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>
31 struct v3_shadow_region * insert_shadow_region(struct v3_vm_info * vm,
32 struct v3_shadow_region * region);
35 static int mem_offset_hypercall(struct guest_info * info, uint_t hcall_id, void * private_data) {
36 PrintDebug("V3Vee: Memory offset hypercall (offset=%p)\n",
37 (void *)(info->vm_info->mem_map.base_region.host_addr));
39 info->vm_regs.rbx = info->vm_info->mem_map.base_region.host_addr;
45 int v3_init_mem_map(struct v3_vm_info * vm) {
46 struct v3_mem_map * map = &(vm->mem_map);
47 addr_t mem_pages = vm->mem_size >> 12;
49 map->shdw_regions.rb_node = NULL;
52 map->hook_hvas = V3_VAddr(V3_AllocPages(vm->num_cores));
55 // There is an underlying region that contains all of the guest memory
56 // PrintDebug("Mapping %d pages of memory (%u bytes)\n", (int)mem_pages, (uint_t)info->mem_size);
58 map->base_region.guest_start = 0;
59 map->base_region.guest_end = mem_pages * PAGE_SIZE_4KB;
60 map->base_region.host_type = SHDW_REGION_ALLOCATED;
61 map->base_region.host_addr = (addr_t)V3_AllocPages(mem_pages);
64 if ((void *)map->base_region.host_addr == NULL) {
65 PrintError("Could not allocate Guest memory\n");
69 //memset(V3_VAddr((void *)map->base_region.host_addr), 0xffffffff, map->base_region.guest_end);
71 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
77 static inline addr_t get_hook_hva(struct guest_info * info) {
78 return (addr_t)(info->vm_info->mem_map.hook_hvas + (PAGE_SIZE_4KB * info->cpu_id));
81 void v3_delete_shadow_map(struct v3_vm_info * vm) {
82 struct rb_node * node = v3_rb_first(&(vm->mem_map.shdw_regions));
83 struct v3_shadow_region * reg;
84 struct rb_node * tmp_node = NULL;
87 reg = rb_entry(node, struct v3_shadow_region, tree_node);
89 node = v3_rb_next(node);
91 v3_delete_shadow_region(vm, reg);
94 V3_FreePage((void *)(vm->mem_map.base_region.host_addr));
95 V3_FreePage(V3_PAddr((void *)(vm->mem_map.hook_hvas)));
101 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
102 addr_t guest_addr_start,
103 addr_t guest_addr_end,
106 struct v3_shadow_region * entry = (struct v3_shadow_region *)V3_Malloc(sizeof(struct v3_shadow_region));
108 entry->guest_start = guest_addr_start;
109 entry->guest_end = guest_addr_end;
110 entry->host_type = SHDW_REGION_ALLOCATED;
111 entry->host_addr = host_addr;
112 entry->write_hook = NULL;
113 entry->read_hook = NULL;
114 entry->priv_data = NULL;
115 entry->core_id = core_id;
117 if (insert_shadow_region(vm, entry)) {
127 int v3_hook_write_mem(struct v3_vm_info * vm, uint16_t core_id,
128 addr_t guest_addr_start, addr_t guest_addr_end, addr_t host_addr,
129 int (*write)(struct guest_info * core, addr_t guest_addr, void * src, uint_t length, void * priv_data),
132 struct v3_shadow_region * entry = (struct v3_shadow_region *)V3_Malloc(sizeof(struct v3_shadow_region));
135 entry->guest_start = guest_addr_start;
136 entry->guest_end = guest_addr_end;
137 entry->host_type = SHDW_REGION_WRITE_HOOK;
138 entry->host_addr = host_addr;
139 entry->write_hook = write;
140 entry->read_hook = NULL;
141 entry->priv_data = priv_data;
142 entry->core_id = core_id;
144 if (insert_shadow_region(vm, entry)) {
152 int v3_hook_full_mem(struct v3_vm_info * vm, uint16_t core_id,
153 addr_t guest_addr_start, addr_t guest_addr_end,
154 int (*read)(struct guest_info * core, addr_t guest_addr, void * dst, uint_t length, void * priv_data),
155 int (*write)(struct guest_info * core, addr_t guest_addr, void * src, uint_t length, void * priv_data),
158 struct v3_shadow_region * entry = (struct v3_shadow_region *)V3_Malloc(sizeof(struct v3_shadow_region));
160 entry->guest_start = guest_addr_start;
161 entry->guest_end = guest_addr_end;
162 entry->host_type = SHDW_REGION_FULL_HOOK;
163 entry->host_addr = (addr_t)NULL;
164 entry->write_hook = write;
165 entry->read_hook = read;
166 entry->priv_data = priv_data;
167 entry->core_id = core_id;
169 if (insert_shadow_region(vm, entry)) {
178 // This will unhook the memory hook registered at start address
179 // We do not support unhooking subregions
180 int v3_unhook_mem(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr_start) {
181 struct v3_shadow_region * reg = v3_get_shadow_region(vm, core_id, guest_addr_start);
183 if ((reg->host_type != SHDW_REGION_FULL_HOOK) ||
184 (reg->host_type != SHDW_REGION_WRITE_HOOK)) {
185 PrintError("Trying to unhook a non hooked memory region (addr=%p)\n", (void *)guest_addr_start);
189 v3_delete_shadow_region(vm, reg);
197 struct v3_shadow_region * __insert_shadow_region(struct v3_vm_info * vm,
198 struct v3_shadow_region * region) {
199 struct rb_node ** p = &(vm->mem_map.shdw_regions.rb_node);
200 struct rb_node * parent = NULL;
201 struct v3_shadow_region * tmp_region;
205 tmp_region = rb_entry(parent, struct v3_shadow_region, tree_node);
207 if (region->guest_end <= tmp_region->guest_start) {
209 } else if (region->guest_start >= tmp_region->guest_end) {
212 if ((region->guest_end != tmp_region->guest_end) ||
213 (region->guest_start != tmp_region->guest_start)) {
214 PrintError("Trying to map a partial overlapped core specific page...\n");
215 return tmp_region; // This is ugly...
216 } else if (region->core_id == tmp_region->core_id) {
218 } else if (region->core_id < tmp_region->core_id) {
226 rb_link_node(&(region->tree_node), parent, p);
233 struct v3_shadow_region * insert_shadow_region(struct v3_vm_info * vm,
234 struct v3_shadow_region * region) {
235 struct v3_shadow_region * ret;
238 if ((ret = __insert_shadow_region(vm, region))) {
242 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.shdw_regions));
246 for (i = 0; i < vm->num_cores; i++) {
247 struct guest_info * info = &(vm->cores[i]);
249 // flush virtual page tables
250 // 3 cases shadow, shadow passthrough, and nested
252 if (info->shdw_pg_mode == SHADOW_PAGING) {
253 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
255 if (mem_mode == PHYSICAL_MEM) {
258 for (cur_addr = region->guest_start;
259 cur_addr < region->guest_end;
260 cur_addr += PAGE_SIZE_4KB) {
261 v3_invalidate_passthrough_addr(info, cur_addr);
264 v3_invalidate_shadow_pts(info);
267 } else if (info->shdw_pg_mode == NESTED_PAGING) {
270 for (cur_addr = region->guest_start;
271 cur_addr < region->guest_end;
272 cur_addr += PAGE_SIZE_4KB) {
274 v3_invalidate_nested_addr(info, cur_addr);
285 int handle_special_page_fault(struct guest_info * info,
286 addr_t fault_gva, addr_t fault_gpa, pf_error_t access_info)
288 struct v3_shadow_region * reg = v3_get_shadow_region(info->vm_info, info->cpu_id, fault_gpa);
290 PrintDebug("Handling Special Page Fault\n");
292 switch (reg->host_type) {
293 case SHDW_REGION_WRITE_HOOK:
294 return v3_handle_mem_wr_hook(info, fault_gva, fault_gpa, reg, access_info);
295 case SHDW_REGION_FULL_HOOK:
296 return v3_handle_mem_full_hook(info, fault_gva, fault_gpa, reg, access_info);
305 int v3_handle_mem_wr_hook(struct guest_info * info, addr_t guest_va, addr_t guest_pa,
306 struct v3_shadow_region * reg, pf_error_t access_info) {
308 addr_t dst_addr = (addr_t)V3_VAddr((void *)v3_get_shadow_addr(reg, info->cpu_id, guest_pa));
310 if (v3_emulate_write_op(info, guest_va, guest_pa, dst_addr,
311 reg->write_hook, reg->priv_data) == -1) {
312 PrintError("Write hook emulation failed\n");
319 int v3_handle_mem_full_hook(struct guest_info * info, addr_t guest_va, addr_t guest_pa,
320 struct v3_shadow_region * reg, pf_error_t access_info) {
322 addr_t op_addr = get_hook_hva(info);
324 if (access_info.write == 1) {
325 if (v3_emulate_write_op(info, guest_va, guest_pa, op_addr,
326 reg->write_hook, reg->priv_data) == -1) {
327 PrintError("Write Full Hook emulation failed\n");
331 if (v3_emulate_read_op(info, guest_va, guest_pa, op_addr,
332 reg->read_hook, reg->write_hook,
333 reg->priv_data) == -1) {
334 PrintError("Read Full Hook emulation failed\n");
344 struct v3_shadow_region * v3_get_shadow_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
345 struct rb_node * n = vm->mem_map.shdw_regions.rb_node;
346 struct v3_shadow_region * reg = NULL;
349 reg = rb_entry(n, struct v3_shadow_region, tree_node);
351 if (guest_addr < reg->guest_start) {
353 } else if (guest_addr >= reg->guest_end) {
356 if ((core_id == reg->core_id) ||
357 (reg->core_id == V3_MEM_CORE_ANY)) {
366 // There is not registered region, so we check if its a valid address in the base region
368 if (guest_addr > vm->mem_map.base_region.guest_end) {
369 PrintError("Guest Address Exceeds Base Memory Size (ga=%p), (limit=%p)\n",
370 (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end);
371 v3_print_mem_map(vm);
376 return &(vm->mem_map.base_region);
380 void v3_delete_shadow_region(struct v3_vm_info * vm, struct v3_shadow_region * reg) {
387 for (i = 0; i < vm->num_cores; i++) {
388 struct guest_info * info = &(vm->cores[i]);
390 // flush virtual page tables
391 // 3 cases shadow, shadow passthrough, and nested
393 if (info->shdw_pg_mode == SHADOW_PAGING) {
394 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
396 if (mem_mode == PHYSICAL_MEM) {
399 for (cur_addr = reg->guest_start;
400 cur_addr < reg->guest_end;
401 cur_addr += PAGE_SIZE_4KB) {
402 v3_invalidate_passthrough_addr(info, cur_addr);
405 v3_invalidate_shadow_pts(info);
408 } else if (info->shdw_pg_mode == NESTED_PAGING) {
411 for (cur_addr = reg->guest_start;
412 cur_addr < reg->guest_end;
413 cur_addr += PAGE_SIZE_4KB) {
415 v3_invalidate_nested_addr(info, cur_addr);
420 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.shdw_regions));
424 // flush virtual page tables
425 // 3 cases shadow, shadow passthrough, and nested
432 addr_t v3_get_shadow_addr(struct v3_shadow_region * reg, uint16_t core_id, addr_t guest_addr) {
434 (reg->host_type != SHDW_REGION_FULL_HOOK)) {
435 return (guest_addr - reg->guest_start) + reg->host_addr;
437 // PrintError("MEM Region Invalid\n");
445 void v3_print_mem_map(struct v3_vm_info * vm) {
446 struct rb_node * node = v3_rb_first(&(vm->mem_map.shdw_regions));
447 struct v3_shadow_region * reg = &(vm->mem_map.base_region);
450 V3_Print("Memory Layout:\n");
453 V3_Print("Base Region: 0x%p - 0x%p -> 0x%p\n",
454 (void *)(reg->guest_start),
455 (void *)(reg->guest_end - 1),
456 (void *)(reg->host_addr));
459 // If the memory map is empty, don't print it
465 reg = rb_entry(node, struct v3_shadow_region, tree_node);
467 V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
468 (void *)(reg->guest_start),
469 (void *)(reg->guest_end - 1),
470 (void *)(reg->host_addr));
472 V3_Print("\t(%s) (WriteHook = 0x%p) (ReadHook = 0x%p)\n",
473 v3_shdw_region_type_to_str(reg->host_type),
474 (void *)(reg->write_hook),
475 (void *)(reg->read_hook));
478 } while ((node = v3_rb_next(node)));
482 static const uchar_t SHDW_REGION_WRITE_HOOK_STR[] = "SHDW_REGION_WRITE_HOOK";
483 static const uchar_t SHDW_REGION_FULL_HOOK_STR[] = "SHDW_REGION_FULL_HOOK";
484 static const uchar_t SHDW_REGION_ALLOCATED_STR[] = "SHDW_REGION_ALLOCATED";
486 const uchar_t * v3_shdw_region_type_to_str(v3_shdw_region_type_t type) {
488 case SHDW_REGION_WRITE_HOOK:
489 return SHDW_REGION_WRITE_HOOK_STR;
490 case SHDW_REGION_FULL_HOOK:
491 return SHDW_REGION_FULL_HOOK_STR;
492 case SHDW_REGION_ALLOCATED:
493 return SHDW_REGION_ALLOCATED_STR;
495 return (uchar_t *)"SHDW_REGION_INVALID";