[Bug ld/18222] [2.25 regression] binutils fails to link libunwind on i686,x86_64, ARM32 and aarch64

[amodra at gmail dot com]

https://sourceware.org/bugzilla/show_bug.cgi?id=18222

Alan Modra <amodra at gmail dot com> changed:

           What    |Removed                     |Added
----------------------------------------------------------------------------
             Status|NEW                         |ASSIGNED
           Assignee|unassigned at sourceware dot org   |amodra at gmail dot com

--- Comment #1 from Alan Modra <amodra at gmail dot com> ---
libunwind is making functions, and one variable, protected visibility,
presumably because that makes the libunwind code run faster.  Well, that's fair
enough.  The variable isn't read-only so the pr18167 hack doesn't work, and
making the variable const doesn't put it in .rodata either.  It instead goes in
.data.rel.ro which isn't marked as read-only, even when -z relro, which
libunwind doesn't happen to use.

The thing is that the variable in question is in fact a constant.  libunwind
doesn't change it from its initial value, nor do any of the testsuite
testcases.  Neither is the variable's address taken.  So there isn't a real
problem here in use of protected variables.  The linker error is therefore
wrong.

Perhaps we should just sweep the protected variable vs. .dynbss problem under
the rug again?

-- 
You are receiving this mail because:
You are on the CC list for the bug.