I'd like to submit this patchset for review, and
hopefully for inclusion into the test source base.
I'm a little reluctanct to do the checkin myself,
at least not without independent review.
The purpose is to fix problems in calling a function
by way of taking its address and calling through
that address. So far I've located those problems in
the mmap and mprotect tests, specifically those which
call the utility function vsrt_creat_exec_file.
There may be other such assumptions in the code base,
but if so, that's for another fix.
They fix failures on mmap_P 3 and 4, and mprotect_P 4.
To quote David Mosberger on the topic (with reference
to a small test program he was supplied):
The test program assumes that taking the address of a
function returns the entry-point address of that function.
This is not so on ia64 (and several other arches, including
VAX, PA-RISC, and PPC, I believe). Instead, what you get
on ia64 is a pointer to a function descriptor:
struct fdesc {
void *ip; /* entry point (code address) */
void *gp; /* global-pointer */
};
The patches as written are totally Itanium-specific
and have no effect if not compiling __ia64__. If this
code applies to other architectures, someone else will
have to make those additions.
I've sent the patchfile as an attachment so it doesn't
suffer line-wrap problems.
Mats
Attachment:
EXEC-FIX.ZIP
Description: Binary data