X-Git-Url: http://v3vee.org/palacios/gitweb/gitweb.cgi?a=blobdiff_plain;f=palacios%2Fsrc%2Fpalacios%2Fvmm_mem.c;h=dd262c0d02c75d1b20d753e5de0654fd4eb76d52;hb=cfd5e43722796b4951faf721c64f3a74ab81ff2b;hp=ab8fc2cbb42341ef60c50d93fff6d0c258bbaefe;hpb=2c0f0d4f13ebf9071bede458a0d682513082beb3;p=palacios.git

diff --git a/palacios/src/palacios/vmm_mem.c b/palacios/src/palacios/vmm_mem.c
index ab8fc2c..dd262c0 100644
--- a/palacios/src/palacios/vmm_mem.c
+++ b/palacios/src/palacios/vmm_mem.c
@@ -41,7 +41,8 @@ static int mem_offset_hypercall(struct guest_info * info, uint_t hcall_id, void
 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");
+    PrintError("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);
 
@@ -50,8 +51,6 @@ static int unhandled_err(struct guest_info * core, addr_t guest_va, addr_t guest
     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;
@@ -60,13 +59,22 @@ int v3_init_mem_map(struct v3_vm_info * vm) {
 
     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;
+
+#ifdef V3_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
+
+    // Clear the memory...
+    memset(V3_VAddr((void *)map->base_region.host_addr), 0, mem_pages * PAGE_SIZE_4KB);
+
 
     map->base_region.flags.read = 1;
     map->base_region.flags.write = 1;
@@ -93,6 +101,7 @@ 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_mem_region, tree_node);
@@ -102,7 +111,7 @@ void v3_delete_mem_map(struct v3_vm_info * vm) {
 	v3_delete_mem_region(vm, reg);
     }
 
-    V3_FreePage((void *)(vm->mem_map.base_region.host_addr));
+    V3_FreePages((void *)(vm->mem_map.base_region.host_addr), mem_pages);
 }
 
 
@@ -136,7 +145,6 @@ int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
 
     entry->host_addr = host_addr;
 
-
     entry->flags.read = 1;
     entry->flags.write = 1;
     entry->flags.exec = 1;
@@ -154,7 +162,7 @@ int v3_add_shadow_mem( struct v3_vm_info * vm, uint16_t core_id,
 
 static inline 
 struct v3_mem_region * __insert_mem_region(struct v3_vm_info * vm, 
-						 struct v3_mem_region * region) {
+					   struct v3_mem_region * region) {
     struct rb_node ** p = &(vm->mem_map.mem_regions.rb_node);
     struct rb_node * parent = NULL;
     struct v3_mem_region * tmp_region;
@@ -288,44 +296,123 @@ struct v3_mem_region * v3_get_mem_region(struct v3_vm_info * vm, uint16_t core_i
 
 
 
-/* Search the "hooked" memory regions for a region that ends after the given address.  If the
- * address is invalid, return NULL. Else, return the first region found or the base region if no
- * region ends after the given address.
+/* 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...
  */
-struct v3_mem_region * v3_get_next_mem_region( struct v3_vm_info * vm, uint16_t core_id, addr_t guest_addr) {
+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;
+
+    if (n == NULL) {
+	return NULL;
+    }
 
-    // Keep going to the right in the tree while the address is greater than the current region's
-    // end address.
     while (n) {
-        reg = rb_entry(n, struct v3_mem_region, tree_node);
-        if (guest_addr >= reg->guest_end) { // reg is [start,end)
-            n = n->rb_right;
-        } else {
-	    if ((core_id == reg->core_id) || (reg->core_id == V3_MEM_CORE_ANY)) {
+
+	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 (reg->core_id == V3_MEM_CORE_ANY) {
+		// found relevant region, it's available on all cores
 		return reg;
-	    } else {
+	    } 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 no registered region, so we check if it's a valid address in the base region
 
-    if (guest_addr >= vm->mem_map.base_region.guest_end) {
-	PrintError("%s: Guest Address Exceeds Base Memory Size (ga=%p), (limit=%p)\n",
-		__FUNCTION__, (void *)guest_addr, (void *)vm->mem_map.base_region.guest_end);
-        v3_print_mem_map(vm);
-        return NULL;
+    if (parent->guest_start > guest_addr) {
+	return parent;
+    } else if (parent->guest_end < guest_addr) {
+	struct rb_node * node = &(parent->tree_node);
+
+	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 &(vm->mem_map.base_region);
+    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;
+    }
+
+
+    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_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
     int i = 0;
 
@@ -333,6 +420,19 @@ void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
 	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]);
 
@@ -366,8 +466,6 @@ void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
 	}
     }
 
-    v3_rb_erase(&(reg->tree_node), &(vm->mem_map.mem_regions));
-
     V3_Free(reg);
 
     // flush virtual page tables 
@@ -375,6 +473,72 @@ void v3_delete_mem_region(struct v3_vm_info * vm, struct v3_mem_region * reg) {
 
 }
 
+// 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->vcpu_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);
+
+		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;
+        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->vcpu_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->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)));
+            return -1;
+    }
+
+    return page_size;
+}
+
 
 
 void v3_print_mem_map(struct v3_vm_info * vm) {