Re: [PATCH v2] lib/decompress_unxz.c: removing all memory helper functions

On 06/12/2012 11:40 AM, Joe Perches wrote:

On Tue, 2012-06-12 at 10:30 -0700, H. Peter Anvin wrote:

quoted

On 06/12/2012 10:26 AM, Nicolas Pitre wrote:

quoted

Instead of copying, maybe this would be much better to make the content 
of lib/string.c usable in a pre-boot environment.

quoted

Not sure about that... a lot of it is #ifndef __HAVE_ARCH_* and you
*really* want those optimizations on some arches.

Can __weak be used?

lib/string.c contains a lot of stuff that are not needed in preboot, and
would increase the kernel image size.  I'm not sure if we will be able
to make it usable in pre-boot and only bringing in what is needed by
preboot, withtut unnecessarily complicated lib/string.c.

Thanks Joe for the suggestion.  I guess we could remove the #ifndef by
renaming all arch specific mem helper functions to arch_<name>, and
declaring them with a weak attribute in mem.c and let the mem.c check to
see if it is provided, Here is an example for memcpy

Architecture specific,

void *arch_memcpy(void *__dest, __const void *__src, size_t __n)
{
...
}


lib/mem.c

extern void *arch_memcpy(void *__dest, __const void *__src, size_t __n)
__attribute __((weak));

void *memcpy(void *__dest, __const void *__src, size_t __n)
{
        if (arch_memcpy) {
		arch_memcpgoogly();
		return;	
	}

}

But this does not solve the memmove issue. Could we always trust
arch_memcpy in all architectures?  I guess the safest way would be to
use memmove in string.c.  Please let me know if you have any thoughts or
suggestions.

Thanks,
Mak.

.