Balanced processor/SOC (?) for small embedded system running linux

Question

So, I know Linux kernel is quite "heavy" when considering lower scale embedded systems, but currently but we're a 2 man team trying to see how to create our own embedded system. I'm the one in charge of all software (the other guy is a HW guy), and thus I would like to re-use existing libraries and frameworks as much as possible, and I would like to bounce off some ideas with gurus around here.

I am fairly comfortable in Linux, but the booting and initialization process is new to me, and I need to dive in to that soon enough. Any book recommendations are welcome as well! I haven't designed any embedded systems before.. Only own some ARM dev boards (beagleboard and raspberry pi).

Current I have prototype of the software running on beagleboard already, and now we're thinking how to minimize the cost, and to create something our own..

It's a system connected to the internet, and I need to run a tiny web server with some scripting support. Performance wise I don't think it needs to be too powerful.

I would like to minimize all bootloader etc work, since I'm a one man SW team, and just concentrate on the application itself. Of course I understand that I need to configure our kernel for this, but this is indeed why I thought selecting some SoC would be good, since they usually have some linux and bootloaders ready..

First I thought that Cirrus EP9301 would be perfect, since it seems to be a good package, and not very expensive.. But it seems that it's already in end-of-life, and also support for this is very bad (people on the cirrus forums constantly complain about it).

Are there some good choices for this kind of project, which would enable us "easily" to get linux kernel up and running, with still maintaining some kind of decent BOM (hopefully 20USD or so) ?

Answer 1

Your hardware guy should already know this, but go with an existing reference design. Take the raspberry pi, the beagleboard/bone, open-rd, or any number of other existing systems and clone the part you need. As a result the linux porting will be a matter of removing what you are not using from the reference design instead of adding new stuff and hoping it works. If you go with flat pack parts you can do the work in your garage, if you go with bgas you need the equipment for that or pay someone to do it. (can you tell yet that I hate bgas?).

Is linux a requirement, if not that opens the door to a lot more devices using freertos or chibios or a number of other solutions. the stm32f4 discovery board for example is $20, uses what can barely be called a microcontroller for all the features it has (cortex-m4). Supposedly possible to run uclinux on a cortex-m, but definitely possible to run any number of rtoses and have an ip stack, etc. stellaris (ti.com) has a number of eval boards, one/some with ethernet already (use as a reference design). You can also take the wiznet approach (or a spi ethernet) and use any microcontroller (puts you into the avr/msp430 level and price range). Bang for buck the cortex-m's are good, arm based so comfortable to work with, etc.

Using linux if you are already not an experienced at porting to an embedded platform, and dont want to learn that on this go around, I would definitely go with a clone of an existing design, leverage as much as you can from a project with folks that are experienced at porting linux to a platform. If need be take an existing board (beagle/raspi/openrd) and go through the motions of porting to the platform with the cheat sheet of having access to an existing port, see if you cant get uboot ported and linux booting, etc. (dont really need uboot at all, that is possibly an unnecessary complication, just get dram up and pass the atags, etc to linux and just branch to it, pretty easy to launch linux from bare metal).

Answer 2

You could probably do worse than taking the Broadcom BCM2835 - used on the Raspberry Pi - as your starting point - especially if you want to avoid kernel and boot-loader work and a source of reference schematics. If this proves too expensive, check out other devices in the Broadcom range.

A few bits of advice

• You probably want some flash rather than a MMC card interface for production use. eMMC is an option. NAND flash is a nightmare due to rapid component obsolescence and the need to get own and dirty with the MTD drivers.
• USB Ethernet will be easier to integrate than a controller hanging off a general purpose bus, but won't perform as well. SmSC seems to be popular source for either

Answer 3

You could also have a look at the work that Olimex is doing with their linux boards. Perhaps even order a som and then combine it with other external components.