#include <palacios/vmm_lowlevel.h>
#include <palacios/vmm_sprintf.h>
#include <palacios/vmm_extensions.h>
+#include <palacios/vmm_timeout.h>
+
#ifdef V3_CONFIG_SVM
#include <palacios/svm.h>
#include <palacios/vmx.h>
#endif
+#ifdef V3_CONFIG_CHECKPOINT
+#include <palacios/vmm_checkpoint.h>
+#endif
+
v3_cpu_arch_t v3_cpu_types[V3_CONFIG_MAX_CPUS];
+v3_cpu_arch_t v3_mach_type = V3_INVALID_CPU;
+
struct v3_os_hooks * os_hooks = NULL;
int v3_dbg_enable = 0;
}
-
-void Init_V3(struct v3_os_hooks * hooks, int num_cpus) {
- int i;
+void Init_V3(struct v3_os_hooks * hooks, char * cpu_mask, int num_cpus) {
+ int i = 0;
+ int minor = 0;
+ int major = 0;
V3_Print("V3 Print statement to fix a Kitten page fault bug\n");
// Set global variables.
os_hooks = hooks;
+ // Determine the global machine type
+ v3_mach_type = V3_INVALID_CPU;
+
for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
v3_cpu_types[i] = V3_INVALID_CPU;
}
V3_init_symmod();
#endif
+#ifdef V3_CONFIG_CHECKPOINT
+ V3_init_checkpoint();
+#endif
-
-#ifdef V3_CONFIG_MULTITHREAD_OS
if ((hooks) && (hooks->call_on_cpu)) {
- for (i = 0; i < num_cpus; i++) {
+ for (i = 0; i < num_cpus; i++) {
+ major = i / 8;
+ minor = i % 8;
- V3_Print("Initializing VMM extensions on cpu %d\n", i);
- hooks->call_on_cpu(i, &init_cpu, (void *)(addr_t)i);
- }
- }
-#else
- init_cpu(0);
-#endif
+ if ((cpu_mask == NULL) || (*(cpu_mask + major) & (0x1 << minor))) {
+ V3_Print("Initializing VMM extensions on cpu %d\n", i);
+ hooks->call_on_cpu(i, &init_cpu, (void *)(addr_t)i);
+ if (v3_mach_type == V3_INVALID_CPU) {
+ v3_mach_type = v3_cpu_types[i];
+ }
+ }
+ }
+ }
}
+
void Shutdown_V3() {
int i;
V3_deinit_symmod();
#endif
+#ifdef V3_CONFIG_CHECKPOINT
+ V3_deinit_checkpoint();
+#endif
+
-#ifdef V3_CONFIG_MULTITHREAD_OS
if ((os_hooks) && (os_hooks->call_on_cpu)) {
for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) {
if (v3_cpu_types[i] != V3_INVALID_CPU) {
}
}
}
-#else
- deinit_cpu(0);
-#endif
}
PrintDebug("virtual core %u (on logical core %u): in start_core (RIP=%p)\n",
core->vcpu_id, core->pcpu_id, (void *)(addr_t)core->rip);
- switch (v3_cpu_types[0]) {
+ switch (v3_mach_type) {
#ifdef V3_CONFIG_SVM
case V3_SVM_CPU:
case V3_SVM_REV3_CPU:
// For the moment very ugly. Eventually we will shift the cpu_mask to an arbitrary sized type...
-#ifdef V3_CONFIG_MULTITHREAD_OS
#define MAX_CORES 32
-#else
-#define MAX_CORES 1
-#endif
int v3_start_vm(struct v3_vm_info * vm, unsigned int cpu_mask) {
uint32_t avail_cores = 0;
int vcore_id = 0;
+
+ if (vm->run_state != VM_STOPPED) {
+ PrintError("VM has already been launched (state=%d)\n", (int)vm->run_state);
+ return -1;
+ }
+
+
+ // Do not run if any core is using shadow paging and we are out of 4 GB bounds
+ for (i=0;i<vm->num_cores;i++) {
+ if (vm->cores[i].shdw_pg_mode == SHADOW_PAGING) {
+ if ((vm->mem_map.base_region.host_addr + vm->mem_size ) >= 0x100000000ULL) {
+ PrintError("Base memory region exceeds 4 GB boundary with shadow paging enabled on core %d.\n",i);
+ PrintError("Any use of non-64 bit mode in the guest is likely to fail in this configuration.\n");
+ PrintError("If you would like to proceed anyway, remove this check and recompile Palacios.\n");
+ PrintError("Alternatively, change this VM to use nested paging.\n");
+ return -1;
+ }
+ }
+ }
+
+
+
/// CHECK IF WE ARE MULTICORE ENABLED....
V3_Print("V3 -- Starting VM (%u cores)\n", vm->num_cores);
int minor = i % 8;
if (core_mask[major] & (0x1 << minor)) {
- avail_cores++;
+ if (v3_cpu_types[i] == V3_INVALID_CPU) {
+ core_mask[major] &= ~(0x1 << minor);
+ } else {
+ avail_cores++;
+ }
}
}
return -1;
}
-#ifdef V3_CONFIG_MULTITHREAD_OS
- // spawn off new threads, for other cores
- for (i = 0, vcore_id = 1; (i < MAX_CORES) && (vcore_id < vm->num_cores); i++) {
+ vm->run_state = VM_RUNNING;
+
+ // Spawn off threads for each core.
+ // We work backwards, so that core 0 is always started last.
+ for (i = 0, vcore_id = vm->num_cores - 1; (i < MAX_CORES) && (vcore_id >= 0); i++) {
int major = 0;
int minor = 0;
struct guest_info * core = &(vm->cores[vcore_id]);
i--; // We reset the logical core idx. Not strictly necessary I guess...
} else {
-
- if (i == V3_Get_CPU()) {
- // We skip the local CPU because it is reserved for vcore 0
- continue;
- }
-
core_idx = i;
}
major = core_idx / 8;
minor = core_idx % 8;
-
if ((core_mask[major] & (0x1 << minor)) == 0) {
PrintError("Logical CPU %d not available for virtual core %d; not started\n",
core_idx, vcore_id);
PrintDebug("run: core=%u, func=0x%p, arg=0x%p, name=%s\n",
core_idx, start_core, core, core->exec_name);
- // TODO: actually manage these threads instead of just launching them
+ core->core_run_state = CORE_STOPPED; // core zero will turn itself on
core->pcpu_id = core_idx;
core->core_thread = V3_CREATE_THREAD_ON_CPU(core_idx, start_core, core, core->exec_name);
return -1;
}
- vcore_id++;
+ vcore_id--;
}
-#endif
-
- sprintf(vm->cores[0].exec_name, "%s", vm->name);
- vm->cores[0].pcpu_id = V3_Get_CPU();
-
- if (start_core(&(vm->cores[0])) != 0) {
- PrintError("Error starting VM core 0\n");
+ if (vcore_id >= 0) {
+ PrintError("Error starting VM: Not enough available CPU cores\n");
v3_stop_vm(vm);
return -1;
}
int v3_move_vm_core(struct v3_vm_info * vm, int vcore_id, int target_cpu) {
struct guest_info * core = NULL;
- if(vcore_id < 0 || vcore_id > vm->num_cores) {
+ if ((vcore_id < 0) || (vcore_id >= vm->num_cores)) {
+ PrintError("Attempted to migrate invalid virtual core (%d)\n", vcore_id);
return -1;
}
core = &(vm->cores[vcore_id]);
- if(target_cpu != core->pcpu_id &&
- core->core_move_state != CORE_MOVE_PENDING){
- core->core_move_state = CORE_MOVE_PENDING;
- core->target_pcpu_id = target_cpu;
- v3_interrupt_cpu(vm, core->pcpu_id, 0);
+ if (target_cpu == core->pcpu_id) {
+ PrintError("Attempted to migrate to local core (%d)\n", target_cpu);
+ // well that was pointless
+ return 0;
+ }
- while(core->core_move_state != CORE_MOVE_DONE){
- v3_yield(NULL);
- }
+ if (core->core_thread == NULL) {
+ PrintError("Attempted to migrate a core without a valid thread context\n");
+ return -1;
}
+ while (v3_raise_barrier(vm, NULL) == -1);
+
+ V3_Print("Performing Migration from %d to %d\n", core->pcpu_id, target_cpu);
+
+ // Double check that we weren't preemptively migrated
+ if (target_cpu != core->pcpu_id) {
+
+ V3_Print("Moving Core\n");
+
+
+#ifdef V3_CONFIG_VMX
+ switch (v3_cpu_types[core->pcpu_id]) {
+ case V3_VMX_CPU:
+ case V3_VMX_EPT_CPU:
+ case V3_VMX_EPT_UG_CPU:
+ PrintDebug("Flushing VMX Guest CPU %d\n", core->vcpu_id);
+ V3_Call_On_CPU(core->pcpu_id, (void (*)(void *))v3_flush_vmx_vm_core, (void *)core);
+ break;
+ default:
+ break;
+ }
+#endif
+
+ if (V3_MOVE_THREAD_TO_CPU(target_cpu, core->core_thread) != 0) {
+ PrintError("Failed to move Vcore %d to CPU %d\n",
+ core->vcpu_id, target_cpu);
+ v3_lower_barrier(vm);
+ return -1;
+ }
+ /* There will be a benign race window here:
+ core->pcpu_id will be set to the target core before its fully "migrated"
+ However the core will NEVER run on the old core again, its just in flight to the new core
+ */
+ core->pcpu_id = target_cpu;
+
+ V3_Print("core now at %d\n", core->pcpu_id);
+ }
+
+ v3_lower_barrier(vm);
+
return 0;
}
int v3_stop_vm(struct v3_vm_info * vm) {
+ if ((vm->run_state != VM_RUNNING) &&
+ (vm->run_state != VM_SIMULATING)) {
+ PrintError("Tried to stop VM in invalid runstate (%d)\n", vm->run_state);
+ return -1;
+ }
+
vm->run_state = VM_STOPPED;
- // force exit all cores via a cross call/IPI
+ // Sanity check to catch any weird execution states
+ if (v3_wait_for_barrier(vm, NULL) == 0) {
+ v3_lower_barrier(vm);
+ }
+
+ // XXX force exit all cores via a cross call/IPI XXX
while (1) {
int i = 0;
break;
}
- v3_yield(NULL);
+ v3_yield(NULL,-1);
}
V3_Print("VM stopped. Returning\n");
return -1;
}
+ vm->run_state = VM_RUNNING;
+
v3_lower_barrier(vm);
+ return 0;
+}
+
+
+
+static int sim_callback(struct guest_info * core, void * private_data) {
+ struct v3_bitmap * timeout_map = private_data;
+
+ v3_bitmap_set(timeout_map, core->vcpu_id);
+
+ V3_Print("Simulation callback activated (guest_rip=%p)\n", (void *)core->rip);
+
+ while (v3_bitmap_check(timeout_map, core->vcpu_id) == 1) {
+ v3_yield(NULL,-1);
+ }
+
+ return 0;
+}
+
+
+
+
+int v3_simulate_vm(struct v3_vm_info * vm, unsigned int msecs) {
+ struct v3_bitmap timeout_map;
+ int i = 0;
+ int all_blocked = 0;
+ uint64_t cycles = 0;
+ uint64_t cpu_khz = V3_CPU_KHZ();
+
+ if (vm->run_state != VM_PAUSED) {
+ PrintError("VM must be paused before simulation begins\n");
+ return -1;
+ }
+
+ /* AT this point VM is paused */
+
+ // initialize bitmap
+ v3_bitmap_init(&timeout_map, vm->num_cores);
+
+
+
+
+ // calculate cycles from msecs...
+ // IMPORTANT: Floating point not allowed.
+ cycles = (msecs * cpu_khz);
+
+
+
+ V3_Print("Simulating %u msecs (%llu cycles) [CPU_KHZ=%llu]\n", msecs, cycles, cpu_khz);
+
+ // set timeout
+
+ for (i = 0; i < vm->num_cores; i++) {
+ if (v3_add_core_timeout(&(vm->cores[i]), cycles, sim_callback, &timeout_map) == -1) {
+ PrintError("Could not register simulation timeout for core %d\n", i);
+ return -1;
+ }
+ }
+
+ V3_Print("timeouts set on all cores\n ");
+
+
+ // Run the simulation
+// vm->run_state = VM_SIMULATING;
vm->run_state = VM_RUNNING;
+ v3_lower_barrier(vm);
+
+
+ V3_Print("Barrier lowered: We are now Simulating!!\n");
+
+ // block until simulation is complete
+ while (all_blocked == 0) {
+ all_blocked = 1;
+
+ for (i = 0; i < vm->num_cores; i++) {
+ if (v3_bitmap_check(&timeout_map, i) == 0) {
+ all_blocked = 0;
+ }
+ }
+
+ if (all_blocked == 1) {
+ break;
+ }
+
+ v3_yield(NULL,-1);
+ }
+
+
+ V3_Print("Simulation is complete\n");
+
+ // Simulation is complete
+ // Reset back to PAUSED state
+
+ v3_raise_barrier_nowait(vm, NULL);
+ vm->run_state = VM_PAUSED;
+
+ v3_bitmap_reset(&timeout_map);
+
+ v3_wait_for_barrier(vm, NULL);
return 0;
+
}
+#ifdef V3_CONFIG_CHECKPOINT
+#include <palacios/vmm_checkpoint.h>
+
+int v3_save_vm(struct v3_vm_info * vm, char * store, char * url) {
+ return v3_chkpt_save_vm(vm, store, url);
+}
+
+
+int v3_load_vm(struct v3_vm_info * vm, char * store, char * url) {
+ return v3_chkpt_load_vm(vm, store, url);
+}
+
+#ifdef V3_CONFIG_LIVE_MIGRATION
+int v3_send_vm(struct v3_vm_info * vm, char * store, char * url) {
+ return v3_chkpt_send_vm(vm, store, url);
+}
+
+
+int v3_receive_vm(struct v3_vm_info * vm, char * store, char * url) {
+ return v3_chkpt_receive_vm(vm, store, url);
+}
+#endif
+
+#endif
+
int v3_free_vm(struct v3_vm_info * vm) {
int i = 0;
// deinitialize guest (free memory, etc...)
+ if ((vm->run_state != VM_STOPPED) &&
+ (vm->run_state != VM_ERROR)) {
+ PrintError("Tried to Free VM in invalid runstate (%d)\n", vm->run_state);
+ return -1;
+ }
+
v3_free_vm_devices(vm);
// free cores
#endif
-#define V3_Yield(addr) \
- do { \
- extern struct v3_os_hooks * os_hooks; \
- if ((os_hooks) && (os_hooks)->yield_cpu) { \
- (os_hooks)->yield_cpu(); \
- } \
- } while (0) \
-void v3_yield_cond(struct guest_info * info) {
+void v3_yield_cond(struct guest_info * info, int usec) {
uint64_t cur_cycle;
cur_cycle = v3_get_host_time(&info->time_state);
if (cur_cycle > (info->yield_start_cycle + info->vm_info->yield_cycle_period)) {
+ //PrintDebug("Conditional Yield (cur_cyle=%p, start_cycle=%p, period=%p)\n",
+ // (void *)cur_cycle, (void *)info->yield_start_cycle,
+ // (void *)info->yield_cycle_period);
+
+ if (usec < 0) {
+ V3_Yield();
+ } else {
+ V3_Sleep(usec);
+ }
- /*
- PrintDebug("Conditional Yield (cur_cyle=%p, start_cycle=%p, period=%p)\n",
- (void *)cur_cycle, (void *)info->yield_start_cycle, (void *)info->yield_cycle_period);
- */
- V3_Yield();
- info->yield_start_cycle = v3_get_host_time(&info->time_state);
+ info->yield_start_cycle += info->vm_info->yield_cycle_period;
}
}
-
+
/*
* unconditional cpu yield
* if the yielding thread is a guest context, the guest quantum is reset on resumption
* Non guest context threads should call this function with a NULL argument
- */
-void v3_yield(struct guest_info * info) {
- V3_Yield();
+ *
+ * usec <0 => the non-timed yield is used
+ * usec >=0 => the timed yield is used, which also usually implies interruptible
+ */
+void v3_yield(struct guest_info * info, int usec) {
+ if (usec < 0) {
+ V3_Yield();
+ } else {
+ V3_Sleep(usec);
+ }
if (info) {
- info->yield_start_cycle = v3_get_host_time(&info->time_state);
+ info->yield_start_cycle += info->vm_info->yield_cycle_period;
}
}
}
-#ifdef V3_CONFIG_MULTITHREAD_OS
void v3_interrupt_cpu(struct v3_vm_info * vm, int logical_cpu, int vector) {
extern struct v3_os_hooks * os_hooks;
(os_hooks)->interrupt_cpu(vm, logical_cpu, vector);
}
}
-#endif
int v3_vm_enter(struct guest_info * info) {
- switch (v3_cpu_types[0]) {
+ switch (v3_mach_type) {
#ifdef V3_CONFIG_SVM
case V3_SVM_CPU:
case V3_SVM_REV3_CPU: