In RISKS 3.53 mlbrown at ncsc-wo writes: >In RISKS 3.51 Bill Janssen writes of errors made in failing to consider >the interaction of the hardware and the software under design. This >failing was all too common in the writing of assembly and machine code >during the early days of programming. Discrete wired machines often had >OP codes that were not generally well known (i.e. the computer designers >kept it secret)... This posting raises another interesting issue; in any system with a long service life, there is a likelihood that the underlying hardware technology will change. Use of anything undocumented on a particular machine is asking for trouble when that machine is replaced with a ``compatible'' one that lacks the undocumented feature. In fact, the undocumented op-codes on 4-pi and 360 were not ``kept secret'' by the machine designers; in many cases they simply were not foreseen. It turned out that the combinations of operations that were performed by certain bit patterns did something useful. The modern microprocessors have this tendency also; witness the plethora of undocumented opcodes on the Z80. The modern mainframe manufacturers have all been burned at one time or another by users who take advantage of undocumented features and then have their programs fail when transported to a ``compatible'' machine using newer technology; the IBM 1401 compatibles brought out by Honeywell after IBM dropped the product line are the classic example. Some of the manufacturers now consider it worth the cost to add logic to verify that a program is using only documented instructions (generate a machine fault rather than an undocumented result); their experience is that documenting something to be forbidden doesn't keep the hackers from using it. There's some justification for the ``everything not permitted is forbidden'' attitude; I've seen mysterious failures years after a machine conversion caused by hardware incompatibilities in little-used areas of the software. I have also discovered successful penetrations of security on systems in which undocumented opcodes allowed user programs to perform privileged operations. I will deliberately refrain from discussing these further since some of the designs thus penetrated are still in service in the field. The goal that the hardware designers should aim for is to provide predictable results under all circumstances, even the cases that are documented to be illegal. Kevin Kenny
I would like to provide some diversion on Jon Jacky's comments. >Date: Mon, 8 Sep 86 16:55:19 PDT >From: firstname.lastname@example.org (Jon Jacky) >Subject: Upside-down F-16's and "Human error" >... should prevent any unreasonable behavior. This way lies >madness. Do we have to include code to prevent the speed from >exceeding 55 mph while taxiing down an interstate highway? I agree with this and subsequent statements about the capabilities of the operator (the pilot). Let's examine a silly analysis of providing that particular code. After you code the routine to prevent the " exceeding the speed", you are going to have to test it. Thus, the F-16 will have to "attempt" to exceed 55 mph on the expressway. Whether the code is there or not, the trooper is still going to give you a ticket. You have already made his day, but no one will believe him without the pilot getting a ticket. Besides he has to make his quota. So you may as well save your money for more important coding. Then the pilot will appear on either 60 minutes or Johnny Carson to explain his side of the problem. The analysis could go further but it belongs in a comedian's dialogue now. I would say that many "unreasonable behavior" situations being analyzed in a silly mode would show that some coding efforts should not be done. You may find out that certain situations cannot be tested in a justifiable fashion. As Jon Jacky and others have concluded, lets be reasonable in the questions responsible people should be addressing vice situations which have little chance of occuring. Good analysis will be better if common sense helps us to priortize these situations.
It seems F-16's are a hot topic everywhere. I think it's novelty thing like computers except for aeronautics. > I am trying to make the point that the gross simplification of > "preventing bomb release while inverted" doesn't map very well to what I > assume the actual goal is: "preventing weapons discharge from damaging > the aircraft". This is yet another instance where the assumptions made > to simplify a real-world situation to manageable size can easily lead to > design "errors", and is an architypical "computer risk" in the use of > relatively simple computer models of reality. > > Wayne Throop <the-known-world>!mcnc!rti-sel!dg_rtp!throopw Excellent point. Several things strike me about this problem. First, the language used by writers up to this point don't use words like "centrifugal force" and "gravity." This worries me about the training of some computer people for jobs like writing mission critical software [Whorf's "If the word does not exist, the concept does not exist."] I am awaiting a paper by Whitehead whch I am told talks about some of this. It can certainly be acknowledged that there are uses which are novel (Spencer cites "lob" bombing, and others cite other reasons [all marginal]) equal concern must be given to straight-and-level flight AND those novel cases. In other words, we have to assume some skill on the part of pilots [Is this arrogance on our part?]. Another problem is that planes and Shuttles do not have the types of sensory mechanisms which living organisms have. What is damage if we cannot "sense it?" Sensing equipment costs weight. I could see some interesting dialogues ala "Dark Star." Another thing is that the people who write simulations seem to have the great difficulty discriminating between the quality of thier simulations and "real world" in the presence of incomplete cues (e.g., G-forces, visual cues, etc.) when solely relying on things like instrument disk [e.g., pilot: "Er, you notice that we are flying on empty tanks?" disturbed pilot expression, programmer: "Ah, it's just a simulation."] Computer people seem to be "ever the optimist." Besides, would you ever get into a real plane with a pilot who's only been in simulators? Most recently, another poster brought up the issue of autonmous weapons. We had a discussion of of this at the last Palo Alto CPSR meeting. Are autonmous weapons moral? If an enemy has a white flag or hand-ups, is the weapon "smart enough" to know the Geneva Convention (or is too moral for programmers of such systems)? On the subject of flight simulators: I visited Singer Link two years ago (We have a DIG 1 system which we are replacing). I "crashed" underneath the earth and the polygon structure became more "visible." It was like being underneath Disneyland. --eugene miya sorry for the length, RISKS covered alot. NASA Ames Research Center President Bay Area ACM SIGGRAPH
Mike Brown wrote: <> Its fine to tell the pilot "Lower your wheels before you land, not <> after" but we still have gear up landings. We should not concern ourselves <> with checking for and preventing any incorrect behavior but we should preclude <> that behavior which will result in damage to or loss of the aircraft or the <> pilot. We do not need to anticipate every possible mistake that he can make <> in this regard either - all we need to do are to identify the hazardous operational modes and prevent <> their occurrence. I disagree that software MUST prevent: what about the case when an aircraft can lower only ONE side of its landing gear???? A belly-up landing is then the only way to go [ assume combat damage, or something, so that the pilot can't eject, and the computer INSISTS on lowering the landing gear whenever you attempt to go under 50 feet, or something stupid like that]. On the other hand, some of the latest experimental planes are totally UNFLYABLE by normal human control -- for those planes, the software better be reliable, because there is no backup!!! Obviously, one can present arguments for each side [human vs computer having the last say -- at TMI, computers were right, but ...] I would say that if humans do override CRITICAL computer control [like TMI], then some means of escalating the attention level must be invoked [ e.g., have the computers automatically notify the NRC]. Again, there's lots of tradeoffs to be made [seriousness of the problem, timeliness of the response necessary, etc.] which means thats there's NO PAT answer in most cases, just hope that people involved in these cases realize the possible consequences of their work. In that case one could argue for professional certification in these fields [ we're software ENGINEERS, right?!? : you wouldn't to go over a bridge built by an uncertified mechanical enginerr, would you?? What if the software he used was written by a flake? ]; if not certification, then perhaps the software should undergo wide scrutiny by independent evaluators [ I'd feel a lot better if I knew that the software controlling nuclear plants had undergone such scrutiny]. Enough said, I believe. Ihor Kinal ihnp4!mtung!ijk.
Reading comments about putting restraints on jet performance within the software reminded me of a conversation I had a few years ago at an air-show. In talking to a pilot who flew F-4's in Vietnam he mentioned that the F-4 specs said a turn exerting more than say 8 G's would cause the wings to "fall off". However in avoiding SAMs or ground-fire they would pull double? this with no such result. My comment to this, is what if a 8G limit had been programmed into the plane (if it had been fly-by-wire). Planes might have been hit and lost which otherwise were saved by violent maneuvers. With a SAM targeted on your jet, nothing could be lost by exceeding the structural limitations of the plane since it was a do-or-die situation. I'm sure 99.99% of the lifetime of a jet is spent within designed specifications, but should software limit the plane the one time a pilot needs to override this constraint?
Please report problems with the web pages to the maintainer