-2
$\begingroup$

Realize I didn't mention x86 or any more details to make this non-specific to any platform, but to just get a general idea of how this is done(this is also a long question with a lot of details).

I know most PC motherboards for desktop (if not all) computers have two sections; north and south bridge, and a chipset connects them together between the CPU's implementation and the buses carrying data around the board.

The x86 CPU has a register, I think, called the accumulator, which has eight bit AL, sixteen bit AX, and thirty-two bit EAX. Data for these registers are typically used for arithmetic operations, general purpose machine instruction, etc.

Also, supposedly for I/O access, such as with Port I/O. There is Port I/O and MMIO (memory-mapped input/output), which are two methods of accessing data outside the CPU.

I have inspected memory maps, but I'm still unclear on whether or not it's part of an instruction fetched from memory (i.e., MOV AL, 0FF) or from some other instruction to actually access hardware.

My basic question is, what specific instructions are used to write and read data from hardware, from a bare machine perspective; no device drivers, no operating system, and no firmware interrupts. This is the very essence of how system software must be made, and there's little complete coverage on how this works from the CPU's register to the hardware itself. I have seen BIOS interrupts, but that abstracts away details. Also, the developer had no idea what happens beyond INT 13.

So, despite my broad question, is there some specific way this is accomplished, either by instruction, correct value set in a register, etc.?

Example: Say I want to make a few beep noises. The sound hardware is responsible for making any noise at all, so I would, from a bare machine perspective, once again, have to directly access the hardware and its components, circuitry, or otherwise internal hardware registers or such to make noise. Is there a complimentary way of doing this that is specifically outlined or guidelined per-platform, or does it vary based on every specific piece of hardware and CPU?

If so, what would help me unravel that mystery to program for it this way? Would the card be, let's say, mapped a special address you write to? Then I would just set a register and the CPU will just send it there with whatever value it contains?

I am confused, and nothing makes clarity of this, because abstractions exist even at the low-level that create a barrier of the unknown.

$\endgroup$
1
$\begingroup$

The Wikipedia article on memory-mapped i/o is pretty good.

Yes, the registers that control the sound card are mapped to specific memory addresses. Reads and writes (loads and stores) to those memory addresses are recognized by the north-bridge as not belonging to the DRAM and are sent to the south-bridge. Based on the address the south-bridge reads and writes the appropriate registers on the sound card.

There are a lot of details that are specific to the processor in question. On x86 some i/o devices are memory mapped and some are port mapped. The port mapped interface uses two special instructions in and out that are like byte loads and stores, but only work with ports. The registers that control the internal speaker (the old-fashioned one that goes "beep", not a modern sound card) is mapped to port address 61h (hexadecimal). And the speaker frequency is controlled by the 8253 timer, the control registers for which are mapped to port addresses 42h and 43h. Here is an example: http://courses.engr.illinois.edu/ece390/books/labmanual/io-devices-speaker.html.

Other i/o devices on an x86 might be memory mapped. For example: the data buffers for the hard-disk controller will almost certainly be memory mapped and not port-mapped.

$\endgroup$

Your Answer

By clicking “Post Your Answer”, you agree to our terms of service, privacy policy and cookie policy

Not the answer you're looking for? Browse other questions tagged or ask your own question.