Re: [Qemu-devel] [PATCH 03/14] ioport: divide ioport_opaque for read/write functions

[Alexander Graf]

On 09.09.2009, at 17:41, 武田 俊也 wrote:

> Now ioport read function and write function that have the same address
> must be registered with one opaque.
> But there may be the case that read-only iodevice and write-only one
> are assigned to same address.
>
> This patch is to divide ioport_opaque for read/write functions.

Why would you need different opaques for read and write functions of  
the same IO port?

> diff -ur a/ioport.c b/ioport.c
> --- a/ioport.c  Tue Sep  8 21:26:50 2009
> +++ b/ioport.c  Wed Sep  9 21:50:06 2009
> @@ -47,7 +47,8 @@
>
> /* XXX: use a two level table to limit memory usage */
>
> -static void *ioport_opaque[MAX_IOPORTS];
> +static void *ioport_read_opaque[MAX_IOPORTS];
> +static void *ioport_write_opaque[MAX_IOPORTS];

This adds quite a lot of memory overhead.

On my x86_64 system that would mean:

#define MAX_IOPORTS     (64 * 1024)

>>> (64 * 1024) * sizeof(void*)
524288

So this simple change adds 512KB of RAM overhead for no obvious  
reason. Maybe you could use the same opaque? :-)

Alex