Embedded Muse 84 Copyright 2003 TGG September 22, 2003
You may redistribute this newsletter for noncommercial purposes. For commercial use contact firstname.lastname@example.org.
EDITOR: Jack Ganssle, email@example.com
- Editorís Notes
- Software WILL Fail
- Joke for the Week
- About The Embedded Muse
After an embarrassingly long hiatus, the result of getting married last May, extended sailing, and work on two books, the Embedded Muse is back. My apologies for the silence, and thanks to all who wrote wondering what was up.
One of those books is The Embedded Systems Dictionary (CMP Books, ISBN 1-57820-120-9), co-authored with Michael Barr (Editor-in-Chief of Embedded Systems Programming magazine). Itís now available at Amazon and all the usual suppliers. We were both rather surprised at the sheer number of terms used in this field! The book has some 4500 entries, and is targeted at developers, their managers, and administrative support folks.
Jim Haflinger wrote in response to Embedded Muse 83 about testing systems through the power-up cycle with a very cool idea: ďA test I have used for power cycling a unit is to have the control line actually turn off the power. All you need is a small circuit that keeps the power on unless a signal is forced from the unit. That way if the unit doesn't power up correctly, it will stay in that state for you to check it out.Ē
Software WILL Fail
I had the honor of giving the keynote speech at last weekís Embedded Systems Conference in Boston. The talk covered software disasters and the lessons we can learn.
But while working on the presentation I wondered why we refuse to acknowledge that software is indeed prone to failure. Developers generally have a subconscious attitude that the software will work, and there will be no unexpected inputs or odd operating conditions. Witness the plethora of buffer overrun conditions that plague many OSes, email and browser programs.
Software is funny stuff. Software engineering even funnier. No other branch of engineering is so intolerant of failure. We can beef up a strut to handle unexpected loads, or use a bigger conductor to handle possible electrical surges. But a single software error brings down the entire application. There are no fuses or weak links we can install to protect a system under stress.
I'll present my Better Firmware Faster seminar in Melbourne and Perth, Australia February 20 and 26th. All are invited. More info here. The early registration discount ends January 20.
WDTs are a well-known and essential ingredient for reliable embedded systems. Yet an astonishing number of systems donít have a WDT. The Clementine spacecraft was lost when the code crashed and dumped all of the satelliteís fuel. The team didnít ďhave timeĒ to write code to use the WDT hardware that existed in the system.
Exception handlers are another tool we can and must use to capture unexpected problems. Yet unimplemented handlers are a common theme in the lore of disaster stories. If weíve bothered to write the handler, in many, many cases that code is poorly tested. The system throws an exception, and the minimally tested handler, meant to save the device, causes a crash.
Memory management units are one of the best defenses against crashes. Run each task in its own partition; the MMU hardware traps errant accesses. The system can process the error and recover... or at least leave debugging breadcrumbs behind. If your system doesnít have an MMU, we can program spare chip selects to tag code that wanders into unused code regions. Yet we donít.
We think of software as being inherently brittle. Maybe itís not; maybe the code is fragile only because we refuse to install these sorts of fuses and weak links, as is done in every other branch of engineering. While writing this a good friend called to discuss some C code, and he very casually mentioned that all of his apps rewrite the output ports every second or so in case ESD flips their state. His code always includes a stack monitor to flag excessive pushing.
Why arenít we all doing this?
Thereís two reasons: laziness/schedule pressure, and lack of knowledge about techniques we can use. The former is problematic and possibly not solvable. We can deal with the latter by inventing cool tricks, and sharing those tricks widely.
A few techniques:
- Add an MMU, as discussed
- Always build in an awesome WDT
- Capture every possible exception Ė and TEST each handler thoroughly
- Monitor stack growth
- Rewrite output ports often
- Use ASSERTS to throw exceptions in any case where an error can possibly occur.
- Apply sanity checks to all inputs, whether from a user or from hardware (like an A/D).
Can you think of others? How else can we build fault-tolerant software, code that is robust enough to survive hardware glitches and outright bugs buried in the firmware? Let me know and Iíll pass them along to the readers of the Muse.
Joke for the Week
Embedded Muse 73 defined the Ultimate Tool Kit: WD-40 and Duct Tape. If it doesn't move, and it's supposed to use the WD-40. If it moves, and it's not supposed to, use the Duct Tape.
Michael Kercsmar wrote about an upgrade to this kit. It includes:
For things that shouldn't move: SuperGlue
For things that should move: A hammer