle_vm_cr_write,
330 #define INT_PENDING_AMD_MSR 0xc0010055
332 v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
333 v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
334 v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
335 v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
336 v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
338 v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
339 v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
340 v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
343 v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
344 v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
346 // Passthrough read operations are ok.
347 v3_hook_msr(core->vm_info, INT_PENDING_AMD_MSR, NULL, v3_msr_unhandled_write, NULL);
352 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
354 PrintDebug(core->vm_info, core, "Allocating VMCB\n");
355 core->vmm_data = (void *)Allocate_VMCB();
357 if (core->vmm_data == NULL) {
358 PrintError(core->vm_info, core, "Could not allocate VMCB, Exiting...\n");
362 if (vm_class == V3_PC_VM) {
363 PrintDebug(core->vm_info, core, "Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
364 Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
366 PrintError(core->vm_info, core, "Invalid VM class\n");
370 core->core_run_state = CORE_STOPPED;
376 int v3_deinit_svm_vmcb(struct guest_info * core) {
377 V3_FreePages(V3_PAddr(core->vmm_data), 1);
382 #ifdef V3_CONFIG_CHECKPOINT
383 int v3_svm_save_core(struct guest_info * core, void * ctx){
385 vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data);
387 // Special case saves of data we need immediate access to
389 V3_CHKPT_SAVE(ctx, "CPL", core->cpl, failout);
390 V3_CHKPT_SAVE(ctx,"STAR", guest_area->star, failout);
391 V3_CHKPT_SAVE(ctx,"CSTAR", guest_area->cstar, failout);
392 V3_CHKPT_SAVE(ctx,"LSTAR", guest_area->lstar, failout);
393 V3_CHKPT_SAVE(ctx,"SFMASK", guest_area->sfmask, failout);
394 V3_CHKPT_SAVE(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout);
395 V3_CHKPT_SAVE(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout);
396 V3_CHKPT_SAVE(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout);
397 V3_CHKPT_SAVE(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout);
399 // and then we save the whole enchilada
400 if (v3_chkpt_save(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) {
401 PrintError(core->vm_info, core, "Could not save SVM vmcb\n");
408 PrintError(core->vm_info, core, "Failed to save SVM state for core\n");
413 int v3_svm_load_core(struct guest_info * core, void * ctx){
416 vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data);
418 // Reload what we special cased, which we will overwrite in a minute
419 V3_CHKPT_LOAD(ctx, "CPL", core->cpl, failout);
420 V3_CHKPT_LOAD(ctx,"STAR", guest_area->star, failout);
421 V3_CHKPT_LOAD(ctx,"CSTAR", guest_area->cstar, failout);
422 V3_CHKPT_LOAD(ctx,"LSTAR", guest_area->lstar, failout);
423 V3_CHKPT_LOAD(ctx,"SFMASK", guest_area->sfmask, failout);
424 V3_CHKPT_LOAD(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout);
425 V3_CHKPT_LOAD(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout);
426 V3_CHKPT_LOAD(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout);
427 V3_CHKPT_LOAD(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout);
429 // and then we load the whole enchilada
430 if (v3_chkpt_load(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) {
431 PrintError(core->vm_info, core, "Could not load SVM vmcb\n");
438 PrintError(core->vm_info, core, "Failed to save SVM state for core\n");
444 static int update_irq_exit_state(struct guest_info * info) {
445 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
447 // Fix for QEMU bug using EVENTINJ as an internal cache
448 guest_ctrl->EVENTINJ.valid = 0;
450 if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
452 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
453 PrintDebug(info->vm_info, info, "INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
456 info->intr_core_state.irq_started = 1;
457 info->intr_core_state.irq_pending = 0;
459 v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
462 if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
463 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
464 PrintDebug(info->vm_info, info, "Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
467 // Interrupt was taken fully vectored
468 info->intr_core_state.irq_started = 0;
470 } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
471 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
472 PrintDebug(info->vm_info, info, "EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
480 static int update_irq_entry_state(struct guest_info * info) {
481 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
484 if (info->intr_core_state.irq_pending == 0) {
485 guest_ctrl->guest_ctrl.V_IRQ = 0;
486 guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
489 if (v3_excp_pending(info)) {
490 uint_t excp = v3_get_excp_number(info);
492 guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
494 if (info->excp_state.excp_error_code_valid) {
495 guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
496 guest_ctrl->EVENTINJ.ev = 1;
497 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
498 PrintDebug(info->vm_info, info, "Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
502 guest_ctrl->EVENTINJ.vector = excp;
504 guest_ctrl->EVENTINJ.valid = 1;
506 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
507 PrintDebug(info->vm_info, info, "<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
508 (int)info->num_exits,
509 guest_ctrl->EVENTINJ.vector,
510 (void *)(addr_t)info->ctrl_regs.cr2,
511 (void *)(addr_t)info->rip);
514 v3_injecting_excp(info, excp);
515 } else if (info->intr_core_state.irq_started == 1) {
516 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
517 PrintDebug(info->vm_info, info, "IRQ pending from previous injection\n");
519 guest_ctrl->guest_ctrl.V_IRQ = 1;
520 guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
522 // We ignore the virtual TPR on this injection
523 // TPR/PPR tests have already been done in the APIC.
524 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
525 guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf;
528 switch (v3_intr_pending(info)) {
529 case V3_EXTERNAL_IRQ: {
530 int irq = v3_get_intr(info);
536 guest_ctrl->guest_ctrl.V_IRQ = 1;
537 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
539 // We ignore the virtual TPR on this injection
540 // TPR/PPR tests have already been done in the APIC.
541 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
542 guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf;
544 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
545 PrintDebug(info->vm_info, info, "Injecting Interrupt %d (EIP=%p)\n",
546 guest_ctrl->guest_ctrl.V_INTR_VECTOR,
547 (void *)(addr_t)info->rip);
550 info->intr_core_state.irq_pending = 1;
551 info->intr_core_state.irq_vector = irq;
556 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
558 case V3_SOFTWARE_INTR:
559 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
561 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
562 PrintDebug(info->vm_info, info, "Injecting software interrupt -- type: %d, vector: %d\n",
563 SVM_INJECTION_SOFT_INTR, info->intr_core_state.swintr_vector);
565 guest_ctrl->EVENTINJ.vector = info->intr_core_state.swintr_vector;
566 guest_ctrl->EVENTINJ.valid = 1;
568 /* reset swintr state */
569 info->intr_core_state.swintr_posted = 0;
570 info->intr_core_state.swintr_vector = 0;
574 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
577 case V3_INVALID_INTR:
588 v3_svm_config_tsc_virtualization(struct guest_info * info) {
589 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
592 if (info->time_state.flags & VM_TIME_TRAP_RDTSC) {
593 ctrl_area->instrs.RDTSC = 1;
594 ctrl_area->svm_instrs.RDTSCP = 1;
596 ctrl_area->instrs.RDTSC = 0;
597 ctrl_area->svm_instrs.RDTSCP = 0;
599 if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) {
600 ctrl_area->TSC_OFFSET = 0;
602 ctrl_area->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
609 * CAUTION and DANGER!!!
611 * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
612 * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
613 * on its contents will cause things to break. The contents at the time of the exit WILL
614 * change before the exit handler is executed.
616 int v3_svm_enter(struct guest_info * info) {
617 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
618 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
619 addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
620 uint64_t guest_cycles = 0;
622 // Conditionally yield the CPU if the timeslice has expired
625 // Update timer devices after being in the VM before doing
626 // IRQ updates, so that any interrupts they raise get seen
628 v3_advance_time(info, NULL);
629 v3_update_timers(info);
631 // disable global interrupts for vm state transition
634 // Synchronize the guest state to the VMCB
635 guest_state->cr0 = info->ctrl_regs.cr0;
636 guest_state->cr2 = info->ctrl_regs.cr2;
637 guest_state->cr3 = info->ctrl_regs.cr3;
638 guest_state->cr4 = info->ctrl_regs.cr4;
639 guest_state->dr6 = info->dbg_regs.dr6;
640 guest_state->dr7 = info->dbg_regs.dr7;
642 // CR8 is now updated by read/writes and it contains the APIC TPR
643 // the V_TPR should be just the class part of that.
644 // This update is here just for completeness. We currently
645 // are ignoring V_TPR on all injections and doing the priority logivc
647 // guest_ctrl->guest_ctrl.V_TPR = ((info->ctrl_regs.apic_tpr) >> 4) & 0xf;
649 //guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
652 guest_state->rflags = info->ctrl_regs.rflags;
653 guest_state->efer = info->ctrl_regs.efer;
655 /* Synchronize MSRs */
656 guest_state->star = info->msrs.star;
657 guest_state->lstar = info->msrs.lstar;
658 guest_state->sfmask = info->msrs.sfmask;
659 guest_state->KernelGsBase = info->msrs.kern_gs_base;
661 guest_state->cpl = info->cpl;
663 v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
665 guest_state->rax = info->vm_regs.rax;
666 guest_state->rip = info->rip;
667 guest_state->rsp = info->vm_regs.rsp;
669 #ifdef V3_CONFIG_SYMCALL
670 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
671 update_irq_entry_state(info);
674 update_irq_entry_state(info);
681 PrintDebug(info->vm_info, info, "SVM Entry to CS=%p rip=%p...\n",
682 (void *)(addr_t)info->segments.cs.base,
683 (void *)(addr_t)info->rip);
686 #ifdef V3_CONFIG_SYMCALL
687 if (info->sym_core_state.symcall_state.sym_call_active == 1) {
688 if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
689 V3_Print(info->vm_info, info, "!!! Injecting Interrupt during Sym call !!!\n");
694 v3_svm_config_tsc_virtualization(info);
696 //V3_Print(info->vm_info, info, "Calling v3_svm_launch\n");
698 uint64_t entry_tsc = 0;
699 uint64_t exit_tsc = 0;
701 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
702 v3_pwrstat_telemetry_enter(info);
705 #ifdef V3_CONFIG_PMU_TELEMETRY
706 v3_pmu_telemetry_enter(info);
712 v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
716 #ifdef V3_CONFIG_PMU_TELEMETRY
717 v3_pmu_telemetry_exit(info);
720 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
721 v3_pwrstat_telemetry_exit(info);
724 guest_cycles = exit_tsc - entry_tsc;
728 //V3_Print(info->vm_info, info, "SVM Returned: Exit Code: %x, guest_rip=%lx\n", (uint32_t)(guest_ctrl->exit_code), (unsigned long)guest_state->rip);
730 v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
732 v3_advance_time(info, &guest_cycles);
736 // Save Guest state from VMCB
737 info->rip = guest_state->rip;
738 info->vm_regs.rsp = guest_state->rsp;
739 info->vm_regs.rax = guest_state->rax;
741 info->cpl = guest_state->cpl;
743 info->ctrl_regs.cr0 = guest_state->cr0;
744 info->ctrl_regs.cr2 = guest_state->cr2;
745 info->ctrl_regs.cr3 = guest_state->cr3;
746 info->ctrl_regs.cr4 = guest_state->cr4;
747 info->dbg_regs.dr6 = guest_state->dr6;
748 info->dbg_regs.dr7 = guest_state->dr7;
750 // We do not track this anymore
751 // V_TPR is ignored and we do the logic in the APIC
752 //info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
754 info->ctrl_regs.rflags = guest_state->rflags;
755 info->ctrl_regs.efer = guest_state->efer;
757 /* Synchronize MSRs */
758 info->msrs.star = guest_state->star;
759 info->msrs.lstar = guest_state->lstar;
760 info->msrs.sfmask = guest_state->sfmask;
761 info->msrs.kern_gs_base = guest_state->KernelGsBase;
763 v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
764 info->cpu_mode = v3_get_vm_cpu_mode(info);
765 info->mem_mode = v3_get_vm_mem_mode(info);
768 // save exit info here
769 exit_code = guest_ctrl->exit_code;
770 exit_info1 = guest_ctrl->exit_info1;
771 exit_info2 = guest_ctrl->exit_info2;
773 #ifdef V3_CONFIG_SYMCALL
774 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
775 update_irq_exit_state(info);
778 update_irq_exit_state(info);
781 // reenable global interrupts after vm exit
784 // Conditionally yield the CPU if the timeslice has expired
787 // This update timers is for time-dependent handlers
788 // if we're slaved to host time
789 v3_advance_time(info, NULL);
790 v3_update_timers(info);
793 int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2);
796 PrintError(info->vm_info, info, "Error in SVM exit handler (ret=%d)\n", ret);
797 PrintError(info->vm_info, info, " last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code);
802 if (info->timeouts.timeout_active) {
803 /* Check to see if any timeouts have expired */
804 v3_handle_timeouts(info, guest_cycles);
812 int v3_start_svm_guest(struct guest_info * info) {
813 // vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
814 // vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
816 PrintDebug(info->vm_info, info, "Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
820 if (info->core_run_state == CORE_STOPPED) {
822 if (info->vcpu_id == 0) {
823 info->core_run_state = CORE_RUNNING;
825 PrintDebug(info->vm_info, info, "SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
827 while (info->core_run_state == CORE_STOPPED) {
829 if (info->vm_info->run_state == VM_STOPPED) {
830 // The VM was stopped before this core was initialized.
835 //PrintDebug(info->vm_info, info, "SVM core %u: still waiting for INIT\n", info->vcpu_id);
838 PrintDebug(info->vm_info, info, "SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id);
840 // We'll be paranoid about race conditions here
841 v3_wait_at_barrier(info);
844 PrintDebug(info->vm_info, info, "SVM core %u(on %u): I am starting at CS=0x%x (base=0x%p, limit=0x%x), RIP=0x%p\n",
845 info->vcpu_id, info->pcpu_id,
846 info->segments.cs.selector, (void *)(info->segments.cs.base),
847 info->segments.cs.limit, (void *)(info->rip));
851 PrintDebug(info->vm_info, info, "SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n",
852 info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
853 //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
858 if (info->vm_info->run_state == VM_STOPPED) {
859 info->core_run_state = CORE_STOPPED;
864 #ifdef V3_CONFIG_PMU_TELEMETRY
865 v3_pmu_telemetry_start(info);
868 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
869 v3_pwrstat_telemetry_start(info);
872 if (v3_svm_enter(info) == -1) {
873 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
875 addr_t linear_addr = 0;
877 info->vm_info->run_state = VM_ERROR;
879 V3_Print(info->vm_info, info, "SVM core %u: SVM ERROR!!\n", info->vcpu_id);
881 v3_print_guest_state(info);
883 V3_Print(info->vm_info, info, "SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code);
885 V3_Print(info->vm_info, info, "SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
886 V3_Print(info->vm_info, info, "SVM core %u: exit_info1 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info1)) + 4));
888 V3_Print(info->vm_info, info, "SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
889 V3_Print(info->vm_info, info, "SVM core %u: exit_info2 high = 0x%.8x\n", info->vcpu_id, *(uint_t *)(((uchar_t *)&(guest_ctrl->exit_info2)) + 4));
891 linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
893 if (info->mem_mode == PHYSICAL_MEM) {
894 v3_gpa_to_hva(info, linear_addr, &host_addr);
895 } else if (info->mem_mode == VIRTUAL_MEM) {
896 v3_gva_to_hva(info, linear_addr, &host_addr);
899 V3_Print(info->vm_info, info, "SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
901 V3_Print(info->vm_info, info, "SVM core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
902 v3_dump_mem((uint8_t *)host_addr, 15);
904 v3_print_stack(info);
909 v3_wait_at_barrier(info);
912 if (info->vm_info->run_state == VM_STOPPED) {
913 info->core_run_state = CORE_STOPPED;
920 if ((info->num_exits % 50000) == 0) {
921 V3_Print(info->vm_info, info, "SVM Exit number %d\n", (uint32_t)info->num_exits);
922 v3_print_guest_state(info);
928 #ifdef V3_CONFIG_PMU_TELEMETRY
929 v3_pmu_telemetry_end(info);
932 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
933 v3_pwrstat_telemetry_end(info);
935 // Need to take down the other cores on error...
943 int v3_reset_svm_vm_core(struct guest_info * core, addr_t rip) {
946 // Write the RIP, CS, and descriptor
947 // assume the rest is already good to go
949 // vector VV -> rip at 0
951 // This means we start executing at linear address VV000
953 // So the selector needs to be VV00
954 // and the base needs to be VV000
957 core->segments.cs.selector = rip << 8;
958 core->segments.cs.limit = 0xffff;
959 core->segments.cs.base = rip << 12;
969 /* Checks machine SVM capability */
970 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
971 int v3_is_svm_capable() {
972 uint_t vm_cr_low = 0, vm_cr_high = 0;
973 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
975 v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
977 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
979 if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
980 V3_Print(VM_NONE, VCORE_NONE, "SVM Not Available\n");
983 v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
985 PrintDebug(VM_NONE, VCORE_NONE, "SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
987 if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
988 V3_Print(VM_NONE, VCORE_NONE, "SVM is available but is disabled.\n");
990 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
992 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
994 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
995 V3_Print(VM_NONE, VCORE_NONE, "SVM BIOS Disabled, not unlockable\n");
997 V3_Print(VM_NONE, VCORE_NONE, "SVM is locked with a key\n");
1002 V3_Print(VM_NONE, VCORE_NONE, "SVM is available and enabled.\n");
1004 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1005 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
1006 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
1007 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
1008 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
1015 static int has_svm_nested_paging() {
1016 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1018 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1020 //PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
1022 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
1023 V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging not supported\n");
1026 V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging supported\n");
1033 void v3_init_svm_cpu(int cpu_id) {
1035 extern v3_cpu_arch_t v3_cpu_types[];
1037 // Enable SVM on the CPU
1038 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
1039 msr.e_reg.low |= EFER_MSR_svm_enable;
1040 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
1042 V3_Print(VM_NONE, VCORE_NONE, "SVM Enabled\n");
1044 // Setup the host state save area
1045 host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4); // need not be shadow-safe, not exposed to guest
1047 if (!host_vmcbs[cpu_id]) {
1048 PrintError(VM_NONE, VCORE_NONE, "Failed to allocate VMCB\n");
1053 // msr.e_reg.high = 0;
1054 //msr.e_reg.low = (uint_t)host_vmcb;
1055 msr.r_reg = host_vmcbs[cpu_id];
1057 PrintDebug(VM_NONE, VCORE_NONE, "Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
1058 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
1061 if (has_svm_nested_paging() == 1) {
1062 v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
1064 v3_cpu_types[cpu_id] = V3_SVM_CPU;
1070 void v3_deinit_svm_cpu(int cpu_id) {
1072 extern v3_cpu_arch_t v3_cpu_types[];
1074 // reset SVM_VM_HSAVE_PA_MSR
1075 // Does setting it to NULL disable??
1077 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
1080 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
1081 msr.e_reg.low &= ~EFER_MSR_svm_enable;
1082 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
1084 v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1086 V3_FreePages((void *)host_vmcbs[cpu_id], 4);
1088 V3_Print(VM_NONE, VCORE_NONE, "Host CPU %d host area freed, and SVM disabled\n", cpu_id);
1143 * Test VMSAVE/VMLOAD Latency
1145 #define vmsave ".byte 0x0F,0x01,0xDB ; "
1146 #define vmload ".byte 0x0F,0x01,0xDA ; "
1148 uint32_t start_lo, start_hi;
1149 uint32_t end_lo, end_hi;
1150 uint64_t start, end;
1152 __asm__ __volatile__ (
1154 "movl %%eax, %%esi ; "
1155 "movl %%edx, %%edi ; "
1156 "movq %%rcx, %%rax ; "
1159 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1160 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1171 PrintDebug(core->vm_info, core, "VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
1173 __asm__ __volatile__ (
1175 "movl %%eax, %%esi ; "
1176 "movl %%edx, %%edi ; "
1177 "movq %%rcx, %%rax ; "
1180 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1181 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1193 PrintDebug(core->vm_info, core, "VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
1195 /* End Latency Test */
1206 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
1207 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
1208 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
1212 guest_state->rsp = vm_info.vm_regs.rsp;
1213 guest_state->rip = vm_info.rip;
1216 /* I pretty much just gutted this from TVMM */
1217 /* Note: That means its probably wrong */
1219 // set the segment registers to mirror ours
1220 guest_state->cs.selector = 1<<3;
1221 guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
1222 guest_state->cs.attrib.fields.S = 1;
1223 guest_state->cs.attrib.fields.P = 1;
1224 guest_state->cs.attrib.fields.db = 1;
1225 guest_state->cs.attrib.fields.G = 1;
1226 guest_state->cs.limit = 0xfffff;
1227 guest_state->cs.base = 0;
1229 struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
1230 for ( i = 0; segregs[i] != NULL; i++) {
1231 struct vmcb_selector * seg = segregs[i];
1233 seg->selector = 2<<3;
1234 seg->attrib.fields.type = 0x2; // Data Segment+read/write
1235 seg->attrib.fields.S = 1;
1236 seg->attrib.fields.P = 1;
1237 seg->attrib.fields.db = 1;
1238 seg->attrib.fields.G = 1;
1239 seg->limit = 0xfffff;
1245 /* JRL THIS HAS TO GO */
1247 // guest_state->tr.selector = GetTR_Selector();
1248 guest_state->tr.attrib.fields.type = 0x9;
1249 guest_state->tr.attrib.fields.P = 1;
1250 // guest_state->tr.limit = GetTR_Limit();
1251 //guest_state->tr.base = GetTR_Base();// - 0x2000;
1259 guest_state->efer |= EFER_MSR_svm_enable;
1260 guest_state->rflags = 0x00000002; // The reserved bit is always 1
1261 ctrl_area->svm_instrs.VMRUN = 1;
1262 guest_state->cr0 = 0x00000001; // PE
1263 ctrl_area->guest_ASID = 1;
1266 // guest_state->cpl = 0;
1272 ctrl_area->cr_writes.cr4 = 1;
1274 ctrl_area->exceptions.de = 1;
1275 ctrl_area->exceptions.df = 1;
1276 ctrl_area->exceptions.pf = 1;
1277 ctrl_area->exceptions.ts = 1;
1278 ctrl_area->exceptions.ss = 1;
1279 ctrl_area->exceptions.ac = 1;
1280 ctrl_area->exceptions.mc = 1;
1281 ctrl_area->exceptions.gp = 1;
1282 ctrl_area->exceptions.ud = 1;
1283 ctrl_area->exceptions.np = 1;
1284 ctrl_area->exceptions.of = 1;
1285 ctrl_area->exceptions.nmi = 1;
1289 ctrl_area->instrs.IOIO_PROT = 1;
1290 ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3); // need not be shadow-safe, not exposed to guest
1292 if (!ctrl_area->IOPM_BASE_PA) {
1293 PrintError(core->vm_info, core, "Cannot allocate IO bitmap\n");
1299 tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
1300 memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
1303 ctrl_area->instrs.INTR = 1;
1310 memset(gdt_buf, 0, 6);
1311 memset(idt_buf, 0, 6);
1314 uint_t gdt_base, idt_base;
1315 ushort_t gdt_limit, idt_limit;
1318 gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
1319 gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
1320 PrintDebug(core->vm_info, core, "GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
1323 idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
1324 idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
1325 PrintDebug(core->vm_info, core, "IDT: base: %x, limit: %x\n",idt_base, idt_limit);
1328 // gdt_base -= 0x2000;
1329 //idt_base -= 0x2000;
1331 guest_state->gdtr.base = gdt_base;
1332 guest_state->gdtr.limit = gdt_limit;
1333 guest_state->idtr.base = idt_base;
1334 guest_state->idtr.limit = idt_limit;
1340 // also determine if CPU supports nested paging
1342 if (vm_info.page_tables) {
1344 // Flush the TLB on entries/exits
1345 ctrl_area->TLB_CONTROL = 1;
1347 // Enable Nested Paging
1348 ctrl_area->NP_ENABLE = 1;
1350 PrintDebug(core->vm_info, core, "NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
1352 // Set the Nested Page Table pointer
1353 ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1356 // ctrl_area->N_CR3 = Get_CR3();
1357 // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1359 guest_state->g_pat = 0x7040600070406ULL;
1361 PrintDebug(core->vm_info, core, "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));
1362 PrintDebug(core->vm_info, core, "Set Guest CR3: lo: 0x%x hi: 0x%x\n", (uint_t)*(&(guest_state->cr3)), (uint_t)*((unsigned char *)&(guest_state->cr3) + 4));
1364 // guest_state->cr0 |= 0x80000000;