I'm working with an embedded device that has memory mapped registers to various bits of hardware.

So I'm trying to emulate it on a desktop by mmap'ing the same address range and then mprotecting it and the setting a sigaction handler to be called if the code attempts to read or write to the addresses in that range.

So I can get as far as that...

..but I don't know how to know detect whether it's a read or a write and find out what value was being written and or return the correct value to a read and then allow the program to proceed.

I suspect the ucontext parameter may help but I'm not sure how.

Any ideas?

It almost seems to me that you might have to look at the instruction that was executing to see if it was a read-type or write-type instruction, by looking at the parameters that are handed to your handler. I believe that the trap will go off based on the memory address that was accessed, no matter how or why. You might well have to figure out what kind of access it was. Messy.

signaction(2) says:
But I don't see anything indicating whether it was read or write. So I agree with sundialscvs, you would have to pull out the code address and decode the instruction. Messy indeed.

Sigh. It was nice dream for a brief moment......

.... I wonder if I could do it in Qemu instead. Hmmmm.

I don't think so: the hardware hands the operating-system a page-fault interrupt that simply informs it that the memory access failed. The OS can determine both what address was referenced and, based on the interrupt-return call-stack entry, where in the world that it was. But you should find, in the source-code of something like QEmu or VirtualBox, code which determines what to do by analyzing the instruction.

If you can't find something that already exists, you could write logic which uses single-byte decision tables to analyze the possibly multi-byte sequences which make up a CPU instruction, to classify it as being either a read or a write ... but there still could be further flies in the ointment: an instruction such as MOVE is both at the same time.

However, since you are apparently trying to create an emulator for a particular case that is well-known to you, perhaps you can suffice with a less-than fully-generalized solution ... one that is good enough for you and that need not be more.