On Thu, Mar 04, 2004 at 03:35:03PM +0000, Nathan Sidwell wrote:
Nick,
This patch implements that functionality in a different manner. Rather
than
lie about .tbss's size, I tweak the couple of places that need to know
it can be ignored for layout purposes.
It appears that there are now some new failures in the linker
testsuite with this patch applied:
This patch adds the equivalent bit into elf.c's segment mapping. With
this patch I find the linkers testsuite has the same failures as without
the patch (on i686-pc-linux-gnu). ok?
I'll investigate those failures I'm seeing -- it might be related
to me using a gcc-3.4-pre compiler. I'd understand if you'd like to
hold off this patch until these other failures are resolved.
This patch is still wrong.
One problem is e.g. mapping following sections:
.text
[ VMA gap here ]
.tbss
.data
.bss
Your change was ignoring .tbss only in one of the 2 places where
last_hdr is set up, but .tbss really cannot be ignored, but rather should
be handled as if it was SEC_LOAD and had zero _raw_size.
Without this, I get e.g. tlsg failures with -z relro patch (which reorders
sections).
Ok to commit?
2004-03-26 Jakub Jelinek <jakub@redhat.com>
* elf.c (map_sections_to_segments): Fix handling of .tbss.