SIGBUS inn-STABLE-20010726

[Alex Kiernan]

Bettina Fink <laura at hydrophil.de> writes:

> /me wrote:
> 
> > leading actors: Solaris 8 (64bit, UltraSPARC-II), inn-STABLE-20010726, 
> > gcc 2.95.3 and a reproduceable crashing innd (core dumped, SIGBUS).
> > 
> > innd starts up, if you telnet to port 119 the banner pops up, you can
> > do everything but "quit", that triggers the crash.
> 
> [...]
> 
> > 6673:   poll(0xFFBEF990, 4, 283000)     (sleeping...)
> > 6673:   poll(0xFFBEF990, 4, 283000)                     = 1
> > 6673:   read(42, " q u i t\r\n", 1023)                  = 6
> > 6673:   write(42, " 2 0 5   .\r\n", 7)                  = 7
> > 6673:       Incurred fault #5, FLTACCESS  %pc = 0x0001F834
> > 6673:         siginfo: SIGBUS BUS_ADRALN addr=0xFFBEF594
> > 6673:       Received signal #10, SIGBUS [default]
> > 6673:         siginfo: SIGBUS BUS_ADRALN addr=0xFFBEF594
> > 6673:           *** process killed ***
> 
> JFTR: It seems to be a compiler problem (gcc 2.95.3), I've tried again
> with an egcs (leftover from squid installation), now it seems to be
> stable.
> 

That makes some sense given the fault you're getting. I'd guess
playing with the compiler optimisation options might well fix it.

Of course its possible that gcc is making a valid optimisation based
on the constraints in the code, but that the constraints are wrong.

The line it happened at would be interesting to take a look at.

-- 
Alex Kiernan, Principal Engineer, Development, Thus PLC