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>
31 uint64_t v3_mem_block_size = V3_CONFIG_MEM_BLOCK_SIZE;
34 struct v3_mem_region * v3_get_base_region(struct v3_vm_info * vm, addr_t gpa) {
35 struct v3_mem_map * map = &(vm->mem_map);
36 uint32_t block_index = gpa / v3_mem_block_size;
38 if (gpa > (map->num_base_regions * v3_mem_block_size) ||
39 (block_index >= map->num_base_regions)) {
40 PrintError(vm, VCORE_NONE, "Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p)\n",
41 (void *)gpa, (void *)vm->mem_size);
48 return &(map->base_regions[block_index]);
52 static int mem_offset_hypercall(struct guest_info * info, uint_t hcall_id, void * private_data) {
54 PrintDebug(info->vm_info, info,"V3Vee: Memory offset hypercall (offset=%p)\n",
55 (void *)(info->vm_info->mem_map.base_region.host_addr));
57 info->vm_regs.rbx = info->vm_info->mem_map.base_region.host_addr;
62 static int unhandled_err(struct guest_info * core, addr_t guest_va, addr_t guest_pa,
63 struct v3_mem_region * reg, pf_error_t access_info) {
65 PrintError(core->vm_info, core, "Unhandled memory access error (gpa=%p, gva=%p, error_code=%d)\n",
66 (void *)guest_pa, (void *)guest_va, *(uint32_t *)&access_info);
68 v3_print_mem_map(core->vm_info);
70 v3_print_guest_state(core);
75 static int gpa_to_node_from_cfg(struct v3_vm_info * vm, addr_t gpa) {
76 v3_cfg_tree_t * layout_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "mem_layout");
77 v3_cfg_tree_t * region_desc = v3_cfg_subtree(layout_cfg, "region");
80 char * start_addr_str = v3_cfg_val(region_desc, "start_addr");
81 char * end_addr_str = v3_cfg_val(region_desc, "end_addr");
82 char * node_id_str = v3_cfg_val(region_desc, "node");
84 addr_t start_addr = 0;
88 if ((!start_addr_str) || (!end_addr_str) || (!node_id_str)) {
89 PrintError(vm, VCORE_NONE, "Invalid memory layout in configuration\n");
93 start_addr = atox(start_addr_str);
94 end_addr = atox(end_addr_str);
95 node_id = atoi(node_id_str);
97 if ((gpa >= start_addr) && (gpa < end_addr)) {
101 region_desc = v3_cfg_next_branch(region_desc);
109 int v3_init_mem_map(struct v3_vm_info * vm) {
110 struct v3_mem_map * map = &(vm->mem_map);
111 addr_t block_pages = v3_mem_block_size >> 12;
114 map->num_base_regions = (vm->mem_size / v3_mem_block_size) + \
115 ((vm->mem_size % v3_mem_block_size) > 0);
118 map->mem_regions.rb_node = NULL;
120 map->base_regions = V3_Malloc(sizeof(struct v3_mem_region) * map->num_base_regions);
122 if (map->base_regions == NULL) {
123 PrintError(vm, VCORE_NONE, "Could not allocate base region array\n");
127 memset(map->base_regions, 0, sizeof(struct v3_mem_region) * map->num_base_regions);
130 for (i = 0; i < map->num_base_regions; i++) {
131 struct v3_mem_region * region = &(map->base_regions[i]);
134 // 2MB page alignment needed for 2MB hardware nested paging
135 region->guest_start = v3_mem_block_size * i;
136 region->guest_end = region->guest_start + v3_mem_block_size;
138 // We assume that the xml config was smart enough to align the layout to the block size
139 // If they didn't we're going to ignore their settings
140 // and use whatever node the first byte of the block is assigned to
141 node_id = gpa_to_node_from_cfg(vm, region->guest_start);
143 V3_Print(vm, VCORE_NONE, "Allocating block %d on node %d\n", i, node_id);
146 region->host_addr = (addr_t)V3_AllocPagesNode(block_pages, node_id);
148 region->host_addr = (addr_t)V3_AllocPages(block_pages);
151 if ((void *)region->host_addr == NULL) {
152 PrintError(vm, VCORE_NONE, "Could not allocate guest memory\n");
156 // Clear the memory...
157 memset(V3_VAddr((void *)region->host_addr), 0, v3_mem_block_size);
159 region->flags.read = 1;
160 region->flags.write = 1;
161 region->flags.exec = 1;
162 region->flags.base = 1;
163 region->flags.alloced = 1;
165 region->unhandled = unhandled_err;
168 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
174 void v3_delete_mem_map(struct v3_vm_info * vm) {
175 struct v3_mem_map * map = &(vm->mem_map);
176 struct rb_node * node = v3_rb_first(&(map->mem_regions));
177 struct v3_mem_region * reg;
178 struct rb_node * tmp_node = NULL;
179 addr_t block_pages = v3_mem_block_size >> 12;
183 reg = rb_entry(node, struct v3_mem_region, tree_node);
185 node = v3_rb_next(node);
187 v3_delete_mem_region(vm, reg);
190 for (i = 0; i < map->num_base_regions; i++) {
191 struct v3_mem_region * region = &(map->base_regions[i]);
192 V3_FreePages((void *)(region->host_addr), block_pages);
195 V3_Free(map->base_regions);
200 struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
201 addr_t guest_addr_start, addr_t guest_addr_end) {
202 struct v3_mem_region * entry = NULL;
204 if (guest_addr_start >= guest_addr_end) {
205 PrintError(vm, VCORE_NONE, "Region start is after region end\n");
209 entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
212 PrintError(vm, VCORE_NONE, "Cannot allocate in creating a memory region\n");
216 memset(entry, 0, sizeof(struct v3_mem_region));
218 entry->guest_start = guest_addr_start;
219 entry->guest_end = guest_addr_end;
220 entry->core_id = core_id;
221 entry->unhandled = unhandled_err;
229 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
230 addr_t guest_addr_start,
231 addr_t guest_addr_end,
234 struct v3_mem_region * entry = NULL;
236 entry = v3_create_mem_region(vm, core_id,
240 entry->host_addr = host_addr;
242 entry->flags.read = 1;
243 entry->flags.write = 1;
244 entry->flags.exec = 1;
245 entry->flags.alloced = 1;
247 if (v3_insert_mem_region(vm, entry) == -1) {
258 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
259 struct v3_mem_region * region) {
260 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
261 struct rb_node * parent = NULL;
262 struct v3_mem_region * tmp_region;
266 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
268 if (region->guest_end <= tmp_region->guest_start) {
270 } else if (region->guest_start >= tmp_region->guest_end) {
273 if ((region->guest_end != tmp_region->guest_end) ||
274 (region->guest_start != tmp_region->guest_start)) {
275 PrintError(vm, VCORE_NONE, "Trying to map a partial overlapped core specific page...\n");
276 return tmp_region; // This is ugly...
277 } else if (region->core_id == tmp_region->core_id) {
278 PrintError(vm, VCORE_NONE, "Trying to map a core-overlapping page\n");
280 } else if (region->core_id < tmp_region->core_id) {
288 rb_link_node(&(region->tree_node), parent, p);
295 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
296 struct v3_mem_region * ret;
300 if ((ret = __insert_mem_region(vm, region))) {
301 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);
305 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
310 for (i = 0; i < vm->num_cores; i++) {
311 struct guest_info * info = &(vm->cores[i]);
313 // flush virtual page tables
314 // 3 cases shadow, shadow passthrough, and nested
316 if (info->shdw_pg_mode == SHADOW_PAGING) {
317 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
319 if (mem_mode == PHYSICAL_MEM) {
320 rc |= v3_invalidate_passthrough_addr_range(info, region->guest_start, region->guest_end-1);
322 rc |= v3_invalidate_shadow_pts(info);
325 } else if (info->shdw_pg_mode == NESTED_PAGING) {
326 rc |= v3_invalidate_nested_addr_range(info, region->guest_start, region->guest_end-1);
336 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
337 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
338 struct v3_mem_region * reg = NULL;
342 reg = rb_entry(n, struct v3_mem_region, tree_node);
344 if (guest_addr < reg->guest_start) {
346 } else if (guest_addr >= reg->guest_end) {
349 if (reg->core_id == V3_MEM_CORE_ANY) {
350 // found relevant region, it's available on all cores
352 } else if (core_id == reg->core_id) {
353 // found relevant region, it's available on the indicated core
355 } else if (core_id < reg->core_id) {
356 // go left, core too big
358 } else if (core_id > reg->core_id) {
359 // go right, core too small
362 PrintDebug(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
369 // There is not registered region, so we check if its a valid address in the base region
371 return v3_get_base_region(vm, guest_addr);
376 /* This returns the next memory region based on a given address.
377 * If the address falls inside a sub region, that region is returned.
378 * If the address falls outside a sub region, the next sub region is returned
379 * NOTE that we have to be careful about core_ids here...
381 static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
382 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
383 struct v3_mem_region * reg = NULL;
384 struct v3_mem_region * parent = NULL;
392 reg = rb_entry(n, struct v3_mem_region, tree_node);
394 if (guest_addr < reg->guest_start) {
396 } else if (guest_addr >= reg->guest_end) {
399 if (reg->core_id == V3_MEM_CORE_ANY) {
400 // found relevant region, it's available on all cores
402 } else if (core_id == reg->core_id) {
403 // found relevant region, it's available on the indicated core
405 } else if (core_id < reg->core_id) {
406 // go left, core too big
408 } else if (core_id > reg->core_id) {
409 // go right, core too small
412 PrintError(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
417 if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
423 if (parent->guest_start > guest_addr) {
425 } else if (parent->guest_end < guest_addr) {
426 struct rb_node * node = &(parent->tree_node);
428 while ((node = v3_rb_next(node)) != NULL) {
429 struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
431 if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
432 (next_reg->core_id == core_id)) {
434 // This check is not strictly necessary, but it makes it clearer
435 if (next_reg->guest_start > guest_addr) {
448 /* Given an address region of memory, find if there are any regions that overlap with it.
449 * This checks that the range lies in a single region, and returns that region if it does,
450 * this can be either the base region or a sub region.
451 * IF there are multiple regions in the range then it returns NULL
453 static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
454 addr_t start_gpa, addr_t end_gpa) {
455 struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
457 if (start_region == NULL) {
458 PrintError(vm, VCORE_NONE, "Invalid memory region\n");
463 if (start_region->guest_end < end_gpa) {
464 // Region ends before range
466 } else if (start_region->flags.base == 0) {
467 // sub region overlaps range
470 // Base region, now we have to scan forward for the next sub region
471 struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
473 if (next_reg == NULL) {
474 // no sub regions after start_addr, base region is ok
476 } else if (next_reg->guest_start >= end_gpa) {
477 // Next sub region begins outside range
485 // Should never get here
493 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
502 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
506 // If the guest isn't running then there shouldn't be anything to invalidate.
507 // Page tables should __always__ be created on demand during execution
508 // NOTE: This is a sanity check, and can be removed if that assumption changes
509 if (vm->run_state != VM_RUNNING) {
516 for (i = 0; i < vm->num_cores; i++) {
517 struct guest_info * info = &(vm->cores[i]);
519 // flush virtual page tables
520 // 3 cases shadow, shadow passthrough, and nested
522 if (info->shdw_pg_mode == SHADOW_PAGING) {
523 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
525 if (mem_mode == PHYSICAL_MEM) {
526 rc |= v3_invalidate_passthrough_addr_range(info,reg->guest_start, reg->guest_end-1);
528 rc |= v3_invalidate_shadow_pts(info);
531 } else if (info->shdw_pg_mode == NESTED_PAGING) {
532 rc |= v3_invalidate_nested_addr_range(info,reg->guest_start, reg->guest_end-1);
538 // flush virtual page tables
539 // 3 cases shadow, shadow passthrough, and nested
541 if (rc) { PrintError(vm, VCORE_NONE, "Error in deleting memory region\n"); }
544 // Determine if a given address can be handled by a large page of the requested size
545 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
548 uint32_t page_size = PAGE_SIZE_4KB;
549 struct v3_mem_region * reg = NULL;
553 if (core->use_large_pages == 1) {
554 pg_start = PAGE_ADDR_4MB(page_addr);
555 pg_end = (pg_start + PAGE_SIZE_4MB);
557 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
559 if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
560 page_size = PAGE_SIZE_4MB;
565 if (core->use_large_pages == 1) {
566 pg_start = PAGE_ADDR_2MB(page_addr);
567 pg_end = (pg_start + PAGE_SIZE_2MB);
569 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
571 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
572 page_size = PAGE_SIZE_2MB;
579 if (core->use_giant_pages == 1) {
580 pg_start = PAGE_ADDR_1GB(page_addr);
581 pg_end = (pg_start + PAGE_SIZE_1GB);
583 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
585 if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
586 page_size = PAGE_SIZE_1GB;
591 if (core->use_large_pages == 1) {
592 pg_start = PAGE_ADDR_2MB(page_addr);
593 pg_end = (pg_start + PAGE_SIZE_2MB);
595 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
597 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
598 page_size = PAGE_SIZE_2MB;
603 PrintError(core->vm_info, core, "Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
612 void v3_print_mem_map(struct v3_vm_info * vm) {
613 struct v3_mem_map * map = &(vm->mem_map);
614 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
615 struct v3_mem_region * reg = NULL;
618 V3_Print(vm, VCORE_NONE, "Memory Layout (all cores):\n");
620 V3_Print(vm, VCORE_NONE, "Base Memory: (%d regions)\n", map->num_base_regions);
622 for (i = 0; i < map->num_base_regions; i++) {
623 reg = &(map->base_regions[i]);
625 V3_Print(vm, VCORE_NONE, "Base Region[%d] (all cores): 0x%p - 0x%p -> 0x%p\n",
627 (void *)(reg->guest_start),
628 (void *)(reg->guest_end - 1),
629 (void *)(reg->host_addr));
633 // If the memory map is empty, don't print it
639 reg = rb_entry(node, struct v3_mem_region, tree_node);
641 V3_Print(vm, VCORE_NONE, "%d: 0x%p - 0x%p -> 0x%p\n", i,
642 (void *)(reg->guest_start),
643 (void *)(reg->guest_end - 1),
644 (void *)(reg->host_addr));
646 V3_Print(vm, VCORE_NONE, "\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
652 } while ((node = v3_rb_next(node)));
658 char *arg = v3_lookup_option("mem_block_size");
661 v3_mem_block_size = atoi(arg);
662 V3_Print(VM_NONE,VCORE_NONE,"memory block size set to %llu bytes\n",v3_mem_block_size);
664 V3_Print(VM_NONE,VCORE_NONE,"default memory block size of %llu bytes is in use\n",v3_mem_block_size);