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\n");
46 v3_print_mem_map(core->vm_info);
48 v3_print_guest_state(core);
53 int v3_init_mem_map(struct v3_vm_info * vm) {
54 struct v3_mem_map * map = &(vm->mem_map);
55 addr_t mem_pages = vm->mem_size >> 12;
57 memset(&(map->base_region), 0, sizeof(struct v3_mem_region));
59 map->mem_regions.rb_node = NULL;
61 // There is an underlying region that contains all of the guest memory
62 // PrintDebug("Mapping %d pages of memory (%u bytes)\n", (int)mem_pages, (uint_t)info->mem_size);
64 // 2MB page alignment needed for 2MB hardware nested paging
65 map->base_region.guest_start = 0;
66 map->base_region.guest_end = mem_pages * PAGE_SIZE_4KB;
68 #ifdef CONFIG_ALIGNED_PG_ALLOC
69 map->base_region.host_addr = (addr_t)V3_AllocAlignedPages(mem_pages, vm->mem_align);
71 map->base_region.host_addr = (addr_t)V3_AllocPages(mem_pages);
74 map->base_region.flags.read = 1;
75 map->base_region.flags.write = 1;
76 map->base_region.flags.exec = 1;
77 map->base_region.flags.base = 1;
78 map->base_region.flags.alloced = 1;
80 map->base_region.unhandled = unhandled_err;
82 if ((void *)map->base_region.host_addr == NULL) {
83 PrintError("Could not allocate Guest memory\n");
87 //memset(V3_VAddr((void *)map->base_region.host_addr), 0xffffffff, map->base_region.guest_end);
89 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
95 void v3_delete_mem_map(struct v3_vm_info * vm) {
96 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
97 struct v3_mem_region * reg;
98 struct rb_node * tmp_node = NULL;
99 addr_t mem_pages = vm->mem_size >> 12;
102 reg = rb_entry(node, struct v3_mem_region, tree_node);
104 node = v3_rb_next(node);
106 v3_delete_mem_region(vm, reg);
109 V3_FreePages((void *)(vm->mem_map.base_region.host_addr), mem_pages);
113 struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
114 addr_t guest_addr_start, addr_t guest_addr_end) {
116 struct v3_mem_region * entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
117 memset(entry, 0, sizeof(struct v3_mem_region));
119 entry->guest_start = guest_addr_start;
120 entry->guest_end = guest_addr_end;
121 entry->core_id = core_id;
122 entry->unhandled = unhandled_err;
130 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
131 addr_t guest_addr_start,
132 addr_t guest_addr_end,
135 struct v3_mem_region * entry = NULL;
137 entry = v3_create_mem_region(vm, core_id,
141 entry->host_addr = host_addr;
143 entry->flags.read = 1;
144 entry->flags.write = 1;
145 entry->flags.exec = 1;
146 entry->flags.alloced = 1;
148 if (v3_insert_mem_region(vm, entry) == -1) {
159 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
160 struct v3_mem_region * region) {
161 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
162 struct rb_node * parent = NULL;
163 struct v3_mem_region * tmp_region;
167 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
169 if (region->guest_end <= tmp_region->guest_start) {
171 } else if (region->guest_start >= tmp_region->guest_end) {
174 if ((region->guest_end != tmp_region->guest_end) ||
175 (region->guest_start != tmp_region->guest_start)) {
176 PrintError("Trying to map a partial overlapped core specific page...\n");
177 return tmp_region; // This is ugly...
178 } else if (region->core_id == tmp_region->core_id) {
180 } else if (region->core_id < tmp_region->core_id) {
188 rb_link_node(&(region->tree_node), parent, p);
195 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
196 struct v3_mem_region * ret;
199 if ((ret = __insert_mem_region(vm, region))) {
203 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
207 for (i = 0; i < vm->num_cores; i++) {
208 struct guest_info * info = &(vm->cores[i]);
210 // flush virtual page tables
211 // 3 cases shadow, shadow passthrough, and nested
213 if (info->shdw_pg_mode == SHADOW_PAGING) {
214 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
216 if (mem_mode == PHYSICAL_MEM) {
219 for (cur_addr = region->guest_start;
220 cur_addr < region->guest_end;
221 cur_addr += PAGE_SIZE_4KB) {
222 v3_invalidate_passthrough_addr(info, cur_addr);
225 v3_invalidate_shadow_pts(info);
228 } else if (info->shdw_pg_mode == NESTED_PAGING) {
231 for (cur_addr = region->guest_start;
232 cur_addr < region->guest_end;
233 cur_addr += PAGE_SIZE_4KB) {
235 v3_invalidate_nested_addr(info, cur_addr);
246 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
247 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
248 struct v3_mem_region * reg = NULL;
252 reg = rb_entry(n, struct v3_mem_region, tree_node);
254 if (guest_addr < reg->guest_start) {
256 } else if (guest_addr >= reg->guest_end) {
259 if (reg->core_id == V3_MEM_CORE_ANY) {
260 // found relevant region, it's available on all cores
262 } else if (core_id == reg->core_id) {
263 // found relevant region, it's available on the indicated core
265 } else if (core_id < reg->core_id) {
266 // go left, core too big
268 } else if (core_id > reg->core_id) {
269 // go right, core too small
272 PrintDebug("v3_get_mem_region: Impossible!\n");
279 // There is not registered region, so we check if its a valid address in the base region
281 if (guest_addr > vm->mem_map.base_region.guest_end) {
282 PrintError("Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p) (core=0x%x)\n",
283 (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end, core_id);
284 v3_print_mem_map(vm);
289 return &(vm->mem_map.base_region);
294 /* This returns the next memory region based on a given address.
295 * If the address falls inside a sub region, that region is returned.
296 * If the address falls outside a sub region, the next sub region is returned
297 * NOTE that we have to be careful about core_ids here...
299 static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
300 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
301 struct v3_mem_region * reg = NULL;
302 struct v3_mem_region * parent = NULL;
306 reg = rb_entry(n, struct v3_mem_region, tree_node);
308 if (guest_addr < reg->guest_start) {
310 } else if (guest_addr >= reg->guest_end) {
313 if (reg->core_id == V3_MEM_CORE_ANY) {
314 // found relevant region, it's available on all cores
316 } else if (core_id == reg->core_id) {
317 // found relevant region, it's available on the indicated core
319 } else if (core_id < reg->core_id) {
320 // go left, core too big
322 } else if (core_id > reg->core_id) {
323 // go right, core too small
326 PrintError("v3_get_mem_region: Impossible!\n");
331 if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
337 if (parent->guest_start > guest_addr) {
339 } else if (parent->guest_end < guest_addr) {
340 struct rb_node * node = &(parent->tree_node);
342 while ((node = v3_rb_next(node)) != NULL) {
343 struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
345 if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
346 (next_reg->core_id == core_id)) {
348 // This check is not strictly necessary, but it makes it clearer
349 if (next_reg->guest_start > guest_addr) {
362 /* Given an address region of memory, find if there are any regions that overlap with it.
363 * This checks that the range lies in a single region, and returns that region if it does,
364 * this can be either the base region or a sub region.
365 * IF there are multiple regions in the range then it returns NULL
367 static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
368 addr_t start_gpa, addr_t end_gpa) {
369 struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
371 if (start_region == NULL) {
372 PrintError("Invalid memory region\n");
377 if (start_region->guest_end < end_gpa) {
378 // Region ends before range
380 } else if (start_region->flags.base == 0) {
381 // sub region overlaps range
384 // Base region, now we have to scan forward for the next sub region
385 struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
387 if (next_reg == NULL) {
388 // no sub regions after start_addr, base region is ok
390 } else if (next_reg->guest_start >= end_gpa) {
391 // Next sub region begins outside range
399 // Should never get here
407 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
414 for (i = 0; i < vm->num_cores; i++) {
415 struct guest_info * info = &(vm->cores[i]);
417 // flush virtual page tables
418 // 3 cases shadow, shadow passthrough, and nested
420 if (info->shdw_pg_mode == SHADOW_PAGING) {
421 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
423 if (mem_mode == PHYSICAL_MEM) {
426 for (cur_addr = reg->guest_start;
427 cur_addr < reg->guest_end;
428 cur_addr += PAGE_SIZE_4KB) {
429 v3_invalidate_passthrough_addr(info, cur_addr);
432 v3_invalidate_shadow_pts(info);
435 } else if (info->shdw_pg_mode == NESTED_PAGING) {
438 for (cur_addr = reg->guest_start;
439 cur_addr < reg->guest_end;
440 cur_addr += PAGE_SIZE_4KB) {
442 v3_invalidate_nested_addr(info, cur_addr);
447 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
451 // flush virtual page tables
452 // 3 cases shadow, shadow passthrough, and nested
456 // Determine if a given address can be handled by a large page of the requested size
457 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
460 uint32_t page_size = PAGE_SIZE_4KB;
461 struct v3_mem_region * reg = NULL;
465 if (core->use_large_pages == 1) {
466 pg_start = PAGE_ADDR_4MB(page_addr);
467 pg_end = (pg_start + PAGE_SIZE_4MB);
469 reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
471 if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
472 page_size = PAGE_SIZE_4MB;
477 if (core->use_large_pages == 1) {
478 pg_start = PAGE_ADDR_2MB(page_addr);
479 pg_end = (pg_start + PAGE_SIZE_2MB);
481 reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
483 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
484 page_size = PAGE_SIZE_2MB;
491 if (core->use_giant_pages == 1) {
492 pg_start = PAGE_ADDR_1GB(page_addr);
493 pg_end = (pg_start + PAGE_SIZE_1GB);
495 reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
497 if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
498 page_size = PAGE_SIZE_1GB;
503 if (core->use_large_pages == 1) {
504 pg_start = PAGE_ADDR_2MB(page_addr);
505 pg_end = (pg_start + PAGE_SIZE_2MB);
507 reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
509 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
510 page_size = PAGE_SIZE_2MB;
515 PrintError("Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
524 void v3_print_mem_map(struct v3_vm_info * vm) {
525 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
526 struct v3_mem_region * reg = &(vm->mem_map.base_region);
529 V3_Print("Memory Layout (all cores):\n");
532 V3_Print("Base Region (all cores): 0x%p - 0x%p -> 0x%p\n",
533 (void *)(reg->guest_start),
534 (void *)(reg->guest_end - 1),
535 (void *)(reg->host_addr));
538 // If the memory map is empty, don't print it
544 reg = rb_entry(node, struct v3_mem_region, tree_node);
546 V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
547 (void *)(reg->guest_start),
548 (void *)(reg->guest_end - 1),
549 (void *)(reg->host_addr));
551 V3_Print("\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
557 } while ((node = v3_rb_next(node)));