newRPL third demo released

[Claudio L.]

(08-29-2014 08:59 PM)BruceH Wrote:  I'm a bit late to this project so apologies for not suggesting this earlier:

I would suggest porting Free RTOS to the calc hardware and then writing newRPL as a process, or set of processes to run under it.

This lets you focus on writing newRPL (which is the fun bit) and easily re-use pre-written code for the 'less fun' things like SD card access, serial port & USB drivers.

Not late at all. The on-target development has just started in June.

I have a few objections to your idea:
1) The Samsung chip in the 50g is not listed as one of the supported targets, so there's no drivers anyway
2) FreeRTOS provides:

• Real time (what do we need it for?)
• Multitasking (what for?)
• Events and queues (?)
  

4) A stack-based calculator uses a different memory management vs. the malloc/free that all kernels require. So we'd have to create 2 different memory managers, or reserve a chunk of ram specifically for the kernel.
5) We already have some drivers from HPGCC3 (LCD, keyboard, timers, SD card)

I had already considered the use of a microkernel and looked at a couple of L4 implementations and eCos, but none really fit the profile.
There's only one thing I'd like to have in a third-party kernel/OS: a complete USB stack (device and host), which would allow to use the host mode and connect devices to the calculator. But the "middleware" is never free. If you know of a tiny kernel with a USB stack for the S3C2410X then I'm interested. I'm willing to port and provide drivers for a kernel, if I can get something good in return.

Otherwise, I have a 50g that's already booting into my own custom firmware, I just finished the MMU programming, CPU clock, and LCD setup (in greyscale). I'm getting ready to add my exception handlers, and write a small memory manager to administer the very scarce ram using the MMU.
Once that memory manager is ready, I can pretty much bring in the demo to the target hardware, and then start the user interface from there.

I'd say by the end of the year I think we can have at least the basic demo we have now, running on-target.

Claudio