Memory-mapped I/O

For more generic meanings of input/output port, see Computer port (hardware).

"MMIO" redirects here. For the airport serving Saltillo, Mexico, assigned the ICAO code MMIO, see Plan de Guadalupe International Airport.

Memory-mapped I/O (MMIO) and port I/O (also called isolated I/O or port-mapped I/O abbreviated PMIO) are two complementary methods of performing input/output between the CPU and peripheral devices in a computer. An alternative approach, not discussed here, is using dedicated I/O processors — commonly known as channels on mainframe computers — that execute their own instructions.

Memory-mapped I/O (not to be confused with memory-mapped file I/O) uses the same address bus to address both memory and I/O devices - the memory and registers of the I/O devices are mapped to (associated with) address values. So when an address is used by the CPU it may refer to a portion of physical RAM, or it can instead refer to memory of the I/O device. Thus, the CPU instructions used to access the memory are also used for accessing devices. Each I/O device monitors the CPU's address bus and responds to any of the CPU's access of address space assigned to that device, connecting the data bus to a desirable device's hardware register. To accommodate the I/O devices, areas of the addresses used by the CPU must be reserved for I/O and not be available for normal physical memory. The reservation might be temporary — the Commodore 64 could bank switch between its I/O devices and regular memory — or permanent.

Port-mapped I/O uses a special class of CPU instructions specifically for performing I/O. This is generally found on Intel microprocessors, specifically the IN and OUT instructions which can read and write one to four bytes (outb, outw, outl) to an I/O device. I/O devices have a separate address space from general memory, either accomplished by an extra "I/O" pin on the CPU's physical interface, or an entire bus dedicated to I/O. Because the address space for I/O is isolated from that for main memory, this is sometimes referred to as isolated I/O.

A device's direct memory access (DMA) is not affected by those CPU-to-device communication methods, especially it is not affected by memory mapping. This is because by definition, DMA is a memory-to-device communication method that bypasses the CPU.

Hardware interrupt is yet another communication method between CPU and peripheral devices. However, it is always treated separately for a number of reasons. It is device-initiated, as opposed to the methods mentioned above, which are CPU-initiated. It is also unidirectional, as information flows only from device to CPU. Lastly, each interrupt line carries only one bit of information with a fixed meaning, namely "an event that requires attention has occurred in a device on this interrupt line".

The main advantage of using port-mapped I/O is on CPUs with a limited addressing capability. Because port-mapped I/O separates I/O access from memory access, the full address space can be used for memory. It is also obvious to a person reading an assembly language program listing (or even, in rare instances, analyzing machine language) when I/O is being performed, due to the special instructions that can only be used for that purpose.

I/O operations can slow the memory access, if the address and data buses are shared. This is because the peripheral device is usually much slower than main memory. In some architectures, port-mapped I/O operates via a dedicated I/O bus, alleviating the problem.

One of them is that, by discarding the extra complexity that port I/O brings, a CPU requires less internal logic and is thus cheaper, faster, easier to build, consumes less power and can be physically smaller; this follows the basic tenets of reduced instruction set computing, and is also advantageous in embedded systems. The other advantage is that, because regular memory instructions are used to address devices, all of the CPU's addressing modes are available for the I/O as well as the memory, and instructions that perform an ALU operation directly on a memory operand — loading an operand from a memory location, storing the result to a memory location, or both — can be used with I/O device registers as well. In contrast, port-mapped I/O instructions are often very limited, often providing only for plain load and store operations between CPU registers and I/O ports, so that, for example, to add a constant to a port-mapped device register would require three instructions: read the port to a CPU register, add the constant to the CPU register, and write the result back to the port.

As 16-bit processors have become obsolete and replaced with 32-bit and 64-bit in general use, reserving ranges of memory address space for I/O is less of a problem, as the memory address space of the processor is usually much larger than the required space for all memory and I/O devices in a system. Therefore, it has become more frequently practical to take advantage of the benefits of memory-mapped I/O. However, even with address space being no longer a major concern, neither I/O mapping method is universally superior to the other, and there will be cases where using port-mapped I/O is still preferable.

Memory-mapped I/O is preferred in x86-based architectures because the instructions that perform port-based I/O are limited to one or two registers: EAX, AX, and AL are the only registers that data can be moved in to or out of, and either a byte-sized immediate value in the instruction or a value in register DX determines which port is the source or destination port of the transfer.^[1][2] Since any general purpose register can send or receive data to or from memory and memory-mapped I/O, memory-mapped I/O uses less instructions and can run faster than port I/O. AMD did not extend the port I/O instructions when defining the x86-64 architecture to support 64-bit ports, so 64-bit transfers cannot be performed using port I/O.^[3]

Memory barriers

Memory-mapped I/O is the cause of memory barriers in older generations of computers – the 640 KiB barrier is due to the IBM PC placing the Upper Memory Area in the 640–1024 KiB range (of its 20-bit memory addressing).

Example

Consider a simple system built around an 8-bit microprocessor. Such a CPU might provide 16-bit address lines, allowing it to address up to 64 kibibytes (KiB) of memory. On such a system, perhaps the first 32 KiB of address space would be allotted to random access memory (RAM), another 16K to read only memory (ROM) and the remainder to a variety of other devices such as timers, counters, video display chips, sound generating devices, and so forth. The hardware of the system is arranged so that devices on the address bus will only respond to particular addresses which are intended for them; all other addresses are ignored. This is the job of the address decoding circuitry, and it is this that establishes the memory map of the system.

Thus we might end up with a memory map like so:

Device	Address range (hexadecimal)	Size
RAM	0000 - 7FFF	32 KiB
General purpose I/O	8000 - 80FF	256 bytes
Sound controller	9000 - 90FF	256 bytes
Video controller/text-mapped display RAM	A000 - A7FF	2 KiB
ROM	C000 - FFFF	16 KiB

Note that this memory map contains gaps; that is also quite common.

Assuming the fourth register of the video controller sets the background colour of the screen, the CPU can set this colour by writing a value to the memory location A003 using its standard memory write instruction. Using the same method, graphs can be displayed on a screen by writing character values into a special area of RAM within the video controller. Prior to cheap RAM that enabled bit-mapped displays, this character cell method was a popular technique for computer video displays (see Text user interface).

Basic types of address decoding

• Exhaustive — 1:1 mapping of unique addresses to one hardware register (physical memory location)
• Partial — n:1 mapping of n unique addresses to one hardware register. Partial decoding allows a memory location to have more than one address, allowing the programmer to reference a memory location using n different addresses. It may also be done just to simplify the decoding hardware, when not all of the CPU's address space is needed. Synonyms: foldback, multiply mapped, partially mapped.
• Linear — Address lines are used directly without any decoding logic. This is done with devices such as RAMs and ROMs that have a sequence of address inputs, and with peripheral chips that have a similar sequence of inputs for addressing a bank of registers. Linear addressing is rarely used alone (only when there are few devices on the bus, as using purely linear addressing for more than one device usually wastes a lot of address space) but instead is combined with one of the other methods to select a device or group of devices within which the linear addressing selects a single register or memory location.

Incomplete address decoding

Addresses may be decoded completely or incompletely by a device.

• Complete decoding involves checking every line of the address bus, causing an open data bus when the CPU accesses an unmapped region of memory. (Note that even with incomplete decoding, decoded partial regions may not be associated with any device, leaving the data bus open when those regions are accessed.)

• Incomplete decoding, or partial decoding, uses simpler and often cheaper logic that examines only some address lines. Such simple decoding circuitry might allow a device to respond to several different addresses, effectively creating virtual copies of the device at different places in the memory map. All of these copies refer to the same real device, so there is no particular advantage in doing this, except to simplify the decoder (or possibly the software that uses the device). This is also known as address aliasing;^[4][5] Aliasing has other meanings in computing. Commonly, the decoding itself is programmable, so the system can reconfigure its own memory map as required, though this is a newer development and generally in conflict with the intent of being cheaper.

Accessing Port I/O via Kernel Device Driver

In Windows-based computers, memory can also be accessed via specific drivers such as DOLLx8KD which gives I/O access in 8, 16 and 32-bit on most Windows platforms starting from Windows 95 up to Windows 7. Installing I/O Port drivers will ensure memory access by activating the drivers with simple DLL calls allowing Port I/O and when not needed, the driver can be closed to ensure unauthorized access to the I/O ports.

References

1. ^ "Intel® 64 and IA-32 Architectures Software Developer’s Manual: Volume 2A: Instruction Set Reference, A-M" (PDF). Intel® 64 and IA-32 Architectures Software Developer’s Manual. Intel Corporation. June 2010. pp. 3–520. http://www.intel.com/Assets/PDF/manual/253666.pdf. Retrieved 2010-08-21. 
2. ^ "Intel® 64 and IA-32 Architectures Software Developer’s Manual: Volume 2B: Instruction Set Reference, N-Z" (PDF). Intel® 64 and IA-32 Architectures Software Developer’s Manual. Intel Corporation. June 2010. pp. 4–22. http://www.intel.com/Assets/PDF/manual/253667.pdf. Retrieved 2010-08-21. 
3. ^ "AMD64 Architecture Programmer's Manual: Volume 3: General-Purpose and System Instructions" (PDF). AMD64 Architecture Programmer's Manual. Advanced Micro Devices. November 2009. pp. 117, 181. http://support.amd.com/us/Processor_TechDocs/24594.pdf. Retrieved 2010-08-21. 
4. ^ Microsoft (December 4, 2001). "Partial Address Decoding and I/O Space in Windows Operating Systems". http://www.microsoft.com/whdc/system/sysinternals/partialaddress.mspx#EFD. 
5. ^ HP. "Address aliasing". http://docs.hp.com/en/A3725-96022/ch03s03.html. 

See also

• mmap, not to be confused with memory-mapped I/O
• Early examples of computers with port-mapped I/O
  • PDP-8
  • Nova
• PDP-11, an early example of a computer architecture using memory-mapped I/O
  • Unibus, a dedicated I/O bus used by the PDP-11
• University lecture notes about computer I/O
• Input/Output Base Address
• Bank switching