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>
42 #include <palacios/vmm_perftune.h>
45 #ifdef V3_CONFIG_CHECKPOINT
46 #include <palacios/vmm_checkpoint.h>
49 #include <palacios/vmm_direct_paging.h>
51 #include <palacios/vmm_ctrl_regs.h>
52 #include <palacios/svm_io.h>
54 #include <palacios/vmm_sprintf.h>
57 #ifndef V3_CONFIG_DEBUG_SVM
59 #define PrintDebug(fmt, args...)
63 uint32_t v3_last_exit;
65 // This is a global pointer to the host's VMCB
66 static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
70 extern void v3_stgi();
71 extern void v3_clgi();
72 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
73 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
76 static vmcb_t * Allocate_VMCB() {
77 vmcb_t * vmcb_page = NULL;
78 addr_t vmcb_pa = (addr_t)V3_AllocPages(1); // need not be shadow safe, not exposed to guest
80 if ((void *)vmcb_pa == NULL) {
81 PrintError(VM_NONE, VCORE_NONE, "Error allocating VMCB\n");
85 vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
87 memset(vmcb_page, 0, 4096);
93 static int v3_svm_handle_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data)
97 // Call arch-independent handler
98 if ((status = v3_handle_efer_write(core, msr, src, priv_data)) != 0) {
104 // Ensure that hardware visible EFER.SVME bit is set (SVM Enable)
105 struct efer_64 * hw_efer = (struct efer_64 *)&(core->ctrl_regs.efer);
113 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
114 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
115 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
120 ctrl_area->svm_instrs.VMRUN = 1;
121 ctrl_area->svm_instrs.VMMCALL = 1;
122 ctrl_area->svm_instrs.VMLOAD = 1;
123 ctrl_area->svm_instrs.VMSAVE = 1;
124 ctrl_area->svm_instrs.STGI = 1;
125 ctrl_area->svm_instrs.CLGI = 1;
126 ctrl_area->svm_instrs.SKINIT = 1;
127 ctrl_area->svm_instrs.ICEBP = 1;
128 ctrl_area->svm_instrs.WBINVD = 1;
129 ctrl_area->svm_instrs.MONITOR = 1;
130 ctrl_area->svm_instrs.MWAIT_always = 1;
131 ctrl_area->svm_instrs.MWAIT_if_armed = 1;
132 ctrl_area->instrs.INVLPGA = 1;
133 ctrl_area->instrs.CPUID = 1;
135 ctrl_area->instrs.HLT = 1;
137 /* Set at VMM launch as needed */
138 ctrl_area->instrs.RDTSC = 0;
139 ctrl_area->svm_instrs.RDTSCP = 0;
141 // guest_state->cr0 = 0x00000001; // PE
144 ctrl_area->exceptions.de = 1;
145 ctrl_area->exceptions.df = 1;
147 ctrl_area->exceptions.ts = 1;
148 ctrl_area->exceptions.ss = 1;
149 ctrl_area->exceptions.ac = 1;
150 ctrl_area->exceptions.mc = 1;
151 ctrl_area->exceptions.gp = 1;
152 ctrl_area->exceptions.ud = 1;
153 ctrl_area->exceptions.np = 1;
154 ctrl_area->exceptions.of = 1;
156 ctrl_area->exceptions.nmi = 1;
160 ctrl_area->instrs.NMI = 1;
161 ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
162 ctrl_area->instrs.INIT = 1;
163 // ctrl_area->instrs.PAUSE = 1;
164 ctrl_area->instrs.shutdown_evts = 1;
167 /* DEBUG FOR RETURN CODE */
168 ctrl_area->exit_code = 1;
171 /* Setup Guest Machine state */
173 core->vm_regs.rsp = 0x00;
176 core->vm_regs.rdx = 0x00000f00;
181 core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
182 core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
183 core->ctrl_regs.efer |= EFER_MSR_svm_enable;
189 core->segments.cs.selector = 0xf000;
190 core->segments.cs.limit = 0xffff;
191 core->segments.cs.base = 0x0000000f0000LL;
193 // (raw attributes = 0xf3)
194 core->segments.cs.type = 0x3;
195 core->segments.cs.system = 0x1;
196 core->segments.cs.dpl = 0x3;
197 core->segments.cs.present = 1;
201 struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
202 &(core->segments.es), &(core->segments.fs),
203 &(core->segments.gs), NULL};
205 for ( i = 0; segregs[i] != NULL; i++) {
206 struct v3_segment * seg = segregs[i];
208 seg->selector = 0x0000;
209 // seg->base = seg->selector << 4;
210 seg->base = 0x00000000;
213 // (raw attributes = 0xf3)
220 core->segments.gdtr.limit = 0x0000ffff;
221 core->segments.gdtr.base = 0x0000000000000000LL;
222 core->segments.idtr.limit = 0x0000ffff;
223 core->segments.idtr.base = 0x0000000000000000LL;
225 core->segments.ldtr.selector = 0x0000;
226 core->segments.ldtr.limit = 0x0000ffff;
227 core->segments.ldtr.base = 0x0000000000000000LL;
228 core->segments.tr.selector = 0x0000;
229 core->segments.tr.limit = 0x0000ffff;
230 core->segments.tr.base = 0x0000000000000000LL;
233 core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
234 core->dbg_regs.dr7 = 0x0000000000000400LL;
237 ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
238 ctrl_area->instrs.IOIO_PROT = 1;
240 ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
241 ctrl_area->instrs.MSR_PROT = 1;
244 PrintDebug(core->vm_info, core, "Exiting on interrupts\n");
245 ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
246 ctrl_area->instrs.INTR = 1;
247 // The above also assures the TPR changes (CR8) are only virtual
250 // However, we need to see TPR writes since they will
251 // affect the virtual apic
252 // we reflect out cr8 to ctrl_regs->apic_tpr
253 ctrl_area->cr_reads.cr8 = 1;
254 ctrl_area->cr_writes.cr8 = 1;
255 // We will do all TPR comparisons in the virtual apic
256 // We also do not want the V_TPR to be able to mask the PIC
257 ctrl_area->guest_ctrl.V_IGN_TPR = 1;
261 v3_hook_msr(core->vm_info, EFER_MSR,
262 &v3_handle_efer_read,
263 &v3_svm_handle_efer_write,
266 if (core->shdw_pg_mode == SHADOW_PAGING) {
267 PrintDebug(core->vm_info, core, "Creating initial shadow page table\n");
269 /* JRL: This is a performance killer, and a simplistic solution */
270 /* We need to fix this */
271 ctrl_area->TLB_CONTROL = 1;
272 ctrl_area->guest_ASID = 1;
275 if (v3_init_passthrough_pts(core) == -1) {
276 PrintError(core->vm_info, core, "Could not initialize passthrough page tables\n");
281 core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
282 PrintDebug(core->vm_info, core, "Created\n");
284 core->ctrl_regs.cr0 |= 0x80000000;
285 core->ctrl_regs.cr3 = core->direct_map_pt;
287 ctrl_area->cr_reads.cr0 = 1;
288 ctrl_area->cr_writes.cr0 = 1;
289 //ctrl_area->cr_reads.cr4 = 1;
290 ctrl_area->cr_writes.cr4 = 1;
291 ctrl_area->cr_reads.cr3 = 1;
292 ctrl_area->cr_writes.cr3 = 1;
295 ctrl_area->instrs.INVLPG = 1;
297 ctrl_area->exceptions.pf = 1;
299 guest_state->g_pat = 0x7040600070406ULL;
302 } else if (core->shdw_pg_mode == NESTED_PAGING) {
303 // Flush the TLB on entries/exits
304 ctrl_area->TLB_CONTROL = 1;
305 ctrl_area->guest_ASID = 1;
307 // Enable Nested Paging
308 ctrl_area->NP_ENABLE = 1;
310 PrintDebug(core->vm_info, core, "NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
312 // Set the Nested Page Table pointer
313 if (v3_init_passthrough_pts(core) == -1) {
314 PrintError(core->vm_info, core, "Could not initialize Nested page tables\n");
318 ctrl_area->N_CR3 = core->direct_map_pt;
320 guest_state->g_pat = 0x7040600070406ULL;
323 /* tell the guest that we don't support SVM */
324 v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
325 &v3_handle_vm_cr_read,
326 &v3_handle_vm_cr_write,
331 #define INT_PENDING_AMD_MSR 0xc0010055
333 v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
334 v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
335 v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
336 v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
337 v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
339 v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
340 v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
341 v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
344 v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
345 v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
347 // Passthrough read operations are ok.
348 v3_hook_msr(core->vm_info, INT_PENDING_AMD_MSR, NULL, v3_msr_unhandled_write, NULL);
353 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
355 PrintDebug(core->vm_info, core, "Allocating VMCB\n");
356 core->vmm_data = (void *)Allocate_VMCB();
358 if (core->vmm_data == NULL) {
359 PrintError(core->vm_info, core, "Could not allocate VMCB, Exiting...\n");
363 if (vm_class == V3_PC_VM) {
364 PrintDebug(core->vm_info, core, "Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
365 Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
367 PrintError(core->vm_info, core, "Invalid VM class\n");
371 core->core_run_state = CORE_STOPPED;
377 int v3_deinit_svm_vmcb(struct guest_info * core) {
378 V3_FreePages(V3_PAddr(core->vmm_data), 1);
383 #ifdef V3_CONFIG_CHECKPOINT
384 int v3_svm_save_core(struct guest_info * core, void * ctx){
386 vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data);
388 // Special case saves of data we need immediate access to
390 V3_CHKPT_SAVE(ctx, "CPL", core->cpl, failout);
391 V3_CHKPT_SAVE(ctx,"STAR", guest_area->star, failout);
392 V3_CHKPT_SAVE(ctx,"CSTAR", guest_area->cstar, failout);
393 V3_CHKPT_SAVE(ctx,"LSTAR", guest_area->lstar, failout);
394 V3_CHKPT_SAVE(ctx,"SFMASK", guest_area->sfmask, failout);
395 V3_CHKPT_SAVE(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout);
396 V3_CHKPT_SAVE(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout);
397 V3_CHKPT_SAVE(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout);
398 V3_CHKPT_SAVE(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout);
400 // and then we save the whole enchilada
401 if (v3_chkpt_save(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) {
402 PrintError(core->vm_info, core, "Could not save SVM vmcb\n");
409 PrintError(core->vm_info, core, "Failed to save SVM state for core\n");
414 int v3_svm_load_core(struct guest_info * core, void * ctx){
417 vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data);
419 // Reload what we special cased, which we will overwrite in a minute
420 V3_CHKPT_LOAD(ctx, "CPL", core->cpl, failout);
421 V3_CHKPT_LOAD(ctx,"STAR", guest_area->star, failout);
422 V3_CHKPT_LOAD(ctx,"CSTAR", guest_area->cstar, failout);
423 V3_CHKPT_LOAD(ctx,"LSTAR", guest_area->lstar, failout);
424 V3_CHKPT_LOAD(ctx,"SFMASK", guest_area->sfmask, failout);
425 V3_CHKPT_LOAD(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout);
426 V3_CHKPT_LOAD(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout);
427 V3_CHKPT_LOAD(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout);
428 V3_CHKPT_LOAD(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout);
430 // and then we load the whole enchilada
431 if (v3_chkpt_load(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) {
432 PrintError(core->vm_info, core, "Could not load SVM vmcb\n");
439 PrintError(core->vm_info, core, "Failed to save SVM state for core\n");
445 static int update_irq_exit_state(struct guest_info * info) {
446 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
448 // Fix for QEMU bug using EVENTINJ as an internal cache
449 guest_ctrl->EVENTINJ.valid = 0;
451 if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
453 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
454 PrintDebug(info->vm_info, info, "INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
457 info->intr_core_state.irq_started = 1;
458 info->intr_core_state.irq_pending = 0;
460 v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
463 if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
464 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
465 PrintDebug(info->vm_info, info, "Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
468 // Interrupt was taken fully vectored
469 info->intr_core_state.irq_started = 0;
471 } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
472 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
473 PrintDebug(info->vm_info, info, "EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
481 static int update_irq_entry_state(struct guest_info * info) {
482 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
485 if (info->intr_core_state.irq_pending == 0) {
486 guest_ctrl->guest_ctrl.V_IRQ = 0;
487 guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
490 if (v3_excp_pending(info)) {
491 uint_t excp = v3_get_excp_number(info);
493 guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
495 if (info->excp_state.excp_error_code_valid) {
496 guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
497 guest_ctrl->EVENTINJ.ev = 1;
498 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
499 PrintDebug(info->vm_info, info, "Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
503 guest_ctrl->EVENTINJ.vector = excp;
505 guest_ctrl->EVENTINJ.valid = 1;
507 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
508 PrintDebug(info->vm_info, info, "<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
509 (int)info->num_exits,
510 guest_ctrl->EVENTINJ.vector,
511 (void *)(addr_t)info->ctrl_regs.cr2,
512 (void *)(addr_t)info->rip);
515 v3_injecting_excp(info, excp);
516 } else if (info->intr_core_state.irq_started == 1) {
517 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
518 PrintDebug(info->vm_info, info, "IRQ pending from previous injection\n");
520 guest_ctrl->guest_ctrl.V_IRQ = 1;
521 guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
523 // We ignore the virtual TPR on this injection
524 // TPR/PPR tests have already been done in the APIC.
525 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
526 guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf;
529 switch (v3_intr_pending(info)) {
530 case V3_EXTERNAL_IRQ: {
531 int irq = v3_get_intr(info);
537 guest_ctrl->guest_ctrl.V_IRQ = 1;
538 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
540 // We ignore the virtual TPR on this injection
541 // TPR/PPR tests have already been done in the APIC.
542 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
543 guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf;
545 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
546 PrintDebug(info->vm_info, info, "Injecting Interrupt %d (EIP=%p)\n",
547 guest_ctrl->guest_ctrl.V_INTR_VECTOR,
548 (void *)(addr_t)info->rip);
551 info->intr_core_state.irq_pending = 1;
552 info->intr_core_state.irq_vector = irq;
557 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
559 case V3_SOFTWARE_INTR:
560 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
562 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
563 PrintDebug(info->vm_info, info, "Injecting software interrupt -- type: %d, vector: %d\n",
564 SVM_INJECTION_SOFT_INTR, info->intr_core_state.swintr_vector);
566 guest_ctrl->EVENTINJ.vector = info->intr_core_state.swintr_vector;
567 guest_ctrl->EVENTINJ.valid = 1;
569 /* reset swintr state */
570 info->intr_core_state.swintr_posted = 0;
571 info->intr_core_state.swintr_vector = 0;
575 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
578 case V3_INVALID_INTR:
589 v3_svm_config_tsc_virtualization(struct guest_info * info) {
590 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
593 if (info->time_state.flags & VM_TIME_TRAP_RDTSC) {
594 ctrl_area->instrs.RDTSC = 1;
595 ctrl_area->svm_instrs.RDTSCP = 1;
597 ctrl_area->instrs.RDTSC = 0;
598 ctrl_area->svm_instrs.RDTSCP = 0;
600 if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) {
601 ctrl_area->TSC_OFFSET = 0;
603 ctrl_area->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
610 * CAUTION and DANGER!!!
612 * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
613 * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
614 * on its contents will cause things to break. The contents at the time of the exit WILL
615 * change before the exit handler is executed.
617 int v3_svm_enter(struct guest_info * info) {
618 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
619 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
620 addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
621 uint64_t guest_cycles = 0;
623 // Conditionally yield the CPU if the timeslice has expired
626 // Update timer devices after being in the VM before doing
627 // IRQ updates, so that any interrupts they raise get seen
629 v3_advance_time(info, NULL);
630 v3_update_timers(info);
632 // disable global interrupts for vm state transition
635 // Synchronize the guest state to the VMCB
636 guest_state->cr0 = info->ctrl_regs.cr0;
637 guest_state->cr2 = info->ctrl_regs.cr2;
638 guest_state->cr3 = info->ctrl_regs.cr3;
639 guest_state->cr4 = info->ctrl_regs.cr4;
640 guest_state->dr6 = info->dbg_regs.dr6;
641 guest_state->dr7 = info->dbg_regs.dr7;
643 // CR8 is now updated by read/writes and it contains the APIC TPR
644 // the V_TPR should be just the class part of that.
645 // This update is here just for completeness. We currently
646 // are ignoring V_TPR on all injections and doing the priority logivc
648 // guest_ctrl->guest_ctrl.V_TPR = ((info->ctrl_regs.apic_tpr) >> 4) & 0xf;
650 //guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
653 guest_state->rflags = info->ctrl_regs.rflags;
654 guest_state->efer = info->ctrl_regs.efer;
656 /* Synchronize MSRs */
657 guest_state->star = info->msrs.star;
658 guest_state->lstar = info->msrs.lstar;
659 guest_state->sfmask = info->msrs.sfmask;
660 guest_state->KernelGsBase = info->msrs.kern_gs_base;
662 guest_state->cpl = info->cpl;
664 v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
666 guest_state->rax = info->vm_regs.rax;
667 guest_state->rip = info->rip;
668 guest_state->rsp = info->vm_regs.rsp;
670 V3_FP_ENTRY_RESTORE(info);
672 #ifdef V3_CONFIG_SYMCALL
673 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
674 update_irq_entry_state(info);
677 update_irq_entry_state(info);
684 PrintDebug(info->vm_info, info, "SVM Entry to CS=%p rip=%p...\n",
685 (void *)(addr_t)info->segments.cs.base,
686 (void *)(addr_t)info->rip);
689 #ifdef V3_CONFIG_SYMCALL
690 if (info->sym_core_state.symcall_state.sym_call_active == 1) {
691 if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
692 V3_Print(info->vm_info, info, "!!! Injecting Interrupt during Sym call !!!\n");
697 v3_svm_config_tsc_virtualization(info);
699 //V3_Print(info->vm_info, info, "Calling v3_svm_launch\n");
701 uint64_t entry_tsc = 0;
702 uint64_t exit_tsc = 0;
704 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
705 v3_pwrstat_telemetry_enter(info);
708 #ifdef V3_CONFIG_PMU_TELEMETRY
709 v3_pmu_telemetry_enter(info);
715 v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
719 #ifdef V3_CONFIG_PMU_TELEMETRY
720 v3_pmu_telemetry_exit(info);
723 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
724 v3_pwrstat_telemetry_exit(info);
727 guest_cycles = exit_tsc - entry_tsc;
731 //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);
733 v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
735 v3_advance_time(info, &guest_cycles);
739 V3_FP_EXIT_SAVE(info);
741 // Save Guest state from VMCB
742 info->rip = guest_state->rip;
743 info->vm_regs.rsp = guest_state->rsp;
744 info->vm_regs.rax = guest_state->rax;
746 info->cpl = guest_state->cpl;
748 info->ctrl_regs.cr0 = guest_state->cr0;
749 info->ctrl_regs.cr2 = guest_state->cr2;
750 info->ctrl_regs.cr3 = guest_state->cr3;
751 info->ctrl_regs.cr4 = guest_state->cr4;
752 info->dbg_regs.dr6 = guest_state->dr6;
753 info->dbg_regs.dr7 = guest_state->dr7;
755 // We do not track this anymore
756 // V_TPR is ignored and we do the logic in the APIC
757 //info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
759 info->ctrl_regs.rflags = guest_state->rflags;
760 info->ctrl_regs.efer = guest_state->efer;
762 /* Synchronize MSRs */
763 info->msrs.star = guest_state->star;
764 info->msrs.lstar = guest_state->lstar;
765 info->msrs.sfmask = guest_state->sfmask;
766 info->msrs.kern_gs_base = guest_state->KernelGsBase;
768 v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
769 info->cpu_mode = v3_get_vm_cpu_mode(info);
770 info->mem_mode = v3_get_vm_mem_mode(info);
773 // save exit info here
774 exit_code = guest_ctrl->exit_code;
775 exit_info1 = guest_ctrl->exit_info1;
776 exit_info2 = guest_ctrl->exit_info2;
778 #ifdef V3_CONFIG_SYMCALL
779 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
780 update_irq_exit_state(info);
783 update_irq_exit_state(info);
786 // reenable global interrupts after vm exit
789 // Conditionally yield the CPU if the timeslice has expired
792 // This update timers is for time-dependent handlers
793 // if we're slaved to host time
794 v3_advance_time(info, NULL);
795 v3_update_timers(info);
798 int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2);
801 PrintError(info->vm_info, info, "Error in SVM exit handler (ret=%d)\n", ret);
802 PrintError(info->vm_info, info, " last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code);
807 if (info->timeouts.timeout_active) {
808 /* Check to see if any timeouts have expired */
809 v3_handle_timeouts(info, guest_cycles);
817 int v3_start_svm_guest(struct guest_info * info) {
818 // vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
819 // vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
821 PrintDebug(info->vm_info, info, "Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
825 if (info->core_run_state == CORE_STOPPED) {
827 if (info->vcpu_id == 0) {
828 info->core_run_state = CORE_RUNNING;
830 PrintDebug(info->vm_info, info, "SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
834 while (info->core_run_state == CORE_STOPPED) {
836 if (info->vm_info->run_state == VM_STOPPED) {
837 // The VM was stopped before this core was initialized.
841 V3_STILL_NO_WORK(info);
843 //PrintDebug(info->vm_info, info, "SVM core %u: still waiting for INIT\n", info->vcpu_id);
846 V3_HAVE_WORK_AGAIN(info);
848 PrintDebug(info->vm_info, info, "SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id);
850 // We'll be paranoid about race conditions here
851 v3_wait_at_barrier(info);
854 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",
855 info->vcpu_id, info->pcpu_id,
856 info->segments.cs.selector, (void *)(info->segments.cs.base),
857 info->segments.cs.limit, (void *)(info->rip));
861 PrintDebug(info->vm_info, info, "SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n",
862 info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
863 //PrintDebugVMCB((vmcb_t*)(info->vmm_data));
868 if (info->vm_info->run_state == VM_STOPPED) {
869 info->core_run_state = CORE_STOPPED;
874 #ifdef V3_CONFIG_PMU_TELEMETRY
875 v3_pmu_telemetry_start(info);
878 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
879 v3_pwrstat_telemetry_start(info);
882 if (v3_svm_enter(info) == -1) {
883 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
885 addr_t linear_addr = 0;
887 info->vm_info->run_state = VM_ERROR;
889 V3_Print(info->vm_info, info, "SVM core %u: SVM ERROR!!\n", info->vcpu_id);
891 v3_print_guest_state(info);
893 V3_Print(info->vm_info, info, "SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code);
895 V3_Print(info->vm_info, info, "SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
896 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));
898 V3_Print(info->vm_info, info, "SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
899 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));
901 linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
903 if (info->mem_mode == PHYSICAL_MEM) {
904 v3_gpa_to_hva(info, linear_addr, &host_addr);
905 } else if (info->mem_mode == VIRTUAL_MEM) {
906 v3_gva_to_hva(info, linear_addr, &host_addr);
909 V3_Print(info->vm_info, info, "SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
911 V3_Print(info->vm_info, info, "SVM core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
912 v3_dump_mem((uint8_t *)host_addr, 15);
914 v3_print_stack(info);
919 v3_wait_at_barrier(info);
922 if (info->vm_info->run_state == VM_STOPPED) {
923 info->core_run_state = CORE_STOPPED;
930 if ((info->num_exits % 50000) == 0) {
931 V3_Print(info->vm_info, info, "SVM Exit number %d\n", (uint32_t)info->num_exits);
932 v3_print_guest_state(info);
938 #ifdef V3_CONFIG_PMU_TELEMETRY
939 v3_pmu_telemetry_end(info);
942 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
943 v3_pwrstat_telemetry_end(info);
945 // Need to take down the other cores on error...
953 int v3_reset_svm_vm_core(struct guest_info * core, addr_t rip) {
956 // Write the RIP, CS, and descriptor
957 // assume the rest is already good to go
959 // vector VV -> rip at 0
961 // This means we start executing at linear address VV000
963 // So the selector needs to be VV00
964 // and the base needs to be VV000
967 core->segments.cs.selector = rip << 8;
968 core->segments.cs.limit = 0xffff;
969 core->segments.cs.base = rip << 12;
979 /* Checks machine SVM capability */
980 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
981 int v3_is_svm_capable() {
982 uint_t vm_cr_low = 0, vm_cr_high = 0;
983 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
985 v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
987 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
989 if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
990 V3_Print(VM_NONE, VCORE_NONE, "SVM Not Available\n");
993 v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
995 PrintDebug(VM_NONE, VCORE_NONE, "SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
997 if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
998 V3_Print(VM_NONE, VCORE_NONE, "SVM is available but is disabled.\n");
1000 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1002 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
1004 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
1005 V3_Print(VM_NONE, VCORE_NONE, "SVM BIOS Disabled, not unlockable\n");
1007 V3_Print(VM_NONE, VCORE_NONE, "SVM is locked with a key\n");
1012 V3_Print(VM_NONE, VCORE_NONE, "SVM is available and enabled.\n");
1014 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1015 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
1016 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
1017 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
1018 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
1025 static int has_svm_nested_paging() {
1026 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1028 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1030 //PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
1032 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
1033 V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging not supported\n");
1036 V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging supported\n");
1043 void v3_init_svm_cpu(int cpu_id) {
1045 extern v3_cpu_arch_t v3_cpu_types[];
1047 // Enable SVM on the CPU
1048 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
1049 msr.e_reg.low |= EFER_MSR_svm_enable;
1050 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
1052 V3_Print(VM_NONE, VCORE_NONE, "SVM Enabled\n");
1054 // Setup the host state save area
1055 host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4); // need not be shadow-safe, not exposed to guest
1057 if (!host_vmcbs[cpu_id]) {
1058 PrintError(VM_NONE, VCORE_NONE, "Failed to allocate VMCB\n");
1063 // msr.e_reg.high = 0;
1064 //msr.e_reg.low = (uint_t)host_vmcb;
1065 msr.r_reg = host_vmcbs[cpu_id];
1067 PrintDebug(VM_NONE, VCORE_NONE, "Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
1068 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
1071 if (has_svm_nested_paging() == 1) {
1072 v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
1074 v3_cpu_types[cpu_id] = V3_SVM_CPU;
1080 void v3_deinit_svm_cpu(int cpu_id) {
1082 extern v3_cpu_arch_t v3_cpu_types[];
1084 // reset SVM_VM_HSAVE_PA_MSR
1085 // Does setting it to NULL disable??
1087 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
1090 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
1091 msr.e_reg.low &= ~EFER_MSR_svm_enable;
1092 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
1094 v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1096 V3_FreePages((void *)host_vmcbs[cpu_id], 4);
1098 V3_Print(VM_NONE, VCORE_NONE, "Host CPU %d host area freed, and SVM disabled\n", cpu_id);
1153 * Test VMSAVE/VMLOAD Latency
1155 #define vmsave ".byte 0x0F,0x01,0xDB ; "
1156 #define vmload ".byte 0x0F,0x01,0xDA ; "
1158 uint32_t start_lo, start_hi;
1159 uint32_t end_lo, end_hi;
1160 uint64_t start, end;
1162 __asm__ __volatile__ (
1164 "movl %%eax, %%esi ; "
1165 "movl %%edx, %%edi ; "
1166 "movq %%rcx, %%rax ; "
1169 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1170 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1181 PrintDebug(core->vm_info, core, "VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
1183 __asm__ __volatile__ (
1185 "movl %%eax, %%esi ; "
1186 "movl %%edx, %%edi ; "
1187 "movq %%rcx, %%rax ; "
1190 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1191 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1203 PrintDebug(core->vm_info, core, "VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
1205 /* End Latency Test */
1216 void Init_VMCB_pe(vmcb_t *vmcb, struct guest_info vm_info) {
1217 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
1218 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
1222 guest_state->rsp = vm_info.vm_regs.rsp;
1223 guest_state->rip = vm_info.rip;
1226 /* I pretty much just gutted this from TVMM */
1227 /* Note: That means its probably wrong */
1229 // set the segment registers to mirror ours
1230 guest_state->cs.selector = 1<<3;
1231 guest_state->cs.attrib.fields.type = 0xa; // Code segment+read
1232 guest_state->cs.attrib.fields.S = 1;
1233 guest_state->cs.attrib.fields.P = 1;
1234 guest_state->cs.attrib.fields.db = 1;
1235 guest_state->cs.attrib.fields.G = 1;
1236 guest_state->cs.limit = 0xfffff;
1237 guest_state->cs.base = 0;
1239 struct vmcb_selector *segregs [] = {&(guest_state->ss), &(guest_state->ds), &(guest_state->es), &(guest_state->fs), &(guest_state->gs), NULL};
1240 for ( i = 0; segregs[i] != NULL; i++) {
1241 struct vmcb_selector * seg = segregs[i];
1243 seg->selector = 2<<3;
1244 seg->attrib.fields.type = 0x2; // Data Segment+read/write
1245 seg->attrib.fields.S = 1;
1246 seg->attrib.fields.P = 1;
1247 seg->attrib.fields.db = 1;
1248 seg->attrib.fields.G = 1;
1249 seg->limit = 0xfffff;
1255 /* JRL THIS HAS TO GO */
1257 // guest_state->tr.selector = GetTR_Selector();
1258 guest_state->tr.attrib.fields.type = 0x9;
1259 guest_state->tr.attrib.fields.P = 1;
1260 // guest_state->tr.limit = GetTR_Limit();
1261 //guest_state->tr.base = GetTR_Base();// - 0x2000;
1269 guest_state->efer |= EFER_MSR_svm_enable;
1270 guest_state->rflags = 0x00000002; // The reserved bit is always 1
1271 ctrl_area->svm_instrs.VMRUN = 1;
1272 guest_state->cr0 = 0x00000001; // PE
1273 ctrl_area->guest_ASID = 1;
1276 // guest_state->cpl = 0;
1282 ctrl_area->cr_writes.cr4 = 1;
1284 ctrl_area->exceptions.de = 1;
1285 ctrl_area->exceptions.df = 1;
1286 ctrl_area->exceptions.pf = 1;
1287 ctrl_area->exceptions.ts = 1;
1288 ctrl_area->exceptions.ss = 1;
1289 ctrl_area->exceptions.ac = 1;
1290 ctrl_area->exceptions.mc = 1;
1291 ctrl_area->exceptions.gp = 1;
1292 ctrl_area->exceptions.ud = 1;
1293 ctrl_area->exceptions.np = 1;
1294 ctrl_area->exceptions.of = 1;
1295 ctrl_area->exceptions.nmi = 1;
1299 ctrl_area->instrs.IOIO_PROT = 1;
1300 ctrl_area->IOPM_BASE_PA = (uint_t)V3_AllocPages(3); // need not be shadow-safe, not exposed to guest
1302 if (!ctrl_area->IOPM_BASE_PA) {
1303 PrintError(core->vm_info, core, "Cannot allocate IO bitmap\n");
1309 tmp_reg.r_reg = ctrl_area->IOPM_BASE_PA;
1310 memset((void*)(tmp_reg.e_reg.low), 0xffffffff, PAGE_SIZE * 2);
1313 ctrl_area->instrs.INTR = 1;
1320 memset(gdt_buf, 0, 6);
1321 memset(idt_buf, 0, 6);
1324 uint_t gdt_base, idt_base;
1325 ushort_t gdt_limit, idt_limit;
1328 gdt_base = *(ulong_t*)((uchar_t*)gdt_buf + 2) & 0xffffffff;
1329 gdt_limit = *(ushort_t*)(gdt_buf) & 0xffff;
1330 PrintDebug(core->vm_info, core, "GDT: base: %x, limit: %x\n", gdt_base, gdt_limit);
1333 idt_base = *(ulong_t*)(idt_buf + 2) & 0xffffffff;
1334 idt_limit = *(ushort_t*)(idt_buf) & 0xffff;
1335 PrintDebug(core->vm_info, core, "IDT: base: %x, limit: %x\n",idt_base, idt_limit);
1338 // gdt_base -= 0x2000;
1339 //idt_base -= 0x2000;
1341 guest_state->gdtr.base = gdt_base;
1342 guest_state->gdtr.limit = gdt_limit;
1343 guest_state->idtr.base = idt_base;
1344 guest_state->idtr.limit = idt_limit;
1350 // also determine if CPU supports nested paging
1352 if (vm_info.page_tables) {
1354 // Flush the TLB on entries/exits
1355 ctrl_area->TLB_CONTROL = 1;
1357 // Enable Nested Paging
1358 ctrl_area->NP_ENABLE = 1;
1360 PrintDebug(core->vm_info, core, "NP_Enable at 0x%x\n", &(ctrl_area->NP_ENABLE));
1362 // Set the Nested Page Table pointer
1363 ctrl_area->N_CR3 |= ((addr_t)vm_info.page_tables & 0xfffff000);
1366 // ctrl_area->N_CR3 = Get_CR3();
1367 // guest_state->cr3 |= (Get_CR3() & 0xfffff000);
1369 guest_state->g_pat = 0x7040600070406ULL;
1371 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));
1372 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));
1374 // guest_state->cr0 |= 0x80000000;