On Mon, Oct 31, 2005 at 02:57:46PM -0800, Matt Mackall wrote:
> On Mon, Oct 31, 2005 at 10:43:01PM +0000, Russell King wrote:
> > On Mon, Oct 31, 2005 at 02:54:51PM -0600, Matt Mackall wrote:
> > > inflate: mark some arrays as INITDATA and define it in in-core callers
> >
> > This breaks ARM. Our decompressor has some rather odd requirements
> > due to the way we support PIC - it's PIC text with fixed data.
> >
> > This means that all fixed initialised data must be "const" or initialised
> > by code. This patch breaks that assertion.
>
> It would have been helpful if you quoted the patch.
That's what threading is for. 8)
> +#ifndef INITDATA
> +#define INITDATA
> +#endif
> ...
> -static const u16 cplens[] = {
> +static INITDATA u16 cplens[] = {
> 3, 4, 5, 6, 7, 8, 9, 10, 11, 13, 15, 17, 19, 23, 27, 31,
> 35, 43, 51, 59, 67, 83, 99, 115, 131, 163, 195, 227, 258, 0, 0
> };
>
> etc..
>
> I think for ARM, we can simply do -DINITDATA=const, yes?
No, unless you want to make this const:
-static u8 window[0x8000]; /* use a statically allocated window */
+static u8 INITDATA window[0x8000]; /* use a statically allocated window */
It shouldn't be marked INITDATA either anyway - it's uninitialised so
it'll end up in the BSS. There is no "discarded at runtime" BSS so
anything you want to place in a non-BSS section has to be initialised.
Of course, if you initialise it, you end up needlessly adding 32K to
the kernel image size...
--
Russell King
Linux kernel 2.6 ARM Linux - http://www.arm.linux.org.uk/
maintainer of: 2.6 Serial core
