12
\$\begingroup\$

In chat, I asked about a possible undocumented core temperature sensor in a specific embedded processor. While the discussion itself was helpful, I was told that questions regarding undocumented features is off-topic on the main site because they would be opinion-based. Regardless of the topicality of the original question I wanted to ask, I would like to understand under what circumstances questions based on undocumented features or behavior are on-topic, if any.

My argument is that many well-known and commonly used features for microprocessors, FPGAs, CPLDs, etc. are undocumented despite being well-known and having established behavior. In fact, for some microprocessors, you cannot even run an emulator for them for any non-trivial program without supporting undocumented opcodes. So would questions regarding, say, undocumented opcodes, sensors, or JTAG ports be off-topic? Are they always off-topic, or only if the undocumented feature requires a significant amount of guesswork or is otherwise not well known?

From what I can see on the rest of meta, even questions regarding hardware which is entirely undocumented is on-topic despite every feature lacking documentation of any kind. I also see a number of well-received questions on the main site asking about undocumented features.

\$\endgroup\$
15
\$\begingroup\$

I think it would be fine. Note that W5VO (a moderator) said that hunting for an undocumented feature would usually be off-topic as a question. But documenting an undocumented feature through direct research (reverse engineering) could legitimately become a self-answered question under the right circumstances.

And I don't think that Marla's comment about it being "opinion-based" is at all relevant. Reproducible research is not opinion.

\$\endgroup\$
  • 1
    \$\begingroup\$ I agree on the reproducible research is not opinion. I see that. In the same conversation I had said that the undocumented was not guaranteed by the manufacturer. \$\endgroup\$ – Marla Nov 17 '18 at 14:48
10
\$\begingroup\$

I'd like to add some points to what Dave Tweed already said in his answer, which I agree with.

  • Reverse engineering usually has lots to do with researching undocumented features, and reverse engineering is a legal mean of obtaining information about a design of a competitor, so it has industrial relevance in the electronics field.

  • Reverse engineering is also a standard practice in cases where you must cope with a design whose blueprints have been lost (say you have a sophisticated and expensive niche instruments to repair/upgrade and the manufacturer has gone bankrupt and cannot be contacted any longer).

  • Not yet/well documented silicon/firmware bugs are undocumented "features" that need to be researched.

  • Often a designer must research whether a component can be used outside the characterization provided by the manufacturer. This is a sort of research on undocumented features. This may involve complex statistical analysis on batches of the part and usually can be outsourced to manufacturers for a price. But sometimes you need to to that yourself because, for example, the manufacturer asks too much or it is not going to do it or it has not the expertise to do that (think of the characterization of a component in extreme environmental conditions, e.g. freezing cold, extreme hot or vacuum).

So, IMO, questions about researching undocumented features/behaviors of a component or a design are well withing the scope of our site and electronics design in general.

\$\endgroup\$
  • 2
    \$\begingroup\$ Finding out about undocumented silicon behavior in particular seems to be an increasingly common task. The silicon vendors just keep rapidly spitting out new parts. "It will be fixed in the next version". Which will have a different bug instead. \$\endgroup\$ – Lundin Nov 22 '18 at 8:37
  • 2
    \$\begingroup\$ I would note that even analog functions can have 'undocumented' or perhaps 'unknowable' features, such as the output of an op-amp (particularly chopper stabililsed) at startup. \$\endgroup\$ – Peter Smith Nov 27 '18 at 11:29

You must log in to answer this question.

Not the answer you're looking for? Browse other questions tagged .