[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Date: Thu, 14 Apr 2016 22:08:47 -0400
From: Rich Felker <dalias@...c.org>
To: musl@...ts.openwall.com
Subject: Re: [PATCH v2] add powerpc64 port
On Thu, Apr 14, 2016 at 07:55:51PM -0500, Bobby Bingham wrote:
> On Thu, Apr 14, 2016 at 03:14:00PM -0400, Rich Felker wrote:
> > On Thu, Apr 14, 2016 at 03:01:38AM -0500, Bobby Bingham wrote:
> > > On Thu, Apr 14, 2016 at 01:05:07AM +0200, Szabolcs Nagy wrote:
> > > > * Bobby Bingham <koorogi@...rogi.info> [2016-04-04 00:26:11 -0500]:
> > > > > +++ b/arch/powerpc64/bits/setjmp.h
> > > > > @@ -0,0 +1 @@
> > > > > +typedef unsigned long long __jmp_buf[66];
> > > >
> > > > hm glibc seems to use long[64] with 16byte alignment,
> > > > is the size diff because of alignment?
> > >
> > > Yes. Though apparently the glibc setjmp asm has code to detect a
> > > misaligned jmp_buf, but its handling of that case ends up overflowing
> > > the jmp_buf.
> > >
> > > I can make some changes to get our jmp_buf down to 65, but the only ways
> > > to get it down to 64 are either with 16 byte alignment, or to have setjmp
> > > spill vector registers to the stack first so it can copy them from there
> > > to the jmp_buf through a gpr.
> > >
> > > How important is it to match glibc here?
> >
> > I think you could avoid the need for alignment or increased buffer
> > size by positioning the vector registers at
> > jmp_buf_end-vector_save_size rounded _down_ to alignment, then
> > positioning the grps around them (so, putting the last gpr at the end
> > rather than before the vectors if the buffer as a whole is
> > misaligned).
>
> Is it valid to do the following?
>
> jmp_buf a, b;
> if (!setjmp(a)) {
> memcpy(b, a, sizeof a);
> longjmp(b, 1);
> }
>
> If that's valid, and the two jmp_bufs might be aligned differently, then
> this wouldn't work.
No, jmp_buf's are not values. You have to pass the same object that
was passed when calling setjmp when you call longjmp. The relevant
text is 7.13.2.1, paragraph 2:
"The longjmp function restores the environment saved by the most
recent invocation of the setjmp macro in the same invocation of the
program with the corresponding jmp_buf argument. If there has been no
^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^^
such invocation, or ... the behavior is undefined."
> > But it might be preferable to have the alignment match ABI too. Is
> > there any way it can be achieved with just things guaranteed to exist
> > by the psABI (is __int128 required by the psABI?) or does it require
> > C11 and/or GNUC attributes to get 16-byte alignment?
>
> The ABI does specify __int128. I can switch jmp_buf to this.
That's probably the right thing to do, then. Anyone else have an
opinion on it?
Rich
Powered by blists - more mailing lists
Confused about mailing lists and their use?
Read about mailing lists on Wikipedia
and check out these
guidelines on proper formatting of your messages.
