Am 11.04.2012 03:06, schrieb David Gibson:
> On Fri, Apr 06, 2012 at 06:17:12PM +0200, Andreas Färber wrote:
>> free() opcode tables. They are being malloc()'ed in create_new_table().
>
> This doesn't seem right. Unless I've missed something in the call
> path, create_new_table() is not called from the instance initializer.
> So surely it should not be torn down in the instance finalizer.
We have this call chain:
helper.c:cpu_ppc_init() -> translate_init.c:cpu_ppc_register_internal()
-> create_ppc_opcodes() -> create_ppc_opcodes() -> { fill_new_table()
(filling it with &invalid_handler),
register_insn() -> { register_[dbl]ind_insn() -> register_ind_in_table()
-> create_new_table(), register_direct_insn() -> insert_in_table()
}(each filling it with non-invalid handlers) }
So you are correct that it is not directly called from the initfn. The
reason not to do that yet is that cpu_ppc_register_internal() still uses
ppc_def_t, which my previous RFC patch series replaces through QOM
subclasses.
Since free() works fine with zero'ed memory such as after the
object_new() in cpu_ppc_init() I still think this patch is fully
correct. But we can postpone it if you prefer.
Andreas
--
SUSE LINUX Products GmbH, Maxfeldstr. 5, 90409 Nürnberg, Germany
GF: Jeff Hawn, Jennifer Guild, Felix Imendörffer; HRB 16746 AG Nürnberg
