Refactoring and additions to direct paging (nested and passthrough)

[palacios.git] / palacios / src / palacios / vmm_mem.c

diff --git a/palacios/src/palacios/vmm_mem.c b/palacios/src/palacios/vmm_mem.c
index d934398..7bf0019 100644 (file)
--- a/palacios/src/palacios/vmm_mem.c
+++ b/palacios/src/palacios/vmm_mem.c
@@ -27,6 +27,7 @@
 #include <palacios/vmm_shadow_paging.h>
 #include <palacios/vmm_direct_paging.h>
 
+#include <interfaces/vmm_numa.h>
 
 uint64_t v3_mem_block_size = V3_CONFIG_MEM_BLOCK_SIZE;
 
@@ -35,7 +36,7 @@ struct v3_mem_region * v3_get_base_region(struct v3_vm_info * vm, addr_t gpa) {
     struct v3_mem_map * map = &(vm->mem_map);
     uint32_t block_index = gpa / v3_mem_block_size;
 
-    if (gpa > (map->num_base_regions * v3_mem_block_size) ||
+    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);
@@ -104,6 +105,36 @@ static int gpa_to_node_from_cfg(struct v3_vm_info * vm, addr_t gpa) {
     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
+       }
+    }
+}
 
 
 int v3_init_mem_map(struct v3_vm_info * vm) {
@@ -141,13 +172,12 @@ int v3_init_mem_map(struct v3_vm_info * vm) {
         node_id = gpa_to_node_from_cfg(vm, region->guest_start);
         
         V3_Print(vm, VCORE_NONE, "Allocating block %d on node %d\n", i, node_id);
-        
-        if (node_id != -1) {
-            region->host_addr = (addr_t)V3_AllocPagesNode(block_pages, node_id);
-        } else {
-            region->host_addr = (addr_t)V3_AllocPages(block_pages);
-        }
 
+       region->host_addr = (addr_t)V3_AllocPagesExtended(block_pages,
+                                                         PAGE_SIZE_4KB,
+                                                         node_id,
+                                                         0); // no constraints 
+                                                            
         if ((void *)region->host_addr == NULL) { 
             PrintError(vm, VCORE_NONE, "Could not allocate guest memory\n");
             return -1;
@@ -155,13 +185,18 @@ int v3_init_mem_map(struct v3_vm_info * vm) {
 
        // Clear the memory...
        memset(V3_VAddr((void *)region->host_addr), 0, v3_mem_block_size);
+       
+       // Note assigned numa ID could be different than our request... 
+       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);
        
+
        region->unhandled = unhandled_err;
     }
 
@@ -317,13 +352,13 @@ int v3_insert_mem_region(struct v3_vm_info * vm, struct v3_mem_region * region)
            v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
            
            if (mem_mode == PHYSICAL_MEM) {
-               rc |= v3_invalidate_passthrough_addr_range(info, region->guest_start, region->guest_end-1);
+             rc |= v3_invalidate_passthrough_addr_range(info, region->guest_start, region->guest_end-1,NULL,NULL);
            } else {
                rc |= v3_invalidate_shadow_pts(info);
            }
            
        } else if (info->shdw_pg_mode == NESTED_PAGING) {
-           rc |= v3_invalidate_nested_addr_range(info, region->guest_start, region->guest_end-1);
+         rc |= v3_invalidate_nested_addr_range(info, region->guest_start, region->guest_end-1,NULL,NULL);
        }
     }
 
@@ -455,7 +490,8 @@ static struct v3_mem_region * get_overlapping_region(struct v3_vm_info * vm, uin
     struct v3_mem_region * start_region = v3_get_mem_region(vm, core_id, start_gpa);
 
     if (start_region == NULL) {
-       PrintError(vm, VCORE_NONE, "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;
     }
 
@@ -523,13 +559,13 @@ void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
            v3_mem_mode_t mem_mode = v3_get_vm_mem_mode(info);
            
            if (mem_mode == PHYSICAL_MEM) {
-             rc |= v3_invalidate_passthrough_addr_range(info,reg->guest_start, reg->guest_end-1);
+             rc |= v3_invalidate_passthrough_addr_range(info,reg->guest_start, reg->guest_end-1,NULL,NULL);
            } else {
              rc |= v3_invalidate_shadow_pts(info);
            }
            
        } else if (info->shdw_pg_mode == NESTED_PAGING) {
-         rc |= v3_invalidate_nested_addr_range(info,reg->guest_start, reg->guest_end-1);
+         rc |= v3_invalidate_nested_addr_range(info,reg->guest_start, reg->guest_end-1,NULL,NULL);
        }
     }