On Wed, Sep 16, 2009 at 11:37 AM, Dave Korn
<dave.korn.cygwin@googlemail.com> wrote:
Richard Guenther wrote:
On Wed, Sep 16, 2009 at 9:34 AM, Dave Korn
gcc/ChangeLog:
* varasm.c (default_encode_section_info): Set
SYMBOL_REF_TLS_MODEL
even for emulated tls.
If there are no regressions, OK for HEAD?
Isn't it enough to remove the targetm.have_tls check? Thus,
DECL_TLS_MODEL
should already be TLS_MODEL_EMULATED, no?
That's what I would have thought at first too, but I read Jakub's
comment
#13(*) as suggesting that we should hardcode TLS_MODEL_EMULATED because
(somewhat to our surprise) the tls model value in the decl looked
strange and
possibly bogus (tls-local-exec?), so it makes sense to override it
here with
what we know is the right value to show to the backend. I haven't
investigated to find out how the decl's tls model got to be that way
and if
it's intended or accidental. (And I'd really rather not get
sidetracked into
investigating all the plumbing of the tls emulation layer when I just
want to
get bootstrap fixed so I can get on with updating to head all the
huge patches
awaiting review that I'm currently carrying around...)
I think the variable should have a proper DECL_TLS_MODEL from the start.
Where does it get (not) set?