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 map->base_region.guest_start = 0;
65 map->base_region.guest_end = mem_pages * PAGE_SIZE_4KB;
67 #ifdef CONFIG_ALIGNED_PG_ALLOC
68 map->base_region.host_addr = (addr_t)V3_AllocAlignedPages(mem_pages, vm->mem_align);
70 map->base_region.host_addr = (addr_t)V3_AllocPages(mem_pages);
73 map->base_region.flags.read = 1;
74 map->base_region.flags.write = 1;
75 map->base_region.flags.exec = 1;
76 map->base_region.flags.base = 1;
77 map->base_region.flags.alloced = 1;
79 map->base_region.unhandled = unhandled_err;
81 if ((void *)map->base_region.host_addr == NULL) {
82 PrintError("Could not allocate Guest memory\n");
86 //memset(V3_VAddr((void *)map->base_region.host_addr), 0xffffffff, map->base_region.guest_end);
88 v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
94 void v3_delete_mem_map(struct v3_vm_info * vm) {
95 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
96 struct v3_mem_region * reg;
97 struct rb_node * tmp_node = NULL;
100 reg = rb_entry(node, struct v3_mem_region, tree_node);
102 node = v3_rb_next(node);
104 v3_delete_mem_region(vm, reg);
107 V3_FreePage((void *)(vm->mem_map.base_region.host_addr));
111 struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
112 addr_t guest_addr_start, addr_t guest_addr_end) {
114 struct v3_mem_region * entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
115 memset(entry, 0, sizeof(struct v3_mem_region));
117 entry->guest_start = guest_addr_start;
118 entry->guest_end = guest_addr_end;
119 entry->core_id = core_id;
120 entry->unhandled = unhandled_err;
128 int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
129 addr_t guest_addr_start,
130 addr_t guest_addr_end,
133 struct v3_mem_region * entry = NULL;
135 entry = v3_create_mem_region(vm, core_id,
139 entry->host_addr = host_addr;
142 entry->flags.read = 1;
143 entry->flags.write = 1;
144 entry->flags.exec = 1;
145 entry->flags.alloced = 1;
147 if (v3_insert_mem_region(vm, entry) == -1) {
158 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
159 struct v3_mem_region * region) {
160 struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
161 struct rb_node * parent = NULL;
162 struct v3_mem_region * tmp_region;
166 tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
168 if (region->guest_end <= tmp_region->guest_start) {
170 } else if (region->guest_start >= tmp_region->guest_end) {
173 if ((region->guest_end != tmp_region->guest_end) ||
174 (region->guest_start != tmp_region->guest_start)) {
175 PrintError("Trying to map a partial overlapped core specific page...\n");
176 return tmp_region; // This is ugly...
177 } else if (region->core_id == tmp_region->core_id) {
179 } else if (region->core_id < tmp_region->core_id) {
187 rb_link_node(&(region->tree_node), parent, p);
194 int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
195 struct v3_mem_region * ret;
198 if ((ret = __insert_mem_region(vm, region))) {
202 v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
206 for (i = 0; i < vm->num_cores; i++) {
207 struct guest_info * info = &(vm->cores[i]);
209 // flush virtual page tables
210 // 3 cases shadow, shadow passthrough, and nested
212 if (info->shdw_pg_mode == SHADOW_PAGING) {
213 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
215 if (mem_mode == PHYSICAL_MEM) {
218 for (cur_addr = region->guest_start;
219 cur_addr < region->guest_end;
220 cur_addr += PAGE_SIZE_4KB) {
221 v3_invalidate_passthrough_addr(info, cur_addr);
224 v3_invalidate_shadow_pts(info);
227 } else if (info->shdw_pg_mode == NESTED_PAGING) {
230 for (cur_addr = region->guest_start;
231 cur_addr < region->guest_end;
232 cur_addr += PAGE_SIZE_4KB) {
234 v3_invalidate_nested_addr(info, cur_addr);
245 struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
246 struct rb_node * n = vm->mem_map.mem_regions.rb_node;
247 struct v3_mem_region * reg = NULL;
251 reg = rb_entry(n, struct v3_mem_region, tree_node);
253 if (guest_addr < reg->guest_start) {
255 } else if (guest_addr >= reg->guest_end) {
258 if (reg->core_id == V3_MEM_CORE_ANY) {
259 // found relevant region, it's available on all cores
261 } else if (core_id == reg->core_id) {
262 // found relevant region, it's available on the indicated core
264 } else if (core_id < reg->core_id) {
265 // go left, core too big
267 } else if (core_id > reg->core_id) {
268 // go right, core too small
271 PrintDebug("v3_get_mem_region: Impossible!\n");
278 // There is not registered region, so we check if its a valid address in the base region
280 if (guest_addr > vm->mem_map.base_region.guest_end) {
281 PrintError("Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p) (core=0x%x)\n",
282 (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end, core_id);
283 v3_print_mem_map(vm);
288 return &(vm->mem_map.base_region);
293 /* Given an address, find the successor region. If the address is within a region, return that
294 * region. Input is an address, because the address may not have a region associated with it.
296 * Returns a region following or touching the given address. If address is invalid, NULL is
297 * returned, else the base region is returned if no region exists at or after the given address.
299 struct v3_mem_region * v3_get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
300 struct rb_node * current_n = vm->mem_map.mem_regions.rb_node;
301 struct rb_node * successor_n = NULL; /* left-most node greater than guest_addr */
302 struct v3_mem_region * current_r = NULL;
304 /* current_n tries to find the region containing guest_addr, going right when smaller and left when
305 * greater. Each time current_n becomes greater than guest_addr, update successor <- current_n.
306 * current_n becomes successively closer to guest_addr than the previous time it was greater
310 /* | is address, ---- is region, + is intersection */
312 current_r = rb_entry(current_n, struct v3_mem_region, tree_node);
313 if (current_r->guest_start > guest_addr) { /* | ---- */
314 successor_n = current_n;
315 current_n = current_n->rb_left;
317 if (current_r->guest_end > guest_addr) {
318 return current_r; /* +--- or --+- */
320 current_n = current_n->rb_right; /* ---- | */
324 /* Address does not have its own region. Check if it's a valid address in the base region */
326 if (guest_addr >= vm->mem_map.base_region.guest_end) {
327 PrintError("%s: Guest Address Exceeds Base Memory Size (ga=%p), (limit=%p)\n",
328 __FUNCTION__, (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end);
329 v3_print_mem_map(vm);
333 return &(vm->mem_map.base_region);
339 void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
346 for (i = 0; i < vm->num_cores; i++) {
347 struct guest_info * info = &(vm->cores[i]);
349 // flush virtual page tables
350 // 3 cases shadow, shadow passthrough, and nested
352 if (info->shdw_pg_mode == SHADOW_PAGING) {
353 v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
355 if (mem_mode == PHYSICAL_MEM) {
358 for (cur_addr = reg->guest_start;
359 cur_addr < reg->guest_end;
360 cur_addr += PAGE_SIZE_4KB) {
361 v3_invalidate_passthrough_addr(info, cur_addr);
364 v3_invalidate_shadow_pts(info);
367 } else if (info->shdw_pg_mode == NESTED_PAGING) {
370 for (cur_addr = reg->guest_start;
371 cur_addr < reg->guest_end;
372 cur_addr += PAGE_SIZE_4KB) {
374 v3_invalidate_nested_addr(info, cur_addr);
379 v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
383 // flush virtual page tables
384 // 3 cases shadow, shadow passthrough, and nested
388 // Determine if a given address can be handled by a large page of the requested size
389 uint32_t v3_get_max_page_size(struct guest_info * core, addr_t fault_addr, uint32_t req_size) {
390 addr_t pg_start = 0UL, pg_end = 0UL; // large page containing the faulting addres
391 struct v3_mem_region * pg_next_reg = NULL; // next immediate mem reg after page start addr
392 uint32_t page_size = PAGE_SIZE_4KB;
394 /* If the guest has been configured for large pages, then we must check for hooked regions of
395 * memory which may overlap with the large page containing the faulting address (due to
396 * potentially differing access policies in place for e.g. i/o devices and APIC). A large page
397 * can be used if a) no region overlaps the page [or b) a region does overlap but fully contains
398 * the page]. The [bracketed] text pertains to the #if 0'd code below, state D. TODO modify this
399 * note if someone decides to enable this optimization. It can be tested with the SeaStar
402 * Examples: (CAPS regions are returned by v3_get_next_mem_region; state A returns the base reg)
404 * |region| |region| 2MiB mapped (state A)
405 * |reg| |REG| 2MiB mapped (state B)
406 * |region| |reg| |REG| |region| |reg| 4KiB mapped (state C)
407 * |reg| |reg| |--REGION---| [2MiB mapped (state D)]
408 * |--------------------------------------------| RAM
410 * |----|----|----|----|----|page|----|----|----| 2MB pages
411 * >>>>>>>>>>>>>>>>>>>> search space
415 // guest page maps to a host page + offset (so when we shift, it aligns with a host page)
418 return PAGE_SIZE_4KB;
420 pg_start = PAGE_ADDR_2MB(fault_addr);
421 pg_end = (pg_start + PAGE_SIZE_2MB);
424 pg_start = PAGE_ADDR_4MB(fault_addr);
425 pg_end = (pg_start + PAGE_SIZE_4MB);
428 pg_start = PAGE_ADDR_1GB(fault_addr);
429 pg_end = (pg_start + PAGE_SIZE_1GB);
432 PrintError("Invalid large page size requested.\n");
436 //PrintDebug("%s: page [%p,%p) contains address\n", __FUNCTION__, (void *)pg_start, (void *)pg_end);
438 pg_next_reg = v3_get_next_mem_region(core->vm_info, core->cpu_id, pg_start);
440 if (pg_next_reg == NULL) {
441 PrintError("%s: Error: address not in base region, %p\n", __FUNCTION__, (void *)fault_addr);
442 return PAGE_SIZE_4KB;
445 if (pg_next_reg->flags.base == 1) {
446 page_size = req_size; // State A
447 //PrintDebug("%s: base region [%p,%p) contains page.\n", __FUNCTION__,
448 // (void *)pg_next_reg->guest_start, (void *)pg_next_reg->guest_end);
450 #if 0 // State B/C and D optimization
451 if ((pg_next_reg->guest_end >= pg_end) &&
452 ((pg_next_reg->guest_start >= pg_end) || (pg_next_reg->guest_start <= pg_start))) {
453 page_size = req_size;
456 PrintDebug("%s: region [%p,%p) %s partially overlap with page\n", __FUNCTION__,
457 (void *)pg_next_reg->guest_start, (void *)pg_next_reg->guest_end,
458 (page_size == req_size) ? "does not" : "does");
461 if (pg_next_reg->guest_start >= pg_end) {
463 page_size = req_size;
466 PrintDebug("%s: region [%p,%p) %s overlap with page\n", __FUNCTION__,
467 (void *)pg_next_reg->guest_start, (void *)pg_next_reg->guest_end,
468 (page_size == req_size) ? "does not" : "does");
476 // For an address on a page of size page_size, compute the actual alignment
477 // of the physical page it maps to
478 uint32_t v3_compute_page_alignment(addr_t page_addr)
480 if (PAGE_OFFSET_1GB(page_addr) == 0) {
481 return PAGE_SIZE_1GB;
482 } else if (PAGE_OFFSET_4MB(page_addr) == 0) {
483 return PAGE_SIZE_4MB;
484 } else if (PAGE_OFFSET_2MB(page_addr) == 0) {
485 return PAGE_SIZE_2MB;
486 } else if (PAGE_OFFSET_4KB(page_addr) == 0) {
487 return PAGE_SIZE_4KB;
489 PrintError("Non-page aligned address passed to %s.\n", __FUNCTION__);
494 void v3_print_mem_map(struct v3_vm_info * vm) {
495 struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
496 struct v3_mem_region * reg = &(vm->mem_map.base_region);
499 V3_Print("Memory Layout (all cores):\n");
502 V3_Print("Base Region (all cores): 0x%p - 0x%p -> 0x%p\n",
503 (void *)(reg->guest_start),
504 (void *)(reg->guest_end - 1),
505 (void *)(reg->host_addr));
508 // If the memory map is empty, don't print it
514 reg = rb_entry(node, struct v3_mem_region, tree_node);
516 V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
517 (void *)(reg->guest_start),
518 (void *)(reg->guest_end - 1),
519 (void *)(reg->host_addr));
521 V3_Print("\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
527 } while ((node = v3_rb_next(node)));