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>
26 #include <palacios/vmm_shadow_paging.h>
27 #include <palacios/vmm_direct_paging.h>
32 static int mem_offset_hypercall(struct guest_info * info, uint_t hcall_id, void * private_data) {
33 PrintDebug("V3Vee: Memory offset hypercall (offset=%p)\n",
34 (void *)(info->vm_info->mem_map.base_region.host_addr));
36 info->vm_regs.rbx = info->vm_info->mem_map.base_region.host_addr;
41 static int unhandled_err(struct guest_info * core, addr_t guest_va, addr_t guest_pa,
42 struct v3_mem_region * reg, pf_error_t access_info) {
44 PrintError("Unhandled memory access error (gpa=%p, gva=%p, error_code=%d)\n",
45 (void *)guest_pa, (void *)guest_va, *(uint32_t *)&access_info);
47 v3_print_mem_map(core->vm_info);
49 v3_print_guest_state(core);
54 int v3_init_mem_map(struct v3_vm_info * vm) {
55 struct v3_mem_map * map = &(vm->mem_map);
56 addr_t mem_pages = vm->mem_size >> 12;
58 memset(&(map->base_region), 0, sizeof(struct v3_mem_region));
60 map->mem_regions.rb_node = NULL;
62 // There is an underlying region that contains all of the guest memory
63 // PrintDebug("Mapping %d pages of memory (%u bytes)\n", (int)mem_pages, (uint_t)info->mem_size);
65 // 2MB page alignment needed for 2MB hardware nested paging
66 map->base_region.guest_start = 0;
67 map->base_region.guest_end = mem_pages * PAGE_SIZE_4KB;
69 #ifdef V3_CONFIG_ALIGNED_PG_ALLOC
70 map->base_region.host_addr = (addr_t)V3_AllocAlignedPages(mem_pages, vm->mem_align);
72 map->base_region.host_addr = (addr_t)V3_AllocPages(mem_pages);
75 map->base_region.flags.read = 1;
76 map->base_region.flags.write = 1;
77 map->base_region.flags.exec = 1;
78 map->base_region.flags.base = 1;
79 map->base_region.flags.alloced = 1;
81 map->base_region.unhandled = unhandled_err;
83 if ((void *)map->base_region.host_addr == NULL) {
84 PrintError("Could not allocate Guest memory\n");
88 //memset(V3_VAddr((void *)map->base_region.host_addr), 0xffffffff, map->base_region.guest_end);
90 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
96 void v3_delete_mem_map(struct v3_vm_info * vm) {
97 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
98 struct v3_mem_region * reg;
99 struct rb_node * tmp_node = NULL;
100 addr_t mem_pages = vm->mem_size >> 12;
103 reg = rb_entry(node, struct v3_mem_region, tree_node);
105 node = v3_rb_next(node);
107 v3_delete_mem_region(vm, reg);
110 V3_FreePages((void *)(vm->mem_map.base_region.host_addr), mem_pages);
114 struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
115 addr_t guest_addr_start, addr_t guest_addr_end) {
117 struct v3_mem_region * entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
118 memset(entry, 0, sizeof(struct v3_mem_region));
120 entry->guest_start = guest_addr_start;
121 entry->guest_end = guest_addr_end;
122 entry->core_id = core_id;
123 entry->unhandled = unhandled_err;
131 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
132 addr_t guest_addr_start,
133 addr_t guest_addr_end,
136 struct v3_mem_region * entry = NULL;
138 entry = v3_create_mem_region(vm, core_id,
142 entry->host_addr = host_addr;
144 entry->flags.read = 1;
145 entry->flags.write = 1;
146 entry->flags.exec = 1;
147 entry->flags.alloced = 1;
149 if (v3_insert_mem_region(vm, entry) == -1) {
160 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
161 struct v3_mem_region * region) {
162 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
163 struct rb_node * parent = NULL;
164 struct v3_mem_region * tmp_region;
168 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
170 if (region->guest_end <= tmp_region->guest_start) {
172 } else if (region->guest_start >= tmp_region->guest_end) {
175 if ((region->guest_end != tmp_region->guest_end) ||
176 (region->guest_start != tmp_region->guest_start)) {
177 PrintError("Trying to map a partial overlapped core specific page...\n");
178 return tmp_region; // This is ugly...
179 } else if (region->core_id == tmp_region->core_id) {
181 } else if (region->core_id < tmp_region->core_id) {
189 rb_link_node(&(region->tree_node), parent, p);
196 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
197 struct v3_mem_region * ret;
200 if ((ret = __insert_mem_region(vm, region))) {
204 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
208 for (i = 0; i < vm->num_cores; i++) {
209 struct guest_info * info = &(vm->cores[i]);
211 // flush virtual page tables
212 // 3 cases shadow, shadow passthrough, and nested
214 if (info->shdw_pg_mode == SHADOW_PAGING) {
215 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
217 if (mem_mode == PHYSICAL_MEM) {
220 for (cur_addr = region->guest_start;
221 cur_addr < region->guest_end;
222 cur_addr += PAGE_SIZE_4KB) {
223 v3_invalidate_passthrough_addr(info, cur_addr);
226 v3_invalidate_shadow_pts(info);
229 } else if (info->shdw_pg_mode == NESTED_PAGING) {
232 for (cur_addr = region->guest_start;
233 cur_addr < region->guest_end;
234 cur_addr += PAGE_SIZE_4KB) {
236 v3_invalidate_nested_addr(info, cur_addr);
247 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
248 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
249 struct v3_mem_region * reg = NULL;
253 reg = rb_entry(n, struct v3_mem_region, tree_node);
255 if (guest_addr < reg->guest_start) {
257 } else if (guest_addr >= reg->guest_end) {
260 if (reg->core_id == V3_MEM_CORE_ANY) {
261 // found relevant region, it's available on all cores
263 } else if (core_id == reg->core_id) {
264 // found relevant region, it's available on the indicated core
266 } else if (core_id < reg->core_id) {
267 // go left, core too big
269 } else if (core_id > reg->core_id) {
270 // go right, core too small
273 PrintDebug("v3_get_mem_region: Impossible!\n");
280 // There is not registered region, so we check if its a valid address in the base region
282 if (guest_addr > vm->mem_map.base_region.guest_end) {
283 PrintError("Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p) (core=0x%x)\n",
284 (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end, core_id);
285 v3_print_mem_map(vm);
290 return &(vm->mem_map.base_region);
295 /* This returns the next memory region based on a given address.
296 * If the address falls inside a sub region, that region is returned.
297 * If the address falls outside a sub region, the next sub region is returned
298 * NOTE that we have to be careful about core_ids here...
300 static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
301 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
302 struct v3_mem_region * reg = NULL;
303 struct v3_mem_region * parent = NULL;
311 reg = rb_entry(n, struct v3_mem_region, tree_node);
313 if (guest_addr < reg->guest_start) {
315 } else if (guest_addr >= reg->guest_end) {
318 if (reg->core_id == V3_MEM_CORE_ANY) {
319 // found relevant region, it's available on all cores
321 } else if (core_id == reg->core_id) {
322 // found relevant region, it's available on the indicated core
324 } else if (core_id < reg->core_id) {
325 // go left, core too big
327 } else if (core_id > reg->core_id) {
328 // go right, core too small
331 PrintError("v3_get_mem_region: Impossible!\n");
336 if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
342 if (parent->guest_start > guest_addr) {
344 } else if (parent->guest_end < guest_addr) {
345 struct rb_node * node = &(parent->tree_node);
347 while ((node = v3_rb_next(node)) != NULL) {
348 struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
350 if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
351 (next_reg->core_id == core_id)) {
353 // This check is not strictly necessary, but it makes it clearer
354 if (next_reg->guest_start > guest_addr) {
367 /* Given an address region of memory, find if there are any regions that overlap with it.
368 * This checks that the range lies in a single region, and returns that region if it does,
369 * this can be either the base region or a sub region.
370 * IF there are multiple regions in the range then it returns NULL
372 static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
373 addr_t start_gpa, addr_t end_gpa) {
374 struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
376 if (start_region == NULL) {
377 PrintError("Invalid memory region\n");
382 if (start_region->guest_end < end_gpa) {
383 // Region ends before range
385 } else if (start_region->flags.base == 0) {
386 // sub region overlaps range
389 // Base region, now we have to scan forward for the next sub region
390 struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
392 if (next_reg == NULL) {
393 // no sub regions after start_addr, base region is ok
395 } else if (next_reg->guest_start >= end_gpa) {
396 // Next sub region begins outside range
404 // Should never get here
412 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
420 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
424 // If the guest isn't running then there shouldn't be anything to invalidate.
425 // Page tables should __always__ be created on demand during execution
426 // NOTE: This is a sanity check, and can be removed if that assumption changes
427 if (vm->run_state != VM_RUNNING) {
432 for (i = 0; i < vm->num_cores; i++) {
433 struct guest_info * info = &(vm->cores[i]);
435 // flush virtual page tables
436 // 3 cases shadow, shadow passthrough, and nested
438 if (info->shdw_pg_mode == SHADOW_PAGING) {
439 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
441 if (mem_mode == PHYSICAL_MEM) {
444 for (cur_addr = reg->guest_start;
445 cur_addr < reg->guest_end;
446 cur_addr += PAGE_SIZE_4KB) {
447 v3_invalidate_passthrough_addr(info, cur_addr);
450 v3_invalidate_shadow_pts(info);
453 } else if (info->shdw_pg_mode == NESTED_PAGING) {
456 for (cur_addr = reg->guest_start;
457 cur_addr < reg->guest_end;
458 cur_addr += PAGE_SIZE_4KB) {
460 v3_invalidate_nested_addr(info, cur_addr);
467 // flush virtual page tables
468 // 3 cases shadow, shadow passthrough, and nested
472 // Determine if a given address can be handled by a large page of the requested size
473 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
476 uint32_t page_size = PAGE_SIZE_4KB;
477 struct v3_mem_region * reg = NULL;
481 if (core->use_large_pages == 1) {
482 pg_start = PAGE_ADDR_4MB(page_addr);
483 pg_end = (pg_start + PAGE_SIZE_4MB);
485 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
487 if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
488 page_size = PAGE_SIZE_4MB;
493 if (core->use_large_pages == 1) {
494 pg_start = PAGE_ADDR_2MB(page_addr);
495 pg_end = (pg_start + PAGE_SIZE_2MB);
497 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
499 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
500 page_size = PAGE_SIZE_2MB;
507 if (core->use_giant_pages == 1) {
508 pg_start = PAGE_ADDR_1GB(page_addr);
509 pg_end = (pg_start + PAGE_SIZE_1GB);
511 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
513 if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
514 page_size = PAGE_SIZE_1GB;
519 if (core->use_large_pages == 1) {
520 pg_start = PAGE_ADDR_2MB(page_addr);
521 pg_end = (pg_start + PAGE_SIZE_2MB);
523 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
525 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
526 page_size = PAGE_SIZE_2MB;
531 PrintError("Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
540 void v3_print_mem_map(struct v3_vm_info * vm) {
541 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
542 struct v3_mem_region * reg = &(vm->mem_map.base_region);
545 V3_Print("Memory Layout (all cores):\n");
548 V3_Print("Base Region (all cores): 0x%p - 0x%p -> 0x%p\n",
549 (void *)(reg->guest_start),
550 (void *)(reg->guest_end - 1),
551 (void *)(reg->host_addr));
554 // If the memory map is empty, don't print it
560 reg = rb_entry(node, struct v3_mem_region, tree_node);
562 V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
563 (void *)(reg->guest_start),
564 (void *)(reg->guest_end - 1),
565 (void *)(reg->host_addr));
567 V3_Print("\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
573 } while ((node = v3_rb_next(node)));