Connecting the debugger to a running image causes it to crash

[Christopher Blackburn]

I'm running OpenVms 8.3 on an ITANIUM processor. We have just ported a large number of programs from ALPHA to this new platform and I'm now trying to fix a few process crashes.
I have noticed that when I connect the VMS debugger to a running image (compiled and linked with DEBUG) the image crashes with an access violation when the debugger continues. It would appear that the debugger has corrupted the process. It does not matter what I do in the debugger just connecting and using GO causes the crash. The error reported in an access violation with an address of 000000000.

If I run the vms process from a decterm and start the debugger that way I can set break points and run the image as per normal. The problem only seems to be when I run DEBUG/KEEP and use the CONNECT command.

Anyone any ideas ?

[Volker Halle]

Chris,

I've tried this on our rx2600 running V8.2 and it worked.

Make sure you have the most recent patches installed, write a simple test program and try to reproduce the problem. Consider to document the error (include a terminal session log file as an attachment).

And log a call with HP.

Volker.

[Christopher Blackburn]

Volker,

I have tried a simple standalone program and the debugger connects correctly without crashing the program. The actual applications I am trying to debug use DEC Window X/MOTIF libraries. I will look for patches and check we have the latest updates.
Many thanks for your suggestions.

[Christopher Blackburn]

Volker,

I have tried a simple test program and I can connect the debugger and use GO to carry on program execution without problems.
I still get the same crash when I connect to one of our running programs. I have attached the output from the VDU debug session.

I noticed a problem raised by H.Bachner on a problem with the ANALYSE/CRASH command on an Itanium platform. The output indicated the following was obtained -

%DEBUG-I-PRETHRDEVNTFAIL, setting a predefined thread event failed
%SYSTEM-F-IMGDMP, dynamic image dump signal at PC=FFFFFFFF80150768, PS=00000018
%SYSTEM-F-ACCVIO, access violation, reason mask=00, virtual address=A725C0F7D9018B78, PC=A725C0F7D9018B78, PS=0000001B
break on unhandled exception preceding 18446744071563446120 in THREAD -2

My crash from the DEBUGGER appears to be in the same address range with an access violation as above.

Are these two problems linked ?

[Volker Halle]

Chris,

virtual addresses in system space (S0) can point anywhere. One would need to symbolize those addresses with SDA and see, if they point to the same module/routine.

In your case, try:

$ ANAL/SYS
SDA> READ/EXEC/NOLOG
SDA> EXA/INS 805F7211

And try to find out from the instruction shown, if an ACCVIO at virtual address 00000087 seems plausible from the decoded instruction. The ACCVIO seems to have happened, when some of your (?) code in SHARE$FLDSHARE_CODE2 has called a routine in system space. It sometimes happens, that running an image with the debugger causes problems not seen, when running the same image without the debugger...

Are you running the debugger in DECwindows Motif mode ? Did you try $ DEFINE DBG$DECW$DISPLAY " " to use the debugger in character cell mode ?

Volker.

[Christopher Blackburn]

Problem was fixed in OpenVms 8.3 on the IA64 platform.