#include <palacios/vmm_shadow_paging.h>
#include <palacios/vmm_direct_paging.h>
-#define MEM_OFFSET_HCALL 0x1000
-static inline
-struct v3_shadow_region * insert_shadow_region(struct v3_vm_info * vm,
- struct v3_shadow_region * region);
-
static int mem_offset_hypercall(struct guest_info * info, uint_t hcall_id, void * private_data) {
PrintDebug("V3Vee: Memory offset hypercall (offset=%p)\n",
return 0;
}
+static int unhandled_err(struct guest_info * core, addr_t guest_va, addr_t guest_pa,
+ struct v3_mem_region * reg, pf_error_t access_info) {
+
+ PrintError("Unhandled memory access error\n");
+
+ v3_print_mem_map(core->vm_info);
+
+ v3_print_guest_state(core);
+
+ return -1;
+}
int v3_init_mem_map(struct v3_vm_info * vm) {
struct v3_mem_map * map = &(vm->mem_map);
addr_t mem_pages = vm->mem_size >> 12;
- map->shdw_regions.rb_node = NULL;
-
-
- map->hook_hvas = V3_VAddr(V3_AllocPages(vm->num_cores));
+ memset(&(map->base_region), 0, sizeof(struct v3_mem_region));
+ map->mem_regions.rb_node = NULL;
// There is an underlying region that contains all of the guest memory
// PrintDebug("Mapping %d pages of memory (%u bytes)\n", (int)mem_pages, (uint_t)info->mem_size);
+ // 2MB page alignment needed for 2MB hardware nested paging
map->base_region.guest_start = 0;
map->base_region.guest_end = mem_pages * PAGE_SIZE_4KB;
- map->base_region.host_type = SHDW_REGION_ALLOCATED;
+
+#ifdef CONFIG_ALIGNED_PG_ALLOC
+ map->base_region.host_addr = (addr_t)V3_AllocAlignedPages(mem_pages, vm->mem_align);
+#else
map->base_region.host_addr = (addr_t)V3_AllocPages(mem_pages);
+#endif
+ map->base_region.flags.read = 1;
+ map->base_region.flags.write = 1;
+ map->base_region.flags.exec = 1;
+ map->base_region.flags.base = 1;
+ map->base_region.flags.alloced = 1;
+
+ map->base_region.unhandled = unhandled_err;
if ((void *)map->base_region.host_addr == NULL) {
PrintError("Could not allocate Guest memory\n");
}
-static inline addr_t get_hook_hva(struct guest_info * info) {
- return (addr_t)(info->vm_info->mem_map.hook_hvas + (PAGE_SIZE_4KB * info->cpu_id));
-}
-
-void v3_delete_shadow_map(struct v3_vm_info * vm) {
- struct rb_node * node = v3_rb_first(&(vm->mem_map.shdw_regions));
- struct v3_shadow_region * reg;
+void v3_delete_mem_map(struct v3_vm_info * vm) {
+ struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
+ struct v3_mem_region * reg;
struct rb_node * tmp_node = NULL;
+ addr_t mem_pages = vm->mem_size >> 12;
while (node) {
- reg = rb_entry(node, struct v3_shadow_region, tree_node);
+ reg = rb_entry(node, struct v3_mem_region, tree_node);
tmp_node = node;
node = v3_rb_next(node);
- v3_delete_shadow_region(vm, reg);
+ v3_delete_mem_region(vm, reg);
}
- V3_FreePage((void *)(vm->mem_map.base_region.host_addr));
- V3_FreePage(V3_PAddr((void *)(vm->mem_map.hook_hvas)));
+ V3_FreePages((void *)(vm->mem_map.base_region.host_addr), mem_pages);
}
-
-
-int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
- addr_t guest_addr_start,
- addr_t guest_addr_end,
- addr_t host_addr)
-{
- struct v3_shadow_region * entry = (struct v3_shadow_region *)V3_Malloc(sizeof(struct v3_shadow_region));
+struct v3_mem_region * v3_create_mem_region(struct v3_vm_info * vm, uint16_t core_id,
+ addr_t guest_addr_start, addr_t guest_addr_end) {
+
+ struct v3_mem_region * entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
+ memset(entry, 0, sizeof(struct v3_mem_region));
entry->guest_start = guest_addr_start;
entry->guest_end = guest_addr_end;
- entry->host_type = SHDW_REGION_ALLOCATED;
- entry->host_addr = host_addr;
- entry->write_hook = NULL;
- entry->read_hook = NULL;
- entry->priv_data = NULL;
entry->core_id = core_id;
+ entry->unhandled = unhandled_err;
- if (insert_shadow_region(vm, entry)) {
- V3_Free(entry);
- return -1;
- }
-
- return 0;
+ return entry;
}
-int v3_hook_write_mem(struct v3_vm_info * vm, uint16_t core_id,
- addr_t guest_addr_start, addr_t guest_addr_end, addr_t host_addr,
- int (*write)(addr_t guest_addr, void * src, uint_t length, void * priv_data),
- void * priv_data) {
- struct v3_shadow_region * entry = (struct v3_shadow_region *)V3_Malloc(sizeof(struct v3_shadow_region));
+int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
+ addr_t guest_addr_start,
+ addr_t guest_addr_end,
+ addr_t host_addr)
+{
+ struct v3_mem_region * entry = NULL;
+ entry = v3_create_mem_region(vm, core_id,
+ guest_addr_start,
+ guest_addr_end);
- entry->guest_start = guest_addr_start;
- entry->guest_end = guest_addr_end;
- entry->host_type = SHDW_REGION_WRITE_HOOK;
entry->host_addr = host_addr;
- entry->write_hook = write;
- entry->read_hook = NULL;
- entry->priv_data = priv_data;
- entry->core_id = core_id;
-
- if (insert_shadow_region(vm, entry)) {
- V3_Free(entry);
- return -1;
- }
-
- return 0;
-}
-
-int v3_hook_full_mem(struct v3_vm_info * vm, uint16_t core_id,
- addr_t guest_addr_start, addr_t guest_addr_end,
- int (*read)(addr_t guest_addr, void * dst, uint_t length, void * priv_data),
- int (*write)(addr_t guest_addr, void * src, uint_t length, void * priv_data),
- void * priv_data) {
-
- struct v3_shadow_region * entry = (struct v3_shadow_region *)V3_Malloc(sizeof(struct v3_shadow_region));
- entry->guest_start = guest_addr_start;
- entry->guest_end = guest_addr_end;
- entry->host_type = SHDW_REGION_FULL_HOOK;
- entry->host_addr = (addr_t)NULL;
- entry->write_hook = write;
- entry->read_hook = read;
- entry->priv_data = priv_data;
- entry->core_id = core_id;
+ entry->flags.read = 1;
+ entry->flags.write = 1;
+ entry->flags.exec = 1;
+ entry->flags.alloced = 1;
- if (insert_shadow_region(vm, entry)) {
+ if (v3_insert_mem_region(vm, entry) == -1) {
V3_Free(entry);
return -1;
}
}
-// This will unhook the memory hook registered at start address
-// We do not support unhooking subregions
-int v3_unhook_mem(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr_start) {
- struct v3_shadow_region * reg = v3_get_shadow_region(vm, core_id, guest_addr_start);
-
- if ((reg->host_type != SHDW_REGION_FULL_HOOK) ||
- (reg->host_type != SHDW_REGION_WRITE_HOOK)) {
- PrintError("Trying to unhook a non hooked memory region (addr=%p)\n", (void *)guest_addr_start);
- return -1;
- }
-
- v3_delete_shadow_region(vm, reg);
-
- return 0;
-}
-
-
static inline
-struct v3_shadow_region * __insert_shadow_region(struct v3_vm_info * vm,
- struct v3_shadow_region * region) {
- struct rb_node ** p = &(vm->mem_map.shdw_regions.rb_node);
+struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm,
+ struct v3_mem_region * region) {
+ struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
struct rb_node * parent = NULL;
- struct v3_shadow_region * tmp_region;
+ struct v3_mem_region * tmp_region;
while (*p) {
parent = *p;
- tmp_region = rb_entry(parent, struct v3_shadow_region, tree_node);
+ tmp_region = rb_entry(parent, struct v3_mem_region, tree_node);
if (region->guest_end <= tmp_region->guest_start) {
p = &(*p)->rb_left;
return tmp_region;
} else if (region->core_id < tmp_region->core_id) {
p = &(*p)->rb_left;
- } else {
+ } else {
p = &(*p)->rb_right;
}
}
}
-static inline
-struct v3_shadow_region * insert_shadow_region(struct v3_vm_info * vm,
- struct v3_shadow_region * region) {
- struct v3_shadow_region * ret;
+
+int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
+ struct v3_mem_region * ret;
int i = 0;
- if ((ret = __insert_shadow_region(vm, region))) {
- return ret;
+ if ((ret = __insert_mem_region(vm, region))) {
+ return -1;
}
- v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.shdw_regions));
+ v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
}
}
- return NULL;
+ return 0;
}
-int handle_special_page_fault(struct guest_info * info,
- addr_t fault_gva, addr_t fault_gpa, pf_error_t access_info)
-{
- struct v3_shadow_region * reg = v3_get_shadow_region(info->vm_info, info->cpu_id, fault_gpa);
-
- PrintDebug("Handling Special Page Fault\n");
-
- switch (reg->host_type) {
- case SHDW_REGION_WRITE_HOOK:
- return v3_handle_mem_wr_hook(info, fault_gva, fault_gpa, reg, access_info);
- case SHDW_REGION_FULL_HOOK:
- return v3_handle_mem_full_hook(info, fault_gva, fault_gpa, reg, access_info);
- default:
- return -1;
- }
-
- return 0;
-
-}
+struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
+ struct rb_node * n = vm->mem_map.mem_regions.rb_node;
+ struct v3_mem_region * reg = NULL;
-int v3_handle_mem_wr_hook(struct guest_info * info, addr_t guest_va, addr_t guest_pa,
- struct v3_shadow_region * reg, pf_error_t access_info) {
+ while (n) {
- addr_t dst_addr = (addr_t)V3_VAddr((void *)v3_get_shadow_addr(reg, info->cpu_id, guest_pa));
+ reg = rb_entry(n, struct v3_mem_region, tree_node);
- if (v3_emulate_write_op(info, guest_va, guest_pa, dst_addr,
- reg->write_hook, reg->priv_data) == -1) {
- PrintError("Write hook emulation failed\n");
- return -1;
+ if (guest_addr < reg->guest_start) {
+ n = n->rb_left;
+ } else if (guest_addr >= reg->guest_end) {
+ n = n->rb_right;
+ } else {
+ if (reg->core_id == V3_MEM_CORE_ANY) {
+ // found relevant region, it's available on all cores
+ return reg;
+ } else if (core_id == reg->core_id) {
+ // found relevant region, it's available on the indicated core
+ return reg;
+ } else if (core_id < reg->core_id) {
+ // go left, core too big
+ n = n->rb_left;
+ } else if (core_id > reg->core_id) {
+ // go right, core too small
+ n = n->rb_right;
+ } else {
+ PrintDebug("v3_get_mem_region: Impossible!\n");
+ return NULL;
+ }
+ }
}
- return 0;
-}
-int v3_handle_mem_full_hook(struct guest_info * info, addr_t guest_va, addr_t guest_pa,
- struct v3_shadow_region * reg, pf_error_t access_info) {
-
- addr_t op_addr = get_hook_hva(info);
+ // There is not registered region, so we check if its a valid address in the base region
- if (access_info.write == 1) {
- if (v3_emulate_write_op(info, guest_va, guest_pa, op_addr,
- reg->write_hook, reg->priv_data) == -1) {
- PrintError("Write Full Hook emulation failed\n");
- return -1;
- }
- } else {
- if (v3_emulate_read_op(info, guest_va, guest_pa, op_addr,
- reg->read_hook, reg->write_hook,
- reg->priv_data) == -1) {
- PrintError("Read Full Hook emulation failed\n");
- return -1;
- }
+ if (guest_addr > vm->mem_map.base_region.guest_end) {
+ PrintError("Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p) (core=0x%x)\n",
+ (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end, core_id);
+ v3_print_mem_map(vm);
+
+ return NULL;
}
- return 0;
+ return &(vm->mem_map.base_region);
}
-struct v3_shadow_region * v3_get_shadow_region(struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
- struct rb_node * n = vm->mem_map.shdw_regions.rb_node;
- struct v3_shadow_region * reg = NULL;
+/* This returns the next memory region based on a given address.
+ * If the address falls inside a sub region, that region is returned.
+ * If the address falls outside a sub region, the next sub region is returned
+ * NOTE that we have to be careful about core_ids here...
+ */
+static struct v3_mem_region * get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
+ struct rb_node * n = vm->mem_map.mem_regions.rb_node;
+ struct v3_mem_region * reg = NULL;
+ struct v3_mem_region * parent = NULL;
while (n) {
- reg = rb_entry(n, struct v3_shadow_region, tree_node);
+
+ reg = rb_entry(n, struct v3_mem_region, tree_node);
if (guest_addr < reg->guest_start) {
n = n->rb_left;
} else if (guest_addr >= reg->guest_end) {
n = n->rb_right;
} else {
- if ((core_id == reg->core_id) ||
- (reg->core_id == V3_MEM_CORE_ANY)) {
- return reg;
- } else {
+ if (reg->core_id == V3_MEM_CORE_ANY) {
+ // found relevant region, it's available on all cores
+ return reg;
+ } else if (core_id == reg->core_id) {
+ // found relevant region, it's available on the indicated core
+ return reg;
+ } else if (core_id < reg->core_id) {
+ // go left, core too big
+ n = n->rb_left;
+ } else if (core_id > reg->core_id) {
+ // go right, core too small
n = n->rb_right;
+ } else {
+ PrintError("v3_get_mem_region: Impossible!\n");
+ return NULL;
}
}
+
+ if ((reg->core_id == core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
+ parent = reg;
+ }
}
- // There is not registered region, so we check if its a valid address in the base region
+ if (parent->guest_start > guest_addr) {
+ return parent;
+ } else if (parent->guest_end < guest_addr) {
+ struct rb_node * node = &(parent->tree_node);
- if (guest_addr > vm->mem_map.base_region.guest_end) {
- PrintError("Guest Address Exceeds Base Memory Size (ga=%p), (limit=%p)\n",
- (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end);
- v3_print_mem_map(vm);
+ while ((node = v3_rb_next(node)) != NULL) {
+ struct v3_mem_region * next_reg = rb_entry(node, struct v3_mem_region, tree_node);
+
+ if ((next_reg->core_id == V3_MEM_CORE_ANY) ||
+ (next_reg->core_id == core_id)) {
+
+ // This check is not strictly necessary, but it makes it clearer
+ if (next_reg->guest_start > guest_addr) {
+ return next_reg;
+ }
+ }
+ }
+ }
+
+ return NULL;
+}
+
+
+
+/* Given an address region of memory, find if there are any regions that overlap with it.
+ * This checks that the range lies in a single region, and returns that region if it does,
+ * this can be either the base region or a sub region.
+ * IF there are multiple regions in the range then it returns NULL
+ */
+static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uint16_t core_id,
+ addr_t start_gpa, addr_t end_gpa) {
+ struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
+
+ if (start_region == NULL) {
+ PrintError("Invalid memory region\n");
return NULL;
}
-
- return &(vm->mem_map.base_region);
+
+
+ if (start_region->guest_end < end_gpa) {
+ // Region ends before range
+ return NULL;
+ } else if (start_region->flags.base == 0) {
+ // sub region overlaps range
+ return start_region;
+ } else {
+ // Base region, now we have to scan forward for the next sub region
+ struct v3_mem_region * next_reg = get_next_mem_region(vm, core_id, start_gpa);
+
+ if (next_reg == NULL) {
+ // no sub regions after start_addr, base region is ok
+ return start_region;
+ } else if (next_reg->guest_start >= end_gpa) {
+ // Next sub region begins outside range
+ return start_region;
+ } else {
+ return NULL;
+ }
+ }
+
+
+ // Should never get here
+ return NULL;
}
-void v3_delete_shadow_region(struct v3_vm_info * vm, struct v3_shadow_region * reg) {
+
+
+
+void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
int i = 0;
if (reg == NULL) {
return;
}
+
+ v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
+
+
+
+ // If the guest isn't running then there shouldn't be anything to invalidate.
+ // Page tables should __always__ be created on demand during execution
+ // NOTE: This is a sanity check, and can be removed if that assumption changes
+ if (vm->run_state != VM_RUNNING) {
+ V3_Free(reg);
+ return;
+ }
+
for (i = 0; i < vm->num_cores; i++) {
struct guest_info * info = &(vm->cores[i]);
}
}
- v3_rb_erase(&(reg->tree_node), &(vm->mem_map.shdw_regions));
-
V3_Free(reg);
// flush virtual page tables
}
+// Determine if a given address can be handled by a large page of the requested size
+uint32_t v3_get_max_page_size(struct guest_info * core, addr_t page_addr, v3_cpu_mode_t mode) {
+ addr_t pg_start = 0;
+ addr_t pg_end = 0;
+ uint32_t page_size = PAGE_SIZE_4KB;
+ struct v3_mem_region * reg = NULL;
+
+ switch (mode) {
+ case PROTECTED:
+ if (core->use_large_pages == 1) {
+ pg_start = PAGE_ADDR_4MB(page_addr);
+ pg_end = (pg_start + PAGE_SIZE_4MB);
+ reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
+ if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
+ page_size = PAGE_SIZE_4MB;
+ }
+ }
+ break;
+ case PROTECTED_PAE:
+ if (core->use_large_pages == 1) {
+ pg_start = PAGE_ADDR_2MB(page_addr);
+ pg_end = (pg_start + PAGE_SIZE_2MB);
-addr_t v3_get_shadow_addr(struct v3_shadow_region * reg, uint16_t core_id, addr_t guest_addr) {
- if ( (reg) &&
- (reg->host_type != SHDW_REGION_FULL_HOOK)) {
- return (guest_addr - reg->guest_start) + reg->host_addr;
- } else {
- // PrintError("MEM Region Invalid\n");
- return 0;
+ reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
+
+ if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
+ page_size = PAGE_SIZE_2MB;
+ }
+ }
+ break;
+ case LONG:
+ case LONG_32_COMPAT:
+ case LONG_16_COMPAT:
+ if (core->use_giant_pages == 1) {
+ pg_start = PAGE_ADDR_1GB(page_addr);
+ pg_end = (pg_start + PAGE_SIZE_1GB);
+
+ reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
+
+ if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
+ page_size = PAGE_SIZE_1GB;
+ break;
+ }
+ }
+
+ if (core->use_large_pages == 1) {
+ pg_start = PAGE_ADDR_2MB(page_addr);
+ pg_end = (pg_start + PAGE_SIZE_2MB);
+
+ reg = get_overlapping_region(core->vm_info, core->cpu_id, pg_start, pg_end);
+
+ if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
+ page_size = PAGE_SIZE_2MB;
+ }
+ }
+ break;
+ default:
+ PrintError("Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
+ return -1;
}
+ return page_size;
}
void v3_print_mem_map(struct v3_vm_info * vm) {
- struct rb_node * node = v3_rb_first(&(vm->mem_map.shdw_regions));
- struct v3_shadow_region * reg = &(vm->mem_map.base_region);
+ struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
+ struct v3_mem_region * reg = &(vm->mem_map.base_region);
int i = 0;
- V3_Print("Memory Layout:\n");
+ V3_Print("Memory Layout (all cores):\n");
- V3_Print("Base Region: 0x%p - 0x%p -> 0x%p\n",
+ V3_Print("Base Region (all cores): 0x%p - 0x%p -> 0x%p\n",
(void *)(reg->guest_start),
(void *)(reg->guest_end - 1),
(void *)(reg->host_addr));
}
do {
- reg = rb_entry(node, struct v3_shadow_region, tree_node);
+ reg = rb_entry(node, struct v3_mem_region, tree_node);
V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
(void *)(reg->guest_start),
(void *)(reg->guest_end - 1),
(void *)(reg->host_addr));
- V3_Print("\t(%s) (WriteHook = 0x%p) (ReadHook = 0x%p)\n",
- v3_shdw_region_type_to_str(reg->host_type),
- (void *)(reg->write_hook),
- (void *)(reg->read_hook));
+ V3_Print("\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
+ reg->flags.value,
+ reg->core_id,
+ reg->unhandled);
i++;
} while ((node = v3_rb_next(node)));
}
-
-static const uchar_t SHDW_REGION_WRITE_HOOK_STR[] = "SHDW_REGION_WRITE_HOOK";
-static const uchar_t SHDW_REGION_FULL_HOOK_STR[] = "SHDW_REGION_FULL_HOOK";
-static const uchar_t SHDW_REGION_ALLOCATED_STR[] = "SHDW_REGION_ALLOCATED";
-
-const uchar_t * v3_shdw_region_type_to_str(v3_shdw_region_type_t type) {
- switch (type) {
- case SHDW_REGION_WRITE_HOOK:
- return SHDW_REGION_WRITE_HOOK_STR;
- case SHDW_REGION_FULL_HOOK:
- return SHDW_REGION_FULL_HOOK_STR;
- case SHDW_REGION_ALLOCATED:
- return SHDW_REGION_ALLOCATED_STR;
- default:
- return (uchar_t *)"SHDW_REGION_INVALID";
- }
-}
-