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>
44 #include <palacios/vmm_bios.h>
47 #ifdef V3_CONFIG_CHECKPOINT
48 #include <palacios/vmm_checkpoint.h>
51 #include <palacios/vmm_direct_paging.h>
53 #include <palacios/vmm_ctrl_regs.h>
54 #include <palacios/svm_io.h>
56 #include <palacios/vmm_sprintf.h>
58 #ifdef V3_CONFIG_MEM_TRACK
59 #include <palacios/vmm_mem_track.h>
62 #ifdef V3_CONFIG_TM_FUNC
63 #include <extensions/trans_mem.h>
66 #ifndef V3_CONFIG_DEBUG_SVM
68 #define PrintDebug(fmt, args...)
73 uint32_t v3_last_exit;
75 // This is a global pointer to the host's VMCB
76 // These are physical addresses
77 static addr_t host_vmcbs[V3_CONFIG_MAX_CPUS] = { [0 ... V3_CONFIG_MAX_CPUS - 1] = 0};
81 extern void v3_stgi();
82 extern void v3_clgi();
83 //extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, uint64_t * fs, uint64_t * gs);
84 extern int v3_svm_launch(vmcb_t * vmcb, struct v3_gprs * vm_regs, vmcb_t * host_vmcb);
88 static vmcb_t * Allocate_VMCB() {
89 vmcb_t * vmcb_page = NULL;
90 addr_t vmcb_pa = (addr_t)V3_AllocPages(1); // need not be shadow safe, not exposed to guest
92 if ((void *)vmcb_pa == NULL) {
93 PrintError(VM_NONE, VCORE_NONE, "Error allocating VMCB\n");
97 vmcb_page = (vmcb_t *)V3_VAddr((void *)vmcb_pa);
99 memset(vmcb_page, 0, 4096);
105 static int v3_svm_handle_efer_write(struct guest_info * core, uint_t msr, struct v3_msr src, void * priv_data)
109 // Call arch-independent handler
110 if ((status = v3_handle_efer_write(core, msr, src, priv_data)) != 0) {
116 // Ensure that hardware visible EFER.SVME bit is set (SVM Enable)
117 struct efer_64 * hw_efer = (struct efer_64 *)&(core->ctrl_regs.efer);
125 * This is invoked both on an initial boot and on a reset
127 * The difference is that on a reset we will not rehook anything
131 static void Init_VMCB_BIOS(vmcb_t * vmcb, struct guest_info * core) {
132 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA(vmcb);
133 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA(vmcb);
136 if (core->core_run_state!=CORE_INVALID && core->core_run_state!=CORE_RESETTING) {
137 PrintError(core->vm_info, core, "Atempt to Init_VMCB_BIOS in invalid state (%d)\n",core->core_run_state);
141 // need to invalidate any shadow page tables early
142 if (core->shdw_pg_mode == SHADOW_PAGING && core->core_run_state==CORE_RESETTING) {
143 if (v3_get_vm_cpu_mode(core) != REAL) {
144 if (v3_invalidate_shadow_pts(core) == -1) {
145 PrintError(core->vm_info,core,"Could not invalidate shadow page tables\n");
151 // Guarantee we are starting from a clean slate
155 ctrl_area->svm_instrs.VMRUN = 1;
156 ctrl_area->svm_instrs.VMMCALL = 1;
157 ctrl_area->svm_instrs.VMLOAD = 1;
158 ctrl_area->svm_instrs.VMSAVE = 1;
159 ctrl_area->svm_instrs.STGI = 1;
160 ctrl_area->svm_instrs.CLGI = 1;
161 ctrl_area->svm_instrs.SKINIT = 1; // secure startup... why
162 ctrl_area->svm_instrs.ICEBP = 1; // in circuit emulator breakpoint
163 ctrl_area->svm_instrs.WBINVD = 1; // write back and invalidate caches... why?
164 ctrl_area->svm_instrs.MONITOR = 1;
165 ctrl_area->svm_instrs.MWAIT_always = 1;
166 ctrl_area->svm_instrs.MWAIT_if_armed = 1;
167 ctrl_area->instrs.INVLPGA = 1; // invalidate page in asid... why?
168 ctrl_area->instrs.CPUID = 1;
170 ctrl_area->instrs.HLT = 1;
172 /* Set at VMM launch as needed */
173 ctrl_area->instrs.RDTSC = 0;
174 ctrl_area->svm_instrs.RDTSCP = 0;
177 #ifdef V3_CONFIG_TM_FUNC
178 v3_tm_set_excp_intercepts(ctrl_area);
182 ctrl_area->instrs.NMI = 1;
183 ctrl_area->instrs.SMI = 0; // allow SMIs to run in guest
184 ctrl_area->instrs.INIT = 1;
185 // ctrl_area->instrs.PAUSE = 1; // do not care as does not halt
186 ctrl_area->instrs.shutdown_evts = 1;
189 /* DEBUG FOR RETURN CODE */
190 ctrl_area->exit_code = 1;
193 /* Setup Guest Machine state */
195 memset(&core->vm_regs,0,sizeof(core->vm_regs));
196 memset(&core->ctrl_regs,0,sizeof(core->ctrl_regs));
197 memset(&core->dbg_regs,0,sizeof(core->dbg_regs));
198 memset(&core->segments,0,sizeof(core->segments));
199 memset(&core->msrs,0,sizeof(core->msrs));
200 memset(&core->fp_state,0,sizeof(core->fp_state));
203 core->intr_core_state.irq_pending=0;
204 core->intr_core_state.irq_started=0;
205 core->intr_core_state.swintr_posted=0;
208 core->excp_state.excp_pending=0;
210 // reset of gprs to expected values at init
211 core->vm_regs.rsp = 0x00;
213 core->vm_regs.rdx = 0x00000f00; // family/stepping/etc
218 core->ctrl_regs.rflags = 0x00000002; // The reserved bit is always 1
220 core->ctrl_regs.cr0 = 0x60010010; // Set the WP flag so the memory hooks work in real-mode
221 core->shdw_pg_state.guest_cr0 = core->ctrl_regs.cr0;
224 core->shdw_pg_state.guest_cr3 = core->ctrl_regs.cr3;
226 core->shdw_pg_state.guest_cr4 = core->ctrl_regs.cr4;
228 core->ctrl_regs.efer |= EFER_MSR_svm_enable ;
229 core->shdw_pg_state.guest_efer.value = core->ctrl_regs.efer;
231 core->segments.cs.selector = 0xf000;
232 core->segments.cs.limit = 0xffff;
233 core->segments.cs.base = 0x0000f0000LL;
235 // (raw attributes = 0xf3)
236 core->segments.cs.type = 0xa;
237 core->segments.cs.system = 0x1;
238 core->segments.cs.dpl = 0x0;
239 core->segments.cs.present = 1;
243 struct v3_segment * segregs [] = {&(core->segments.ss), &(core->segments.ds),
244 &(core->segments.es), &(core->segments.fs),
245 &(core->segments.gs), NULL};
247 for ( i = 0; segregs[i] != NULL; i++) {
248 struct v3_segment * seg = segregs[i];
250 seg->selector = 0x0000;
251 // seg->base = seg->selector << 4;
252 seg->base = 0x00000000;
255 // (raw attributes = 0xf3)
262 core->segments.gdtr.selector = 0x0000;
263 core->segments.gdtr.limit = 0x0000ffff;
264 core->segments.gdtr.base = 0x0000000000000000LL;
265 core->segments.gdtr.dpl = 0x0;
267 core->segments.idtr.selector = 0x0000;
268 core->segments.idtr.limit = 0x0000ffff;
269 core->segments.idtr.base = 0x0000000000000000LL;
270 core->segments.ldtr.limit = 0x0000ffff;
271 core->segments.ldtr.base = 0x0000000000000000LL;
272 core->segments.ldtr.system = 0;
273 core->segments.ldtr.type = 0x2;
274 core->segments.ldtr.dpl = 0x0;
276 core->segments.tr.selector = 0x0000;
277 core->segments.tr.limit = 0x0000ffff;
278 core->segments.tr.base = 0x0000000000000000LL;
279 core->segments.tr.system = 0;
280 core->segments.tr.type = 0x3;
281 core->segments.tr.dpl = 0x0;
283 core->dbg_regs.dr6 = 0x00000000ffff0ff0LL;
284 core->dbg_regs.dr7 = 0x0000000000000400LL;
287 ctrl_area->IOPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->io_map.arch_data);
288 ctrl_area->instrs.IOIO_PROT = 1;
290 ctrl_area->MSRPM_BASE_PA = (addr_t)V3_PAddr(core->vm_info->msr_map.arch_data);
291 ctrl_area->instrs.MSR_PROT = 1;
294 ctrl_area->guest_ctrl.V_INTR_MASKING = 1;
295 ctrl_area->instrs.INTR = 1;
296 // The above also assures the TPR changes (CR8) are only virtual
299 // However, we need to see TPR writes since they will
300 // affect the virtual apic
301 // we reflect out cr8 to ctrl_regs->apic_tpr
302 ctrl_area->cr_reads.cr8 = 1;
303 ctrl_area->cr_writes.cr8 = 1;
304 // We will do all TPR comparisons in the virtual apic
305 // We also do not want the V_TPR to be able to mask the PIC
306 ctrl_area->guest_ctrl.V_IGN_TPR = 1;
310 if (core->core_run_state == CORE_INVALID) {
311 v3_hook_msr(core->vm_info, EFER_MSR,
312 &v3_handle_efer_read,
313 &v3_svm_handle_efer_write,
317 if (core->shdw_pg_mode == SHADOW_PAGING) {
319 /* JRL: This is a performance killer, and a simplistic solution */
320 /* We need to fix this */
321 ctrl_area->TLB_CONTROL = 1;
322 ctrl_area->guest_ASID = 1;
325 if (core->core_run_state == CORE_INVALID) {
326 if (v3_init_passthrough_pts(core) == -1) {
327 PrintError(core->vm_info, core, "Could not initialize passthrough page tables\n");
330 // the shadow page tables are OK since we have not initialized hem yet
333 // invalidation of shadow page tables happened earlier in this function
336 core->shdw_pg_state.guest_cr0 = 0x0000000000000010LL;
338 core->ctrl_regs.cr0 |= 0x80000000;
340 v3_activate_passthrough_pt(core);
342 ctrl_area->cr_reads.cr0 = 1;
343 ctrl_area->cr_writes.cr0 = 1;
344 //intercept cr4 read so shadow pager can use PAE independently of guest
345 ctrl_area->cr_reads.cr4 = 1;
346 ctrl_area->cr_writes.cr4 = 1;
347 ctrl_area->cr_reads.cr3 = 1;
348 ctrl_area->cr_writes.cr3 = 1;
351 ctrl_area->instrs.INVLPG = 1;
353 ctrl_area->exceptions.pf = 1;
355 guest_state->g_pat = 0x7040600070406ULL;
358 } else if (core->shdw_pg_mode == NESTED_PAGING) {
359 // Flush the TLB on entries/exits
360 ctrl_area->TLB_CONTROL = 1;
361 ctrl_area->guest_ASID = 1;
363 // Enable Nested Paging
364 ctrl_area->NP_ENABLE = 1;
366 PrintDebug(core->vm_info, core, "NP_Enable at 0x%p\n", (void *)&(ctrl_area->NP_ENABLE));
368 // Set the Nested Page Table pointer
369 if (core->core_run_state == CORE_INVALID) {
370 if (v3_init_passthrough_pts(core) == -1) {
371 PrintError(core->vm_info, core, "Could not initialize Nested page tables\n");
375 // the existing nested page tables will work fine
378 ctrl_area->N_CR3 = core->direct_map_pt;
380 guest_state->g_pat = 0x7040600070406ULL;
383 /* tell the guest that we don't support SVM */
384 if (core->core_run_state == CORE_INVALID) {
385 v3_hook_msr(core->vm_info, SVM_VM_CR_MSR,
386 &v3_handle_vm_cr_read,
387 &v3_handle_vm_cr_write,
391 if (core->core_run_state == CORE_INVALID) {
392 #define INT_PENDING_AMD_MSR 0xc0010055
394 v3_hook_msr(core->vm_info, IA32_STAR_MSR, NULL, NULL, NULL);
395 v3_hook_msr(core->vm_info, IA32_LSTAR_MSR, NULL, NULL, NULL);
396 v3_hook_msr(core->vm_info, IA32_FMASK_MSR, NULL, NULL, NULL);
397 v3_hook_msr(core->vm_info, IA32_KERN_GS_BASE_MSR, NULL, NULL, NULL);
398 v3_hook_msr(core->vm_info, IA32_CSTAR_MSR, NULL, NULL, NULL);
400 v3_hook_msr(core->vm_info, SYSENTER_CS_MSR, NULL, NULL, NULL);
401 v3_hook_msr(core->vm_info, SYSENTER_ESP_MSR, NULL, NULL, NULL);
402 v3_hook_msr(core->vm_info, SYSENTER_EIP_MSR, NULL, NULL, NULL);
405 v3_hook_msr(core->vm_info, FS_BASE_MSR, NULL, NULL, NULL);
406 v3_hook_msr(core->vm_info, GS_BASE_MSR, NULL, NULL, NULL);
408 // Passthrough read operations are ok.
409 v3_hook_msr(core->vm_info, INT_PENDING_AMD_MSR, NULL, v3_msr_unhandled_write, NULL);
416 int v3_init_svm_vmcb(struct guest_info * core, v3_vm_class_t vm_class) {
418 PrintDebug(core->vm_info, core, "Allocating VMCB\n");
419 core->vmm_data = (void *)Allocate_VMCB();
421 if (core->vmm_data == NULL) {
422 PrintError(core->vm_info, core, "Could not allocate VMCB, Exiting...\n");
426 if (vm_class == V3_PC_VM) {
427 PrintDebug(core->vm_info, core, "Initializing VMCB (addr=%p)\n", (void *)core->vmm_data);
428 Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
430 PrintError(core->vm_info, core, "Invalid VM class\n");
434 core->core_run_state = CORE_STOPPED;
440 int v3_deinit_svm_vmcb(struct guest_info * core) {
441 if (core->vmm_data) {
442 V3_FreePages(V3_PAddr(core->vmm_data), 1);
448 static int svm_handle_standard_reset(struct guest_info *core)
450 if (core->core_run_state != CORE_RESETTING) {
454 PrintDebug(core->vm_info,core,"Handling standard reset (guest state before follows)\n");
456 #ifdef V3_CONFIG_DEBUG_SVM
457 v3_print_guest_state(core);
460 // wait until all resetting cores get here (ROS or whole VM)
461 v3_counting_barrier(&core->vm_info->reset_barrier);
463 // I could be a ROS core, or I could be in a non-HVM
464 // either way, if I'm core 0, I'm the leader
465 if (core->vcpu_id==0) {
466 core->vm_info->run_state = VM_RESETTING;
467 // copy bioses again because some,
468 // like seabios, assume
469 // this should also blow away the BDA and EBDA
470 PrintDebug(core->vm_info,core,"Clear memory (%p bytes)\n",(void*)core->vm_info->mem_size);
471 if (v3_set_gpa_memory(core, 0, core->vm_info->mem_size, 0)!=core->vm_info->mem_size) {
472 PrintError(core->vm_info,core,"Clear of memory failed\n");
474 PrintDebug(core->vm_info,core,"Copying bioses\n");
475 if (v3_setup_bioses(core->vm_info, core->vm_info->cfg_data->cfg)) {
476 PrintError(core->vm_info,core,"Setup of bioses failed\n");
480 Init_VMCB_BIOS((vmcb_t*)(core->vmm_data), core);
482 PrintDebug(core->vm_info,core,"InitVMCB done\n");
485 core->cpu_mode = REAL;
486 core->mem_mode = PHYSICAL_MEM;
489 PrintDebug(core->vm_info,core,"Machine reset to REAL/PHYSICAL\n");
491 memset(V3_VAddr((void*)(host_vmcbs[V3_Get_CPU()])),0,4096*4); // good measure...
493 // core zero will be restarted by the main execution loop
494 core->core_run_state = CORE_STOPPED;
496 if (core->vcpu_id==0) {
497 core->vm_info->run_state = VM_RUNNING;
500 #ifdef V3_CONFIG_DEBUG_SVM
501 PrintDebug(core->vm_info,core,"VMCB state at end of reset\n");
502 PrintDebugVMCB((vmcb_t*)(core->vmm_data));
503 PrintDebug(core->vm_info,core,"Guest state at end of reset\n");
504 v3_print_guest_state(core);
507 // wait until we are all ready to go
508 v3_counting_barrier(&core->vm_info->reset_barrier);
510 PrintDebug(core->vm_info,core,"Returning with request for recycle loop\n");
512 return 1; // reboot is occuring
516 #ifdef V3_CONFIG_CHECKPOINT
517 int v3_svm_save_core(struct guest_info * core, void * ctx){
519 vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data);
521 // Special case saves of data we need immediate access to
523 V3_CHKPT_SAVE(ctx, "CPL", core->cpl, failout);
524 V3_CHKPT_SAVE(ctx,"STAR", guest_area->star, failout);
525 V3_CHKPT_SAVE(ctx,"CSTAR", guest_area->cstar, failout);
526 V3_CHKPT_SAVE(ctx,"LSTAR", guest_area->lstar, failout);
527 V3_CHKPT_SAVE(ctx,"SFMASK", guest_area->sfmask, failout);
528 V3_CHKPT_SAVE(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout);
529 V3_CHKPT_SAVE(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout);
530 V3_CHKPT_SAVE(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout);
531 V3_CHKPT_SAVE(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout);
533 // and then we save the whole enchilada
534 if (v3_chkpt_save(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) {
535 PrintError(core->vm_info, core, "Could not save SVM vmcb\n");
542 PrintError(core->vm_info, core, "Failed to save SVM state for core\n");
547 int v3_svm_load_core(struct guest_info * core, void * ctx){
550 vmcb_saved_state_t * guest_area = GET_VMCB_SAVE_STATE_AREA(core->vmm_data);
552 // Reload what we special cased, which we will overwrite in a minute
553 V3_CHKPT_LOAD(ctx, "CPL", core->cpl, failout);
554 V3_CHKPT_LOAD(ctx,"STAR", guest_area->star, failout);
555 V3_CHKPT_LOAD(ctx,"CSTAR", guest_area->cstar, failout);
556 V3_CHKPT_LOAD(ctx,"LSTAR", guest_area->lstar, failout);
557 V3_CHKPT_LOAD(ctx,"SFMASK", guest_area->sfmask, failout);
558 V3_CHKPT_LOAD(ctx,"KERNELGSBASE", guest_area->KernelGsBase, failout);
559 V3_CHKPT_LOAD(ctx,"SYSENTER_CS", guest_area->sysenter_cs, failout);
560 V3_CHKPT_LOAD(ctx,"SYSENTER_ESP", guest_area->sysenter_esp, failout);
561 V3_CHKPT_LOAD(ctx,"SYSENTER_EIP", guest_area->sysenter_eip, failout);
563 // and then we load the whole enchilada
564 if (v3_chkpt_load(ctx, "VMCB_DATA", PAGE_SIZE, core->vmm_data)) {
565 PrintError(core->vm_info, core, "Could not load SVM vmcb\n");
572 PrintError(core->vm_info, core, "Failed to save SVM state for core\n");
578 static int update_irq_exit_state(struct guest_info * info) {
579 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
581 // Fix for QEMU bug using EVENTINJ as an internal cache
582 guest_ctrl->EVENTINJ.valid = 0;
584 if ((info->intr_core_state.irq_pending == 1) && (guest_ctrl->guest_ctrl.V_IRQ == 0)) {
586 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
587 PrintDebug(info->vm_info, info, "INTAK cycle completed for irq %d\n", info->intr_core_state.irq_vector);
590 info->intr_core_state.irq_started = 1;
591 info->intr_core_state.irq_pending = 0;
593 v3_injecting_intr(info, info->intr_core_state.irq_vector, V3_EXTERNAL_IRQ);
596 if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 0)) {
597 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
598 PrintDebug(info->vm_info, info, "Interrupt %d taken by guest\n", info->intr_core_state.irq_vector);
601 // Interrupt was taken fully vectored
602 info->intr_core_state.irq_started = 0;
604 } else if ((info->intr_core_state.irq_started == 1) && (guest_ctrl->exit_int_info.valid == 1)) {
605 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
606 PrintDebug(info->vm_info, info, "EXIT INT INFO is set (vec=%d)\n", guest_ctrl->exit_int_info.vector);
614 static int update_irq_entry_state(struct guest_info * info) {
615 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
618 if (info->intr_core_state.irq_pending == 0) {
619 guest_ctrl->guest_ctrl.V_IRQ = 0;
620 guest_ctrl->guest_ctrl.V_INTR_VECTOR = 0;
623 if (v3_excp_pending(info)) {
624 uint_t excp = v3_get_excp_number(info);
626 guest_ctrl->EVENTINJ.type = SVM_INJECTION_EXCEPTION;
628 if (info->excp_state.excp_error_code_valid) {
629 guest_ctrl->EVENTINJ.error_code = info->excp_state.excp_error_code;
630 guest_ctrl->EVENTINJ.ev = 1;
631 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
632 PrintDebug(info->vm_info, info, "Injecting exception %d with error code %x\n", excp, guest_ctrl->EVENTINJ.error_code);
636 guest_ctrl->EVENTINJ.vector = excp;
638 guest_ctrl->EVENTINJ.valid = 1;
640 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
641 PrintDebug(info->vm_info, info, "<%d> Injecting Exception %d (CR2=%p) (EIP=%p)\n",
642 (int)info->num_exits,
643 guest_ctrl->EVENTINJ.vector,
644 (void *)(addr_t)info->ctrl_regs.cr2,
645 (void *)(addr_t)info->rip);
648 v3_injecting_excp(info, excp);
649 } else if (info->intr_core_state.irq_started == 1) {
650 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
651 PrintDebug(info->vm_info, info, "IRQ pending from previous injection\n");
653 guest_ctrl->guest_ctrl.V_IRQ = 1;
654 guest_ctrl->guest_ctrl.V_INTR_VECTOR = info->intr_core_state.irq_vector;
656 // We ignore the virtual TPR on this injection
657 // TPR/PPR tests have already been done in the APIC.
658 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
659 guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf;
662 switch (v3_intr_pending(info)) {
663 case V3_EXTERNAL_IRQ: {
664 int irq = v3_get_intr(info);
670 guest_ctrl->guest_ctrl.V_IRQ = 1;
671 guest_ctrl->guest_ctrl.V_INTR_VECTOR = irq;
673 // We ignore the virtual TPR on this injection
674 // TPR/PPR tests have already been done in the APIC.
675 guest_ctrl->guest_ctrl.V_IGN_TPR = 1;
676 guest_ctrl->guest_ctrl.V_INTR_PRIO = info->intr_core_state.irq_vector >> 4 ; // 0xf;
678 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
679 PrintDebug(info->vm_info, info, "Injecting Interrupt %d (EIP=%p)\n",
680 guest_ctrl->guest_ctrl.V_INTR_VECTOR,
681 (void *)(addr_t)info->rip);
684 info->intr_core_state.irq_pending = 1;
685 info->intr_core_state.irq_vector = irq;
691 guest_ctrl->EVENTINJ.type = SVM_INJECTION_NMI;
693 case V3_SOFTWARE_INTR:
694 guest_ctrl->EVENTINJ.type = SVM_INJECTION_SOFT_INTR;
695 #ifdef V3_CONFIG_DEBUG_INTERRUPTS
696 PrintDebug(info->vm_info, info, "Injecting software interrupt -- type: %d, vector: %d\n",
697 SVM_INJECTION_SOFT_INTR, info->intr_core_state.swintr_vector);
699 guest_ctrl->EVENTINJ.vector = info->intr_core_state.swintr_vector;
700 guest_ctrl->EVENTINJ.valid = 1;
702 /* reset swintr state */
703 info->intr_core_state.swintr_posted = 0;
704 info->intr_core_state.swintr_vector = 0;
707 guest_ctrl->EVENTINJ.type = SVM_INJECTION_IRQ;
710 case V3_INVALID_INTR:
721 v3_svm_config_tsc_virtualization(struct guest_info * info) {
722 vmcb_ctrl_t * ctrl_area = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
725 if (info->time_state.flags & VM_TIME_TRAP_RDTSC) {
726 ctrl_area->instrs.RDTSC = 1;
727 ctrl_area->svm_instrs.RDTSCP = 1;
729 ctrl_area->instrs.RDTSC = 0;
730 ctrl_area->svm_instrs.RDTSCP = 0;
732 if (info->time_state.flags & VM_TIME_TSC_PASSTHROUGH) {
733 ctrl_area->TSC_OFFSET = 0;
735 ctrl_area->TSC_OFFSET = v3_tsc_host_offset(&info->time_state);
744 * CAUTION and DANGER!!!
746 * The VMCB CANNOT(!!) be accessed outside of the clgi/stgi calls inside this function
747 * When exectuing a symbiotic call, the VMCB WILL be overwritten, so any dependencies
748 * on its contents will cause things to break. The contents at the time of the exit WILL
749 * change before the exit handler is executed.
751 int v3_svm_enter(struct guest_info * info) {
752 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
753 vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
754 addr_t exit_code = 0, exit_info1 = 0, exit_info2 = 0;
755 uint64_t guest_cycles = 0;
758 // Conditionally yield the CPU if the timeslice has expired
761 #ifdef V3_CONFIG_MEM_TRACK
762 v3_mem_track_entry(info);
765 // Update timer devices after being in the VM before doing
766 // IRQ updates, so that any interrupts they raise get seen
769 v3_advance_time(info, NULL);
771 v3_update_timers(info);
774 // disable global interrupts for vm state transition
777 // Synchronize the guest state to the VMCB
778 guest_state->cr0 = info->ctrl_regs.cr0;
779 guest_state->cr2 = info->ctrl_regs.cr2;
780 guest_state->cr3 = info->ctrl_regs.cr3;
781 guest_state->cr4 = info->ctrl_regs.cr4;
782 guest_state->dr6 = info->dbg_regs.dr6;
783 guest_state->dr7 = info->dbg_regs.dr7;
785 // CR8 is now updated by read/writes and it contains the APIC TPR
786 // the V_TPR should be just the class part of that.
787 // This update is here just for completeness. We currently
788 // are ignoring V_TPR on all injections and doing the priority logivc
790 // guest_ctrl->guest_ctrl.V_TPR = ((info->ctrl_regs.apic_tpr) >> 4) & 0xf;
792 //guest_ctrl->guest_ctrl.V_TPR = info->ctrl_regs.cr8 & 0xff;
795 guest_state->rflags = info->ctrl_regs.rflags;
799 guest_state->efer = info->ctrl_regs.efer;
801 /* Synchronize MSRs */
802 guest_state->star = info->msrs.star;
803 guest_state->lstar = info->msrs.lstar;
804 guest_state->sfmask = info->msrs.sfmask;
805 guest_state->KernelGsBase = info->msrs.kern_gs_base;
807 guest_state->cpl = info->cpl;
809 v3_set_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
811 guest_state->rax = info->vm_regs.rax;
812 guest_state->rip = info->rip;
813 guest_state->rsp = info->vm_regs.rsp;
815 V3_FP_ENTRY_RESTORE(info);
817 #ifdef V3_CONFIG_SYMCALL
818 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
819 update_irq_entry_state(info);
823 update_irq_entry_state(info);
826 #ifdef V3_CONFIG_TM_FUNC
827 v3_tm_check_intr_state(info, guest_ctrl, guest_state);
834 PrintDebug(info->vm_info, info, "SVM Entry to CS=%p rip=%p...\n",
835 (void *)(addr_t)info->segments.cs.base,
836 (void *)(addr_t)info->rip);
839 #ifdef V3_CONFIG_SYMCALL
840 if (info->sym_core_state.symcall_state.sym_call_active == 1) {
841 if (guest_ctrl->guest_ctrl.V_IRQ == 1) {
842 V3_Print(info->vm_info, info, "!!! Injecting Interrupt during Sym call !!!\n");
847 v3_svm_config_tsc_virtualization(info);
849 //V3_Print(info->vm_info, info, "Calling v3_svm_launch\n");
851 uint64_t entry_tsc = 0;
852 uint64_t exit_tsc = 0;
854 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
855 v3_pwrstat_telemetry_enter(info);
858 #ifdef V3_CONFIG_PMU_TELEMETRY
859 v3_pmu_telemetry_enter(info);
865 v3_svm_launch((vmcb_t *)V3_PAddr(info->vmm_data), &(info->vm_regs), (vmcb_t *)host_vmcbs[V3_Get_CPU()]);
869 #ifdef V3_CONFIG_PMU_TELEMETRY
870 v3_pmu_telemetry_exit(info);
873 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
874 v3_pwrstat_telemetry_exit(info);
877 guest_cycles = exit_tsc - entry_tsc;
881 //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);
883 v3_last_exit = (uint32_t)(guest_ctrl->exit_code);
885 v3_advance_time(info, &guest_cycles);
889 V3_FP_EXIT_SAVE(info);
891 // Save Guest state from VMCB
892 info->rip = guest_state->rip;
893 info->vm_regs.rsp = guest_state->rsp;
894 info->vm_regs.rax = guest_state->rax;
896 info->cpl = guest_state->cpl;
898 info->ctrl_regs.cr0 = guest_state->cr0;
899 info->ctrl_regs.cr2 = guest_state->cr2;
900 info->ctrl_regs.cr3 = guest_state->cr3;
901 info->ctrl_regs.cr4 = guest_state->cr4;
902 info->dbg_regs.dr6 = guest_state->dr6;
903 info->dbg_regs.dr7 = guest_state->dr7;
905 // We do not track this anymore
906 // V_TPR is ignored and we do the logic in the APIC
907 //info->ctrl_regs.cr8 = guest_ctrl->guest_ctrl.V_TPR;
909 info->ctrl_regs.rflags = guest_state->rflags;
910 info->ctrl_regs.efer = guest_state->efer;
912 /* Synchronize MSRs */
913 info->msrs.star = guest_state->star;
914 info->msrs.lstar = guest_state->lstar;
915 info->msrs.sfmask = guest_state->sfmask;
916 info->msrs.kern_gs_base = guest_state->KernelGsBase;
918 v3_get_vmcb_segments((vmcb_t*)(info->vmm_data), &(info->segments));
919 info->cpu_mode = v3_get_vm_cpu_mode(info);
920 info->mem_mode = v3_get_vm_mem_mode(info);
923 // save exit info here
924 exit_code = guest_ctrl->exit_code;
925 exit_info1 = guest_ctrl->exit_info1;
926 exit_info2 = guest_ctrl->exit_info2;
928 #ifdef V3_CONFIG_SYMCALL
929 if (info->sym_core_state.symcall_state.sym_call_active == 0) {
930 update_irq_exit_state(info);
933 update_irq_exit_state(info);
936 // reenable global interrupts after vm exit
939 // Conditionally yield the CPU if the timeslice has expired
942 // This update timers is for time-dependent handlers
943 // if we're slaved to host time
944 v3_advance_time(info, NULL);
945 v3_update_timers(info);
949 int ret = v3_handle_svm_exit(info, exit_code, exit_info1, exit_info2);
952 PrintError(info->vm_info, info, "Error in SVM exit handler (ret=%d)\n", ret);
953 PrintError(info->vm_info, info, " last Exit was %d (exit code=0x%llx)\n", v3_last_exit, (uint64_t) exit_code);
960 if (info->timeouts.timeout_active) {
961 /* Check to see if any timeouts have expired */
962 v3_handle_timeouts(info, guest_cycles);
965 #ifdef V3_CONFIG_MEM_TRACK
966 v3_mem_track_exit(info);
973 int v3_start_svm_guest(struct guest_info * info) {
977 // vmcb_saved_state_t * guest_state = GET_VMCB_SAVE_STATE_AREA((vmcb_t*)(info->vmm_data));
978 // vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
980 PrintDebug(info->vm_info, info, "Starting SVM core %u (on logical core %u)\n", info->vcpu_id, info->pcpu_id);
983 #ifdef V3_CONFIG_MULTIBOOT
984 if (v3_setup_multiboot_core_for_boot(info)) {
985 PrintError(info->vm_info, info, "Failed to setup Multiboot core...\n");
991 if (v3_setup_hvm_hrt_core_for_boot(info)) {
992 PrintError(info->vm_info, info, "Failed to setup HRT core...\n");
999 if (info->core_run_state == CORE_STOPPED) {
1001 if (info->vcpu_id == 0) {
1002 info->core_run_state = CORE_RUNNING;
1004 PrintDebug(info->vm_info, info, "SVM core %u (on %u): Waiting for core initialization\n", info->vcpu_id, info->pcpu_id);
1008 // Compiler must not optimize away this read
1009 while (*((volatile int *)(&info->core_run_state)) == CORE_STOPPED) {
1011 if (info->vm_info->run_state == VM_STOPPED) {
1012 // The VM was stopped before this core was initialized.
1016 V3_STILL_NO_WORK(info);
1018 //PrintDebug(info->vm_info, info, "SVM core %u: still waiting for INIT\n", info->vcpu_id);
1021 V3_HAVE_WORK_AGAIN(info);
1023 PrintDebug(info->vm_info, info, "SVM core %u(on %u) initialized\n", info->vcpu_id, info->pcpu_id);
1025 // We'll be paranoid about race conditions here
1026 v3_wait_at_barrier(info);
1034 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",
1035 info->vcpu_id, info->pcpu_id,
1036 info->segments.cs.selector, (void *)(info->segments.cs.base),
1037 info->segments.cs.limit, (void *)(info->rip));
1041 PrintDebug(info->vm_info, info, "SVM core %u: Launching SVM VM (vmcb=%p) (on cpu %u)\n",
1042 info->vcpu_id, (void *)info->vmm_data, info->pcpu_id);
1044 #ifdef V3_CONFIG_DEBUG_SVM
1045 PrintDebugVMCB((vmcb_t*)(info->vmm_data));
1048 v3_start_time(info);
1051 if (info->vm_info->run_state == VM_STOPPED) {
1052 info->core_run_state = CORE_STOPPED;
1057 #ifdef V3_CONFIG_HVM
1058 if (v3_handle_hvm_reset(info) > 0) {
1063 #ifdef V3_CONFIG_MULTIBOOT
1064 if (v3_handle_multiboot_reset(info) > 0) {
1069 if (svm_handle_standard_reset(info) > 0 ) {
1075 #ifdef V3_CONFIG_PMU_TELEMETRY
1076 v3_pmu_telemetry_start(info);
1079 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
1080 v3_pwrstat_telemetry_start(info);
1083 if (v3_svm_enter(info) == -1 ) {
1084 vmcb_ctrl_t * guest_ctrl = GET_VMCB_CTRL_AREA((vmcb_t*)(info->vmm_data));
1086 addr_t linear_addr = 0;
1088 info->vm_info->run_state = VM_ERROR;
1090 V3_Print(info->vm_info, info, "SVM core %u: SVM ERROR!!\n", info->vcpu_id);
1092 v3_print_guest_state(info);
1094 V3_Print(info->vm_info, info, "SVM core %u: SVM Exit Code: %p\n", info->vcpu_id, (void *)(addr_t)guest_ctrl->exit_code);
1096 V3_Print(info->vm_info, info, "SVM core %u: exit_info1 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info1));
1097 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));
1099 V3_Print(info->vm_info, info, "SVM core %u: exit_info2 low = 0x%.8x\n", info->vcpu_id, *(uint_t*)&(guest_ctrl->exit_info2));
1100 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));
1102 linear_addr = get_addr_linear(info, info->rip, &(info->segments.cs));
1104 if (info->mem_mode == PHYSICAL_MEM) {
1105 v3_gpa_to_hva(info, linear_addr, &host_addr);
1106 } else if (info->mem_mode == VIRTUAL_MEM) {
1107 v3_gva_to_hva(info, linear_addr, &host_addr);
1110 V3_Print(info->vm_info, info, "SVM core %u: Host Address of rip = 0x%p\n", info->vcpu_id, (void *)host_addr);
1112 V3_Print(info->vm_info, info, "SVM core %u: Instr (15 bytes) at %p:\n", info->vcpu_id, (void *)host_addr);
1113 v3_dump_mem((uint8_t *)host_addr, 15);
1115 v3_print_stack(info);
1120 v3_wait_at_barrier(info);
1123 if (info->vm_info->run_state == VM_STOPPED) {
1124 info->core_run_state = CORE_STOPPED;
1131 if ((info->num_exits % 50000) == 0) {
1132 V3_Print(info->vm_info, info, "SVM Exit number %d\n", (uint32_t)info->num_exits);
1133 v3_print_guest_state(info);
1139 #ifdef V3_CONFIG_PMU_TELEMETRY
1140 v3_pmu_telemetry_end(info);
1143 #ifdef V3_CONFIG_PWRSTAT_TELEMETRY
1144 v3_pwrstat_telemetry_end(info);
1146 // Need to take down the other cores on error...
1154 int v3_reset_svm_vm_core(struct guest_info * core, addr_t rip) {
1157 // Write the RIP, CS, and descriptor
1158 // assume the rest is already good to go
1160 // vector VV -> rip at 0
1162 // This means we start executing at linear address VV000
1164 // So the selector needs to be VV00
1165 // and the base needs to be VV000
1168 core->segments.cs.selector = rip << 8;
1169 core->segments.cs.limit = 0xffff;
1170 core->segments.cs.base = rip << 12;
1180 /* Checks machine SVM capability */
1181 /* Implemented from: AMD Arch Manual 3, sect 15.4 */
1182 int v3_is_svm_capable() {
1183 uint_t vm_cr_low = 0, vm_cr_high = 0;
1184 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1186 v3_cpuid(CPUID_EXT_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1188 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_ecx=0x%x\n", ecx);
1190 if ((ecx & CPUID_EXT_FEATURE_IDS_ecx_svm_avail) == 0) {
1191 V3_Print(VM_NONE, VCORE_NONE, "SVM Not Available\n");
1194 v3_get_msr(SVM_VM_CR_MSR, &vm_cr_high, &vm_cr_low);
1196 PrintDebug(VM_NONE, VCORE_NONE, "SVM_VM_CR_MSR = 0x%x 0x%x\n", vm_cr_high, vm_cr_low);
1198 if ((vm_cr_low & SVM_VM_CR_MSR_svmdis) == 1) {
1199 V3_Print(VM_NONE, VCORE_NONE, "SVM is available but is disabled.\n");
1201 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1203 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
1205 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_svml) == 0) {
1206 V3_Print(VM_NONE, VCORE_NONE, "SVM BIOS Disabled, not unlockable\n");
1208 V3_Print(VM_NONE, VCORE_NONE, "SVM is locked with a key\n");
1213 V3_Print(VM_NONE, VCORE_NONE, "SVM is available and enabled.\n");
1215 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1216 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_eax=0x%x\n", eax);
1217 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ebx=0x%x\n", ebx);
1218 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_ecx=0x%x\n", ecx);
1219 PrintDebug(VM_NONE, VCORE_NONE, "CPUID_SVM_REV_AND_FEATURE_IDS_edx=0x%x\n", edx);
1226 static int has_svm_nested_paging() {
1227 uint32_t eax = 0, ebx = 0, ecx = 0, edx = 0;
1229 v3_cpuid(CPUID_SVM_REV_AND_FEATURE_IDS, &eax, &ebx, &ecx, &edx);
1231 //PrintDebug(VM_NONE, VCORE_NONE, "CPUID_EXT_FEATURE_IDS_edx=0x%x\n", edx);
1233 if ((edx & CPUID_SVM_REV_AND_FEATURE_IDS_edx_np) == 0) {
1234 V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging not supported\n");
1237 V3_Print(VM_NONE, VCORE_NONE, "SVM Nested Paging supported\n");
1244 void v3_init_svm_cpu(int cpu_id) {
1246 extern v3_cpu_arch_t v3_cpu_types[];
1248 // Enable SVM on the CPU
1249 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
1250 msr.e_reg.low |= EFER_MSR_svm_enable;
1251 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
1253 V3_Print(VM_NONE, VCORE_NONE, "SVM Enabled\n");
1255 // Setup the host state save area
1256 host_vmcbs[cpu_id] = (addr_t)V3_AllocPages(4); // need not be shadow-safe, not exposed to guest
1258 if (!host_vmcbs[cpu_id]) {
1259 PrintError(VM_NONE, VCORE_NONE, "Failed to allocate VMCB\n");
1264 // msr.e_reg.high = 0;
1265 //msr.e_reg.low = (uint_t)host_vmcb;
1266 msr.r_reg = host_vmcbs[cpu_id];
1268 PrintDebug(VM_NONE, VCORE_NONE, "Host State being saved at %p\n", (void *)host_vmcbs[cpu_id]);
1269 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
1272 if (has_svm_nested_paging() == 1) {
1273 v3_cpu_types[cpu_id] = V3_SVM_REV3_CPU;
1275 v3_cpu_types[cpu_id] = V3_SVM_CPU;
1281 void v3_deinit_svm_cpu(int cpu_id) {
1283 extern v3_cpu_arch_t v3_cpu_types[];
1285 // reset SVM_VM_HSAVE_PA_MSR
1286 // Does setting it to NULL disable??
1288 v3_set_msr(SVM_VM_HSAVE_PA_MSR, msr.e_reg.high, msr.e_reg.low);
1291 v3_get_msr(EFER_MSR, &(msr.e_reg.high), &(msr.e_reg.low));
1292 msr.e_reg.low &= ~EFER_MSR_svm_enable;
1293 v3_set_msr(EFER_MSR, 0, msr.e_reg.low);
1295 v3_cpu_types[cpu_id] = V3_INVALID_CPU;
1297 V3_FreePages((void *)host_vmcbs[cpu_id], 4);
1299 V3_Print(VM_NONE, VCORE_NONE, "Host CPU %d host area freed, and SVM disabled\n", cpu_id);
1354 * Test VMSAVE/VMLOAD Latency
1356 #define vmsave ".byte 0x0F,0x01,0xDB ; "
1357 #define vmload ".byte 0x0F,0x01,0xDA ; "
1359 uint32_t start_lo, start_hi;
1360 uint32_t end_lo, end_hi;
1361 uint64_t start, end;
1363 __asm__ __volatile__ (
1365 "movl %%eax, %%esi ; "
1366 "movl %%edx, %%edi ; "
1367 "movq %%rcx, %%rax ; "
1370 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1371 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1382 PrintDebug(core->vm_info, core, "VMSave Cycle Latency: %d\n", (uint32_t)(end - start));
1384 __asm__ __volatile__ (
1386 "movl %%eax, %%esi ; "
1387 "movl %%edx, %%edi ; "
1388 "movq %%rcx, %%rax ; "
1391 : "=D"(start_hi), "=S"(start_lo), "=a"(end_lo),"=d"(end_hi)
1392 : "c"(host_vmcb[cpu_id]), "0"(0), "1"(0), "2"(0), "3"(0)
1404 PrintDebug(core->vm_info, core, "VMLoad Cycle Latency: %d\n", (uint32_t)(end - start));
1406 /* End Latency Test */