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".
21 #include <palacios/vmm_shadow_paging.h>
24 #include <palacios/vmm.h>
25 #include <palacios/vm_guest_mem.h>
26 #include <palacios/vmm_decoder.h>
27 #include <palacios/vmm_ctrl_regs.h>
29 #include <palacios/vmm_hashtable.h>
31 #include <palacios/vmm_direct_paging.h>
36 #ifdef CONFIG_SHADOW_PAGING_TELEMETRY
37 #include <palacios/vmm_telemetry.h>
40 #ifdef CONFIG_SYMBIOTIC_SWAP
41 #include <palacios/vmm_sym_swap.h>
44 #ifndef CONFIG_DEBUG_SHADOW_PAGING
46 #define PrintDebug(fmt, args...)
51 static struct hashtable * master_shdw_pg_table = NULL;
53 static uint_t shdw_pg_hash_fn(addr_t key) {
54 char * name = (char *)key;
55 return v3_hash_buffer((uint8_t *)name, strlen(name));
58 static int shdw_pg_eq_fn(addr_t key1, addr_t key2) {
59 char * name1 = (char *)key1;
60 char * name2 = (char *)key2;
62 return (strcmp(name1, name2) == 0);
66 int V3_init_shdw_paging() {
67 extern struct v3_shdw_pg_impl * __start__v3_shdw_pg_impls[];
68 extern struct v3_shdw_pg_impl * __stop__v3_shdw_pg_impls[];
69 struct v3_shdw_pg_impl ** tmp_impl = __start__v3_shdw_pg_impls;
72 master_shdw_pg_table = v3_create_htable(0, shdw_pg_hash_fn, shdw_pg_eq_fn);
75 while (tmp_impl != __stop__v3_shdw_pg_impls) {
76 V3_Print("Registering Shadow Paging Impl (%s)\n", (*tmp_impl)->name);
78 if (v3_htable_search(master_shdw_pg_table, (addr_t)((*tmp_impl)->name))) {
79 PrintError("Multiple instances of shadow paging impl (%s)\n", (*tmp_impl)->name);
83 if (v3_htable_insert(master_shdw_pg_table,
84 (addr_t)((*tmp_impl)->name),
85 (addr_t)(*tmp_impl)) == 0) {
86 PrintError("Could not register shadow paging impl (%s)\n", (*tmp_impl)->name);
90 tmp_impl = &(__start__v3_shdw_pg_impls[++i]);
103 #ifdef CONFIG_SHADOW_PAGING_TELEMETRY
104 static void telemetry_cb(struct v3_vm_info * vm, void * private_data, char * hdr) {
106 for (i = 0; i < vm->num_cores; i++) {
107 struct guest_info * core = &(vm->cores[i]);
109 V3_Print("%s Guest Page faults: %d\n", hdr, core->shdw_pg_state.guest_faults);
116 int v3_init_shdw_pg_state(struct guest_info * core) {
117 struct v3_shdw_pg_state * state = &(core->shdw_pg_state);
118 struct v3_shdw_pg_impl * impl = core->vm_info->shdw_impl.current_impl;
121 state->guest_cr3 = 0;
122 state->guest_cr0 = 0;
123 state->guest_efer.value = 0x0LL;
126 if (impl->local_init(core) == -1) {
127 PrintError("Error in Shadow paging local initialization (%s)\n", impl->name);
132 #ifdef CONFIG_SHADOW_PAGING_TELEMETRY
133 v3_add_telemetry_cb(core->vm_info, telemetry_cb, NULL);
141 int v3_init_shdw_impl(struct v3_vm_info * vm) {
142 struct v3_shdw_impl_state * impl_state = &(vm->shdw_impl);
143 v3_cfg_tree_t * pg_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "paging");
144 char * type_name = v3_cfg_val(pg_cfg, NULL);
145 char * impl_name = v3_cfg_val(pg_cfg, "mode");
146 struct v3_shdw_pg_impl * impl = NULL;
148 if (type_name && (strcasecmp(type_name, "shadow") != 0)) {
149 PrintDebug("Shadow paging not specified for VM.\n");
153 V3_Print("Initialization of Shadow Paging implementation\n");
155 impl = (struct v3_shdw_pg_impl *)v3_htable_search(master_shdw_pg_table, (addr_t)impl_name);
158 PrintError("Could not find shadow paging impl (%s)\n", impl_name);
162 impl_state->current_impl = impl;
164 if (impl->init(vm, pg_cfg) == -1) {
165 PrintError("Could not initialize Shadow paging implemenation (%s)\n", impl->name);
176 // Reads the guest CR3 register
177 // creates new shadow page tables
178 // updates the shadow CR3 register to point to the new pts
179 int v3_activate_shadow_pt(struct guest_info * core) {
180 struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
181 struct v3_shdw_pg_impl * impl = state->current_impl;
182 return impl->activate_shdw_pt(core);
187 // This must flush any caches
188 // and reset the cr3 value to the correct value
189 int v3_invalidate_shadow_pts(struct guest_info * core) {
190 struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
191 struct v3_shdw_pg_impl * impl = state->current_impl;
192 return impl->invalidate_shdw_pt(core);
196 int v3_handle_shadow_pagefault(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
198 if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) {
199 // If paging is not turned on we need to handle the special cases
200 return v3_handle_passthrough_pagefault(core, fault_addr, error_code);
201 } else if (v3_get_vm_mem_mode(core) == VIRTUAL_MEM) {
202 struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
203 struct v3_shdw_pg_impl * impl = state->current_impl;
205 return impl->handle_pagefault(core, fault_addr, error_code);
207 PrintError("Invalid Memory mode\n");
213 int v3_handle_shadow_invlpg(struct guest_info * core) {
215 struct x86_instr dec_instr;
219 if (v3_get_vm_mem_mode(core) != VIRTUAL_MEM) {
220 // Paging must be turned on...
221 // should handle with some sort of fault I think
222 PrintError("ERROR: INVLPG called in non paged mode\n");
226 if (v3_get_vm_mem_mode(core) == PHYSICAL_MEM) {
227 ret = v3_read_gpa_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr);
229 ret = v3_read_gva_memory(core, get_addr_linear(core, core->rip, &(core->segments.cs)), 15, instr);
233 PrintError("Could not read instruction into buffer\n");
237 if (v3_decode(core, (addr_t)instr, &dec_instr) == -1) {
238 PrintError("Decoding Error\n");
242 if ((dec_instr.op_type != V3_OP_INVLPG) ||
243 (dec_instr.num_operands != 1) ||
244 (dec_instr.dst_operand.type != MEM_OPERAND)) {
245 PrintError("Decoder Error: Not a valid INVLPG instruction...\n");
249 vaddr = dec_instr.dst_operand.operand;
251 core->rip += dec_instr.instr_length;
254 struct v3_shdw_impl_state * state = &(core->vm_info->shdw_impl);
255 struct v3_shdw_pg_impl * impl = state->current_impl;
257 return impl->handle_invlpg(core, vaddr);
266 int v3_inject_guest_pf(struct guest_info * core, addr_t fault_addr, pf_error_t error_code) {
267 core->ctrl_regs.cr2 = fault_addr;
269 #ifdef CONFIG_SHADOW_PAGING_TELEMETRY
270 core->shdw_pg_state.guest_faults++;
273 return v3_raise_exception_with_error(core, PF_EXCEPTION, *(uint_t *)&error_code);
277 int v3_is_guest_pf(pt_access_status_t guest_access, pt_access_status_t shadow_access) {
278 /* basically the reasoning is that there can be multiple reasons for a page fault:
279 If there is a permissions failure for a page present in the guest _BUT_
280 the reason for the fault was that the page is not present in the shadow,
281 _THEN_ we have to map the shadow page in and reexecute, this will generate
282 a permissions fault which is _THEN_ valid to send to the guest
283 _UNLESS_ both the guest and shadow have marked the page as not present
287 if (guest_access != PT_ACCESS_OK) {
288 // Guest Access Error
290 if ((shadow_access != PT_ACCESS_NOT_PRESENT) &&
291 (guest_access != PT_ACCESS_NOT_PRESENT)) {
292 // aka (guest permission error)
297 if ((shadow_access == PT_ACCESS_NOT_PRESENT) &&
298 (guest_access == PT_ACCESS_NOT_PRESENT)) {
299 // Page tables completely blank, handle guest first
304 if (guest_access == PT_ACCESS_NOT_PRESENT) {
305 // Page tables completely blank, handle guest first
309 // Otherwise we'll handle the guest fault later...?