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>
28 #include <palacios/vmm_timeout.h>
32 #include <palacios/svm.h>
35 #include <palacios/vmx.h>
38 #ifdef V3_CONFIG_CHECKPOINT
39 #include <palacios/vmm_checkpoint.h>
43 v3_cpu_arch_t v3_cpu_types[V3_CONFIG_MAX_CPUS];
44 v3_cpu_arch_t v3_mach_type = V3_INVALID_CPU;
46 struct v3_os_hooks * os_hooks = NULL;
47 int v3_dbg_enable = 0;
52 static void init_cpu(void * arg) {
53 uint32_t cpu_id = (uint32_t)(addr_t)arg;
56 if (v3_is_svm_capable()) {
57 PrintDebug("Machine is SVM Capable\n");
58 v3_init_svm_cpu(cpu_id);
63 if (v3_is_vmx_capable()) {
64 PrintDebug("Machine is VMX Capable\n");
65 v3_init_vmx_cpu(cpu_id);
70 PrintError("CPU has no virtualization Extensions\n");
75 static void deinit_cpu(void * arg) {
76 uint32_t cpu_id = (uint32_t)(addr_t)arg;
79 switch (v3_cpu_types[cpu_id]) {
83 PrintDebug("Deinitializing SVM CPU %d\n", cpu_id);
84 v3_deinit_svm_cpu(cpu_id);
90 case V3_VMX_EPT_UG_CPU:
91 PrintDebug("Deinitializing VMX CPU %d\n", cpu_id);
92 v3_deinit_vmx_cpu(cpu_id);
97 PrintError("CPU has no virtualization Extensions\n");
103 void Init_V3(struct v3_os_hooks * hooks, char * cpu_mask, int num_cpus) {
108 V3_Print("V3 Print statement to fix a Kitten page fault bug\n");
110 // Set global variables.
113 // Determine the global machine type
114 v3_mach_type = V3_INVALID_CPU;
116 for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
117 v3_cpu_types[i] = V3_INVALID_CPU;
120 // Register all the possible device types
123 // Register all shadow paging handlers
124 V3_init_shdw_paging();
126 // Register all extensions
127 V3_init_extensions();
130 #ifdef V3_CONFIG_SYMMOD
134 #ifdef V3_CONFIG_CHECKPOINT
135 V3_init_checkpoint();
138 if ((hooks) && (hooks->call_on_cpu)) {
140 for (i = 0; i < num_cpus; i++) {
144 if ((cpu_mask == NULL) || (*(cpu_mask + major) & (0x1 << minor))) {
145 V3_Print("Initializing VMM extensions on cpu %d\n", i);
146 hooks->call_on_cpu(i, &init_cpu, (void *)(addr_t)i);
148 if (v3_mach_type == V3_INVALID_CPU) {
149 v3_mach_type = v3_cpu_types[i];
162 V3_deinit_shdw_paging();
164 V3_deinit_extensions();
166 #ifdef V3_CONFIG_SYMMOD
170 #ifdef V3_CONFIG_CHECKPOINT
171 V3_deinit_checkpoint();
175 if ((os_hooks) && (os_hooks->call_on_cpu)) {
176 for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
177 if (v3_cpu_types[i] != V3_INVALID_CPU) {
178 V3_Call_On_CPU(i, deinit_cpu, (void *)(addr_t)i);
179 //deinit_cpu((void *)(addr_t)i);
187 v3_cpu_arch_t v3_get_cpu_type(int cpu_id) {
188 return v3_cpu_types[cpu_id];
192 struct v3_vm_info * v3_create_vm(void * cfg, void * priv_data, char * name) {
193 struct v3_vm_info * vm = v3_config_guest(cfg, priv_data);
196 PrintError("Could not configure guest\n");
200 V3_Print("CORE 0 RIP=%p\n", (void *)(addr_t)(vm->cores[0].rip));
204 } else if (strlen(name) >= 128) {
205 PrintError("VM name is too long. Will be truncated to 128 chars.\n");
208 memset(vm->name, 0, 128);
209 strncpy(vm->name, name, 127);
217 static int start_core(void * p)
219 struct guest_info * core = (struct guest_info *)p;
222 PrintDebug("virtual core %u (on logical core %u): in start_core (RIP=%p)\n",
223 core->vcpu_id, core->pcpu_id, (void *)(addr_t)core->rip);
225 switch (v3_mach_type) {
228 case V3_SVM_REV3_CPU:
229 return v3_start_svm_guest(core);
235 case V3_VMX_EPT_UG_CPU:
236 return v3_start_vmx_guest(core);
240 PrintError("Attempting to enter a guest on an invalid CPU\n");
248 // For the moment very ugly. Eventually we will shift the cpu_mask to an arbitrary sized type...
252 int v3_start_vm(struct v3_vm_info * vm, unsigned int cpu_mask) {
254 uint8_t * core_mask = (uint8_t *)&cpu_mask; // This is to make future expansion easier
255 uint32_t avail_cores = 0;
259 if (vm->run_state == VM_RUNNING ||
260 vm->run_state == VM_INVALID ||
261 vm->run_state == VM_ERROR) {
262 PrintError("VM has already been launched (state=%d)\n", (int)vm->run_state);
266 /// CHECK IF WE ARE MULTICORE ENABLED....
268 V3_Print("V3 -- Starting VM (%u cores)\n", vm->num_cores);
269 V3_Print("CORE 0 RIP=%p\n", (void *)(addr_t)(vm->cores[0].rip));
272 // Check that enough cores are present in the mask to handle vcores
273 for (i = 0; i < MAX_CORES; i++) {
277 if (core_mask[major] & (0x1 << minor)) {
278 if (v3_cpu_types[i] == V3_INVALID_CPU) {
279 core_mask[major] &= ~(0x1 << minor);
287 if (vm->num_cores > avail_cores) {
288 PrintError("Attempted to start a VM with too many cores (vm->num_cores = %d, avail_cores = %d, MAX=%d)\n",
289 vm->num_cores, avail_cores, MAX_CORES);
293 vm->run_state = VM_RUNNING;
295 // Spawn off threads for each core.
296 // We work backwards, so that core 0 is always started last.
297 for (i = 0, vcore_id = vm->num_cores - 1; (i < MAX_CORES) && (vcore_id >= 0); i++) {
300 struct guest_info * core = &(vm->cores[vcore_id]);
301 char * specified_cpu = v3_cfg_val(core->core_cfg_data, "target_cpu");
302 uint32_t core_idx = 0;
304 if (specified_cpu != NULL) {
305 core_idx = atoi(specified_cpu);
307 if ((core_idx < 0) || (core_idx >= MAX_CORES)) {
308 PrintError("Target CPU out of bounds (%d) (MAX_CORES=%d)\n", core_idx, MAX_CORES);
311 i--; // We reset the logical core idx. Not strictly necessary I guess...
316 major = core_idx / 8;
317 minor = core_idx % 8;
319 if ((core_mask[major] & (0x1 << minor)) == 0) {
320 PrintError("Logical CPU %d not available for virtual core %d; not started\n",
323 if (specified_cpu != NULL) {
324 PrintError("CPU was specified explicitly (%d). HARD ERROR\n", core_idx);
332 PrintDebug("Starting virtual core %u on logical core %u\n",
335 sprintf(core->exec_name, "%s-%u", vm->name, vcore_id);
337 PrintDebug("run: core=%u, func=0x%p, arg=0x%p, name=%s\n",
338 core_idx, start_core, core, core->exec_name);
340 core->core_run_state = CORE_STOPPED; // core zero will turn itself on
341 core->pcpu_id = core_idx;
342 core->core_thread = V3_CREATE_THREAD_ON_CPU(core_idx, start_core, core, core->exec_name);
344 if (core->core_thread == NULL) {
345 PrintError("Thread launch failed\n");
354 PrintError("Error starting VM: Not enough available CPU cores\n");
365 int v3_reset_vm_core(struct guest_info * core, addr_t rip) {
367 switch (v3_cpu_types[core->pcpu_id]) {
370 case V3_SVM_REV3_CPU:
371 PrintDebug("Resetting SVM Guest CPU %d\n", core->vcpu_id);
372 return v3_reset_svm_vm_core(core, rip);
377 case V3_VMX_EPT_UG_CPU:
378 PrintDebug("Resetting VMX Guest CPU %d\n", core->vcpu_id);
379 return v3_reset_vmx_vm_core(core, rip);
383 PrintError("CPU has no virtualization Extensions\n");
392 /* move a virtual core to different physical core */
393 int v3_move_vm_core(struct v3_vm_info * vm, int vcore_id, int target_cpu) {
394 struct guest_info * core = NULL;
396 if ((vcore_id < 0) || (vcore_id >= vm->num_cores)) {
397 PrintError("Attempted to migrate invalid virtual core (%d)\n", vcore_id);
401 core = &(vm->cores[vcore_id]);
403 if (target_cpu == core->pcpu_id) {
404 PrintError("Attempted to migrate to local core (%d)\n", target_cpu);
405 // well that was pointless
409 if (core->core_thread == NULL) {
410 PrintError("Attempted to migrate a core without a valid thread context\n");
414 while (v3_raise_barrier(vm, NULL) == -1);
416 V3_Print("Performing Migration from %d to %d\n", core->pcpu_id, target_cpu);
418 // Double check that we weren't preemptively migrated
419 if (target_cpu != core->pcpu_id) {
421 V3_Print("Moving Core\n");
425 switch (v3_cpu_types[core->pcpu_id]) {
428 case V3_VMX_EPT_UG_CPU:
429 PrintDebug("Flushing VMX Guest CPU %d\n", core->vcpu_id);
430 V3_Call_On_CPU(core->pcpu_id, (void (*)(void *))v3_flush_vmx_vm_core, (void *)core);
437 if (V3_MOVE_THREAD_TO_CPU(target_cpu, core->core_thread) != 0) {
438 PrintError("Failed to move Vcore %d to CPU %d\n",
439 core->vcpu_id, target_cpu);
440 v3_lower_barrier(vm);
444 /* There will be a benign race window here:
445 core->pcpu_id will be set to the target core before its fully "migrated"
446 However the core will NEVER run on the old core again, its just in flight to the new core
448 core->pcpu_id = target_cpu;
450 V3_Print("core now at %d\n", core->pcpu_id);
453 v3_lower_barrier(vm);
460 int v3_stop_vm(struct v3_vm_info * vm) {
462 vm->run_state = VM_STOPPED;
464 // Sanity check to catch any weird execution states
465 if (v3_wait_for_barrier(vm, NULL) == 0) {
466 v3_lower_barrier(vm);
469 // XXX force exit all cores via a cross call/IPI XXX
473 int still_running = 0;
475 for (i = 0; i < vm->num_cores; i++) {
476 if (vm->cores[i].core_run_state != CORE_STOPPED) {
481 if (still_running == 0) {
488 V3_Print("VM stopped. Returning\n");
494 int v3_pause_vm(struct v3_vm_info * vm) {
496 if (vm->run_state != VM_RUNNING) {
497 PrintError("Tried to pause a VM that was not running\n");
501 while (v3_raise_barrier(vm, NULL) == -1);
503 vm->run_state = VM_PAUSED;
509 int v3_continue_vm(struct v3_vm_info * vm) {
511 if (vm->run_state != VM_PAUSED) {
512 PrintError("Tried to continue a VM that was not paused\n");
516 vm->run_state = VM_RUNNING;
518 v3_lower_barrier(vm);
525 static int sim_callback(struct guest_info * core, void * private_data) {
526 struct v3_bitmap * timeout_map = private_data;
528 v3_bitmap_set(timeout_map, core->vcpu_id);
530 V3_Print("Simulation callback activated (guest_rip=%p)\n", (void *)core->rip);
532 while (v3_bitmap_check(timeout_map, core->vcpu_id) == 1) {
542 int v3_simulate_vm(struct v3_vm_info * vm, unsigned int msecs) {
543 struct v3_bitmap timeout_map;
547 uint64_t cpu_khz = V3_CPU_KHZ();
549 if (vm->run_state != VM_PAUSED) {
550 PrintError("VM must be paused before simulation begins\n");
554 /* AT this point VM is paused */
557 v3_bitmap_init(&timeout_map, vm->num_cores);
562 // calculate cycles from msecs...
563 // IMPORTANT: Floating point not allowed.
564 cycles = (msecs * cpu_khz);
568 V3_Print("Simulating %u msecs (%llu cycles) [CPU_KHZ=%llu]\n", msecs, cycles, cpu_khz);
572 for (i = 0; i < vm->num_cores; i++) {
573 if (v3_add_core_timeout(&(vm->cores[i]), cycles, sim_callback, &timeout_map) == -1) {
574 PrintError("Could not register simulation timeout for core %d\n", i);
579 V3_Print("timeouts set on all cores\n ");
582 // Run the simulation
583 // vm->run_state = VM_SIMULATING;
584 vm->run_state = VM_RUNNING;
585 v3_lower_barrier(vm);
588 V3_Print("Barrier lowered: We are now Simulating!!\n");
590 // block until simulation is complete
591 while (all_blocked == 0) {
594 for (i = 0; i < vm->num_cores; i++) {
595 if (v3_bitmap_check(&timeout_map, i) == 0) {
600 if (all_blocked == 1) {
608 V3_Print("Simulation is complete\n");
610 // Simulation is complete
611 // Reset back to PAUSED state
613 v3_raise_barrier_nowait(vm, NULL);
614 vm->run_state = VM_PAUSED;
616 v3_bitmap_reset(&timeout_map);
618 v3_wait_for_barrier(vm, NULL);
624 #ifdef V3_CONFIG_CHECKPOINT
625 #include <palacios/vmm_checkpoint.h>
627 int v3_save_vm(struct v3_vm_info * vm, char * store, char * url) {
628 return v3_chkpt_save_vm(vm, store, url);
632 int v3_load_vm(struct v3_vm_info * vm, char * store, char * url) {
633 return v3_chkpt_load_vm(vm, store, url);
638 int v3_free_vm(struct v3_vm_info * vm) {
640 // deinitialize guest (free memory, etc...)
642 v3_free_vm_devices(vm);
645 for (i = 0; i < vm->num_cores; i++) {
646 v3_free_core(&(vm->cores[i]));
650 v3_free_vm_internal(vm);
662 v3_cpu_mode_t v3_get_host_cpu_mode() {
672 cr4 = (struct cr4_32 *)&(cr4_val);
675 return PROTECTED_PAE;
683 v3_cpu_mode_t v3_get_host_cpu_mode() {
693 void v3_yield_cond(struct guest_info * info, int usec) {
695 cur_cycle = v3_get_host_time(&info->time_state);
697 if (cur_cycle > (info->yield_start_cycle + info->vm_info->yield_cycle_period)) {
698 //PrintDebug("Conditional Yield (cur_cyle=%p, start_cycle=%p, period=%p)\n",
699 // (void *)cur_cycle, (void *)info->yield_start_cycle,
700 // (void *)info->yield_cycle_period);
705 V3_Yield_Timed(usec);
708 info->yield_start_cycle += info->vm_info->yield_cycle_period;
714 * unconditional cpu yield
715 * if the yielding thread is a guest context, the guest quantum is reset on resumption
716 * Non guest context threads should call this function with a NULL argument
718 * usec <0 => the non-timed yield is used
719 * usec >=0 => the timed yield is used, which also usually implies interruptible
721 void v3_yield(struct guest_info * info, int usec) {
725 V3_Yield_Timed(usec);
729 info->yield_start_cycle += info->vm_info->yield_cycle_period;
736 void v3_print_cond(const char * fmt, ...) {
737 if (v3_dbg_enable == 1) {
742 vsnprintf(buf, 2048, fmt, ap);
751 void v3_interrupt_cpu(struct v3_vm_info * vm, int logical_cpu, int vector) {
752 extern struct v3_os_hooks * os_hooks;
754 if ((os_hooks) && (os_hooks)->interrupt_cpu) {
755 (os_hooks)->interrupt_cpu(vm, logical_cpu, vector);
761 int v3_vm_enter(struct guest_info * info) {
762 switch (v3_mach_type) {
765 case V3_SVM_REV3_CPU:
766 return v3_svm_enter(info);
772 case V3_VMX_EPT_UG_CPU:
773 return v3_vmx_enter(info);
777 PrintError("Attemping to enter a guest on an invalid CPU\n");