Conference turris::decladebug

Check note 796 to see if this applies here.

PeterT

    No the problem only occurs in dxladebug not in decldeabug. There is
    something in the UI.
    
    -ed halekty
    

Ed,

What version of Digital UNIX and ladebug are these customers running? Any
chance we can get a reproducer to try out?

Jack Mahoney
Ladebug

Here's some additional info on yet another instance of the problem for
a different customer.

He can reproduce the problem at will with either dxdecladebug or the
3rd-party 'Greenhill' debugger (GUI) on a large C/FORTRAN application,
or on a small hello-world program in the following environments:

  Unix v4.0b
  Alphastation250
  ZLX-L1
  dxdecladebug v4.0-25

  Unix v4.0b
  Alphastation500
  ZLXP2-EZ5
  dxdecladebug v4.0-25

He starts dxdecladebug, but while app is loading the console suddenly
returns to the blue CDE screen;
the console is hung, no keyboard response at all;
he can still log in via telnet -- network resources still working --
and says 'ps' shows dtwm is gone, so the window manager apparently crashed;
nothing appended to ~/.dt/errorlog;

Thanks in advance for any pointers or suggestions !

                        Steve Sheppard
                        Digital Unix / Ultrix Applications Support Team
                        [email protected]

I've seen similar problems, it may be the db_enable problem which I've mentioned
before, or it may be a similar problem.  I've seen many of them on the
various build levels, that it's getting hard to remember which problem
is which.  
Go to super user and check these variables with the debugger,
tcp_noackwar    // if 1, set to 0.  Mostly this seems to kill off rlogins and
	     	// other windows after a period of no activity.

db_enable       //  if 1, may crash the system while debugging, set to 0 to 
		// avoid the crash.

Note that this did not crash all systems, ie a Flamingo had no problem
with this.  But maybe use of dxladebug aggravates it enough to cause it 
to crash finally.

There is yet another problem which I've seen on Steel BL4, which may also
be a problem in PTB, though I haven't seen it.  It sounds much like the
one described, where the console seems to lock up, but you can get into 
the machine from another system.  I don't know the particulars of this
one, or if there is a specific QAR, but I do know that it had something 
to do with configuring STREAMS when the kernel is built.  It seemed to 
be a malloc problem, where one process was just chewing up memory.  
It may be something similar here, if not the same exact problem.

For the db_enable or tcp_noackwar problems, you can do the following:

On the running kernel you can change this with ladebug or dbx

ladebug -k /vmunix
assign db_enable = 0
q

To patch this on the disk copy (so you don't have to do the above
every time you reboot) you must use dbx.  Ladebug will not patch the 
disk copy (a bug that needs fixing)

dbx /vmunix
patch db_enable = 0
q

and then you should be able to debug without a problem.

There are actually some machines that do not fall down with this problem,
but the majority of newer systems do.

PeterT

The problem in .4 was db_enable.   dxladebug caused
the Xserver (Xdec) to die.  The system was up and available
over the network until the customer tried to get a login
prompt on the console while it was in text mode.  Then
the system crashed.

Thanks for the pointers,

John

    I just got an update for this system crashing problem, but it 
    still crashed my machine. I'll be working with the kernel person
    responsible to get it resolved.  It seems to affect different 
    systems in different ways, and the ones it seems to spare entirely
    are the older machines, which makes me start to wonder if it's
    a timing issue.  Or maybe dependent on the rev of the Alpha chip.
    It hits my EV5 and EV56 machines, but seems to spare the Flamingos
    and Sandpipers, which are EV4's at the most.
    
    PeterT
    

    
    This problem sounds familiar.  Jeff Denham of kernel group has
    worked on similiar customer problem last month.  I will send 
    this info to him to see if it sounds familiar.
    
    Wanfang