Validate full offset for shared memories (#2472)

This commit fixes a bug in wasmparser where the full offset in a memarg
was not validated when the memarg was used with an atomic instruction.
This is an internal bug that wasn't caught during the implementation of
the memory64 proposal which is necessary to faithfully validate the
input module. The fix here is to delegate to the normal `check_memarg`
helper which ensures that all the normal rules apply in addition to the
extra rule for atomics.

Author: alexcrichton

crates/wasmparser/src/validator/operators.rs

@@ -1162,7 +1162,7 @@ where
                 "atomic instructions must always specify maximum alignment"
             );
         }
-        self.check_memory_index(memarg.memory)
+        self.check_memarg(memarg)
     }
 
     /// Validates a block type, primarily with various in-flight proposals.

tests/cli/threads.wast

@@ -9,3 +9,13 @@
     )
   )
   "must always specify maximum alignment")
+
+(assert_invalid
+  (module
+    (memory 1)
+    (func (param i32) (result i32)
+      local.get 0
+      i32.atomic.load offset=0x1_0000_0000
+    )
+  )
+  "offset out of range")

tests/snapshots/cli/threads.wast.json

@@ -7,6 +7,13 @@
       "filename": "threads.0.wasm",
       "module_type": "binary",
       "text": "must always specify maximum alignment"
+    },
+    {
+      "type": "assert_invalid",
+      "line": 14,
+      "filename": "threads.1.wasm",
+      "module_type": "binary",
+      "text": "offset out of range"
     }
   ]
 }