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);
108 // This code parallels that in vmm_shadow_paging.c:v3_init_shdw_impl()
109 // and vmm_config.c:determine_paging_mode. The determination of which
110 // paging mode will be used is determined much later than the allocation of
111 // the guest memory regions, so we need to do this here to decide if they
112 // need to be below 4 GB or not.
113 static int will_use_shadow_paging(struct v3_vm_info *vm)
115 v3_cfg_tree_t * pg_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "paging");
116 char * pg_mode = v3_cfg_val(pg_cfg, "mode");
118 if (pg_mode == NULL) {
119 return 1; // did not ask, get shadow
121 if (strcasecmp(pg_mode, "nested") == 0) {
122 extern v3_cpu_arch_t v3_mach_type;
123 if ((v3_mach_type == V3_SVM_REV3_CPU) ||
124 (v3_mach_type == V3_VMX_EPT_CPU) ||
125 (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
126 return 0; // ask for nested, get nested
128 return 1; // ask for nested, get shadow
130 } else if (strcasecmp(pg_mode, "shadow") != 0) {
131 return 1; // ask for shadow, get shadow
133 return 1; // ask for something else, get shadow
139 int v3_init_mem_map(struct v3_vm_info * vm) {
140 struct v3_mem_map * map = &(vm->mem_map);
141 addr_t block_pages = v3_mem_block_size >> 12;
144 map->num_base_regions = (vm->mem_size / v3_mem_block_size) + \
145 ((vm->mem_size % v3_mem_block_size) > 0);
148 map->mem_regions.rb_node = NULL;
150 map->base_regions = V3_Malloc(sizeof(struct v3_mem_region) * map->num_base_regions);
152 if (map->base_regions == NULL) {
153 PrintError(vm, VCORE_NONE, "Could not allocate base region array\n");
157 memset(map->base_regions, 0, sizeof(struct v3_mem_region) * map->num_base_regions);
160 for (i = 0; i < map->num_base_regions; i++) {
161 struct v3_mem_region * region = &(map->base_regions[i]);
164 // 2MB page alignment needed for 2MB hardware nested paging
165 region->guest_start = v3_mem_block_size * i;
166 region->guest_end = region->guest_start + v3_mem_block_size;
168 // We assume that the xml config was smart enough to align the layout to the block size
169 // If they didn't we're going to ignore their settings
170 // and use whatever node the first byte of the block is assigned to
171 node_id = gpa_to_node_from_cfg(vm, region->guest_start);
173 V3_Print(vm, VCORE_NONE, "Allocating block %d on node %d\n", i, node_id);
175 region->host_addr = (addr_t)V3_AllocPagesExtended(block_pages,
178 will_use_shadow_paging(vm) ?
179 V3_ALLOC_PAGES_CONSTRAINT_4GB : 0 );
181 if ((void *)region->host_addr == NULL) {
182 PrintError(vm, VCORE_NONE, "Could not allocate guest memory\n");
186 // Clear the memory...
187 memset(V3_VAddr((void *)region->host_addr), 0, v3_mem_block_size);
189 region->flags.read = 1;
190 region->flags.write = 1;
191 region->flags.exec = 1;
192 region->flags.base = 1;
193 region->flags.alloced = 1;
195 region->unhandled = unhandled_err;
198 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
204 void v3_delete_mem_map(struct v3_vm_info * vm) {
205 struct v3_mem_map * map = &(vm->mem_map);
206 struct rb_node * node = v3_rb_first(&(map->mem_regions));
207 struct v3_mem_region * reg;
208 struct rb_node * tmp_node = NULL;
209 addr_t block_pages = v3_mem_block_size >> 12;
213 reg = rb_entry(node, struct v3_mem_region, tree_node);
215 node = v3_rb_next(node);
217 v3_delete_mem_region(vm, reg);
220 for (i = 0; i < map->num_base_regions; i++) {
221 struct v3_mem_region * region = &(map->base_regions[i]);
222 V3_FreePages((void *)(region->host_addr), block_pages);
225 V3_Free(map->base_regions);
230 struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
231 addr_t guest_addr_start, addr_t guest_addr_end) {
232 struct v3_mem_region * entry = NULL;
234 if (guest_addr_start >= guest_addr_end) {
235 PrintError(vm, VCORE_NONE, "Region start is after region end\n");
239 entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
242 PrintError(vm, VCORE_NONE, "Cannot allocate in creating a memory region\n");
246 memset(entry, 0, sizeof(struct v3_mem_region));
248 entry->guest_start = guest_addr_start;
249 entry->guest_end = guest_addr_end;
250 entry->core_id = core_id;
251 entry->unhandled = unhandled_err;
259 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
260 addr_t guest_addr_start,
261 addr_t guest_addr_end,
264 struct v3_mem_region * entry = NULL;
266 entry = v3_create_mem_region(vm, core_id,
270 entry->host_addr = host_addr;
272 entry->flags.read = 1;
273 entry->flags.write = 1;
274 entry->flags.exec = 1;
275 entry->flags.alloced = 1;
277 if (v3_insert_mem_region(vm, entry) == -1) {
288 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
289 struct v3_mem_region * region) {
290 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
291 struct rb_node * parent = NULL;
292 struct v3_mem_region * tmp_region;
296 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
298 if (region->guest_end <= tmp_region->guest_start) {
300 } else if (region->guest_start >= tmp_region->guest_end) {
303 if ((region->guest_end != tmp_region->guest_end) ||
304 (region->guest_start != tmp_region->guest_start)) {
305 PrintError(vm, VCORE_NONE, "Trying to map a partial overlapped core specific page...\n");
306 return tmp_region; // This is ugly...
307 } else if (region->core_id == tmp_region->core_id) {
308 PrintError(vm, VCORE_NONE, "Trying to map a core-overlapping page\n");
310 } else if (region->core_id < tmp_region->core_id) {
318 rb_link_node(&(region->tree_node), parent, p);
325 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
326 struct v3_mem_region * ret;
330 if ((ret = __insert_mem_region(vm, region))) {
331 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);
335 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
340 for (i = 0; i < vm->num_cores; i++) {
341 struct guest_info * info = &(vm->cores[i]);
343 // flush virtual page tables
344 // 3 cases shadow, shadow passthrough, and nested
346 if (info->shdw_pg_mode == SHADOW_PAGING) {
347 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
349 if (mem_mode == PHYSICAL_MEM) {
350 rc |= v3_invalidate_passthrough_addr_range(info, region->guest_start, region->guest_end-1);
352 rc |= v3_invalidate_shadow_pts(info);
355 } else if (info->shdw_pg_mode == NESTED_PAGING) {
356 rc |= v3_invalidate_nested_addr_range(info, region->guest_start, region->guest_end-1);
366 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
367 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
368 struct v3_mem_region * reg = NULL;
372 reg = rb_entry(n, struct v3_mem_region, tree_node);
374 if (guest_addr < reg->guest_start) {
376 } else if (guest_addr >= reg->guest_end) {
379 if (reg->core_id == V3_MEM_CORE_ANY) {
380 // found relevant region, it's available on all cores
382 } else if (core_id == reg->core_id) {
383 // found relevant region, it's available on the indicated core
385 } else if (core_id < reg->core_id) {
386 // go left, core too big
388 } else if (core_id > reg->core_id) {
389 // go right, core too small
392 PrintDebug(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
399 // There is not registered region, so we check if its a valid address in the base region
401 return v3_get_base_region(vm, guest_addr);
406 /* This returns the next memory region based on a given address.
407 * If the address falls inside a sub region, that region is returned.
408 * If the address falls outside a sub region, the next sub region is returned
409 * NOTE that we have to be careful about core_ids here...
411 static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
412 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
413 struct v3_mem_region * reg = NULL;
414 struct v3_mem_region * parent = NULL;
422 reg = rb_entry(n, struct v3_mem_region, tree_node);
424 if (guest_addr < reg->guest_start) {
426 } else if (guest_addr >= reg->guest_end) {
429 if (reg->core_id == V3_MEM_CORE_ANY) {
430 // found relevant region, it's available on all cores
432 } else if (core_id == reg->core_id) {
433 // found relevant region, it's available on the indicated core
435 } else if (core_id < reg->core_id) {
436 // go left, core too big
438 } else if (core_id > reg->core_id) {
439 // go right, core too small
442 PrintError(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
447 if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
453 if (parent->guest_start > guest_addr) {
455 } else if (parent->guest_end < guest_addr) {
456 struct rb_node * node = &(parent->tree_node);
458 while ((node = v3_rb_next(node)) != NULL) {
459 struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
461 if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
462 (next_reg->core_id == core_id)) {
464 // This check is not strictly necessary, but it makes it clearer
465 if (next_reg->guest_start > guest_addr) {
478 /* Given an address region of memory, find if there are any regions that overlap with it.
479 * This checks that the range lies in a single region, and returns that region if it does,
480 * this can be either the base region or a sub region.
481 * IF there are multiple regions in the range then it returns NULL
483 static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
484 addr_t start_gpa, addr_t end_gpa) {
485 struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
487 if (start_region == NULL) {
488 PrintError(vm, VCORE_NONE, "Invalid memory region\n");
493 if (start_region->guest_end < end_gpa) {
494 // Region ends before range
496 } else if (start_region->flags.base == 0) {
497 // sub region overlaps range
500 // Base region, now we have to scan forward for the next sub region
501 struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
503 if (next_reg == NULL) {
504 // no sub regions after start_addr, base region is ok
506 } else if (next_reg->guest_start >= end_gpa) {
507 // Next sub region begins outside range
515 // Should never get here
523 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
532 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
536 // If the guest isn't running then there shouldn't be anything to invalidate.
537 // Page tables should __always__ be created on demand during execution
538 // NOTE: This is a sanity check, and can be removed if that assumption changes
539 if (vm->run_state != VM_RUNNING) {
546 for (i = 0; i < vm->num_cores; i++) {
547 struct guest_info * info = &(vm->cores[i]);
549 // flush virtual page tables
550 // 3 cases shadow, shadow passthrough, and nested
552 if (info->shdw_pg_mode == SHADOW_PAGING) {
553 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
555 if (mem_mode == PHYSICAL_MEM) {
556 rc |= v3_invalidate_passthrough_addr_range(info,reg->guest_start, reg->guest_end-1);
558 rc |= v3_invalidate_shadow_pts(info);
561 } else if (info->shdw_pg_mode == NESTED_PAGING) {
562 rc |= v3_invalidate_nested_addr_range(info,reg->guest_start, reg->guest_end-1);
568 // flush virtual page tables
569 // 3 cases shadow, shadow passthrough, and nested
571 if (rc) { PrintError(vm, VCORE_NONE, "Error in deleting memory region\n"); }
574 // Determine if a given address can be handled by a large page of the requested size
575 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
578 uint32_t page_size = PAGE_SIZE_4KB;
579 struct v3_mem_region * reg = NULL;
583 if (core->use_large_pages == 1) {
584 pg_start = PAGE_ADDR_4MB(page_addr);
585 pg_end = (pg_start + PAGE_SIZE_4MB);
587 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
589 if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
590 page_size = PAGE_SIZE_4MB;
595 if (core->use_large_pages == 1) {
596 pg_start = PAGE_ADDR_2MB(page_addr);
597 pg_end = (pg_start + PAGE_SIZE_2MB);
599 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
601 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
602 page_size = PAGE_SIZE_2MB;
609 if (core->use_giant_pages == 1) {
610 pg_start = PAGE_ADDR_1GB(page_addr);
611 pg_end = (pg_start + PAGE_SIZE_1GB);
613 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
615 if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
616 page_size = PAGE_SIZE_1GB;
621 if (core->use_large_pages == 1) {
622 pg_start = PAGE_ADDR_2MB(page_addr);
623 pg_end = (pg_start + PAGE_SIZE_2MB);
625 reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
627 if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
628 page_size = PAGE_SIZE_2MB;
633 PrintError(core->vm_info, core, "Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
642 void v3_print_mem_map(struct v3_vm_info * vm) {
643 struct v3_mem_map * map = &(vm->mem_map);
644 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
645 struct v3_mem_region * reg = NULL;
648 V3_Print(vm, VCORE_NONE, "Memory Layout (all cores):\n");
650 V3_Print(vm, VCORE_NONE, "Base Memory: (%d regions)\n", map->num_base_regions);
652 for (i = 0; i < map->num_base_regions; i++) {
653 reg = &(map->base_regions[i]);
655 V3_Print(vm, VCORE_NONE, "Base Region[%d] (all cores): 0x%p - 0x%p -> 0x%p\n",
657 (void *)(reg->guest_start),
658 (void *)(reg->guest_end - 1),
659 (void *)(reg->host_addr));
663 // If the memory map is empty, don't print it
669 reg = rb_entry(node, struct v3_mem_region, tree_node);
671 V3_Print(vm, VCORE_NONE, "%d: 0x%p - 0x%p -> 0x%p\n", i,
672 (void *)(reg->guest_start),
673 (void *)(reg->guest_end - 1),
674 (void *)(reg->host_addr));
676 V3_Print(vm, VCORE_NONE, "\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
682 } while ((node = v3_rb_next(node)));
688 char *arg = v3_lookup_option("mem_block_size");
691 v3_mem_block_size = atoi(arg);
692 V3_Print(VM_NONE,VCORE_NONE,"memory block size set to %llu bytes\n",v3_mem_block_size);
694 V3_Print(VM_NONE,VCORE_NONE,"default memory block size of %llu bytes is in use\n",v3_mem_block_size);