Rewrite interrupt handling

[jdemeyer]

There are lots of things to be improved in the interrupt handling routines in c_lib/src/interrupt.c and c_lib/include/interrupt.h.

Major changes planned:

• DONE: make sig_on() have function syntax so that we can declare it cdef int sig_on() except 0. See #10115 for the syntax changes.
• DONE: do not save signals in sigsetjmp() (by giving a second argument of 0 instead of 1). This speeds up a sig_on/sig_off loop from 382 clock cycles to 30 clock cycles on a Core(TM)2 Duo CPU T5870 @ 2.00GHz running Linux 2.6.34 glibc 2.11.2.
• DONE: using sigaction() instead of signal() since that has more well-defined semantics.
• DONE: handle SIGINT differently from other signals (other signals are urgent and cannot be ignored. SIGINT on the other hand does not need to handled immediately, but we have to be careful for race conditions).
• DONE: allow sig_on() and sig_off() to be nested.
• DONE: implement sig_retry() for retrying failed computations (this is useful for PARI, see #10018).
• DONE: have an interface for more general errors which are not signals. This can then be used by PARI, NTL and possibly other C libraries (various tickets).
• DONE: clean up old, unused code.
• DONE: testing interrupt handling: #10030.
• DONE: fix breakage because of this patch: #10061.
• DONE: documentation: #10109.
• DONE: Block interrupts during malloc: #10258.
• DONE: eliminate race condition when a SIGINT arrives before sig_on() or during sig_on().

Other potentially related tickets:

• #800 (make _sig_on and _sig_off faster when stacked)
• #9640 (Change PARI error catching mechanism)
• #9564 (libsingular exponentiation can not be interrupted)
• #7879 (Remove unnecessary signal handling for low prec mpfr operations) --- hopefully sig_on() and sig_off() can be made very fast such that this shouldn't be an issue anymore.
• #7794 (PolynomialRing_integral_domain ignores Ctrl-C and segfaults)
• #5313 (patch singular so that when it runs out of memory the error message says "singular" in it)
• #7702 (Handle interrupts better in the notebook)
• #3423 (Make Pari error messages more informative)
• #10126 (Fix error handing in Matrix_rational_dense._invert_pari())
• #10818 (EclLib? should allow signals to make LISP code interruptable)

[leif]

Author of what? ;-)

[vbraun]

Definite improvement! We should put it in the very next Sage-4.6.2.alpha to give it as much exposure as possible since it touches a couple of core C files.

[jdemeyer]

Replying to vbraun:

Definite improvement! We should put it in the very next Sage-4.6.2.alpha to give it as much exposure as possible since it touches a couple of core C files.

Alternatively, I am considering making 4.6.2 a pretty straightforward release and then putting various "big" tickets such as this one, #9433, #10572 in the next release which would then be sage-4.7.

[vbraun]

I'm trying out Sage-4.6.2.rc0 in Fedora 14 i386 inside VirtualBox, and it doesn't catch the SIGILL. Bactrace is below. Everything else works fine. Also, no such issue on my native Fedora 14 x86_64.

I'm not sure who is to blame here, could be something with the virtual machine. I did write a SIGILL handler in Perl and that works, so VirtualBox is at least in principle able to have guests raise and handle SIGILL. Though there is also some forking involved...

The only thing thats special about SIGILL that comes to mind is that it might not be automatically be reset to SIG_DFL upon entrance. But I don't see how that could cause this. Here is my testcase:

[vbraun@localhost ~]$ sage -gdb
----------------------------------------------------------------------
| Sage Version 4.6.2.rc0, Release Date: 2011-02-18                   |
| Type notebook() for the GUI, and license() for information.        |
----------------------------------------------------------------------
**********************************************************************
*                                                                    *
* Warning: this is a prerelease version, and it may be unstable.     *
*                                                                    *
**********************************************************************
/home/vbraun/Sage/sage/local/bin/sage-ipython
GNU gdb (GDB) Fedora (7.2-41.fc14)
Copyright (C) 2010 Free Software Foundation, Inc.
License GPLv3+: GNU GPL version 3 or later <http://gnu.org/licenses/gpl.html>
This is free software: you are free to change and redistribute it.
There is NO WARRANTY, to the extent permitted by law.  Type "show copying"
and "show warranty" for details.
This GDB was configured as "i686-redhat-linux-gnu".
For bug reporting instructions, please see:
<http://www.gnu.org/software/gdb/bugs/>...
Reading symbols from /mnt/sage/vbraun/sage-4.6.2.rc0/local/bin/python...done.
[Thread debugging using libthread_db enabled]
Python 2.6.4 (r264:75706, Feb 19 2011, 16:56:24) 
[GCC 4.5.1 20100924 (Red Hat 4.5.1-4)] on linux2
Type "help", "copyright", "credits" or "license" for more information.
Traceback (most recent call last):
  File "/usr/share/gdb/auto-load/usr/lib/libstdc++.so.6.0.14-gdb.py", line 59, in <module>
    from libstdcxx.v6.printers import register_libstdcxx_printers
  File "/usr/lib/../share/gcc-4.5.1/python/libstdcxx/v6/printers.py", line 19, in <module>
    import itertools
ImportError: No module named itertools
Detaching after fork from child process 10247.

So far, so good. Now run:

sage: import sage.tests.interrupt
sage: sage.tests.interrupt.test_signal_ill()
Detaching after fork from child process 10248.

Program received signal SIGILL, Illegal instruction.
0x05dac6b1 in _sig_on_prejmp (__pyx_self=<value optimized out>, __pyx_args=0xb7fa702c, __pyx_kwds=0x0) at /home/vbraun/Sage/sage/local/include/csage/interrupt.h:192
192	        _signals.sig_on_count++;
Missing separate debuginfos, use: debuginfo-install expat-2.0.1-10.fc13.i686 fontconfig-2.8.0-2.fc14.i686 glibc-2.13-1.i686 keyutils-libs-1.2-6.fc12.i686 krb5-libs-1.8.2-8.fc14.i686 libcom_err-1.41.12-6.fc14.i686 libgcc-4.5.1-4.fc14.i686 libgfortran-4.5.1-4.fc14.i686 libselinux-2.0.96-6.fc14.1.i686 libstdc++-4.5.1-4.fc14.i686 ncurses-libs-5.7-9.20100703.fc14.i686 nss-softokn-freebl-3.12.9-2.fc14.i686 openssl-1.0.0d-1.fc14.i686
(gdb) l
187	    fprintf(stderr, "sig_on (counter = %i) at %s:%i\n", _signals.sig_on_count+1, file, line);
188	    fflush(stderr);
189	#endif
190	    if (_signals.sig_on_count > 0)
191	    {
192	        _signals.sig_on_count++;
193	        return 1;
194	    }
195	
196	    /* At this point, _signals.sig_on_count == 0 */
(gdb) bt
#0  0x05dac6b1 in _sig_on_prejmp (__pyx_self=<value optimized out>, __pyx_args=0xb7fa702c, __pyx_kwds=0x0)
    at /home/vbraun/Sage/sage/local/include/csage/interrupt.h:192
#1  __pyx_pf_4sage_5tests_9interrupt_test_signal_ill (__pyx_self=<value optimized out>, __pyx_args=0xb7fa702c, __pyx_kwds=0x0) at sage/tests/interrupt.c:2878
#2  0x0017c398 in PyCFunction_Call (func=0xaad686c, arg=0xb7fa702c, kw=0x0) at Objects/methodobject.c:85
#3  0x001de9dd in call_function (f=0x820c4d4, throwflag=0) at Python/ceval.c:3706
#4  PyEval_EvalFrameEx (f=0x820c4d4, throwflag=0) at Python/ceval.c:2389
#5  0x001e06e7 in PyEval_EvalCodeEx (co=0xaa5fd58, globals=0x83cf3e4, locals=0x83cf3e4, args=0x0, argcount=0, kws=0x0, kwcount=0, defs=0x0, defcount=0, 
    closure=0x0) at Python/ceval.c:2968
#6  0x001e0803 in PyEval_EvalCode (co=0xaa5fd58, globals=0x83cf3e4, locals=0x83cf3e4) at Python/ceval.c:522
#7  0x001deccb in exec_statement (f=0x81630ec, throwflag=0) at Python/ceval.c:4401
#8  PyEval_EvalFrameEx (f=0x81630ec, throwflag=0) at Python/ceval.c:1717
#9  0x001e06e7 in PyEval_EvalCodeEx (co=0x837aa88, globals=0x8370714, locals=0x0, args=0x8162c08, argcount=2, kws=0x8162c10, kwcount=0, defs=0x0, defcount=0, 
    closure=0x0) at Python/ceval.c:2968
#10 0x001dea92 in fast_function (f=0x8162abc, throwflag=0) at Python/ceval.c:3802

[jdemeyer]

Volker, do you have strace installed? If so, you could try

$ mkdir trace
$ strace -ff -o trace/sage ./sage -c 'import sage.tests.interrupt; sage.tests.interrupt.test_signal_ill()'

and send me the trace/ directory.

And maybe do the same for test_signal_bus() to compare with.

[vbraun]

Sure, no problem. I did strace ill, bus, and fpe and uploaded the trace/ directory to ​http://www.stp.dias.ie/~vbraun/trace.tar.bz2

[vbraun]

Hold your horses! Its working now, and SIGILL is caught correctly. I did rebuild the sage library in the meantime, so I probably miscompiled something before.

[jdemeyer]

In any case, based on your trace, it looks like everything is working correctly. I see the following:

--- SIGILL (Illegal instruction) @ 0 (0) ---
rt_sigprocmask(SIG_SETMASK, [], NULL, 8) = 0
write(2, "Traceback (most recent call last"..., 35) = 35
write(2, "  File \"/home/vbraun/Sage/sage/l"..., 74) = 74

So, the program receives SIGILL and prints a Traceback, which is exactly what should happen.

I cannot understand why at some point it did not work for you...

[vbraun]

After many more rebuilds I found the problem: On i386 (and only i386) the Parma Polyhedra library (PPL, #10039) tests if SSE is available. It does this during the library initialization by installing a SIGILL handler and then issuing SSE instructions. It then resets the handler to signal(SIGILL, SIG_DFL). Clearly, this deactivates Sage's SIGILL handler.

Since we don't use PPL's floating point arithmetics I'll make sure to configure it in a way that it does not test for SSE availability.

[jdemeyer]

Replying to vbraun:

After many more rebuilds I found the problem: On i386 (and only i386) the Parma Polyhedra library (PPL, #10039) tests if SSE is available. It does this during the library initialization by installing a SIGILL handler and then issuing SSE instructions. It then resets the handler to signal(SIGILL, SIG_DFL). Clearly, this deactivates Sage's SIGILL handler.

I consider this to be a bug in PPL. Using sigaction() or signal(), it is possible to save an existing signal handler and then restore it afterwards.

[vbraun]

I agree and will report it upstream.