Skip to contentSkip to conversation

Problem with Guile on ia64 (possibly "getcontext" related)

DISCUSSION Rob Browning 2 participants 4 messages Jul 15, 2006 – Jul 19, 2006

lore
Share & tools
lore
#0 lore raw

Guile 1.6.8 is crashing on ia64 with an illegal instruction whenever it tries to execute call-with-current-continuation. For those who don't know, call-with-current-continuation captures the current state of the computation; it's a language feature somewhat similar to setjmp.

The crash seems to be right around an asm "getcontext" call, and after looking around, I found these two links:

http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/144939
http://lists.freebsd.org/pipermail/freebsd-sparc64/2006-January/003739.html

They appear claim that adding a dummy (non-executed) setjmp just after the getcontext call will fix the problem because gcc recognizes setjmp, but not getcontext, and makes appropriate arrangements.

So I tried adding a dummy setjmp just after the getcontext call, and that does fix the problem.

However, even though this worked, I'd still prefer to have a better idea that this is the right fix before adopting it. Is that likely?

The relevant function is scm_make_continuation which you can find here:

http://cvs.savannah.gnu.org/viewcvs/guile/guile-core/libguile/continuations.c?rev=1.38.2.7&root=guile&only_with_tag=branch_release-1-6&view=markup

Versions:

Debian libc6.1 2.3.6-15
Debian gcc (GCC) 4.1.2 20060708 (prerelease) (Debian 4.1.1-8)

Thanks

Signature · 3 lines
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4
#1 lore raw
Quoted 20 lines · Rob Browning (on Sat, 15 Jul 2006 15:24:31 -0700) wrote:

Guile 1.6.8 is crashing on ia64 with an illegal instruction whenever it tries to execute call-with-current-continuation. For those who don't know, call-with-current-continuation captures the current state of the computation; it's a language feature somewhat similar to setjmp.

The crash seems to be right around an asm "getcontext" call, and after looking around, I found these two links:

http://blade.nagaokaut.ac.jp/cgi-bin/scat.rb/ruby/ruby-talk/144939
http://lists.freebsd.org/pipermail/freebsd-sparc64/2006-January/003739.html

They appear claim that adding a dummy (non-executed) setjmp just after the getcontext call will fix the problem because gcc recognizes setjmp, but not getcontext, and makes appropriate arrangements.

So I tried adding a dummy setjmp just after the getcontext call, and that does fix the problem.

Quoted 7 lines

However, even though this worked, I'd still prefer to have a better idea that this is the right fix before adopting it. Is that likely?

Quoted 1 line

The relevant function is scm_make_continuation which you can find here:

http://cvs.savannah.gnu.org/viewcvs/guile/guile-core/libguile/continuations.c?rev=1.38.2.7&root=guile&only_with_tag=branch_release-1-6&view=markup

Wild guess, based on no data. Add '__attribute__ ((returns_twice))' to the definition of ia64_getcontext. That should remove the need to use setjmp.

#2 lore raw
In reply to Rob Browning
Quoted 3 lines · Keith Owens

Wild guess, based on no data. Add '__attribute__ ((returns_twice))' to the definition of ia64_getcontext. That should remove the need to use setjmp.

That also seems to fix the crash.

So would it be correct to say that the likely problem is just that because Guile calls getcontext via asm, gcc doesn't know that the call needs to be handled specially?

Thanks

Signature · 3 lines
Rob Browning
rlb @defaultvalue.org and @debian.org; previously @cs.utexas.edu
GPG starting 2002-11-03 = 14DD 432F AE39 534D B592  F9A0 25C8 D377 8C7E 73A4
#3 lore raw
In reply to Rob Browning
Quoted 7 lines · Rob Browning (on Tue, 18 Jul 2006 18:40:07 -0700) wrote:
Quoted 3 lines · Keith Owens

Wild guess, based on no data. Add '__attribute__ ((returns_twice))' to the definition of ia64_getcontext. That should remove the need to use setjmp.

That also seems to fix the crash.

Good.

Quoted 3 lines

So would it be correct to say that the likely problem is just that because Guile calls getcontext via asm, gcc doesn't know that the call needs to be handled specially?

Exactly right.