[PATCH 01/12] ARM zImage: do memmap if possible

[Sascha Hauer]

Hi Peter,

On Mon, Mar 29, 2010 at 08:47:57PM +0200, Peter Korsgaard wrote:
> >>>>> "Sascha" == Sascha Hauer <s.hauer at pengutronix.de> writes:
> 
> Hi,
> 
>  Sascha>  	zimage = xmalloc(header.end);
>  Sascha>  	memcpy(zimage, &header, sizeof(header));
>  Sascha> -
>  Sascha> -	ret = read(fd, zimage + sizeof(header), header.end - sizeof(header));
>  Sascha> -	if (ret < header.end - sizeof(header)) {
>  Sascha> -		printf("could not read %s\n", argv[1]);
>  Sascha> -		goto err_out1;
>  Sascha> +	m = memmap(fd, PROT_READ);
>  Sascha> +	if (m != (void *)-1) {
>  Sascha> +		memcpy(zimage, m, header.end);
> 
> Here you're rereading the header as well, instead of everything past it
> like the normal case.

Yes, mostly because I was too lazy to calculate the offsets.

> 
> Does this give a significant boost? read() from something memmap'able
> (like NOR) boils down to a memcpy as well, right?

Hmm, with memmap it's 149ms and with read 156ms in my case. I think this
is because with memmap it goes down to a single memcpy whereas with read
we memcpy in chunks of 4k. I implemented this out of reflex, but maybe
it's really not worth the added complexity.

Sascha

-- 
Pengutronix e.K.                           |                             |
Industrial Linux Solutions                 | http://www.pengutronix.de/  |
Peiner Str. 6-8, 31137 Hildesheim, Germany | Phone: +49-5121-206917-0    |
Amtsgericht Hildesheim, HRA 2686           | Fax:   +49-5121-206917-5555 |