X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm.c;h=e16e2fa60f580e7b36f00f8e3fe1f66b393b7d4d;hb=928297a7cf5ffbf3a210fcaf7556e681bbb436d5;hp=a162da5ab3dc644a5bf3178f0dfa137c12a1f75b;hpb=2fff50d3e72abf29655326449ed4dc5cf6e8f429;p=palacios.git diff --git a/palacios/src/palacios/vmm.c b/palacios/src/palacios/vmm.c index a162da5..e16e2fa 100644 --- a/palacios/src/palacios/vmm.c +++ b/palacios/src/palacios/vmm.c @@ -18,6 +18,7 @@ */ #include +#include #include #include #include @@ -28,6 +29,9 @@ #include #include #include +#include +#include +#include #ifdef V3_CONFIG_SVM #include @@ -49,10 +53,11 @@ int v3_dbg_enable = 0; - static void init_cpu(void * arg) { uint32_t cpu_id = (uint32_t)(addr_t)arg; + v3_init_fp(); + #ifdef V3_CONFIG_SVM if (v3_is_svm_capable()) { PrintDebug(VM_NONE, VCORE_NONE, "Machine is SVM Capable\n"); @@ -98,7 +103,21 @@ static void deinit_cpu(void * arg) { PrintError(VM_NONE, VCORE_NONE, "CPU has no virtualization Extensions\n"); break; } + + v3_deinit_fp(); + +} + + +static int in_long_mode() +{ + uint32_t high, low; + + v3_get_msr(0xc0000080,&high,&low); // EFER + + return ((low & 0x500)== 0x500); // LMA and LME set } + void Init_V3(struct v3_os_hooks * hooks, char * cpu_mask, int num_cpus, char *options) { int i = 0; @@ -107,9 +126,24 @@ void Init_V3(struct v3_os_hooks * hooks, char * cpu_mask, int num_cpus, char *op V3_Print(VM_NONE, VCORE_NONE, "V3 Print statement to fix a Kitten page fault bug\n"); + + +#ifndef __V3_64BIT__ +#error Palacios does not support compilation for a 32 bit host OS!!!! +#else + if (!in_long_mode()) { + PrintError(VM_NONE,VCORE_NONE,"Palacios supports execution only in long mode (64 bit).\n"); + return; + } +#endif + // Set global variables. os_hooks = hooks; + if (num_cpus>V3_CONFIG_MAX_CPUS) { + PrintError(VM_NONE,VCORE_NONE, "Requesting as many as %d cpus, but Palacios is compiled for a maximum of %d. Only the first %d cpus will be considered\n", num_cpus, V3_CONFIG_MAX_CPUS, V3_CONFIG_MAX_CPUS); + } + // Determine the global machine type v3_mach_type = V3_INVALID_CPU; @@ -120,12 +154,19 @@ void Init_V3(struct v3_os_hooks * hooks, char * cpu_mask, int num_cpus, char *op // Parse host-os defined options into an easily-accessed format. v3_parse_options(options); + // Memory manager initialization + v3_init_mem(); + // Register all the possible device types V3_init_devices(); // Register all shadow paging handlers V3_init_shdw_paging(); +#ifdef V3_CONFIG_SWAPPING + v3_init_swapping(); +#endif + // Initialize the cpu_mapper framework (must be before extensions) V3_init_cpu_mapper(); @@ -152,7 +193,7 @@ void Init_V3(struct v3_os_hooks * hooks, char * cpu_mask, int num_cpus, char *op if ((hooks) && (hooks->call_on_cpu)) { - for (i = 0; i < num_cpus; i++) { + for (i = 0; i < num_cpus && i < V3_CONFIG_MAX_CPUS; i++) { major = i / 8; minor = i % 8; @@ -173,19 +214,9 @@ void Init_V3(struct v3_os_hooks * hooks, char * cpu_mask, int num_cpus, char *op void Shutdown_V3() { int i; - V3_deinit_devices(); - V3_deinit_shdw_paging(); - - V3_deinit_extensions(); - -#ifdef V3_CONFIG_SYMMOD - V3_deinit_symmod(); -#endif - -#ifdef V3_CONFIG_CHECKPOINT - V3_deinit_checkpoint(); -#endif + // Reverse order of Init_V3 + // bring down CPUs if ((os_hooks) && (os_hooks->call_on_cpu)) { for (i = 0; i < V3_CONFIG_MAX_CPUS; i++) { @@ -196,6 +227,37 @@ void Shutdown_V3() { } } +#ifdef V3_CONFIG_CHECKPOINT + V3_deinit_checkpoint(); +#endif + +#ifdef V3_CONFIG_SYMMOD + V3_deinit_symmod(); +#endif + + V3_disable_scheduler(); + + V3_disable_cpu_mapper(); + + V3_deinit_extensions(); + + V3_deinit_scheduling(); + + V3_deinit_cpu_mapper(); + +#ifdef V3_CONFIG_SWAPPING + v3_deinit_swapping(); +#endif + + V3_deinit_shdw_paging(); + + V3_deinit_devices(); + + v3_deinit_mem(); + + v3_deinit_options(); + + } @@ -223,6 +285,11 @@ struct v3_vm_info * v3_create_vm(void * cfg, void * priv_data, char * name) { memset(vm->name, 0, 128); strncpy(vm->name, name, 127); + if(v3_cpu_mapper_register_vm(vm) == -1) { + + PrintError(vm, VCORE_NONE,"Error registering VM with cpu_mapper\n"); + } + /* * Register this VM with the palacios scheduler. It will ask for admission * prior to launch. @@ -278,25 +345,11 @@ 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, VCORE_NONE, "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;inum_cores;i++) { - if (vm->cores[i].shdw_pg_mode == SHADOW_PAGING) { - if ((vm->mem_map.base_region.host_addr + vm->mem_size ) >= 0x100000000ULL) { - PrintError(vm, VCORE_NONE, "Base memory region exceeds 4 GB boundary with shadow paging enabled on core %d.\n",i); - PrintError(vm, VCORE_NONE, "Any use of non-64 bit mode in the guest is likely to fail in this configuration.\n"); - PrintError(vm, VCORE_NONE, "If you would like to proceed anyway, remove this check and recompile Palacios.\n"); - PrintError(vm, VCORE_NONE, "Alternatively, change this VM to use nested paging.\n"); - return -1; - } - } - } /// CHECK IF WE ARE MULTICORE ENABLED.... @@ -320,21 +373,16 @@ int v3_start_vm(struct v3_vm_info * vm, unsigned int cpu_mask) { vm->avail_cores = avail_cores; - if (v3_scheduler_admit_vm(vm) != 0){ - PrintError(vm, VCORE_NONE,"Error admitting VM %s for scheduling", vm->name); + if (v3_cpu_mapper_admit_vm(vm,cpu_mask) != 0){ + PrintError(vm, VCORE_NONE,"Error admitting VM %s for mapping", vm->name); } - if (v3_cpu_mapper_admit_vm(vm) != 0){ - PrintError(vm, VCORE_NONE,"Error admitting VM %s for mapping", vm->name); + if (v3_scheduler_admit_vm(vm) != 0){ + PrintError(vm, VCORE_NONE,"Error admitting VM %s for scheduling", vm->name); } vm->run_state = VM_RUNNING; - if(v3_cpu_mapper_register_vm(vm,cpu_mask) == -1) { - - PrintError(vm, VCORE_NONE,"Error registering VM with cpu_mapper\n"); - } - for (vcore_id = 0; vcore_id < vm->num_cores; vcore_id++) { @@ -348,7 +396,14 @@ int v3_start_vm(struct v3_vm_info * vm, unsigned int cpu_mask) { PrintDebug(vm, VCORE_NONE, "run: core=%u, func=0x%p, arg=0x%p, name=%s\n", core->pcpu_id, start_core, core, core->exec_name); - core->core_run_state = CORE_STOPPED; // core zero will turn itself on + if (core->core_run_state==CORE_INVALID) { + // launch of a fresh VM + core->core_run_state = CORE_STOPPED; + // core zero will turn itself on + } else { + // this is a resume - use whatever its current run_state is + } + core->core_thread = V3_CREATE_THREAD_ON_CPU(core->pcpu_id, start_core, core, core->exec_name); if (core->core_thread == NULL) { @@ -461,10 +516,124 @@ int v3_move_vm_core(struct v3_vm_info * vm, int vcore_id, int target_cpu) { return 0; } +/* move a memory region to memory with affinity for a specific physical core */ +int v3_move_vm_mem(struct v3_vm_info * vm, void *gpa, int target_cpu) { + int old_node; + int new_node; + struct v3_mem_region *reg; + void *new_hpa; + int num_pages; + void *old_hpa; + int i; + + old_node = v3_numa_gpa_to_node(vm,(addr_t)gpa); + + if (old_node<0) { + PrintError(vm, VCORE_NONE, "Cannot determine current node of gpa %p\n",gpa); + return -1; + } + + new_node = v3_numa_cpu_to_node(target_cpu); + + if (new_node<0) { + PrintError(vm, VCORE_NONE, "Cannot determine current node of cpu %d\n",target_cpu); + return -1; + } + + if (new_node==old_node) { + PrintDebug(vm, VCORE_NONE, "Affinity is already established - ignoring request\n"); + return 0; + } + + // We are now going to change the universe, so + // we'll barrier everyone first + + while (v3_raise_barrier(vm, NULL) == -1); + + // get region + + reg = v3_get_mem_region(vm, V3_MEM_CORE_ANY, (addr_t) gpa); + + if (!reg) { + PrintError(vm, VCORE_NONE, "Attempt to migrate non-existent memory\n"); + goto out_fail; + } + + if (!(reg->flags.base) || !(reg->flags.alloced)) { + PrintError(vm, VCORE_NONE, "Attempt to migrate invalid region: base=%d alloced=%d\n", reg->flags.base, reg->flags.alloced); + goto out_fail; + } + + // we now have the allocated base region corresponding to - and not a copy + // we will rewrite this region after moving its contents + + // first, let's double check that we are in fact changing the numa_id... + + if (reg->numa_id==new_node) { + PrintDebug(vm, VCORE_NONE, "Affinity for this base region is already established - ignoring...\n"); + goto out_success; + } + + // region uses exclusive addressing [guest_start,guest_end) + num_pages = (reg->guest_end-reg->guest_start)/PAGE_SIZE; + + new_hpa = V3_AllocPagesExtended(num_pages, + PAGE_SIZE_4KB, + new_node, + 0); // no constraints given new shadow pager impl + + if (!new_hpa) { + PrintError(vm, VCORE_NONE, "Cannot allocate memory for new base region...\n"); + goto out_fail; + } + + // Note, assumes virtual contiguity in the host OS... + memcpy(V3_VAddr((void*)new_hpa), V3_VAddr((void*)(reg->host_addr)), num_pages*PAGE_SIZE); + + old_hpa = (void*)(reg->host_addr); + old_node = (int)(reg->numa_id); + reg->host_addr = (addr_t)new_hpa; + reg->numa_id = v3_numa_hpa_to_node((addr_t)new_hpa); + + // flush all page tables / kill all humans + + for (i=0;inum_cores;i++) { + if (vm->cores[i].shdw_pg_mode==SHADOW_PAGING) { + v3_invalidate_shadow_pts(&(vm->cores[i])); + } else if (vm->cores[i].shdw_pg_mode==NESTED_PAGING) { + // nested invalidator uses inclusive addressing [start,end], not [start,end) + v3_invalidate_nested_addr_range(&(vm->cores[i]),reg->guest_start,reg->guest_end-1,NULL,NULL); + } else { + PrintError(vm,VCORE_NONE, "Cannot determine how to invalidate paging structures! Reverting to previous region.\n"); + // We'll restore things... + reg->host_addr = (addr_t) old_hpa; + reg->numa_id = old_node; + V3_FreePages(new_hpa,num_pages); + goto out_fail; + } + } + + // Now the old region can go away... + V3_FreePages(old_hpa,num_pages); + + PrintDebug(vm,VCORE_NONE,"Migration of memory complete - new region is %p to %p\n", + (void*)(reg->host_addr),(void*)(reg->host_addr+num_pages*PAGE_SIZE-1)); + + out_success: + v3_lower_barrier(vm); + return 0; + + + out_fail: + v3_lower_barrier(vm); + return -1; +} int v3_stop_vm(struct v3_vm_info * vm) { + struct guest_info * running_core; + if ((vm->run_state != VM_RUNNING) && (vm->run_state != VM_SIMULATING)) { PrintError(vm, VCORE_NONE,"Tried to stop VM in invalid runstate (%d)\n", vm->run_state); @@ -486,6 +655,7 @@ int v3_stop_vm(struct v3_vm_info * vm) { for (i = 0; i < vm->num_cores; i++) { if (vm->cores[i].core_run_state != CORE_STOPPED) { + running_core = &vm->cores[i]; still_running = 1; } } @@ -494,7 +664,7 @@ int v3_stop_vm(struct v3_vm_info * vm) { break; } - v3_yield(NULL,-1); + v3_scheduler_stop_core(running_core); } V3_Print(vm, VCORE_NONE,"VM stopped. Returning\n"); @@ -542,6 +712,7 @@ static int sim_callback(struct guest_info * core, void * private_data) { V3_Print(core->vm_info, core, "Simulation callback activated (guest_rip=%p)\n", (void *)core->rip); while (v3_bitmap_check(timeout_map, core->vcpu_id) == 1) { + // We spin here if there is noone to yield to v3_yield(NULL,-1); } @@ -612,7 +783,8 @@ int v3_simulate_vm(struct v3_vm_info * vm, unsigned int msecs) { if (all_blocked == 1) { break; } - + + // Intentionally spin if there is no one to yield to v3_yield(NULL,-1); } @@ -633,59 +805,76 @@ int v3_simulate_vm(struct v3_vm_info * vm, unsigned int msecs) { } -int v3_get_state_vm(struct v3_vm_info *vm, struct v3_vm_state *s) +int v3_get_state_vm(struct v3_vm_info *vm, + struct v3_vm_base_state *base, + struct v3_vm_core_state *core, + struct v3_vm_mem_state *mem) { - uint32_t i; - uint32_t numcores = s->num_vcores > vm->num_cores ? vm->num_cores : s->num_vcores; - - switch (vm->run_state) { - case VM_INVALID: s->state = V3_VM_INVALID; break; - case VM_RUNNING: s->state = V3_VM_RUNNING; break; - case VM_STOPPED: s->state = V3_VM_STOPPED; break; - case VM_PAUSED: s->state = V3_VM_PAUSED; break; - case VM_ERROR: s->state = V3_VM_ERROR; break; - case VM_SIMULATING: s->state = V3_VM_SIMULATING; break; - default: s->state = V3_VM_UNKNOWN; break; - } + uint32_t i; + uint32_t numcores = core->num_vcores > vm->num_cores ? vm->num_cores : core->num_vcores; + uint32_t numregions = mem->num_regions > vm->mem_map.num_base_regions ? vm->mem_map.num_base_regions : mem->num_regions; + extern uint64_t v3_mem_block_size; + + switch (vm->run_state) { + case VM_INVALID: base->state = V3_VM_INVALID; break; + case VM_RUNNING: base->state = V3_VM_RUNNING; break; + case VM_STOPPED: base->state = V3_VM_STOPPED; break; + case VM_PAUSED: base->state = V3_VM_PAUSED; break; + case VM_ERROR: base->state = V3_VM_ERROR; break; + case VM_SIMULATING: base->state = V3_VM_SIMULATING; break; + default: base->state = V3_VM_UNKNOWN; break; + } - s->mem_base_paddr = (void*)(vm->mem_map.base_region.host_addr); - s->mem_size = vm->mem_size; - - s->num_vcores = numcores; - - for (i=0;icores[i].core_run_state) { - case CORE_INVALID: s->vcore[i].state = V3_VCORE_INVALID; break; - case CORE_RUNNING: s->vcore[i].state = V3_VCORE_RUNNING; break; - case CORE_STOPPED: s->vcore[i].state = V3_VCORE_STOPPED; break; - default: s->vcore[i].state = V3_VCORE_UNKNOWN; break; - } - switch (vm->cores[i].cpu_mode) { - case REAL: s->vcore[i].cpu_mode = V3_VCORE_CPU_REAL; break; - case PROTECTED: s->vcore[i].cpu_mode = V3_VCORE_CPU_PROTECTED; break; - case PROTECTED_PAE: s->vcore[i].cpu_mode = V3_VCORE_CPU_PROTECTED_PAE; break; - case LONG: s->vcore[i].cpu_mode = V3_VCORE_CPU_LONG; break; - case LONG_32_COMPAT: s->vcore[i].cpu_mode = V3_VCORE_CPU_LONG_32_COMPAT; break; - case LONG_16_COMPAT: s->vcore[i].cpu_mode = V3_VCORE_CPU_LONG_16_COMPAT; break; - default: s->vcore[i].cpu_mode = V3_VCORE_CPU_UNKNOWN; break; - } - switch (vm->cores[i].shdw_pg_mode) { - case SHADOW_PAGING: s->vcore[i].mem_state = V3_VCORE_MEM_STATE_SHADOW; break; - case NESTED_PAGING: s->vcore[i].mem_state = V3_VCORE_MEM_STATE_NESTED; break; - default: s->vcore[i].mem_state = V3_VCORE_MEM_STATE_UNKNOWN; break; - } - switch (vm->cores[i].mem_mode) { - case PHYSICAL_MEM: s->vcore[i].mem_mode = V3_VCORE_MEM_MODE_PHYSICAL; break; - case VIRTUAL_MEM: s->vcore[i].mem_mode=V3_VCORE_MEM_MODE_VIRTUAL; break; - default: s->vcore[i].mem_mode=V3_VCORE_MEM_MODE_UNKNOWN; break; - } - - s->vcore[i].pcore=vm->cores[i].pcpu_id; - s->vcore[i].last_rip=(void*)(vm->cores[i].rip); - s->vcore[i].num_exits=vm->cores[i].num_exits; - } + for (i=0;icores[i].core_run_state) { + case CORE_INVALID: core->vcore[i].state = V3_VCORE_INVALID; break; + case CORE_RUNNING: core->vcore[i].state = V3_VCORE_RUNNING; break; + case CORE_STOPPED: core->vcore[i].state = V3_VCORE_STOPPED; break; + default: core->vcore[i].state = V3_VCORE_UNKNOWN; break; + } + switch (vm->cores[i].cpu_mode) { + case REAL: core->vcore[i].cpu_mode = V3_VCORE_CPU_REAL; break; + case PROTECTED: core->vcore[i].cpu_mode = V3_VCORE_CPU_PROTECTED; break; + case PROTECTED_PAE: core->vcore[i].cpu_mode = V3_VCORE_CPU_PROTECTED_PAE; break; + case LONG: core->vcore[i].cpu_mode = V3_VCORE_CPU_LONG; break; + case LONG_32_COMPAT: core->vcore[i].cpu_mode = V3_VCORE_CPU_LONG_32_COMPAT; break; + case LONG_16_COMPAT: core->vcore[i].cpu_mode = V3_VCORE_CPU_LONG_16_COMPAT; break; + default: core->vcore[i].cpu_mode = V3_VCORE_CPU_UNKNOWN; break; + } + switch (vm->cores[i].shdw_pg_mode) { + case SHADOW_PAGING: core->vcore[i].mem_state = V3_VCORE_MEM_STATE_SHADOW; break; + case NESTED_PAGING: core->vcore[i].mem_state = V3_VCORE_MEM_STATE_NESTED; break; + default: core->vcore[i].mem_state = V3_VCORE_MEM_STATE_UNKNOWN; break; + } + switch (vm->cores[i].mem_mode) { + case PHYSICAL_MEM: core->vcore[i].mem_mode = V3_VCORE_MEM_MODE_PHYSICAL; break; + case VIRTUAL_MEM: core->vcore[i].mem_mode=V3_VCORE_MEM_MODE_VIRTUAL; break; + default: core->vcore[i].mem_mode=V3_VCORE_MEM_MODE_UNKNOWN; break; + } + + core->vcore[i].pcore=vm->cores[i].pcpu_id; + core->vcore[i].last_rip=(void*)(vm->cores[i].rip); + core->vcore[i].num_exits=vm->cores[i].num_exits; + } + + core->num_vcores=numcores; + + for (i=0;imem_map.num_base_regions;i++) { + mem->region[i].host_paddr = (void*)(vm->mem_map.base_regions[i].host_addr); + mem->region[i].size = v3_mem_block_size; +#ifdef V3_CONFIG_SWAPPING + mem->region[i].swapped = vm->mem_map.base_regions[i].flags.swapped; + mem->region[i].pinned = vm->mem_map.base_regions[i].flags.pinned; +#else + mem->region[i].swapped = 0; + mem->region[i].pinned = 0; +#endif - return 0; + } + + mem->num_regions=numregions; + + return 0; } @@ -729,10 +918,12 @@ int v3_free_vm(struct v3_vm_info * vm) { // free cores for (i = 0; i < vm->num_cores; i++) { + v3_scheduler_free_core(&(vm->cores[i])); v3_free_core(&(vm->cores[i])); } // free vm + v3_scheduler_free_vm(vm); v3_free_vm_internal(vm); v3_free_config(vm);