RFC: make linking BC-compiled classes more lazy - part one

[Andrew Haley]

Robert Schuster writes:
 > Hi Andrew,
 > thanks for reviewing this. I will fix all the issues you mentioned.
 > 
 > And for the records: The patch Andrew reviewed was *not* the one I posted
 > lately. Instead it is the verifier fix and both parts of my linker patch.
 > 
 > Thanks again for going through this rather big patch.
 > 
 > >>-  if (field_type == NULL)
 > >>-    throw new java::lang::NoClassDefFoundError(field_name->toString());
 > >>+  // Note: This call will resolve the primitive type names ("Z", "B", ...) to
 > >>+  // their Java counterparts ("boolean", "byte", ...) if accessed via
 > >>+  // field_type->name later. Using these variants of the type name is in turn
 > >>+  // important for the find_field_helper function. However if the class
 > > 
 > > 
 > > Two spaces after each full stop, please.  GNU coding convention.
 > Just a natural language issue: What is a full stop ?

One of these: "."

 > >>       if (target_class == NULL)
 > >>-	throw new java::lang::NoClassDefFoundError 
 > >>-	  (_Jv_NewStringUTF (sym.class_name->chars()));
 > >>-      
 > >>+        {
 > >>+          ffi_closure *closure =
 > >>+            (ffi_closure *) _Jv_Malloc( sizeof( ffi_closure ));
 > >>+          ffi_cif *cif = (ffi_cif *) _Jv_Malloc( sizeof( ffi_cif ));
 > >>+
 > >>+          // Pretends that we want to call a void (*) (void) function via
 > >>+          // ffi_call.
 > >>+          ffi_type **arg_types = (ffi_type **) _Jv_Malloc( sizeof( ffi_type * ));
 > > 
 > > 
 > > We need to mprotect(PROT_READ | PROT_WRITE | PROT_EXEC) this memory or
 > > alloc it from the heap.

 > I am new to this. The specs for mprotect say that the memory region
 > has to be aligned at the page size. How is this being done
 > efficiently and portably? I am thinking about using the modulo
 > division, the address and the page size. Is that the way to go?

Either:

	  /* Align to a multiple of PAGESIZE, assumed to be a power of two */
	  p = (void*)(((intptr_t) p + PAGESIZE-1) & ~(PAGESIZE-1));

or -- better -- allocate a Java array of bytes and put your closure in
there.  This works because memory allocated by the gc already has
PROT_EXEC set.  

You'll need to put a reference to the array somewhere the gc can see
it.  Either a static data item or a field of a Java class.  It's
better not to allocate memory that the gc can't see, because it causes
memory leaks.

Andrew.