X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm.c;h=e9a3a9df7d1f22fe2af3179cb60d92376e9714b5;hb=d597d493eb303496f3bb32e2a73a094a8a20a022;hp=a7e83cc1bd1ff78e19208795d6b304f79c3ac7cf;hpb=268cdc635ece417c3472295c4c81591564c6e0df;p=palacios.git diff --git a/palacios/src/palacios/vmm.c b/palacios/src/palacios/vmm.c index a7e83cc..e9a3a9d 100644 --- a/palacios/src/palacios/vmm.c +++ b/palacios/src/palacios/vmm.c @@ -21,41 +21,45 @@ #include #include #include -#include #include #include #include +#include +#include -#ifdef CONFIG_SVM + +#ifdef V3_CONFIG_SVM #include #endif -#ifdef CONFIG_VMX +#ifdef V3_CONFIG_VMX #include #endif -#ifdef CONFIG_VNET -#include +#ifdef V3_CONFIG_CHECKPOINT +#include #endif -v3_cpu_arch_t v3_cpu_types[CONFIG_MAX_CPUS]; -struct v3_os_hooks * os_hooks = NULL; +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; + static void init_cpu(void * arg) { uint32_t cpu_id = (uint32_t)(addr_t)arg; -#ifdef CONFIG_SVM +#ifdef V3_CONFIG_SVM if (v3_is_svm_capable()) { PrintDebug("Machine is SVM Capable\n"); v3_init_svm_cpu(cpu_id); } else #endif -#ifdef CONFIG_VMX +#ifdef V3_CONFIG_VMX if (v3_is_vmx_capable()) { PrintDebug("Machine is VMX Capable\n"); v3_init_vmx_cpu(cpu_id); @@ -67,40 +71,49 @@ static void init_cpu(void * arg) { } } -#if 0 + static void deinit_cpu(void * arg) { -// uint32_t cpu_id = (uint32_t)(addr_t)arg; + uint32_t cpu_id = (uint32_t)(addr_t)arg; - #ifdef CONFIG_SVM - if (v3_is_svm_capable()) { - PrintDebug("Machine is SVM Capable\n"); - //v3_deinit_svm_cpu(cpu_id); - } else + switch (v3_cpu_types[cpu_id]) { +#ifdef V3_CONFIG_SVM + case V3_SVM_CPU: + case V3_SVM_REV3_CPU: + PrintDebug("Deinitializing SVM CPU %d\n", cpu_id); + v3_deinit_svm_cpu(cpu_id); + break; #endif -#ifdef CONFIG_VMX - if (v3_is_vmx_capable()) { - PrintDebug("Machine is VMX Capable\n"); - //v3_deinit_vmx_cpu(cpu_id); - - } else +#ifdef V3_CONFIG_VMX + case V3_VMX_CPU: + case V3_VMX_EPT_CPU: + case V3_VMX_EPT_UG_CPU: + PrintDebug("Deinitializing VMX CPU %d\n", cpu_id); + v3_deinit_vmx_cpu(cpu_id); + break; #endif - { - PrintError("CPU has no virtualization Extensions\n"); + case V3_INVALID_CPU: + default: + PrintError("CPU has no virtualization Extensions\n"); + break; } } -#endif -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; - for (i = 0; i < CONFIG_MAX_CPUS; i++) { + // 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; } @@ -110,70 +123,63 @@ void Init_V3(struct v3_os_hooks * hooks, int num_cpus) { // Register all shadow paging handlers V3_init_shdw_paging(); + // Register all extensions + V3_init_extensions(); -#ifdef CONFIG_SYMMOD - V3_init_symmod(); -#endif -#ifdef CONFIG_INSTRUMENT_VMM - v3_init_instrumentation(); +#ifdef V3_CONFIG_SYMMOD + V3_init_symmod(); #endif - -#ifdef CONFIG_VNET - v3_init_vnet(); +#ifdef V3_CONFIG_CHECKPOINT + V3_init_checkpoint(); #endif - -#ifdef 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; + int i; V3_deinit_devices(); V3_deinit_shdw_paging(); -#if 0 + V3_deinit_extensions(); -#ifdef CONFIG_SYMMOD +#ifdef V3_CONFIG_SYMMOD V3_deinit_symmod(); #endif -#ifdef CONFIG_INSTRUMENT_VMM - v3_deinit_instrumentation(); +#ifdef V3_CONFIG_CHECKPOINT + V3_deinit_checkpoint(); #endif -#ifdef CONFIG_VNET - v3_deinit_vnet(); -#endif -#ifdef CONFIG_MULTITHREAD_OS - if ((hooks) && (hooks->call_on_cpu)) { - for (i = 0; i < CONFIG_MAX_CPUS; i++) { + 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) { - deinit_cpu(i); + V3_Call_On_CPU(i, deinit_cpu, (void *)(addr_t)i); + //deinit_cpu((void *)(addr_t)i); } } } -#else - deinit_cpu(0); -#endif - -#endif } @@ -186,14 +192,13 @@ v3_cpu_arch_t v3_get_cpu_type(int cpu_id) { struct v3_vm_info * v3_create_vm(void * cfg, void * priv_data, char * name) { struct v3_vm_info * vm = v3_config_guest(cfg, priv_data); - V3_Print("CORE 0 RIP=%p\n", (void *)(addr_t)(vm->cores[0].rip)); - - if (vm == NULL) { PrintError("Could not configure guest\n"); return NULL; } + V3_Print("CORE 0 RIP=%p\n", (void *)(addr_t)(vm->cores[0].rip)); + if (name == NULL) { name = "[V3_VM]"; } else if (strlen(name) >= 128) { @@ -207,26 +212,27 @@ struct v3_vm_info * v3_create_vm(void * cfg, void * priv_data, char * name) { } + + static int start_core(void * p) { struct guest_info * core = (struct guest_info *)p; - PrintDebug("core %u: in start_core (RIP=%p)\n", - core->cpu_id, (void *)(addr_t)core->rip); - + 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); - // JRL: Whoa WTF? cpu_types are tied to the vcoreID???? - switch (v3_cpu_types[core->cpu_id]) { -#ifdef CONFIG_SVM + switch (v3_mach_type) { +#ifdef V3_CONFIG_SVM case V3_SVM_CPU: case V3_SVM_REV3_CPU: return v3_start_svm_guest(core); break; #endif -#if CONFIG_VMX +#if V3_CONFIG_VMX case V3_VMX_CPU: case V3_VMX_EPT_CPU: + case V3_VMX_EPT_UG_CPU: return v3_start_vmx_guest(core); break; #endif @@ -240,88 +246,111 @@ static int start_core(void * p) // For the moment very ugly. Eventually we will shift the cpu_mask to an arbitrary sized type... -#ifdef 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 i; -#ifdef CONFIG_MULTITHREAD_OS - int vcore_id = 0; -#endif uint8_t * core_mask = (uint8_t *)&cpu_mask; // This is to make future expansion easier 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; + } /// CHECK IF WE ARE MULTICORE ENABLED.... V3_Print("V3 -- Starting VM (%u cores)\n", vm->num_cores); V3_Print("CORE 0 RIP=%p\n", (void *)(addr_t)(vm->cores[0].rip)); + // Check that enough cores are present in the mask to handle vcores for (i = 0; i < MAX_CORES; i++) { int major = i / 8; 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++; + } } - } - + + if (vm->num_cores > avail_cores) { - PrintError("Attempted to start a VM with too many cores (vm->num_cores = %d, avail_cores = %d, MAX=%d)\n", vm->num_cores, avail_cores, MAX_CORES); + PrintError("Attempted to start a VM with too many cores (vm->num_cores = %d, avail_cores = %d, MAX=%d)\n", + vm->num_cores, avail_cores, MAX_CORES); return -1; } + vm->run_state = VM_RUNNING; -#ifdef 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++) { - int major = i / 8; - int minor = i % 8; - void * core_thread = NULL; + // 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]); + char * specified_cpu = v3_cfg_val(core->core_cfg_data, "target_cpu"); + uint32_t core_idx = 0; + + if (specified_cpu != NULL) { + core_idx = atoi(specified_cpu); + + if ((core_idx < 0) || (core_idx >= MAX_CORES)) { + PrintError("Target CPU out of bounds (%d) (MAX_CORES=%d)\n", core_idx, MAX_CORES); + } - if (i == V3_Get_CPU()) { - // We skip the local CPU, because it is reserved for vcore 0 - continue; + i--; // We reset the logical core idx. Not strictly necessary I guess... + } else { + core_idx = i; } + major = core_idx / 8; + minor = core_idx % 8; if ((core_mask[major] & (0x1 << minor)) == 0) { - // cpuid not set in cpu_mask + PrintError("Logical CPU %d not available for virtual core %d; not started\n", + core_idx, vcore_id); + + if (specified_cpu != NULL) { + PrintError("CPU was specified explicitly (%d). HARD ERROR\n", core_idx); + v3_stop_vm(vm); + return -1; + } + continue; - } + } PrintDebug("Starting virtual core %u on logical core %u\n", - vcore_id, i); + vcore_id, core_idx); sprintf(core->exec_name, "%s-%u", vm->name, vcore_id); PrintDebug("run: core=%u, func=0x%p, arg=0x%p, name=%s\n", - i, start_core, core, core->exec_name); + core_idx, start_core, core, core->exec_name); - // TODO: actually manage these threads instead of just launching them - core_thread = V3_CREATE_THREAD_ON_CPU(i, start_core, core, core->exec_name); + 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); - if (core_thread == NULL) { + if (core->core_thread == NULL) { PrintError("Thread launch failed\n"); + v3_stop_vm(vm); return -1; } - vcore_id++; + vcore_id--; } -#endif - - sprintf(vm->cores[0].exec_name, "%s", vm->name); - 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; } @@ -331,13 +360,111 @@ int v3_start_vm(struct v3_vm_info * vm, unsigned int cpu_mask) { } +int v3_reset_vm_core(struct guest_info * core, addr_t rip) { + + switch (v3_cpu_types[core->pcpu_id]) { +#ifdef V3_CONFIG_SVM + case V3_SVM_CPU: + case V3_SVM_REV3_CPU: + PrintDebug("Resetting SVM Guest CPU %d\n", core->vcpu_id); + return v3_reset_svm_vm_core(core, rip); +#endif +#ifdef V3_CONFIG_VMX + case V3_VMX_CPU: + case V3_VMX_EPT_CPU: + case V3_VMX_EPT_UG_CPU: + PrintDebug("Resetting VMX Guest CPU %d\n", core->vcpu_id); + return v3_reset_vmx_vm_core(core, rip); +#endif + case V3_INVALID_CPU: + default: + PrintError("CPU has no virtualization Extensions\n"); + break; + } + + return -1; +} + + + +/* move a virtual core to different physical core */ +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)) { + PrintError("Attempted to migrate invalid virtual core (%d)\n", vcore_id); + return -1; + } + + core = &(vm->cores[vcore_id]); + + if (target_cpu == core->pcpu_id) { + PrintError("Attempted to migrate to local core (%d)\n", target_cpu); + // well that was pointless + return 0; + } + + 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) { 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; @@ -353,9 +480,7 @@ int v3_stop_vm(struct v3_vm_info * vm) { break; } - V3_Print("Yielding\n"); - - v3_yield(NULL); + v3_yield(NULL,-1); } V3_Print("VM stopped. Returning\n"); @@ -364,6 +489,150 @@ int v3_stop_vm(struct v3_vm_info * vm) { } +int v3_pause_vm(struct v3_vm_info * vm) { + + if (vm->run_state != VM_RUNNING) { + PrintError("Tried to pause a VM that was not running\n"); + return -1; + } + + while (v3_raise_barrier(vm, NULL) == -1); + + vm->run_state = VM_PAUSED; + + return 0; +} + + +int v3_continue_vm(struct v3_vm_info * vm) { + + if (vm->run_state != VM_PAUSED) { + PrintError("Tried to continue a VM that was not paused\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 + +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); +} +#endif + + int v3_free_vm(struct v3_vm_info * vm) { int i = 0; // deinitialize guest (free memory, etc...) @@ -416,28 +685,25 @@ v3_cpu_mode_t v3_get_host_cpu_mode() { #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_Yield_Timed(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; } } @@ -446,12 +712,19 @@ void v3_yield_cond(struct guest_info * info) { * 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_Yield_Timed(usec); + } if (info) { - info->yield_start_cycle = v3_get_host_time(&info->time_state); + info->yield_start_cycle += info->vm_info->yield_cycle_period; } } @@ -472,7 +745,6 @@ void v3_print_cond(const char * fmt, ...) { } -#ifdef CONFIG_MULTITHREAD_OS void v3_interrupt_cpu(struct v3_vm_info * vm, int logical_cpu, int vector) { extern struct v3_os_hooks * os_hooks; @@ -481,21 +753,21 @@ void v3_interrupt_cpu(struct v3_vm_info * vm, int logical_cpu, int vector) { (os_hooks)->interrupt_cpu(vm, logical_cpu, vector); } } -#endif int v3_vm_enter(struct guest_info * info) { - switch (v3_cpu_types[info->cpu_id]) { -#ifdef CONFIG_SVM + switch (v3_mach_type) { +#ifdef V3_CONFIG_SVM case V3_SVM_CPU: case V3_SVM_REV3_CPU: return v3_svm_enter(info); break; #endif -#if CONFIG_VMX +#if V3_CONFIG_VMX case V3_VMX_CPU: case V3_VMX_EPT_CPU: + case V3_VMX_EPT_UG_CPU: return v3_vmx_enter(info); break; #endif