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 // Clear the memory...
77 memset(V3_VAddr((void *)map->base_region.host_addr), 0, mem_pages * PAGE_SIZE_4KB);
80 map->base_region.flags.read = 1;
81 map->base_region.flags.write = 1;
82 map->base_region.flags.exec = 1;
83 map->base_region.flags.base = 1;
84 map->base_region.flags.alloced = 1;
86 map->base_region.unhandled = unhandled_err;
88 if ((void *)map->base_region.host_addr == NULL) {
89 PrintError("Could not allocate Guest memory\n");
93 //memset(V3_VAddr((void *)map->base_region.host_addr), 0xffffffff, map->base_region.guest_end);
95 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
101 void v3_delete_mem_map(struct v3_vm_info * vm) {
102 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
103 struct v3_mem_region * reg;
104 struct rb_node * tmp_node = NULL;
105 addr_t mem_pages = vm->mem_size >> 12;
108 reg = rb_entry(node, struct v3_mem_region, tree_node);
110 node = v3_rb_next(node);
112 v3_delete_mem_region(vm, reg);
115 V3_FreePages((void *)(vm->mem_map.base_region.host_addr), mem_pages);
119 struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
120 addr_t guest_addr_start, addr_t guest_addr_end) {
121 struct v3_mem_region * entry = NULL;
123 if (guest_addr_start >= guest_addr_end) {
124 PrintError("Region start is after region end\n");
128 entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
129 memset(entry, 0, sizeof(struct v3_mem_region));
131 entry->guest_start = guest_addr_start;
132 entry->guest_end = guest_addr_end;
133 entry->core_id = core_id;
134 entry->unhandled = unhandled_err;
142 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
143 addr_t guest_addr_start,
144 addr_t guest_addr_end,
147 struct v3_mem_region * entry = NULL;
149 entry = v3_create_mem_region(vm, core_id,
153 entry->host_addr = host_addr;
155 entry->flags.read = 1;
156 entry->flags.write = 1;
157 entry->flags.exec = 1;
158 entry->flags.alloced = 1;
160 if (v3_insert_mem_region(vm, entry) == -1) {
171 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
172 struct v3_mem_region * region) {
173 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
174 struct rb_node * parent = NULL;
175 struct v3_mem_region * tmp_region;
179 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
181 if (region->guest_end <= tmp_region->guest_start) {
183 } else if (region->guest_start >= tmp_region->guest_end) {
186 if ((region->guest_end != tmp_region->guest_end) ||
187 (region->guest_start != tmp_region->guest_start)) {
188 PrintError("Trying to map a partial overlapped core specific page...\n");
189 return tmp_region; // This is ugly...
190 } else if (region->core_id == tmp_region->core_id) {
192 } else if (region->core_id < tmp_region->core_id) {
200 rb_link_node(&(region->tree_node), parent, p);
207 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
208 struct v3_mem_region * ret;
211 if ((ret = __insert_mem_region(vm, region))) {
215 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
219 for (i = 0; i < vm->num_cores; i++) {
220 struct guest_info * info = &(vm->cores[i]);
222 // flush virtual page tables
223 // 3 cases shadow, shadow passthrough, and nested
225 if (info->shdw_pg_mode == SHADOW_PAGING) {
226 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
228 if (mem_mode == PHYSICAL_MEM) {
231 for (cur_addr = region->guest_start;
232 cur_addr < region->guest_end;
233 cur_addr += PAGE_SIZE_4KB) {
234 v3_invalidate_passthrough_addr(info, cur_addr);
237 v3_invalidate_shadow_pts(info);
240 } else if (info->shdw_pg_mode == NESTED_PAGING) {
243 for (cur_addr = region->guest_start;
244 cur_addr < region->guest_end;
245 cur_addr += PAGE_SIZE_4KB) {
247 v3_invalidate_nested_addr(info, cur_addr);
258 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
259 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
260 struct v3_mem_region * reg = NULL;
264 reg = rb_entry(n, struct v3_mem_region, tree_node);
266 if (guest_addr < reg->guest_start) {
268 } else if (guest_addr >= reg->guest_end) {
271 if (reg->core_id == V3_MEM_CORE_ANY) {
272 // found relevant region, it's available on all cores
274 } else if (core_id == reg->core_id) {
275 // found relevant region, it's available on the indicated core
277 } else if (core_id < reg->core_id) {
278 // go left, core too big
280 } else if (core_id > reg->core_id) {
281 // go right, core too small
284 PrintDebug("v3_get_mem_region: Impossible!\n");
291 // There is not registered region, so we check if its a valid address in the base region
293 if (guest_addr > vm->mem_map.base_region.guest_end) {
294 PrintError("Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p) (core=0x%x)\n",
295 (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end, core_id);
296 v3_print_mem_map(vm);
301 return &(vm->mem_map.base_region);
306 /* This returns the next memory region based on a given address.
307 * If the address falls inside a sub region, that region is returned.
308 * If the address falls outside a sub region, the next sub region is returned
309 * NOTE that we have to be careful about core_ids here...
311 static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
312 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
313 struct v3_mem_region * reg = NULL;
314 struct v3_mem_region * parent = NULL;
322 reg = rb_entry(n, struct v3_mem_region, tree_node);
324 if (guest_addr < reg->guest_start) {
326 } else if (guest_addr >= reg->guest_end) {
329 if (reg->core_id == V3_MEM_CORE_ANY) {
330 // found relevant region, it's available on all cores
332 } else if (core_id == reg->core_id) {
333 // found relevant region, it's available on the indicated core
335 } else if (core_id < reg->core_id) {
336 // go left, core too big
338 } else if (core_id > reg->core_id) {
339 // go right, core too small
342 PrintError("v3_get_mem_region: Impossible!\n");
347 if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
353 if (parent->guest_start > guest_addr) {
355 } else if (parent->guest_end < guest_addr) {
356 struct rb_node * node = &(parent->tree_node);
358 while ((node = v3_rb_next(node)) != NULL) {
359 struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
361 if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
362 (next_reg->core_id == core_id)) {
364 // This check is not strictly necessary, but it makes it clearer
365 if (next_reg->guest_start > guest_addr) {
378 /* Given an address region of memory, find if there are any regions that overlap with it.
379 * This checks that the range lies in a single region, and returns that region if it does,
380 * this can be either the base region or a sub region.
381 * IF there are multiple regions in the range then it returns NULL
383 static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
384 addr_t start_gpa, addr_t end_gpa) {
385 struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
387 if (start_region == NULL) {
388 PrintError("Invalid memory region\n");
393 if (start_region->guest_end < end_gpa) {
394 // Region ends before range
396 } else if (start_region->flags.base == 0) {
397 // sub region overlaps range
400 // Base region, now we have to scan forward for the next sub region
401 struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
403 if (next_reg == NULL) {
404 // no sub regions after start_addr, base region is ok
406 } else if (next_reg->guest_start >= end_gpa) {
407 // Next sub region begins outside range
415 // Should never get here
423 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
431 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
435 // If the guest isn't running then there shouldn't be anything to invalidate.
436 // Page tables should __always__ be created on demand during execution
437 // NOTE: This is a sanity check, and can be removed if that assumption changes
438 if (vm->run_state != VM_RUNNING) {
443 for (i = 0; i < vm->num_cores; i++) {
444 struct guest_info * info = &(vm->cores[i]);
446 // flush virtual page tables
447 // 3 cases shadow, shadow passthrough, and nested
449 if (info->shdw_pg_mode == SHADOW_PAGING) {
450 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
452 if (mem_mode == PHYSICAL_MEM) {
455 for (cur_addr = reg->guest_start;
456 cur_addr < reg->guest_end;
457 cur_addr += PAGE_SIZE_4KB) {
458 v3_invalidate_passthrough_addr(info, cur_addr);
461 v3_invalidate_shadow_pts(info);
464 } else if (info->shdw_pg_mode == NESTED_PAGING) {
467 for (cur_addr = reg->guest_start;
468 cur_addr < reg->guest_end;
469 cur_addr += PAGE_SIZE_4KB) {
471 v3_invalidate_nested_addr(info, cur_addr);
478 // flush virtual page tables
479 // 3 cases shadow, shadow passthrough, and nested
483 // Determine if a given address can be handled by a large page of the requested size
484 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
487 uint32_t page_size = PAGE_SIZE_4KB;
488 struct v3_mem_region * reg = NULL;
492 if (core->use_large_pages == 1) {
493 pg_start = PAGE_ADDR_4MB(page_addr);
494 pg_end = (pg_start + PAGE_SIZE_4MB);
496 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
498 if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
499 page_size = PAGE_SIZE_4MB;
504 if (core->use_large_pages == 1) {
505 pg_start = PAGE_ADDR_2MB(page_addr);
506 pg_end = (pg_start + PAGE_SIZE_2MB);
508 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
510 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
511 page_size = PAGE_SIZE_2MB;
518 if (core->use_giant_pages == 1) {
519 pg_start = PAGE_ADDR_1GB(page_addr);
520 pg_end = (pg_start + PAGE_SIZE_1GB);
522 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
524 if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
525 page_size = PAGE_SIZE_1GB;
530 if (core->use_large_pages == 1) {
531 pg_start = PAGE_ADDR_2MB(page_addr);
532 pg_end = (pg_start + PAGE_SIZE_2MB);
534 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
536 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
537 page_size = PAGE_SIZE_2MB;
542 PrintError("Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
551 void v3_print_mem_map(struct v3_vm_info * vm) {
552 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
553 struct v3_mem_region * reg = &(vm->mem_map.base_region);
556 V3_Print("Memory Layout (all cores):\n");
559 V3_Print("Base Region (all cores): 0x%p - 0x%p -> 0x%p\n",
560 (void *)(reg->guest_start),
561 (void *)(reg->guest_end - 1),
562 (void *)(reg->host_addr));
565 // If the memory map is empty, don't print it
571 reg = rb_entry(node, struct v3_mem_region, tree_node);
573 V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
574 (void *)(reg->guest_start),
575 (void *)(reg->guest_end - 1),
576 (void *)(reg->host_addr));
578 V3_Print("\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
584 } while ((node = v3_rb_next(node)));