SDK • Re: How debug a Pico without the elf?

I don't see any 0x4001xxxx addresses? or any 32bit numbers. Those are the GPIO addresses right? maybe I have the wrong options still

There aren't any 32-bit numbers in ARM assembler, especially CortexM0 which only has 16-bit instructions and so doesn't have any instructions with immediate fields greater than 8-bit wide (apart from jumps). 32-bit constants can be loaded by a sequence of instructions, or from constant pools. With more up-scale ARMs, an arbitrary 32-bit load can be done with a pair of 32-bit instructions (With 16-bit immediates in them) which the compilers will often pick, but on CortexM0 this only makes sense for "nice" values that can be made with something like 8 bits and a shift.

So your GPIO address is probably in a constant pool. These will be harder to spot in a disassembly without symbols - the disassembler will probably have disassembled them as nonsense instructions. Constant pools are a group of 32-bit values, placed at 32-bit aligned addresses directly after a piece of code where they are used: the constant is loaded with a PC-relative load instruction.

If you know what value you are looking for, it's probably easier to find it in a hex dump than the disassembly - then go and look at the disassembly for how it's used. Note that the value loaded for the address of an I/O register may not be the exact value needing to be used: it will often load the base address of a group of I/O registers into a (CPU) register and then use an indexed load to add the small offset to the specific (I/O) register needed - this isn't directly driven by the C source, but it's a combination of the compiler's optimisation and the SDK writers building the header files to make it easy for the compiler to spot the possibility.

Here's a with-symbols disassembly of the SDK's gpio_set_function()You can see the constant pool at 0x20002bc0 containing two constants - 4001c004, being GPIO0_CTRL and 40014000 being the base of the pad control registers, out of which it again wants the GPIO0 value at offset 4 from the base, but this time gets that with the str r1,[r0,#4] instruction - it adds the 4 as an offset rather than having it as a part of the constant. In this example, that's a completely arbitrary choice by the compiler - either way round is the same size because each constant is only used once - but it becomes more important in a larger function where the constant can be re-used.

And the corresponding source:

Statistics: Posted by arg001 — Tue Mar 19, 2024 9:33 am

---