Some products are already running full Linux (with BusyBox) on a PPC with 16 or 64MB DRAM, 8 or 16MB flash and optional PCMCIA or CF. And talking about costs - the PCMCIA connector itself with eject mechanism costs as a very nice procesor.

I haven't worked with any ARM9 processor, but are waiting for a development kit with an ARM9-based module to initially take a closer look at the Linux support. It is not uncommon that the quality of the drivers are said to be "excellent" when in reality one or more of the shipped drivers are only marginally working.

I have seen the reverse: I have generally seen C as adequate for products down to x kB where x is depending on what chip family that gets selected but normally is a one-digit value.

I can directly agree and say that I'm not fond of a positive reset.

One interesting thing with being able to build parts of the source code on a Linux machine is that large parts of the code (the decision and computation parts) can be compiled and run every night on the machine where the source code repository is. This doesn't test problems only relevant when cross-compiling and running on the real hardware, but quite a lot of logic problems or changes can be automagically detected and reported that the target hardware may lack the resources to cath using automatic regression tests. Single logic can be regression-tested in the target hardware by externally (or with internal stubs) feeding test vectors, but some module blocks may have too large state memory to allow regression testing in finite time by just feeding external state information and the tett code to inject test data to internal modules may require too much space.

The only real problem with running tests on a 32- or 64-bit machine is that most 8-bit compiers normally skips the integer promotion to instead concentrate of efficiency. Because of this, a move to a 16-bit, 32-bit or 64-bit compiler may in some cases result in very different results even if the code is correct according to the integer-promption rules used in the target build.