Re: [RFC][PATCH] make PowerPC64 default to nonexecutable stack

[Ryan Arnold <rsa at us dot ibm dot com>]

On Tue, 2010-11-02 at 23:36 -0400, Ulrich Drepper wrote:
> Putting a definition in a common header only really makes sense if the
> value will be changed at some point or you can make a change in one
> place only.  Neither is the case here.  Just add the definition in
> each header.  It's not going to change and it is a machine-specific
> value.

Ok, this makes sense.

> Also, no reason to use a symbol in the reserved namespace.
> stackinfo.h is not installed.  Use DEFAULT_STACK_PERMS or something
> like that.

I'll make this change.

> And why that configure override?  If you say it's not needed and gcc
> doesn't generate it, why should this change?  Seems to be completely
> unnecessary.

The configure fragment is there to select the correct test-case based on
the default of the compiler used to build GLIBC.  The testcase
elf/check-execstack.c assumes PT_GNU_STACK will be present and will
report an error if it finds otherwise. The testcase
sysdeps/powerpc/powerpc64/check-execstack.c assumes PT_GNU_STACK will
not be present and the test binary's stack permissions are checked to
see if the stack is nonexecutable.

I realize that this may select the wrong testcase if --build != --host,
whereby BUILD_CC might default to emitting PT_GNU_STACK when CC doesn't.
I'll try to get AC_TRY_COMMAND to use BUILD_CC instead of CC.  I'm not
sure how to move this selection mechanism outside of config.h.in
cleanly.

Ryan