[Info-vax] Something is happening at VSI

[Simon Clubley]

On 2024-04-05, Simon Clubley <clubley at remove_me.eisner.decus.org-Earth.UFP> wrote:
> On 2024-04-05, motk <meh at meh.meh> wrote:
>> On 4/04/2024 10:43 pm, Simon Clubley wrote:
>>>> Am 04.04.2024 um 09:52 schrieb motk:
>>>>> On 4/3/24 10:20, motk wrote:
>>>>> %SYSTEM-F-ACCVIO, access violation, reason mask=00, virtual
>>>>> address=000000000000000C, PC=0000000000000002, PS=7AD5D8EE
>>> 
>>> Assuming the posted output has not been edited, running a user-mode program
>>> resulted in the process being killed. That means it failed in either
>>> supervisor mode or executive mode. That means VSI need to be told _privately_
>>> about the sequence to reproduce (if it can be discovered) so that they can
>>> see if the sequence can be modified to actually exploit the system.
>>
>> I can guarantee that output wasn't mangled in any way. I've been trying 
>> to reproduce it with no success so far. It was an extremely surprising 
>> result though; would it have hit a log somewhere?
>>
>
> The exit status on the accounting log may be interesting.
>
> Try a "$ acc/since=<whatever>/full" and once you have found the correct
> session, have a look at the exit status.
>
> Assuming the current mode bits on x86-64 in the PS register match those
> on Alpha, then, assuming I am decoding the bitfield correctly, it looks
> like an executive mode failure (ie: a failure in RMS itself). :-(
>
> Can someone confirm I am decoding the current mode bits in the above
> PS register correctly ?
>

Another idea: If this really is an executive mode failure, I wonder if
setting BUGCHECKFATAL to 1 would be useful here ?

Alex: What this would do is to turn the failure into a failure that
crashes the system (and writes a dumpfile, assuming the VSI virtual
image is setup correctly) instead of just deleting the current process.

It would also mean anything in memory (including command history, etc)
would be written to the dumpfile, so make sure there's nothing private
in the memory of your system before performing more tests.

What you could do then is to compress the dumpfile and send it to VSI
privately via some means they give you.

BTW, another idea: does x86-64 VMS currently write an entry into the
errorlog on an executive mode bugcheck ? I wonder if it would be useful
to check the errorlog to see if there is anything useful there from the
previous failure ?

Simon.

PS: pointed message to VSI management: A hobbyist has just appeared to
find a process-deleting bug within VMS that your testing has missed so far.
_This_ is an example of why the hobbyist program is important, and of
benefit, to you.

-- 
Simon Clubley, clubley at remove_me.eisner.decus.org-Earth.UFP
Walking destinations on a map are further away than they appear.