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".
20 #include <palacios/vmm.h>
21 #include <palacios/vmm_intr.h>
22 #include <palacios/vmm_config.h>
23 #include <palacios/vm_guest.h>
24 #include <palacios/vmm_ctrl_regs.h>
25 #include <palacios/vmm_lowlevel.h>
26 #include <palacios/vmm_sprintf.h>
27 #include <palacios/vmm_extensions.h>
30 #include <palacios/svm.h>
33 #include <palacios/vmx.h>
36 #ifdef V3_CONFIG_CHECKPOINT
37 #include <palacios/vmm_checkpoint.h>
41 v3_cpu_arch_t v3_cpu_types[V3_CONFIG_MAX_CPUS];
42 struct v3_os_hooks * os_hooks = NULL;
43 int v3_dbg_enable = 0;
48 static void init_cpu(void * arg) {
49 uint32_t cpu_id = (uint32_t)(addr_t)arg;
52 if (v3_is_svm_capable()) {
53 PrintDebug("Machine is SVM Capable\n");
54 v3_init_svm_cpu(cpu_id);
59 if (v3_is_vmx_capable()) {
60 PrintDebug("Machine is VMX Capable\n");
61 v3_init_vmx_cpu(cpu_id);
66 PrintError("CPU has no virtualization Extensions\n");
71 static void deinit_cpu(void * arg) {
72 uint32_t cpu_id = (uint32_t)(addr_t)arg;
75 switch (v3_cpu_types[cpu_id]) {
79 PrintDebug("Deinitializing SVM CPU %d\n", cpu_id);
80 v3_deinit_svm_cpu(cpu_id);
86 case V3_VMX_EPT_UG_CPU:
87 PrintDebug("Deinitializing VMX CPU %d\n", cpu_id);
88 v3_deinit_vmx_cpu(cpu_id);
93 PrintError("CPU has no virtualization Extensions\n");
100 void Init_V3(struct v3_os_hooks * hooks, int num_cpus) {
103 V3_Print("V3 Print statement to fix a Kitten page fault bug\n");
105 // Set global variables.
108 for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
109 v3_cpu_types[i] = V3_INVALID_CPU;
112 // Register all the possible device types
115 // Register all shadow paging handlers
116 V3_init_shdw_paging();
118 // Register all extensions
119 V3_init_extensions();
122 #ifdef V3_CONFIG_SYMMOD
126 #ifdef V3_CONFIG_CHECKPOINT
127 V3_init_checkpoint();
133 if ((hooks) && (hooks->call_on_cpu)) {
135 for (i = 0; i < num_cpus; i++) {
137 V3_Print("Initializing VMM extensions on cpu %d\n", i);
138 hooks->call_on_cpu(i, &init_cpu, (void *)(addr_t)i);
150 V3_deinit_shdw_paging();
152 V3_deinit_extensions();
154 #ifdef V3_CONFIG_SYMMOD
158 #ifdef V3_CONFIG_CHECKPOINT
159 V3_deinit_checkpoint();
163 if ((os_hooks) && (os_hooks->call_on_cpu)) {
164 for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
165 if (v3_cpu_types[i] != V3_INVALID_CPU) {
166 V3_Call_On_CPU(i, deinit_cpu, (void *)(addr_t)i);
167 //deinit_cpu((void *)(addr_t)i);
175 v3_cpu_arch_t v3_get_cpu_type(int cpu_id) {
176 return v3_cpu_types[cpu_id];
180 struct v3_vm_info * v3_create_vm(void * cfg, void * priv_data, char * name) {
181 struct v3_vm_info * vm = v3_config_guest(cfg, priv_data);
184 PrintError("Could not configure guest\n");
188 V3_Print("CORE 0 RIP=%p\n", (void *)(addr_t)(vm->cores[0].rip));
192 } else if (strlen(name) >= 128) {
193 PrintError("VM name is too long. Will be truncated to 128 chars.\n");
196 memset(vm->name, 0, 128);
197 strncpy(vm->name, name, 127);
205 static int start_core(void * p)
207 struct guest_info * core = (struct guest_info *)p;
210 PrintDebug("virtual core %u (on logical core %u): in start_core (RIP=%p)\n",
211 core->vcpu_id, core->pcpu_id, (void *)(addr_t)core->rip);
213 switch (v3_cpu_types[0]) {
216 case V3_SVM_REV3_CPU:
217 return v3_start_svm_guest(core);
223 case V3_VMX_EPT_UG_CPU:
224 return v3_start_vmx_guest(core);
228 PrintError("Attempting to enter a guest on an invalid CPU\n");
236 // For the moment very ugly. Eventually we will shift the cpu_mask to an arbitrary sized type...
240 int v3_start_vm(struct v3_vm_info * vm, unsigned int cpu_mask) {
242 uint8_t * core_mask = (uint8_t *)&cpu_mask; // This is to make future expansion easier
243 uint32_t avail_cores = 0;
246 /// CHECK IF WE ARE MULTICORE ENABLED....
248 V3_Print("V3 -- Starting VM (%u cores)\n", vm->num_cores);
249 V3_Print("CORE 0 RIP=%p\n", (void *)(addr_t)(vm->cores[0].rip));
252 // Check that enough cores are present in the mask to handle vcores
253 for (i = 0; i < MAX_CORES; i++) {
257 if (core_mask[major] & (0x1 << minor)) {
258 if (v3_cpu_types[i] == V3_INVALID_CPU) {
259 core_mask[major] &= ~(0x1 << minor);
267 if (vm->num_cores > avail_cores) {
268 PrintError("Attempted to start a VM with too many cores (vm->num_cores = %d, avail_cores = %d, MAX=%d)\n",
269 vm->num_cores, avail_cores, MAX_CORES);
273 // Spawn off threads for each core.
274 // We work backwards, so that core 0 is always started last.
275 for (i = 0, vcore_id = vm->num_cores - 1; (i < MAX_CORES) && (vcore_id >= 0); i++) {
278 struct guest_info * core = &(vm->cores[vcore_id]);
279 char * specified_cpu = v3_cfg_val(core->core_cfg_data, "target_cpu");
280 uint32_t core_idx = 0;
282 if (specified_cpu != NULL) {
283 core_idx = atoi(specified_cpu);
285 if ((core_idx < 0) || (core_idx >= MAX_CORES)) {
286 PrintError("Target CPU out of bounds (%d) (MAX_CORES=%d)\n", core_idx, MAX_CORES);
289 i--; // We reset the logical core idx. Not strictly necessary I guess...
294 major = core_idx / 8;
295 minor = core_idx % 8;
297 if ((core_mask[major] & (0x1 << minor)) == 0) {
298 PrintError("Logical CPU %d not available for virtual core %d; not started\n",
301 if (specified_cpu != NULL) {
302 PrintError("CPU was specified explicitly (%d). HARD ERROR\n", core_idx);
310 PrintDebug("Starting virtual core %u on logical core %u\n",
313 sprintf(core->exec_name, "%s-%u", vm->name, vcore_id);
315 PrintDebug("run: core=%u, func=0x%p, arg=0x%p, name=%s\n",
316 core_idx, start_core, core, core->exec_name);
318 core->pcpu_id = core_idx;
319 core->core_thread = V3_CREATE_THREAD_ON_CPU(core_idx, start_core, core, core->exec_name);
321 if (core->core_thread == NULL) {
322 PrintError("Thread launch failed\n");
331 PrintError("Error starting VM: Not enough available CPU cores\n");
342 int v3_reset_vm_core(struct guest_info * core, addr_t rip) {
344 switch (v3_cpu_types[core->pcpu_id]) {
347 case V3_SVM_REV3_CPU:
348 PrintDebug("Resetting SVM Guest CPU %d\n", core->vcpu_id);
349 return v3_reset_svm_vm_core(core, rip);
354 case V3_VMX_EPT_UG_CPU:
355 PrintDebug("Resetting VMX Guest CPU %d\n", core->vcpu_id);
356 return v3_reset_vmx_vm_core(core, rip);
360 PrintError("CPU has no virtualization Extensions\n");
369 /* move a virtual core to different physical core */
370 int v3_move_vm_core(struct v3_vm_info * vm, int vcore_id, int target_cpu) {
371 struct guest_info * core = NULL;
373 if ((vcore_id < 0) || (vcore_id >= vm->num_cores)) {
374 PrintError("Attempted to migrate invalid virtual core (%d)\n", vcore_id);
378 core = &(vm->cores[vcore_id]);
380 if (target_cpu == core->pcpu_id) {
381 PrintError("Attempted to migrate to local core (%d)\n", target_cpu);
382 // well that was pointless
386 if (core->core_thread == NULL) {
387 PrintError("Attempted to migrate a core without a valid thread context\n");
391 while (v3_raise_barrier(vm, NULL) == -1);
393 V3_Print("Performing Migration from %d to %d\n", core->pcpu_id, target_cpu);
395 // Double check that we weren't preemptively migrated
396 if (target_cpu != core->pcpu_id) {
398 V3_Print("Moving Core\n");
402 switch (v3_cpu_types[core->pcpu_id]) {
405 case V3_VMX_EPT_UG_CPU:
406 PrintDebug("Flushing VMX Guest CPU %d\n", core->vcpu_id);
407 V3_Call_On_CPU(core->pcpu_id, (void (*)(void *))v3_flush_vmx_vm_core, (void *)core);
414 if (V3_MOVE_THREAD_TO_CPU(target_cpu, core->core_thread) != 0) {
415 PrintError("Failed to move Vcore %d to CPU %d\n",
416 core->vcpu_id, target_cpu);
417 v3_lower_barrier(vm);
421 /* There will be a benign race window here:
422 core->pcpu_id will be set to the target core before its fully "migrated"
423 However the core will NEVER run on the old core again, its just in flight to the new core
425 core->pcpu_id = target_cpu;
427 V3_Print("core now at %d\n", core->pcpu_id);
430 v3_lower_barrier(vm);
437 int v3_stop_vm(struct v3_vm_info * vm) {
439 vm->run_state = VM_STOPPED;
441 // force exit all cores via a cross call/IPI
445 int still_running = 0;
447 for (i = 0; i < vm->num_cores; i++) {
448 if (vm->cores[i].core_run_state != CORE_STOPPED) {
453 if (still_running == 0) {
460 V3_Print("VM stopped. Returning\n");
466 int v3_pause_vm(struct v3_vm_info * vm) {
468 if (vm->run_state != VM_RUNNING) {
469 PrintError("Tried to pause a VM that was not running\n");
473 while (v3_raise_barrier(vm, NULL) == -1);
475 vm->run_state = VM_PAUSED;
481 int v3_continue_vm(struct v3_vm_info * vm) {
483 if (vm->run_state != VM_PAUSED) {
484 PrintError("Tried to continue a VM that was not paused\n");
488 v3_lower_barrier(vm);
490 vm->run_state = VM_RUNNING;
495 #ifdef V3_CONFIG_CHECKPOINT
496 #include <palacios/vmm_checkpoint.h>
498 int v3_save_vm(struct v3_vm_info * vm, char * store, char * url) {
499 return v3_chkpt_save_vm(vm, store, url);
503 int v3_load_vm(struct v3_vm_info * vm, char * store, char * url) {
504 return v3_chkpt_load_vm(vm, store, url);
509 int v3_free_vm(struct v3_vm_info * vm) {
511 // deinitialize guest (free memory, etc...)
513 v3_free_vm_devices(vm);
516 for (i = 0; i < vm->num_cores; i++) {
517 v3_free_core(&(vm->cores[i]));
521 v3_free_vm_internal(vm);
533 v3_cpu_mode_t v3_get_host_cpu_mode() {
543 cr4 = (struct cr4_32 *)&(cr4_val);
546 return PROTECTED_PAE;
554 v3_cpu_mode_t v3_get_host_cpu_mode() {
561 #define V3_Yield(addr) \
563 extern struct v3_os_hooks * os_hooks; \
564 if ((os_hooks) && (os_hooks)->yield_cpu) { \
565 (os_hooks)->yield_cpu(); \
571 void v3_yield_cond(struct guest_info * info) {
573 cur_cycle = v3_get_host_time(&info->time_state);
575 if (cur_cycle > (info->yield_start_cycle + info->vm_info->yield_cycle_period)) {
576 //PrintDebug("Conditional Yield (cur_cyle=%p, start_cycle=%p, period=%p)\n",
577 // (void *)cur_cycle, (void *)info->yield_start_cycle,
578 // (void *)info->yield_cycle_period);
581 info->yield_start_cycle = v3_get_host_time(&info->time_state);
587 * unconditional cpu yield
588 * if the yielding thread is a guest context, the guest quantum is reset on resumption
589 * Non guest context threads should call this function with a NULL argument
591 void v3_yield(struct guest_info * info) {
595 info->yield_start_cycle = v3_get_host_time(&info->time_state);
602 void v3_print_cond(const char * fmt, ...) {
603 if (v3_dbg_enable == 1) {
608 vsnprintf(buf, 2048, fmt, ap);
617 void v3_interrupt_cpu(struct v3_vm_info * vm, int logical_cpu, int vector) {
618 extern struct v3_os_hooks * os_hooks;
620 if ((os_hooks) && (os_hooks)->interrupt_cpu) {
621 (os_hooks)->interrupt_cpu(vm, logical_cpu, vector);
627 int v3_vm_enter(struct guest_info * info) {
628 switch (v3_cpu_types[0]) {
631 case V3_SVM_REV3_CPU:
632 return v3_svm_enter(info);
638 case V3_VMX_EPT_UG_CPU:
639 return v3_vmx_enter(info);
643 PrintError("Attemping to enter a guest on an invalid CPU\n");