palacios/src/palacios/vmm_shadow_paging.c

   1 /*
   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.
   5  *
   6  * The V3VEE Project is a joint project between Northwestern University
   7  * and the University of New Mexico.  You can find out more at
   8  * http://www.v3vee.org
   9  *
  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.
  13  *
  14  * Author: Jack Lange <jarusl@cs.northwestern.edu>
  15  *
  16  * This is free software.  You are permitted to use,
  17  * redistribute, and modify it as specified in the file "V3VEE_LICENSE".
  18  */
  19
  20
  21 #include <palacios/vmm_shadow_paging.h>
  22
  23
  24 #include <palacios/vmm.h>
  25 #include <palacios/vm_guest_mem.h>
  26 #include <palacios/vmm_decoder.h>
  27 #include <palacios/vmm_ctrl_regs.h>
  28
  29 #include <palacios/vmm_hashtable.h>
  30
  31 #include <palacios/vmm_direct_paging.h>
  32
  33
  34
  35
  36 #ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
  37 #include <palacios/vmm_telemetry.h>
  38 #endif
  39
  40 #ifdef V3_CONFIG_SYMBIOTIC_SWAP
  41 #include <palacios/vmm_sym_swap.h>
  42 #endif
  43
  44 #ifndef V3_CONFIG_DEBUG_SHADOW_PAGING
  45 #undef PrintDebug
  46 #define PrintDebug(fmt, args...)
  47 #endif
  48
  49
  50 static const char default_strategy[] = "VTLB";
  51
  52
  53 static struct hashtable * master_shdw_pg_table = NULL;
  54
  55
  56 struct event_callback {
  57     int (*callback)(struct guest_info *core, struct v3_shdw_pg_event *event, void *priv_data);
  58     void *priv_data;
  59
  60     struct list_head node;
  61 };
  62
  63 static uint_t shdw_pg_hash_fn(addr_t key) {
  64     char * name = (char *)key;
  65     return v3_hash_buffer((uint8_t *)name, strlen(name));
  66 }
  67
  68 static int shdw_pg_eq_fn(addr_t key1, addr_t key2) {
  69     char * name1 = (char *)key1;
  70     char * name2 = (char *)key2;
  71
  72     return (strcmp(name1, name2) == 0);
  73 }
  74
  75 static int have_callbacks(struct guest_info *core)
  76 {
  77     return !list_empty(&(core->vm_info->shdw_impl.event_callback_list));
  78 }
  79
  80 static void dispatch_event(struct guest_info *core, struct v3_shdw_pg_event *event)
  81 {
  82     struct event_callback *cb,*temp;
  83
  84     list_for_each_entry_safe(cb,
  85                              temp,
  86                              &(core->vm_info->shdw_impl.event_callback_list),
  87                              node) {
  88         cb->callback(core,event,cb->priv_data);
  89     }
  90 }
  91
  92
  93 int V3_init_shdw_paging() {
  94     extern struct v3_shdw_pg_impl * __start__v3_shdw_pg_impls[];
  95     extern struct v3_shdw_pg_impl * __stop__v3_shdw_pg_impls[];
  96     struct v3_shdw_pg_impl ** tmp_impl = __start__v3_shdw_pg_impls;
  97     int i = 0;
  98
  99     master_shdw_pg_table = v3_create_htable(0, shdw_pg_hash_fn, shdw_pg_eq_fn);
 100
 101
 102     while (tmp_impl != __stop__v3_shdw_pg_impls) {
 103         V3_Print(VM_NONE, VCORE_NONE, "Registering Shadow Paging Impl (%s)\n", (*tmp_impl)->name);
 104
 105         if (v3_htable_search(master_shdw_pg_table, (addr_t)((*tmp_impl)->name))) {
 106             PrintError(VM_NONE, VCORE_NONE, "Multiple instances of shadow paging impl (%s)\n", (*tmp_impl)->name);
 107             return -1;
 108         }
 109
 110         if (v3_htable_insert(master_shdw_pg_table,
 111                              (addr_t)((*tmp_impl)->name),
 112                              (addr_t)(*tmp_impl)) == 0) {
 113             PrintError(VM_NONE, VCORE_NONE, "Could not register shadow paging impl (%s)\n", (*tmp_impl)->name);
 114             return -1;
 115         }
 116
 117         tmp_impl = &(__start__v3_shdw_pg_impls[++i]);
 118     }
 119
 120     return 0;
 121 }
 122
 123 int V3_deinit_shdw_paging() {
 124     v3_free_htable(master_shdw_pg_table, 0, 0);
 125     return 0;
 126 }
 127
 128
 129
 130 /***
 131  ***  There be dragons
 132  ***/
 133
 134
 135 #ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
 136 static void telemetry_cb(struct v3_vm_info * vm, void * private_data, char * hdr) {
 137     int i = 0;
 138     for (i = 0; i < vm->num_cores; i++) {
 139         struct guest_info * core = &(vm->cores[i]);
 140
 141         V3_Print(vm, core, "%s Guest Page faults: %d\n", hdr, core->shdw_pg_state.guest_faults);
 142     }
 143 }
 144 #endif
 145
 146
 147
 148 int v3_init_shdw_pg_state(struct guest_info * core) {
 149     struct v3_shdw_pg_state * state = &(core->shdw_pg_state);
 150     struct v3_shdw_pg_impl * impl = core->vm_info->shdw_impl.current_impl;
 151
 152
 153     state->guest_cr3 = 0;
 154     state->guest_cr0 = 0;
 155     state->guest_efer.value = 0x0LL;
 156
 157     if (impl->local_init(core) == -1) {
 158         PrintError(core->vm_info, core, "Error in Shadow paging local initialization (%s)\n", impl->name);
 159         return -1;
 160     }
 161
 162
 163 #ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
 164     v3_add_telemetry_cb(core->vm_info, telemetry_cb, NULL);
 165 #endif
 166
 167
 168     return 0;
 169 }
 170
 171
 172 int v3_deinit_shdw_pg_state(struct guest_info * core) {
 173     struct v3_shdw_pg_impl * impl = NULL;
 174
 175     if (!core || !core->vm_info) {
 176         return -1;
 177     }
 178
 179     impl = core->vm_info->shdw_impl.current_impl;
 180
 181     if (impl && impl->local_deinit(core) == -1) {
 182         PrintError(core->vm_info, core, "Error deinitializing shadow paging state\n");
 183         return -1;
 184     }
 185
 186
 187     return 0;
 188 }
 189
 190
 191
 192 int v3_init_shdw_impl(struct v3_vm_info * vm) {
 193     struct v3_shdw_impl_state * impl_state = &(vm->shdw_impl);
 194     v3_cfg_tree_t * pg_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "paging");
 195     char * pg_mode = v3_cfg_val(pg_cfg, "mode");
 196     char * pg_strat = v3_cfg_val(pg_cfg, "strategy");
 197     struct v3_shdw_pg_impl * impl = NULL;
 198
 199     PrintDebug(vm, VCORE_NONE, "Checking if shadow paging requested.\n");
 200     if (pg_mode == NULL) {
 201         V3_Print(vm, VCORE_NONE, "No paging mode specified, assuming shadow with defaults\n");
 202         pg_mode = "shadow";
 203     } else {
 204         if (strcasecmp(pg_mode, "nested") == 0) {
 205             // this check is repeated here (compare to vmm_config's determine paging mode) since
 206             // shadow paging initialization *precedes* per-core pre-config.
 207             extern v3_cpu_arch_t v3_mach_type;
 208             if ((v3_mach_type == V3_SVM_REV3_CPU) ||
 209                 (v3_mach_type == V3_VMX_EPT_CPU) ||
 210                 (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
 211                 PrintDebug(vm, VCORE_NONE, "Nested paging specified on machine that supports it - not initializing shadow paging\n");
 212                 return 0;
 213             } else {
 214                 V3_Print(vm, VCORE_NONE, "Nested paging specified but machine does not support it - falling back to shadow paging with defaults\n");
 215                 pg_mode = "shadow";
 216             }
 217         } else if (strcasecmp(pg_mode, "shadow") != 0) {
 218             V3_Print(vm, VCORE_NONE, "Unknown paging mode '%s' specified - falling back to shadow paging with defaults\n",pg_mode);
 219             pg_mode = "shadow";
 220         }
 221     }
 222
 223     if (pg_strat == NULL) {
 224         pg_strat = (char *)default_strategy;
 225     }
 226
 227     V3_Print(vm, VCORE_NONE,"Initialization of Shadow Paging implementation\n");
 228
 229     impl = (struct v3_shdw_pg_impl *)v3_htable_search(master_shdw_pg_table, (addr_t)pg_strat);
 230
 231     if (impl == NULL) {
 232         PrintError(vm, VCORE_NONE, "Could not find shadow paging impl (%s)\n", pg_strat);
 233         return -1;
 234     }
 235
 236     INIT_LIST_HEAD(&(impl_state->event_callback_list));
 237
 238     impl_state->current_impl = impl;
 239
 240     if (impl->init(vm, pg_cfg) == -1) {
 241         PrintError(vm, VCORE_NONE, "Could not initialize Shadow paging implemenation (%s)\n", impl->name);
 242         return -1;
 243     }
 244
 245
 246
 247     return 0;
 248 }
 249
 250 int v3_deinit_shdw_impl(struct v3_vm_info * vm) {
 251     struct v3_shdw_pg_impl * impl = vm->shdw_impl.current_impl;
 252     struct event_callback *cb,*temp;
 253
 254     if (impl == NULL) {
 255         // Shadow paging not implemented
 256         return 0;
 257     }
 258
 259     if (impl->deinit(vm) == -1) {
 260         PrintError(vm, VCORE_NONE,"Error deinitializing shadow paging implementation\n");
 261         return -1;
 262     }
 263
 264     list_for_each_entry_safe(cb,
 265                              temp,
 266                              &(vm->shdw_impl.event_callback_list),
 267                              node) {
 268         list_del(&(cb->node));
 269         V3_Free(cb);
 270     }
 271
 272     return 0;
 273 }
 274
 275
 276 // Reads the guest CR3 register
 277 // creates new shadow page tables
 278 // updates the shadow CR3 register to point to the new pts
 279 int v3_activate_shadow_pt(struct guest_info * core) {
 280     struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
 281     struct v3_shdw_pg_impl * impl = state->current_impl;
 282
 283     if (!have_callbacks(core)) {
 284         return impl->activate_shdw_pt(core);
 285     } else {
 286         int rc;
 287         struct v3_shdw_pg_event event_pre={SHADOW_ACTIVATE,SHADOW_PREIMPL,0,{0,0,0,0,0,0}};
 288         struct v3_shdw_pg_event event_post={SHADOW_ACTIVATE,SHADOW_POSTIMPL,0,{0,0,0,0,0,0}};
 289
 290         dispatch_event(core,&event_pre);
 291
 292         rc =impl->activate_shdw_pt(core);
 293
 294         dispatch_event(core,&event_post);
 295
 296         return rc;
 297     }
 298 }
 299
 300
 301
 302 // This must flush any caches
 303 // and reset the cr3 value to the correct value
 304 int v3_invalidate_shadow_pts(struct guest_info * core) {
 305     struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
 306     struct v3_shdw_pg_impl * impl = state->current_impl;
 307
 308     if (!have_callbacks(core)) {
 309         return impl->invalidate_shdw_pt(core);
 310     } else {
 311         int rc;
 312         struct v3_shdw_pg_event event_pre={SHADOW_INVALIDATE,SHADOW_PREIMPL,0,{0,0,0,0,0,0}};
 313         struct v3_shdw_pg_event event_post={SHADOW_INVALIDATE,SHADOW_POSTIMPL,0,{0,0,0,0,0,0}};
 314
 315         dispatch_event(core,&event_pre);
 316
 317         rc = impl->invalidate_shdw_pt(core);
 318
 319         dispatch_event(core,&event_post);
 320
 321         return rc;
 322     }
 323 }
 324
 325
 326 int v3_handle_shadow_pagefault(struct guest_info * core, addr_t fault_addr, pf_error_t error_code)
 327 {
 328     int rc;
 329
 330
 331     if (have_callbacks(core)) {
 332         struct v3_shdw_pg_event event={SHADOW_PAGEFAULT,SHADOW_PREIMPL,fault_addr,error_code};
 333         dispatch_event(core,&event);
 334     }
 335
 336     if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) {
 337         // If paging is not turned on we need to handle the special cases
 338       rc = v3_handle_passthrough_pagefault(core, fault_addr, error_code,NULL,NULL);
 339     } else if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) {
 340         struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
 341         struct v3_shdw_pg_impl * impl = state->current_impl;
 342
 343         rc = impl->handle_pagefault(core, fault_addr, error_code);
 344     } else {
 345         PrintError(core->vm_info, core, "Invalid Memory mode\n");
 346         rc = -1;
 347     }
 348
 349     if (have_callbacks(core)) {
 350         struct v3_shdw_pg_event event={SHADOW_PAGEFAULT,SHADOW_POSTIMPL,fault_addr,error_code};
 351         dispatch_event(core,&event);
 352     }
 353
 354     return rc;
 355 }
 356
 357
 358 int v3_handle_shadow_invlpg(struct guest_info * core) {
 359     uchar_t instr[15];
 360     struct x86_instr dec_instr;
 361     int ret = 0;
 362     addr_t vaddr = 0;
 363
 364     if (v3_get_vm_mem_mode(core) != VIRTUAL_MEM) {
 365         // Paging must be turned on...
 366         // should handle with some sort of fault I think
 367         PrintError(core->vm_info, core, "ERROR: INVLPG called in non paged mode\n");
 368         return -1;
 369     }
 370
 371     if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) {
 372         ret = v3_read_gpa_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr);
 373     } else {
 374         ret = v3_read_gva_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr);
 375     }
 376
 377     if (ret == -1) {
 378         PrintError(core->vm_info, core, "Could not read instruction into buffer\n");
 379         return -1;
 380     }
 381
 382     if (v3_decode(core, (addr_t)instr, &dec_instr) == -1) {
 383         PrintError(core->vm_info, core, "Decoding Error\n");
 384         return -1;
 385     }
 386
 387     if ((dec_instr.op_type != V3_OP_INVLPG) ||
 388         (dec_instr.num_operands != 1) ||
 389         (dec_instr.dst_operand.type != MEM_OPERAND)) {
 390         PrintError(core->vm_info, core, "Decoder Error: Not a valid INVLPG instruction...\n");
 391         return -1;
 392     }
 393
 394     vaddr = dec_instr.dst_operand.operand;
 395
 396     core->rip += dec_instr.instr_length;
 397
 398     {
 399         struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
 400         struct v3_shdw_pg_impl * impl = state->current_impl;
 401         int rc;
 402
 403         if (have_callbacks(core)) {
 404             struct v3_shdw_pg_event event={SHADOW_INVLPG,SHADOW_PREIMPL,vaddr,{0,0,0,0,0,0}};
 405             dispatch_event(core,&event);
 406         }
 407
 408         rc=impl->handle_invlpg(core, vaddr);
 409
 410         if (have_callbacks(core)) {
 411             struct v3_shdw_pg_event event={SHADOW_INVLPG,SHADOW_POSTIMPL,vaddr,{0,0,0,0,0,0}};
 412             dispatch_event(core,&event);
 413         }
 414
 415         return rc;
 416     }
 417 }
 418
 419
 420
 421
 422
 423
 424 int v3_inject_guest_pf(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
 425     core->ctrl_regs.cr2 = fault_addr;
 426
 427 #ifdef V3_CONFIG_SHADOW_PAGING_TELEMETRY
 428     core->shdw_pg_state.guest_faults++;
 429 #endif
 430
 431     return v3_raise_exception_with_error(core, PF_EXCEPTION, *(uint_t *)&error_code);
 432 }
 433
 434
 435 int v3_is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access) {
 436     /* basically the reasoning is that there can be multiple reasons for a page fault:
 437        If there is a permissions failure for a page present in the guest _BUT_
 438        the reason for the fault was that the page is not present in the shadow,
 439        _THEN_ we have to map the shadow page in and reexecute, this will generate
 440        a permissions fault which is _THEN_ valid to send to the guest
 441        _UNLESS_ both the guest and shadow have marked the page as not present
 442
 443        whew...
 444     */
 445     if (guest_access != PT_ACCESS_OK) {
 446         // Guest Access Error
 447
 448         if ((shadow_access != PT_ACCESS_NOT_PRESENT) &&
 449             (guest_access != PT_ACCESS_NOT_PRESENT)) {
 450             // aka (guest permission error)
 451             return 1;
 452         }
 453
 454         /*
 455           if ((shadow_access == PT_ACCESS_NOT_PRESENT) &&
 456           (guest_access == PT_ACCESS_NOT_PRESENT)) {
 457           // Page tables completely blank, handle guest first
 458           return 1;
 459           }
 460         */
 461
 462         if (guest_access == PT_ACCESS_NOT_PRESENT) {
 463             // Page tables completely blank, handle guest first
 464             return 1;
 465         }
 466
 467         // Otherwise we'll handle the guest fault later...?
 468     }
 469
 470     return 0;
 471 }
 472
 473
 474 int v3_register_shadow_paging_event_callback(struct v3_vm_info *vm,
 475                                              int (*callback)(struct guest_info *core,
 476                                                              struct v3_shdw_pg_event *event,
 477                                                              void      *priv_data),
 478                                              void *priv_data)
 479 {
 480     struct event_callback *ec = V3_Malloc(sizeof(struct event_callback));
 481
 482     if (!ec) {
 483         PrintError(vm, VCORE_NONE, "Unable to allocate for a shadow paging event callback\n");
 484         return -1;
 485     }
 486
 487     ec->callback = callback;
 488     ec->priv_data = priv_data;
 489
 490     list_add(&(ec->node),&(vm->shdw_impl.event_callback_list));
 491
 492     return 0;
 493
 494 }
 495
 496 int v3_unregister_shadow_paging_event_callback(struct v3_vm_info *vm,
 497                                                int (*callback)(struct guest_info *core,
 498                                                                struct v3_shdw_pg_event *event,
 499                                                                void      *priv_data),
 500                                                void *priv_data)
 501 {
 502     struct event_callback *cb,*temp;
 503
 504     list_for_each_entry_safe(cb,
 505                              temp,
 506                              &(vm->shdw_impl.event_callback_list),
 507                              node) {
 508         if ((callback == cb->callback) && (priv_data == cb->priv_data)) {
 509             list_del(&(cb->node));
 510             V3_Free(cb);
 511             return 0;
 512         }
 513     }
 514
 515     PrintError(vm, VCORE_NONE, "No callback found!\n");
 516
 517     return -1;
 518 }
 519
 520