Note: SA_SIGINFO has now been implemented by Jérémie Koenig. It will be uploaded in Debian eglibc 2.13-19.
IRC, #hurd, August / September 2010:
<giselher> Hy, I came across SA_SIGINFO in cherokee, I have the void sighandler(int num) prototype but how do I add the sa_handler field? <pinotree> if SA_SIGACTION is not defined, then you use sa_handler instead of sa_sigaction, and not add SA_SIGINFO in the sa_flags <giselher> SA_SIGINFO is not defined <pinotree> s/SA_SIGACTION/SA_SIGINFO/ above, yes <giselher> K <giselher> I am not sure if I fully understand this, there is the line "act.sa_flags = SA_SIGINFO" and how do I have to change that >_> <pinotree> can you paste the source in a pastebin? <giselher> k <giselher> http://archhurd.pastebin.com/N8BCnG6g at line 790 <pinotree> something along the lines of http://www.archhurd.pastebin.com/tdpcFD5G <pinotree> note that in the handler the siginfo_t parameter is used, which cannot be done if SA_SIGINFO is not defined <pinotree> (that code still won't compile, yet) <giselher> btw: is there a reason why SA_SIGINFO is not implemented? <giselher> the guildlines only say "It's not implemented" <azeem> 09:43 < azeem> signal stuff is tricky :-/ <azeem> basically it was pending on a complete rewrite by Roland, which never occured <youpi> I have an almost complete implementation, just not finished yet <youpi> (only the siginfo part) <azeem> nobody really groked that code for years until youpi showed up, but he added partial support AFAIK, not having much time on his hand <azeem> ah, he's here <azeem> :) <giselher> oh, should I just wait ? <youpi> no <giselher> k <youpi> there are OSes which don't have SA_SIGINFO <youpi> just cope with them: use sa_handler instead of sa_sigaction, and don't set SA_SIGINFO <youpi> (i.e. replace with 0 in your example) <giselher> ok <youpi> when SA_SIGINFO becomes available, it'll just be used
IRC, freenode, #hurd, 2011-08-20:
< youpi> erf, tcpwrappers will need si_pid < jkoenig> I could implement it not too far away in the future, we just need a version of msg_sig_post() with a siginfo argument or something. < youpi> I can also see a lot of packages using SA_SIGINFO for no reason... < youpi> (probably copy/pasty code) < youpi> sa.sa_flags = SA_SIGINFO; < youpi> sa.sa_handler = parse_config; < youpi> void parse_config(int) < youpi> yay < youpi> if(siginf->si_signo == SIGXCPU) < youpi> fprintf(stderr, "Exceeded CPU usage.\n"); < youpi> ... < youpi> jkoenig: actually most package don't actually use the SA_SIGINFO they request... < youpi> jkoenig: si_pid should get us almost all actually used coverage < youpi> I've seen only one example using si_errno < jkoenig> ok < youpi> oh, it's actually supported by your patch < youpi> (errno) < jkoenig> but I guess since implementing si_pid will require a new RPC, we might as well plan for the rest < youpi> jkoenig: indeed < jkoenig> youpi, hmm I doubt it's properly filled in in all circumstances? < youpi> ok, well, we'll see < pinotree> jkoenig: if it can be of help, boost::unit_test queries various fields of siginfo_t depending on the signal < pinotree> jkoenig: also, pulseaudio uses siginfo_t for remapping faulting memory on SIGBUS < jkoenig> pinotree, oh ok good to know < pinotree> *faulty < youpi> jkoenig: well, I guess you had checked that the si_addr field is correct in a few simple testcase :) < jkoenig> hmm I think so, yes < jkoenig> I ran like, "* (char *) 0x12345678;" or something IIRC < youpi> ok < jkoenig> I seem to remember mach generated SIGBUS instead of SIGSEGV depending on the upper bit, or something (I can't quite remember) < jkoenig> but when sigsegv was generated si_addr was right. < pinotree> jkoenig: (see boost/test/impl/execution_monitor.ipp in boost sources) < pinotree> maybe you can try the unit tests for boost::unit_tests, if any :) < pinotree> (while src/pulsecore/memtrap.c in PA) * pinotree stops doing MrObvious™