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_mem.h>
21 #include <palacios/vmm.h>
22 #include <palacios/vmm_util.h>
23 #include <palacios/vmm_emulator.h>
24 #include <palacios/vm_guest.h>
25 #include <palacios/vmm_debug.h>
27 #include <palacios/vmm_shadow_paging.h>
28 #include <palacios/vmm_direct_paging.h>
33 static int mem_offset_hypercall(struct guest_info * info, uint_t hcall_id, void * private_data) {
34 PrintDebug("V3Vee: Memory offset hypercall (offset=%p)\n",
35 (void *)(info->vm_info->mem_map.base_region.host_addr));
37 info->vm_regs.rbx = info->vm_info->mem_map.base_region.host_addr;
42 static int unhandled_err(struct guest_info * core, addr_t guest_va, addr_t guest_pa,
43 struct v3_mem_region * reg, pf_error_t access_info) {
45 PrintError("Unhandled memory access error (gpa=%p, gva=%p, error_code=%d)\n",
46 (void *)guest_pa, (void *)guest_va, *(uint32_t *)&access_info);
48 v3_print_mem_map(core->vm_info);
50 v3_print_guest_state(core);
55 int v3_init_mem_map(struct v3_vm_info * vm) {
56 struct v3_mem_map * map = &(vm->mem_map);
57 addr_t mem_pages = vm->mem_size >> 12;
59 memset(&(map->base_region), 0, sizeof(struct v3_mem_region));
61 map->mem_regions.rb_node = NULL;
63 // There is an underlying region that contains all of the guest memory
64 // PrintDebug("Mapping %d pages of memory (%u bytes)\n", (int)mem_pages, (uint_t)info->mem_size);
66 // 2MB page alignment needed for 2MB hardware nested paging
67 map->base_region.guest_start = 0;
68 map->base_region.guest_end = mem_pages * PAGE_SIZE_4KB;
70 #ifdef V3_CONFIG_ALIGNED_PG_ALLOC
71 map->base_region.host_addr = (addr_t)V3_AllocAlignedPages(mem_pages, vm->mem_align);
73 map->base_region.host_addr = (addr_t)V3_AllocPages(mem_pages);
76 if ((void*)map->base_region.host_addr == NULL) {
77 PrintError("Could not allocate guest memory\n");
81 // Clear the memory...
82 memset(V3_VAddr((void *)map->base_region.host_addr), 0, mem_pages * PAGE_SIZE_4KB);
85 map->base_region.flags.read = 1;
86 map->base_region.flags.write = 1;
87 map->base_region.flags.exec = 1;
88 map->base_region.flags.base = 1;
89 map->base_region.flags.alloced = 1;
91 map->base_region.unhandled = unhandled_err;
93 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
99 void v3_delete_mem_map(struct v3_vm_info * vm) {
100 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
101 struct v3_mem_region * reg;
102 struct rb_node * tmp_node = NULL;
103 addr_t mem_pages = vm->mem_size >> 12;
106 reg = rb_entry(node, struct v3_mem_region, tree_node);
108 node = v3_rb_next(node);
110 v3_delete_mem_region(vm, reg);
113 V3_FreePages((void *)(vm->mem_map.base_region.host_addr), mem_pages);
117 struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
118 addr_t guest_addr_start, addr_t guest_addr_end) {
119 struct v3_mem_region * entry = NULL;
121 if (guest_addr_start >= guest_addr_end) {
122 PrintError("Region start is after region end\n");
126 entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
127 memset(entry, 0, sizeof(struct v3_mem_region));
129 entry->guest_start = guest_addr_start;
130 entry->guest_end = guest_addr_end;
131 entry->core_id = core_id;
132 entry->unhandled = unhandled_err;
140 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
141 addr_t guest_addr_start,
142 addr_t guest_addr_end,
145 struct v3_mem_region * entry = NULL;
147 entry = v3_create_mem_region(vm, core_id,
151 entry->host_addr = host_addr;
153 entry->flags.read = 1;
154 entry->flags.write = 1;
155 entry->flags.exec = 1;
156 entry->flags.alloced = 1;
158 if (v3_insert_mem_region(vm, entry) == -1) {
169 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
170 struct v3_mem_region * region) {
171 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
172 struct rb_node * parent = NULL;
173 struct v3_mem_region * tmp_region;
177 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
179 if (region->guest_end <= tmp_region->guest_start) {
181 } else if (region->guest_start >= tmp_region->guest_end) {
184 if ((region->guest_end != tmp_region->guest_end) ||
185 (region->guest_start != tmp_region->guest_start)) {
186 PrintError("Trying to map a partial overlapped core specific page...\n");
187 return tmp_region; // This is ugly...
188 } else if (region->core_id == tmp_region->core_id) {
190 } else if (region->core_id < tmp_region->core_id) {
198 rb_link_node(&(region->tree_node), parent, p);
205 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
206 struct v3_mem_region * ret;
209 if ((ret = __insert_mem_region(vm, region))) {
213 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
217 for (i = 0; i < vm->num_cores; i++) {
218 struct guest_info * info = &(vm->cores[i]);
220 // flush virtual page tables
221 // 3 cases shadow, shadow passthrough, and nested
223 if (info->shdw_pg_mode == SHADOW_PAGING) {
224 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
226 if (mem_mode == PHYSICAL_MEM) {
229 for (cur_addr = region->guest_start;
230 cur_addr < region->guest_end;
231 cur_addr += PAGE_SIZE_4KB) {
232 v3_invalidate_passthrough_addr(info, cur_addr);
235 v3_invalidate_shadow_pts(info);
238 } else if (info->shdw_pg_mode == NESTED_PAGING) {
241 for (cur_addr = region->guest_start;
242 cur_addr < region->guest_end;
243 cur_addr += PAGE_SIZE_4KB) {
245 v3_invalidate_nested_addr(info, cur_addr);
256 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
257 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
258 struct v3_mem_region * reg = NULL;
262 reg = rb_entry(n, struct v3_mem_region, tree_node);
264 if (guest_addr < reg->guest_start) {
266 } else if (guest_addr >= reg->guest_end) {
269 if (reg->core_id == V3_MEM_CORE_ANY) {
270 // found relevant region, it's available on all cores
272 } else if (core_id == reg->core_id) {
273 // found relevant region, it's available on the indicated core
275 } else if (core_id < reg->core_id) {
276 // go left, core too big
278 } else if (core_id > reg->core_id) {
279 // go right, core too small
282 PrintDebug("v3_get_mem_region: Impossible!\n");
289 // There is not registered region, so we check if its a valid address in the base region
291 if (guest_addr > vm->mem_map.base_region.guest_end) {
292 PrintError("Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p) (core=0x%x)\n",
293 (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end, core_id);
294 v3_print_mem_map(vm);
299 return &(vm->mem_map.base_region);
304 /* This returns the next memory region based on a given address.
305 * If the address falls inside a sub region, that region is returned.
306 * If the address falls outside a sub region, the next sub region is returned
307 * NOTE that we have to be careful about core_ids here...
309 static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
310 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
311 struct v3_mem_region * reg = NULL;
312 struct v3_mem_region * parent = NULL;
320 reg = rb_entry(n, struct v3_mem_region, tree_node);
322 if (guest_addr < reg->guest_start) {
324 } else if (guest_addr >= reg->guest_end) {
327 if (reg->core_id == V3_MEM_CORE_ANY) {
328 // found relevant region, it's available on all cores
330 } else if (core_id == reg->core_id) {
331 // found relevant region, it's available on the indicated core
333 } else if (core_id < reg->core_id) {
334 // go left, core too big
336 } else if (core_id > reg->core_id) {
337 // go right, core too small
340 PrintError("v3_get_mem_region: Impossible!\n");
345 if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
351 if (parent->guest_start > guest_addr) {
353 } else if (parent->guest_end < guest_addr) {
354 struct rb_node * node = &(parent->tree_node);
356 while ((node = v3_rb_next(node)) != NULL) {
357 struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
359 if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
360 (next_reg->core_id == core_id)) {
362 // This check is not strictly necessary, but it makes it clearer
363 if (next_reg->guest_start > guest_addr) {
376 /* Given an address region of memory, find if there are any regions that overlap with it.
377 * This checks that the range lies in a single region, and returns that region if it does,
378 * this can be either the base region or a sub region.
379 * IF there are multiple regions in the range then it returns NULL
381 static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
382 addr_t start_gpa, addr_t end_gpa) {
383 struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
385 if (start_region == NULL) {
386 PrintError("Invalid memory region\n");
391 if (start_region->guest_end < end_gpa) {
392 // Region ends before range
394 } else if (start_region->flags.base == 0) {
395 // sub region overlaps range
398 // Base region, now we have to scan forward for the next sub region
399 struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
401 if (next_reg == NULL) {
402 // no sub regions after start_addr, base region is ok
404 } else if (next_reg->guest_start >= end_gpa) {
405 // Next sub region begins outside range
413 // Should never get here
421 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
429 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
433 // If the guest isn't running then there shouldn't be anything to invalidate.
434 // Page tables should __always__ be created on demand during execution
435 // NOTE: This is a sanity check, and can be removed if that assumption changes
436 if (vm->run_state != VM_RUNNING) {
441 for (i = 0; i < vm->num_cores; i++) {
442 struct guest_info * info = &(vm->cores[i]);
444 // flush virtual page tables
445 // 3 cases shadow, shadow passthrough, and nested
447 if (info->shdw_pg_mode == SHADOW_PAGING) {
448 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
450 if (mem_mode == PHYSICAL_MEM) {
453 for (cur_addr = reg->guest_start;
454 cur_addr < reg->guest_end;
455 cur_addr += PAGE_SIZE_4KB) {
456 v3_invalidate_passthrough_addr(info, cur_addr);
459 v3_invalidate_shadow_pts(info);
462 } else if (info->shdw_pg_mode == NESTED_PAGING) {
465 for (cur_addr = reg->guest_start;
466 cur_addr < reg->guest_end;
467 cur_addr += PAGE_SIZE_4KB) {
469 v3_invalidate_nested_addr(info, cur_addr);
476 // flush virtual page tables
477 // 3 cases shadow, shadow passthrough, and nested
481 // Determine if a given address can be handled by a large page of the requested size
482 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
485 uint32_t page_size = PAGE_SIZE_4KB;
486 struct v3_mem_region * reg = NULL;
490 if (core->use_large_pages == 1) {
491 pg_start = PAGE_ADDR_4MB(page_addr);
492 pg_end = (pg_start + PAGE_SIZE_4MB);
494 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
496 if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
497 page_size = PAGE_SIZE_4MB;
502 if (core->use_large_pages == 1) {
503 pg_start = PAGE_ADDR_2MB(page_addr);
504 pg_end = (pg_start + PAGE_SIZE_2MB);
506 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
508 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
509 page_size = PAGE_SIZE_2MB;
516 if (core->use_giant_pages == 1) {
517 pg_start = PAGE_ADDR_1GB(page_addr);
518 pg_end = (pg_start + PAGE_SIZE_1GB);
520 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
522 if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
523 page_size = PAGE_SIZE_1GB;
528 if (core->use_large_pages == 1) {
529 pg_start = PAGE_ADDR_2MB(page_addr);
530 pg_end = (pg_start + PAGE_SIZE_2MB);
532 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
534 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
535 page_size = PAGE_SIZE_2MB;
540 PrintError("Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
549 void v3_print_mem_map(struct v3_vm_info * vm) {
550 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
551 struct v3_mem_region * reg = &(vm->mem_map.base_region);
554 V3_Print("Memory Layout (all cores):\n");
557 V3_Print("Base Region (all cores): 0x%p - 0x%p -> 0x%p\n",
558 (void *)(reg->guest_start),
559 (void *)(reg->guest_end - 1),
560 (void *)(reg->host_addr));
563 // If the memory map is empty, don't print it
569 reg = rb_entry(node, struct v3_mem_region, tree_node);
571 V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
572 (void *)(reg->guest_start),
573 (void *)(reg->guest_end - 1),
574 (void *)(reg->host_addr));
576 V3_Print("\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
582 } while ((node = v3_rb_next(node)));