Volume 3, Number 3 Copyright 1998 TGG January 30, 1998

You may redistribute this newsletter for noncommercial purposes. For commercial use contact jack@ganssle.com.

EDITOR: Jack Ganssle, jack@ganssle.com

- Editor's Notes
- Magic and Delays
- More Dumb Mistakes
- Embedded Seminar in San Jose
- Thought for the Week
- About The Embedded Muse

Editor's Notes

I’m doing my “The Best Ideas for Developing Better Firmware Faster” seminar in San Jose on February 26. More info follows in this newsletter, or pop me an email.

The folks at Miller Freeman have asked me to give the keynote speech at the Embedded Systems Conference on March 31 in Chicago. These conferences are always a ton of fun; I look forward to seeing some of you all there.

Magic and Delays

Miguel Flores contributed a “Dumb Mistakes” piece (following) about the perils of using “magic numbers”, in this case in delay routines. It’s scary how many times people get into trouble with delays, and also with constants tuned to “make the damn thing work”. In his case both conditions combine to form a disaster.

If you’ve been in this business for a while you’ve surely seen the perils of delay routines. It’s almost amusing, in a bittersweet way, to watch programmers wrestle with delays, tuning them to get the system working.

Magic numbers are all too often the shortcuts we use to dodge deep understanding of a problem. We tune a delay, or a constant, coming up with a bit of what is truly magic just to get it out the door.

If we, the people developing the system, don’t have a deep understanding for the reasons behind EVERY DESIGN DECISION WE MAKE then we’re surely invoking magic, just as much as the necromancer of old. Magic numbers are as effective as Tarot cards.

Remember the old adage: problems that magically go away have a habit of magically reappearing.

Understand before coding. Magic has no place in embedded systems.

More Dumb Mistakes

From Miguel Flores

Here is one I had to solve recently that has a good lesson. The boards I mentioned above have an ISA bus interface with the host PC. We have a message protocol for sending messages between the host and the board's microcontroller. The traffic across the ISA bus is controlled by a hardware register containing status bits for words coming and going, and a ready bit. The ready bit is supposed to indicate that the board is ready to do ISA bus data transfers and is tied to an I/O pin on the microcontroller. The board has an interrupt for data coming, and one for data going.

That's the theory. In practice, the I/O pins from the microcontroller are bi-directional. After reset the I/O pins are tri-stated and act as inputs. This includes the ISA ready bit I/O. It is pulled up to inactive so at reset it goes inactive (not ready for ISA I/O). The original developer (in fairness, under considerable schedule pressure) got this whole scheme to work with a little extra (read: magic) delay in the Unix driver after it resets the board. Fine, ship it.

So, next, we make a new version of one of these boards, and eliminate a 15 MHz clock circuit used to drive the microcontroller. Instead, we use an external 10 MHz clock which is used elsewhere on the board anyway. The microcontroller runs slower, but it is still fast enough to do its work. This should not affect the ISA bus messaging since it uses that ISA bus status register. Well, by now you can guess that the Unix driver could not find the slower board after reset.

Puzzled, I first supplied the board with a 15 MHz clock where the 10 MHz should go. This would make the board run like it used to when it worked. OK, binary search. Try 12.5 MHz. Works. Try 11.5 MHz. Works. Try 10.5 MHz. Fails. Try 10.6 MHz. Works. Oh no, a timing dependency. In comes the o-scope and some instrumented firmware and printed copies of I/O drivers.

To make a long story short, the board's ISA ready register was being set within the first couple dozen instructions after reset (about 50 microseconds), even though the firmware was not ready until 5 milliseconds later (what with interrupt vectors and all). The magic delay in the Unix driver was not long enough for the slower microcontroller clock speed, so, the first word on the ISA bus after reset goes into the bit bucket. The board runs, but Unix can't talk to it.

The assembly language start up routine for the firmware, copied from other projects, assumed it knew the I/O configuration of the microcontroller, and set the direction of the I/O pins without first initializing the data that would be driven on the output pins. This included the ISA ready bit in the wrong state. This explained the need for the magic Unix driver delay after reset.

The lesson here is to not make or impose any unnecessary assumptions about the hardware or software. The start up firmware has nothing to do with any of the microcontroller I/O's, so don't mess with them. And further, make sure the I/O data has been initialized before driving the output pins. Good thing this equipment is not connected to any mechanical stuff. Finally, if you need magic to make your system work (like our magic delay in the Unix driver), then you don't fully understand how or why it works.

Embedded Seminar! San Jose

I'm conducting a full-day embedded seminar in San Jose on February 26. It's called "The Best Ideas for Developing Better Firmware Faster", and is for the developer who is honestly looking for new ideas, but who wants to cut through the academic fluff of formal methodologies and immediately find better ways to work.

The focus is uniquely on embedded systems. I'll talk about ways to link the hardware and software, to identify and stamp out bugs, to manage risk, and to meet impossible deadlines.

Thought for the Week

Write in C (Sing to the Beatle’s tune "Let it Be")

When I find my code in tons of trouble,
Friends and colleagues come to me,
Speaking words of wisdom:
"Write in C."

As the deadline fast approaches,
And bugs are all that I can see,
Somewhere, someone whispers:
"Write in C."

Write in C, Write in C,
Write in C, oh, Write in C.
LOGO's dead and buried,
Write in C.

I used to write a lot of FORTRAN,
For science it worked flawlessly.
Try using it for graphics!
Write in C.

If you've just spent nearly 30 hours,
Debugging some assembly,
Soon you will be glad to
Write in C.

Write in C, Write in C,
Write in C, yeah, Write in C.
BASIC's not the answer.
Write in C.

Write in C, Write in C
Write in C, oh, Write in C.
Pascal won't quite cut it.
Write in C.