Conference turris::digital_unix

8589.0. "sendsig: can't grow stack?" by CHEFS::DAVIDSONS () Sun Jan 26 1997 04:13

When a single threaded C++ program is linked -threads (required by a
library it uses) the following happens:

- parent (4419) forks child (4831)
- child immediatly calls gethostbyname(3), never returns!
- child receives signal 11 (segmentation violation) repeatedly
- child grows to ~33MB RSS
- child terminates because it can't grow stack

Any ideas how to debug/solve this?
     
Thanks,
	Stuart.
    
(environment: DU 3.2D-1, C++ 5.1, lazy swap, 96MB RAM, 192MB swap 25% used)
    

*** increase limits in parent shell

limit -h datasize unlimited
limit -h stacksize unlimited
limit -h memoryuse unlimited
limit -h addressspace unlimited
limit datasize unlimited
limit stacksize unlimited
limit memoryuse unlimited
limit addressspace unlimited

*** confirm limits increased in parent process

(dbx) set $pid=4419
(dbx) px thread.stack.uthread.utask.uu_rlimit[2].rlim_max
0x7fffffffffffffff
(dbx) px thread.stack.uthread.utask.uu_rlimit[2].rlim_cur
0x40000000
(dbx) px thread.stack.uthread.utask.uu_rlimit[3].rlim_max
0x7fffffffffffffff
(dbx) px thread.stack.uthread.utask.uu_rlimit[3].rlim_cur
0x2000000
(dbx) px thread.stack.uthread.utask.uu_rlimit[7].rlim_max
0x4000000000
(dbx) px thread.stack.uthread.utask.uu_rlimit[7].rlim_cur
0x4000000000
(dbx) px vm_tune.vt_maxvas
0x4000000000
    
# ps -o sig,sigmask,sigcatch,wchan -p 4419
 PENDING  BLOCKED   CAUGHT WCHAN
       0        0      efa socket

# ps -o sig,sigmask,sigcatch,wchan -p 4831
 PENDING  BLOCKED   CAUGHT WCHAN
       0        0      afa pg_wait

*** trace (freeware tool) o/p (last 2 lines repeated until child terminates)

Tracing process /proc/04831
[4831]: stat ("/etc/svc.conf", 0x3ffc0099478) = 0 [ , <8.3075.12323 -rwxr-xr-x
1
 bin bin 2045 854114780,789580057,789580057> ]
[4831]: open ("/etc/hosts", O_RDONLY, 0666) = 12
[4831]: SIGNAL [11 SIGSEGV]
[4831]: sigprocmask (3, 0x0, 0x0) = 1024
[4831]: SIGNAL [11 SIGSEGV]
[4831]: sigprocmask (3, 0x0, 0x0) = 1024

*** (the param's to sigprocmask from within gethostbyname doen't look valid)

*** child terminates

sendsig: can't grow stack, pid 4831, proc cm6.playback.nss, sig 11, pc
0x3ff8056
fefc
    
                            

    you should apply patch OSFXXX-00106. 
    
    This should solve your problem
     Pit

    I checked guru::patches/osf/v3.2de1/README but could not find
    patch 00106.
    
    Can you give me a pointer?
    
    Thanks,
    	Stuart.

    "patch 106" is for the V4.0* stream.

It's possible that your "single-threaded" application is running into conflict
with threads from some library that you're using.  

.0> When a single threaded C++ program is linked -threads

Could you post the exact compile and link command(s)?


				Webb

    OK, I'll post the compile & link commands but 'ps -m' shows only
    one thread.
     
    Stuart.

.5> only one thread.

Oh.  Nevermind, then.

If the argument that you're passing to gethostbyname(3) looks OK, then I'd
suspect it was QAR time (or something similar).


				Webb

    What's the 'recursive SEGV problem' mentioned in note 8587?
    (looks like my problem)
    
    How does this happen?
    How can I debug it?
    
    Thanks,
    	Stuart.

.8> What's the 'recursive SEGV problem' mentioned in note 8587?

A recursive SEGV results when the attempt to handle a SEGV results in another
SEGV.
    
.8> How does this happen?

Generally, it's the result of a bug in a signal handler routine.  It can also
result if data on which a signal handler routine depends becomes corrupted.

.8> How can I debug it?

Running the program under a debugger (such as Ladebug) is usually a good
technique.  The debugger can stop the program when the first SEGV occurs, which
may give you a good idea of what's going on...


				Webb