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(info->vm_info, info,"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(core->vm_info, core, "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(info->vm_info, info, "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(vm, VCORE_NONE,"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(vm, VCORE_NONE, "Region start is after region end\n");
126 entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
129 PrintError(vm, VCORE_NONE, "Cannot allocate in creating a memory region\n");
133 memset(entry, 0, sizeof(struct v3_mem_region));
135 entry->guest_start = guest_addr_start;
136 entry->guest_end = guest_addr_end;
137 entry->core_id = core_id;
138 entry->unhandled = unhandled_err;
146 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
147 addr_t guest_addr_start,
148 addr_t guest_addr_end,
151 struct v3_mem_region * entry = NULL;
153 entry = v3_create_mem_region(vm, core_id,
157 entry->host_addr = host_addr;
159 entry->flags.read = 1;
160 entry->flags.write = 1;
161 entry->flags.exec = 1;
162 entry->flags.alloced = 1;
164 if (v3_insert_mem_region(vm, entry) == -1) {
175 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
176 struct v3_mem_region * region) {
177 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
178 struct rb_node * parent = NULL;
179 struct v3_mem_region * tmp_region;
183 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
185 if (region->guest_end <= tmp_region->guest_start) {
187 } else if (region->guest_start >= tmp_region->guest_end) {
190 if ((region->guest_end != tmp_region->guest_end) ||
191 (region->guest_start != tmp_region->guest_start)) {
192 PrintError(vm, VCORE_NONE, "Trying to map a partial overlapped core specific page...\n");
193 return tmp_region; // This is ugly...
194 } else if (region->core_id == tmp_region->core_id) {
195 PrintError(vm, VCORE_NONE, "Trying to map a core-overlapping page\n");
197 } else if (region->core_id < tmp_region->core_id) {
205 rb_link_node(&(region->tree_node), parent, p);
212 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
213 struct v3_mem_region * ret;
217 if ((ret = __insert_mem_region(vm, region))) {
218 PrintError(vm, VCORE_NONE, "Internal insert failed returned region is from 0x%p to 0x%p on vcore %d\n", (void*)(ret->guest_start), (void*)(ret->guest_end), ret->core_id);
222 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
227 for (i = 0; i < vm->num_cores; i++) {
228 struct guest_info * info = &(vm->cores[i]);
230 // flush virtual page tables
231 // 3 cases shadow, shadow passthrough, and nested
233 if (info->shdw_pg_mode == SHADOW_PAGING) {
234 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
236 if (mem_mode == PHYSICAL_MEM) {
237 rc |= v3_invalidate_passthrough_addr_range(info, region->guest_start, region->guest_end-1);
239 rc |= v3_invalidate_shadow_pts(info);
242 } else if (info->shdw_pg_mode == NESTED_PAGING) {
243 rc |= v3_invalidate_nested_addr_range(info, region->guest_start, region->guest_end-1);
253 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
254 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
255 struct v3_mem_region * reg = NULL;
259 reg = rb_entry(n, struct v3_mem_region, tree_node);
261 if (guest_addr < reg->guest_start) {
263 } else if (guest_addr >= reg->guest_end) {
266 if (reg->core_id == V3_MEM_CORE_ANY) {
267 // found relevant region, it's available on all cores
269 } else if (core_id == reg->core_id) {
270 // found relevant region, it's available on the indicated core
272 } else if (core_id < reg->core_id) {
273 // go left, core too big
275 } else if (core_id > reg->core_id) {
276 // go right, core too small
279 PrintDebug(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
286 // There is not registered region, so we check if its a valid address in the base region
288 if (guest_addr > vm->mem_map.base_region.guest_end) {
289 PrintError(vm, VCORE_NONE, "Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p) (core=0x%x)\n",
290 (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end, core_id);
291 v3_print_mem_map(vm);
296 return &(vm->mem_map.base_region);
301 /* This returns the next memory region based on a given address.
302 * If the address falls inside a sub region, that region is returned.
303 * If the address falls outside a sub region, the next sub region is returned
304 * NOTE that we have to be careful about core_ids here...
306 static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
307 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
308 struct v3_mem_region * reg = NULL;
309 struct v3_mem_region * parent = NULL;
317 reg = rb_entry(n, struct v3_mem_region, tree_node);
319 if (guest_addr < reg->guest_start) {
321 } else if (guest_addr >= reg->guest_end) {
324 if (reg->core_id == V3_MEM_CORE_ANY) {
325 // found relevant region, it's available on all cores
327 } else if (core_id == reg->core_id) {
328 // found relevant region, it's available on the indicated core
330 } else if (core_id < reg->core_id) {
331 // go left, core too big
333 } else if (core_id > reg->core_id) {
334 // go right, core too small
337 PrintError(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
342 if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
348 if (parent->guest_start > guest_addr) {
350 } else if (parent->guest_end < guest_addr) {
351 struct rb_node * node = &(parent->tree_node);
353 while ((node = v3_rb_next(node)) != NULL) {
354 struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
356 if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
357 (next_reg->core_id == core_id)) {
359 // This check is not strictly necessary, but it makes it clearer
360 if (next_reg->guest_start > guest_addr) {
373 /* Given an address region of memory, find if there are any regions that overlap with it.
374 * This checks that the range lies in a single region, and returns that region if it does,
375 * this can be either the base region or a sub region.
376 * IF there are multiple regions in the range then it returns NULL
378 static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
379 addr_t start_gpa, addr_t end_gpa) {
380 struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
382 if (start_region == NULL) {
383 PrintError(vm, VCORE_NONE, "Invalid memory region\n");
388 if (start_region->guest_end < end_gpa) {
389 // Region ends before range
391 } else if (start_region->flags.base == 0) {
392 // sub region overlaps range
395 // Base region, now we have to scan forward for the next sub region
396 struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
398 if (next_reg == NULL) {
399 // no sub regions after start_addr, base region is ok
401 } else if (next_reg->guest_start >= end_gpa) {
402 // Next sub region begins outside range
410 // Should never get here
418 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
427 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
431 // If the guest isn't running then there shouldn't be anything to invalidate.
432 // Page tables should __always__ be created on demand during execution
433 // NOTE: This is a sanity check, and can be removed if that assumption changes
434 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) {
451 rc |= v3_invalidate_passthrough_addr_range(info,reg->guest_start, reg->guest_end-1);
453 rc |= v3_invalidate_shadow_pts(info);
456 } else if (info->shdw_pg_mode == NESTED_PAGING) {
457 rc |= v3_invalidate_nested_addr_range(info,reg->guest_start, reg->guest_end-1);
463 // flush virtual page tables
464 // 3 cases shadow, shadow passthrough, and nested
466 if (rc) { PrintError(vm, VCORE_NONE, "Error in deleting memory region\n"); }
469 // Determine if a given address can be handled by a large page of the requested size
470 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
473 uint32_t page_size = PAGE_SIZE_4KB;
474 struct v3_mem_region * reg = NULL;
478 if (core->use_large_pages == 1) {
479 pg_start = PAGE_ADDR_4MB(page_addr);
480 pg_end = (pg_start + PAGE_SIZE_4MB);
482 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
484 if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
485 page_size = PAGE_SIZE_4MB;
490 if (core->use_large_pages == 1) {
491 pg_start = PAGE_ADDR_2MB(page_addr);
492 pg_end = (pg_start + PAGE_SIZE_2MB);
494 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
496 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
497 page_size = PAGE_SIZE_2MB;
504 if (core->use_giant_pages == 1) {
505 pg_start = PAGE_ADDR_1GB(page_addr);
506 pg_end = (pg_start + PAGE_SIZE_1GB);
508 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
510 if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
511 page_size = PAGE_SIZE_1GB;
516 if (core->use_large_pages == 1) {
517 pg_start = PAGE_ADDR_2MB(page_addr);
518 pg_end = (pg_start + PAGE_SIZE_2MB);
520 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
522 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
523 page_size = PAGE_SIZE_2MB;
528 PrintError(core->vm_info, core, "Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
537 void v3_print_mem_map(struct v3_vm_info * vm) {
538 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
539 struct v3_mem_region * reg = &(vm->mem_map.base_region);
542 V3_Print(vm, VCORE_NONE, "Memory Layout (all cores):\n");
545 V3_Print(vm, VCORE_NONE, "Base Region (all cores): 0x%p - 0x%p -> 0x%p\n",
546 (void *)(reg->guest_start),
547 (void *)(reg->guest_end - 1),
548 (void *)(reg->host_addr));
551 // If the memory map is empty, don't print it
557 reg = rb_entry(node, struct v3_mem_region, tree_node);
559 V3_Print(vm, VCORE_NONE, "%d: 0x%p - 0x%p -> 0x%p\n", i,
560 (void *)(reg->guest_start),
561 (void *)(reg->guest_end - 1),
562 (void *)(reg->host_addr));
564 V3_Print(vm, VCORE_NONE, "\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
570 } while ((node = v3_rb_next(node)));