#include <palacios/vmm_util.h>
#include <palacios/vmm_emulator.h>
#include <palacios/vm_guest.h>
+#include <palacios/vmm_debug.h>
#include <palacios/vmm_shadow_paging.h>
#include <palacios/vmm_direct_paging.h>
+#include <interfaces/vmm_numa.h>
+
+#ifdef V3_CONFIG_SWAPPING
+#include <palacios/vmm_swapping.h>
+#endif
+
+uint64_t v3_mem_block_size = V3_CONFIG_MEM_BLOCK_SIZE;
+
+
+
+
+struct v3_mem_region * v3_get_base_region(struct v3_vm_info * vm, addr_t gpa) {
+
+ //PrintDebug(VM_NONE, VCORE_NONE, "get_base_region called");
+ struct v3_mem_map * map = &(vm->mem_map);
+ uint32_t block_index = gpa / v3_mem_block_size;
+ struct v3_mem_region *reg;
+ if ((gpa >= (map->num_base_regions * v3_mem_block_size)) ||
+ (block_index >= map->num_base_regions)) {
+ PrintError(vm, VCORE_NONE, "Guest Address Exceeds Base Memory Size (ga=0x%p), (limit=0x%p)\n",
+ (void *)gpa, (void *)vm->mem_size);
+ v3_print_mem_map(vm);
+
+ return NULL;
+ }
+
+ reg = &(map->base_regions[block_index]);
+
+#ifdef V3_CONFIG_SWAPPING
+ if(vm->swap_state.enable_swapping) {
+ if (reg->flags.swapped) {
+ if (v3_swap_in_region(vm,reg)) {
+ PrintError(vm, VCORE_NONE, "Unable to swap in region GPA=%p..%p!!!\n",(void*)reg->guest_start,(void*)reg->guest_end);
+ v3_print_mem_map(vm);
+ return NULL;
+ }
+ }
+ }
+ v3_touch_region(vm,reg);
+#endif
+
+ return reg;
+}
static int mem_offset_hypercall(struct guest_info * info, uint_t hcall_id, void * private_data) {
- PrintDebug("V3Vee: Memory offset hypercall (offset=%p)\n",
+ /*
+ PrintDebug(info->vm_info, info,"V3Vee: Memory offset hypercall (offset=%p)\n",
(void *)(info->vm_info->mem_map.base_region.host_addr));
info->vm_regs.rbx = info->vm_info->mem_map.base_region.host_addr;
-
- return 0;
+ */
+ return -1;
}
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 (gpa=%p, gva=%p, error_code=%d)\n",
+ PrintError(core->vm_info, core, "Unhandled memory access error (gpa=%p, gva=%p, error_code=%d)\n",
(void *)guest_pa, (void *)guest_va, *(uint32_t *)&access_info);
v3_print_mem_map(core->vm_info);
return -1;
}
+static int gpa_to_node_from_cfg(struct v3_vm_info * vm, addr_t gpa) {
+ v3_cfg_tree_t * layout_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "mem_layout");
+ v3_cfg_tree_t * region_desc = v3_cfg_subtree(layout_cfg, "region");
+
+ while (region_desc) {
+ char * start_addr_str = v3_cfg_val(region_desc, "start_addr");
+ char * end_addr_str = v3_cfg_val(region_desc, "end_addr");
+ char * node_id_str = v3_cfg_val(region_desc, "node");
+
+ addr_t start_addr = 0;
+ addr_t end_addr = 0;
+ int node_id = 0;
+
+ if ((!start_addr_str) || (!end_addr_str) || (!node_id_str)) {
+ PrintError(vm, VCORE_NONE, "Invalid memory layout in configuration\n");
+ return -1;
+ }
+
+ start_addr = atox(start_addr_str);
+ end_addr = atox(end_addr_str);
+ node_id = atoi(node_id_str);
+
+ if ((gpa >= start_addr) && (gpa < end_addr)) {
+ return node_id;
+ }
+
+ region_desc = v3_cfg_next_branch(region_desc);
+ }
+
+ return -1;
+}
+
+//
+// This code parallels that in vmm_shadow_paging.c:v3_init_shdw_impl()
+// and vmm_config.c:determine_paging_mode. The determination of which
+// paging mode will be used is determined much later than the allocation of
+// the guest memory regions, so we need to do this here to decide if they
+// need to be below 4 GB or not.
+static int will_use_shadow_paging(struct v3_vm_info *vm)
+{
+ v3_cfg_tree_t * pg_cfg = v3_cfg_subtree(vm->cfg_data->cfg, "paging");
+ char * pg_mode = v3_cfg_val(pg_cfg, "mode");
+
+ if (pg_mode == NULL) {
+ return 1; // did not ask, get shadow
+ } else {
+ if (strcasecmp(pg_mode, "nested") == 0) {
+ extern v3_cpu_arch_t v3_mach_type;
+ if ((v3_mach_type == V3_SVM_REV3_CPU) ||
+ (v3_mach_type == V3_VMX_EPT_CPU) ||
+ (v3_mach_type == V3_VMX_EPT_UG_CPU)) {
+ return 0; // ask for nested, get nested
+ } else {
+ return 1; // ask for nested, get shadow
+ }
+ } else if (strcasecmp(pg_mode, "shadow") != 0) {
+ return 1; // ask for shadow, get shadow
+ } else {
+ return 1; // ask for something else, get shadow
+ }
+ }
+}
+
+#define CEIL_DIV(x,y) (((x)/(y)) + !!((x)%(y)))
+
+
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;
+ addr_t block_pages = v3_mem_block_size >> 12;
+ int i = 0;
+ uint64_t num_base_regions_host_mem;
- memset(&(map->base_region), 0, sizeof(struct v3_mem_region));
+ map->num_base_regions = CEIL_DIV(vm->mem_size, v3_mem_block_size);
- map->mem_regions.rb_node = NULL;
+ num_base_regions_host_mem=map->num_base_regions; // without swapping
- // 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);
+ PrintDebug(VM_NONE, VCORE_NONE, "v3_init_mem_map: num_base_regions:%d",map->num_base_regions);
- // 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->mem_regions.rb_node = NULL;
-#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);
+#ifdef V3_CONFIG_SWAPPING
+ if (vm->swap_state.enable_swapping) {
+ num_base_regions_host_mem = CEIL_DIV(vm->swap_state.host_mem_size, v3_mem_block_size);
+ }
#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;
+ PrintDebug(VM_NONE, VCORE_NONE, "v3_init_mem_map: %llu base regions will be allocated of %llu base regions in guest\n",
+ (uint64_t)num_base_regions_host_mem, (uint64_t)map->num_base_regions);
- map->base_region.unhandled = unhandled_err;
-
- if ((void *)map->base_region.host_addr == NULL) {
- PrintError("Could not allocate Guest memory\n");
+ map->base_regions = V3_VMalloc(sizeof(struct v3_mem_region) * map->num_base_regions);
+ if (map->base_regions == NULL) {
+ PrintError(vm, VCORE_NONE, "Could not allocate base region array\n");
return -1;
}
+
+ memset(map->base_regions, 0, sizeof(struct v3_mem_region) * map->num_base_regions);
+
+ for (i = 0; i < map->num_base_regions; i++) {
+
+
+ struct v3_mem_region * region = &(map->base_regions[i]);
+ int node_id = -1;
+
+ // 2MB page alignment needed for 2MB hardware nested paging
+ // If swapping is enabled, the host memory will be allocated to low address regions at initialization
+ region->guest_start = v3_mem_block_size * i;
+ region->guest_end = region->guest_start + v3_mem_block_size;
+
+ // We assume that the xml config was smart enough to align the layout to the block size
+ // If they didn't we're going to ignore their settings
+ // and use whatever node the first byte of the block is assigned to
+ node_id = gpa_to_node_from_cfg(vm, region->guest_start);
+
+
+ if (i < num_base_regions_host_mem) {
+ //The regions within num_base_regions_in_mem are allocated in host memory
+ V3_Print(vm, VCORE_NONE, "Allocating block %d on node %d\n", i, node_id);
+
+#ifdef V3_CONFIG_SWAPPING
+ // nothing to do - memset will have done it.
+#endif
+
+ region->host_addr = (addr_t)V3_AllocPagesExtended(block_pages,
+ PAGE_SIZE_4KB,
+ node_id,
+ vm->resource_control.pg_filter_func,
+ vm->resource_control.pg_filter_state);
+
+ if ((void *)region->host_addr == NULL) {
+ PrintError(vm, VCORE_NONE, "Could not allocate guest memory\n");
+ return -1;
+ }
+
+ // Clear the memory...
+ memset(V3_VAddr((void *)region->host_addr), 0, v3_mem_block_size);
+
+ } else {
+
+#ifdef V3_CONFIG_SWAPPING
+ if(vm->swap_state.enable_swapping) {
+ // The regions beyond num_base_regions_in_mem are allocated on disk to start
+ region->flags.swapped = 1;
+ region->host_addr=(addr_t) 0;
+ // other flags / state correctly set up by zeroing the region earlier
+ }
+#endif
+
+ }
+
+
+ // Note assigned numa ID could be different than our request...
+ // Also note that when swapping is used, the numa info will
+ // reflect the numa id of address 0x0 for unallocated regions
+ //
+ region->numa_id = v3_numa_hpa_to_node(region->host_addr);
+
+ region->flags.read = 1;
+ region->flags.write = 1;
+ region->flags.exec = 1;
+ region->flags.base = 1;
+ region->flags.alloced = 1;
+ region->flags.limit32 = will_use_shadow_paging(vm);
- //memset(V3_VAddr((void *)map->base_region.host_addr), 0xffffffff, map->base_region.guest_end);
+ region->unhandled = unhandled_err;
+ }
v3_register_hypercall(vm, MEM_OFFSET_HCALL, mem_offset_hypercall, NULL);
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_map * map = &(vm->mem_map);
+ struct rb_node * node = v3_rb_first(&(map->mem_regions));
struct v3_mem_region * reg;
struct rb_node * tmp_node = NULL;
- addr_t mem_pages = vm->mem_size >> 12;
-
+ addr_t block_pages = v3_mem_block_size >> 12;
+ int i = 0;
+
while (node) {
reg = rb_entry(node, struct v3_mem_region, tree_node);
tmp_node = node;
v3_delete_mem_region(vm, reg);
}
- V3_FreePages((void *)(vm->mem_map.base_region.host_addr), mem_pages);
+ for (i = 0; i < map->num_base_regions; i++) {
+ struct v3_mem_region * region = &(map->base_regions[i]);
+#ifdef V3_CONFIG_SWAPPING
+ if (vm->swap_state.enable_swapping) {
+ if (!region->flags.swapped) {
+ V3_FreePages((void *)(region->host_addr), block_pages);
+ } // otherwise this is not allocated space
+ }
+#else
+ V3_FreePages((void *)(region->host_addr), block_pages);
+#endif
+ }
+
+ V3_VFree(map->base_regions);
}
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));
+ struct v3_mem_region * entry = NULL;
+
+ if (guest_addr_start >= guest_addr_end) {
+ PrintError(vm, VCORE_NONE, "Region start is after region end\n");
+ return NULL;
+ }
+
+ entry = (struct v3_mem_region *)V3_Malloc(sizeof(struct v3_mem_region));
+
+ if (!entry) {
+ PrintError(vm, VCORE_NONE, "Cannot allocate in creating a memory region\n");
+ return NULL;
+ }
+
memset(entry, 0, sizeof(struct v3_mem_region));
entry->guest_start = guest_addr_start;
} else {
if ((region->guest_end != tmp_region->guest_end) ||
(region->guest_start != tmp_region->guest_start)) {
- PrintError("Trying to map a partial overlapped core specific page...\n");
+ PrintError(vm, VCORE_NONE, "Trying to map a partial overlapped core specific page...\n");
return tmp_region; // This is ugly...
} else if (region->core_id == tmp_region->core_id) {
+ PrintError(vm, VCORE_NONE, "Trying to map a core-overlapping page\n");
return tmp_region;
} else if (region->core_id < tmp_region->core_id) {
p = &(*p)->rb_left;
int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region) {
struct v3_mem_region * ret;
int i = 0;
+ int rc;
if ((ret = __insert_mem_region(vm, region))) {
+ 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);
return -1;
}
v3_rb_insert_color(&(region->tree_node), &(vm->mem_map.mem_regions));
+ rc = 0;
for (i = 0; i < vm->num_cores; i++) {
struct guest_info * info = &(vm->cores[i]);
v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
if (mem_mode == PHYSICAL_MEM) {
- addr_t cur_addr;
-
- for (cur_addr = region->guest_start;
- cur_addr < region->guest_end;
- cur_addr += PAGE_SIZE_4KB) {
- v3_invalidate_passthrough_addr(info, cur_addr);
- }
+ rc |= v3_invalidate_passthrough_addr_range(info, region->guest_start, region->guest_end-1,NULL,NULL);
} else {
- v3_invalidate_shadow_pts(info);
+ rc |= v3_invalidate_shadow_pts(info);
}
} else if (info->shdw_pg_mode == NESTED_PAGING) {
- addr_t cur_addr;
-
- for (cur_addr = region->guest_start;
- cur_addr < region->guest_end;
- cur_addr += PAGE_SIZE_4KB) {
-
- v3_invalidate_nested_addr(info, cur_addr);
- }
+ rc |= v3_invalidate_nested_addr_range(info, region->guest_start, region->guest_end-1,NULL,NULL);
}
}
- return 0;
+ return rc;
}
// go right, core too small
n = n->rb_right;
} else {
- PrintDebug("v3_get_mem_region: Impossible!\n");
+ PrintDebug(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
return NULL;
}
}
// There is not registered region, so we check if its a valid address in the base region
- 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 &(vm->mem_map.base_region);
+ return v3_get_base_region(vm, guest_addr);
}
struct v3_mem_region * reg = NULL;
struct v3_mem_region * parent = NULL;
+ if (n == NULL) {
+ return NULL;
+ }
+
while (n) {
reg = rb_entry(n, struct v3_mem_region, tree_node);
// go right, core too small
n = n->rb_right;
} else {
- PrintError("v3_get_mem_region: Impossible!\n");
+ PrintError(vm, VCORE_NONE, "v3_get_mem_region: Impossible!\n");
return NULL;
}
}
struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
if (start_region == NULL) {
- PrintError("Invalid memory region\n");
+ PrintError(vm, VCORE_NONE, "No overlapping region for core=%d, start_gpa=%p\n", core_id, (void*)start_gpa);
+ v3_print_mem_map(vm);
return NULL;
}
void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
int i = 0;
+ int rc;
if (reg == NULL) {
return;
return;
}
+ rc = 0;
+
for (i = 0; i < vm->num_cores; i++) {
struct guest_info * info = &(vm->cores[i]);
v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
if (mem_mode == PHYSICAL_MEM) {
- addr_t cur_addr;
-
- for (cur_addr = reg->guest_start;
- cur_addr < reg->guest_end;
- cur_addr += PAGE_SIZE_4KB) {
- v3_invalidate_passthrough_addr(info, cur_addr);
- }
+ rc |= v3_invalidate_passthrough_addr_range(info,reg->guest_start, reg->guest_end-1,NULL,NULL);
} else {
- v3_invalidate_shadow_pts(info);
+ rc |= v3_invalidate_shadow_pts(info);
}
} else if (info->shdw_pg_mode == NESTED_PAGING) {
- addr_t cur_addr;
-
- for (cur_addr = reg->guest_start;
- cur_addr < reg->guest_end;
- cur_addr += PAGE_SIZE_4KB) {
-
- v3_invalidate_nested_addr(info, cur_addr);
- }
+ rc |= v3_invalidate_nested_addr_range(info,reg->guest_start, reg->guest_end-1,NULL,NULL);
}
}
// flush virtual page tables
// 3 cases shadow, shadow passthrough, and nested
+ if (rc) { PrintError(vm, VCORE_NONE, "Error in deleting memory region\n"); }
}
// Determine if a given address can be handled by a large page of the requested size
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);
+ reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
if ((reg) && ((reg->host_addr % PAGE_SIZE_4MB) == 0)) {
page_size = PAGE_SIZE_4MB;
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);
+ reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
if ((reg) && ((reg->host_addr % PAGE_SIZE_2MB) == 0)) {
page_size = PAGE_SIZE_2MB;
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);
+ reg = get_overlapping_region(core->vm_info, core->vcpu_id, pg_start, pg_end);
if ((reg) && ((reg->host_addr % PAGE_SIZE_1GB) == 0)) {
page_size = PAGE_SIZE_1GB;
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);
+ reg = get_overlapping_region(core->vm_info, core->vcpu_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)));
+ PrintError(core->vm_info, core, "Invalid CPU mode: %s\n", v3_cpu_mode_to_str(v3_get_vm_cpu_mode(core)));
return -1;
}
void v3_print_mem_map(struct v3_vm_info * vm) {
+ struct v3_mem_map * map = &(vm->mem_map);
struct rb_node * node = v3_rb_first(&(vm->mem_map.mem_regions));
- struct v3_mem_region * reg = &(vm->mem_map.base_region);
+ struct v3_mem_region * reg = NULL;
int i = 0;
- V3_Print("Memory Layout (all cores):\n");
+ V3_Print(vm, VCORE_NONE, "Memory Layout (all cores):\n");
+ V3_Print(vm, VCORE_NONE, "Base Memory: (%d regions)\n", map->num_base_regions);
+
+ for (i = 0; i < map->num_base_regions; i++) {
+ reg = &(map->base_regions[i]);
- 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));
+ V3_Print(vm, VCORE_NONE, "Base Region[%d] (all cores): 0x%p - 0x%p -> 0x%p\n",
+ i,
+ (void *)(reg->guest_start),
+ (void *)(reg->guest_end - 1),
+ (void *)(reg->host_addr));
+ }
// If the memory map is empty, don't print it
if (node == NULL) {
do {
reg = rb_entry(node, struct v3_mem_region, tree_node);
- V3_Print("%d: 0x%p - 0x%p -> 0x%p\n", i,
+ V3_Print(vm, VCORE_NONE, "%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(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
+ V3_Print(vm, VCORE_NONE, "\t(flags=0x%x) (core=0x%x) (unhandled = 0x%p)\n",
reg->flags.value,
reg->core_id,
reg->unhandled);
} while ((node = v3_rb_next(node)));
}
+
+void v3_init_mem()
+{
+ char *arg = v3_lookup_option("mem_block_size");
+
+ if (arg) {
+ v3_mem_block_size = atoi(arg);
+ V3_Print(VM_NONE,VCORE_NONE,"memory block size set to %llu bytes\n",v3_mem_block_size);
+ } else {
+ V3_Print(VM_NONE,VCORE_NONE,"default memory block size of %llu bytes is in use\n",v3_mem_block_size);
+ }
+}
+
+void v3_deinit_mem()
+{
+ // currently nothing
+}
