| |
|
|
|
7.13 I/O and the Cache
It goes without saying that the CPU cannot cache values for memory-mapped I/O ports. If a port is an input port, caching the data from that port would always return the first value read; subsequent reads would read the value in the cache rather than the possible (volatile) data at the input port. Similarly, with a write-back cache mechanism, writes to an output port may never reach that port (i.e., the CPU may save up several writes in the cache and send the last such write to the actual I/O port). Therefore, there must be some mechanism to tell the CPU not to cache up accesses to certain memory locations.
The solution is in the virtual memory subsystem of the 80x86. The 80x86's page table entries contain information that the CPU can use to determine whether it is okay to map data from a page in memory to cache. If this flag is set one way, then the cache operates normally; if the flag is set the other way, then the CPU does not cache up accesses to that page.
Unfortunately, the granularity (that is, the minimum size) of this access is the 4K page. So if you need to map 16 device registers into memory somewhere and cannot cache them, you must actually consume 4K of the address space to hold these 16 locations. Fortunately, there is a lot of room in the 4 GByte virtual address space and there aren't that many peripheral devices that need to be mapped into the memory address space. So assigning these device addresses sparsely in the memory map will not present too many problems.
7.14 Protected Mode Operation
Windows and Linux employ the 80x86's protected mode of operation. In this mode of operation, direct access to devices is restricted to the operating system and certain privileged programs. Standard applications, even those written in assembly language, are not so privileged. If you write a simple program that attempts to send data to an I/O port via an IN or an OUT instruction, the system will generate an illegal access exception and halt your program. Unless you're willing to write a device driver for your operating system, you'll probably not be able to access the I/O devices directly.
Like Windows, Linux does not allow an arbitrary application program to access I/O ports as it pleases. Only programs with "super-user" (root) priviledges may do so. For limited I/O access, it is possible to use the Linux IOPERM system call to make certain I/O ports accessible from user applications (note that only a process with super-user priviledges may call IOPERM, but that program may then invoke a standard user application and the application it runs will have access to the specified ports). For more details, Linux users should read the "man" page on "ioperm".
This chapter has provided an introduction to I/O in a very general, architectural sense. It hasn't spent too much time discussing the particular peripheral devices present in a typical PC. This is an intended omission; there is no need to confuse readers with information they can't use. Furthermore, as manufacturers introduce new PCs they are removing many of the common peripherals like parallel and serial ports that are relatively easy to program in assembly language. They are replacing these devices with complex peripherals like USB and Firewire. Unfortunately, programming these newer peripheral devices is well beyond the scope of this text (Microsoft's USB code, for example, is well over 100 pages of C++ code).
Those who are interested in additional information about programming standard PC peripherals may want to consult one of the many excellent hardware references available for the PC or take a look at the DOS/16-bit version of this text.
IN and OUT aren't the only instructions that you cannot execute in an application running under protected mode. The system considers many instructions to be "privileged" and will abort your program if you attempt to use these instructions. The CLI and STI instructions are good examples. If you attempt to execute either of these instructions, the system will stop your program.
Some instructions will execute in an application, but behave differently than they do when the operating system executes them. The PUSHFD and POPFD instructions are good examples. These instruction push and pop the interrupt enable flag (among others). Therefore, you could use PUSHFD to push the flags on the stack, pop this double word off the stack and clear the bit associated with the interrupt flag, push the value back onto the stack and then use POPFD to restore the flags (and, in the process, clear the interrupt flag). This would seem like a sneaky way around clearing the interrupt flag. The CPU must allow applications to push and pop the flags for other reasons. However, for various security reasons the CPU cannot allow applications to manipulate the interrupt disable flag. Therefore, the POPFD instruction behaves a little differently in an application that it does when the operating system executes it. In an application, the CPU ignores the interrupt flag bit it pops off the stack. In operating system ("kernel") mode, popping the flags register does restore the interrupt flag.
7.15 Device Drivers
If Linux and Windows don't allow direct access to peripheral devices, how does a program communicate with these devices? Clearly this can be done since applications interact with real-world devices all the time. If you reread the previous section carefully, you'll note that it doesn't claim that programs can't access the devices, it only states that user application programs are denied such access. Specially written modules, known as device drivers, are able to access I/O ports by special permission from the operating system. Writing device drivers is well beyond the scope of this chapter (though it will make an excellent subject for a later volume in this text). Nevertheless, an understanding of how device drivers work may help you understand the possibilities and limitations of I//O under a "protected mode" operating system.
A device driver is a special type of program that connects to the operating system. The device driver must follow some special protocols and it must make some special calls to the operating system that are not available to standard applications. Further, in order to install a device driver in your system you must have administrator privileges (device drivers create all kinds of security and resource allocation problems; you can't have every hacker in the world taking advantage of rogue device drivers running on your system). Therefore, "whipping out a device driver" is not a trivial process and application programs cannot load and unload arbitrary drivers at will.
Fortunately, there are only a limited number of devices you'd typically find on a PC, therefore you only need a limited number of device drivers. You would typically install a device driver in the operating system the same time you install the device (or when you install the operating system if the device is built into the PC). About the only time you'd really need to write your own device driver is when you build your own device or in some special instance when you need to take advantage of some devices capabilities that the standard device drivers don't allow for.
One big advantage to the device driver mechanism is that the operating system (or device vendors) must provide a reasonable set of device drivers or the system will never become popular (one of the reasons Microsoft and IBM's OS/2 operating system was never successful was the dearth of device drivers). This means that applications can easily manipulate lots of devices without the application programmer having to know much about the device itself; the real work has been taken care of by the operating system.
The device driver model does have a few drawbacks, however. The device driver model is great for low-speed devices, where the OS and device driver can respond to the device much more quickly than the device requires. The device driver model is also great for medium and high-speed devices where the system transmits large blocks of data in one direction at a time; in such a situation the application can pass a large block of data to the operating system and the OS can transmit this data to the device (or conversely, read a large block of data from the device and place it in an application-supplied buffer). One problem with the device driver model is that it does not support medium and high-speed data transfers that require a high degree of interaction between the device and the application.
The problem is that calling the operating system is an expensive process. Whenever an application makes a call to the OS to transmit data to the device it could actually take hundreds of microseconds, if not milliseconds, before the device driver actually sees the data. If the interaction between the device and the application requires a constant flurry of bytes moving back and forth, there will be a big delay if each transfer has to go through the operating system. For such applications you will need to write a special device driver to handle the transactions directly in the driver rather than continually returning to the application.
7.16 Putting It All Together
Although the CPU is where all the computation takes place in a computer system, that computation would be for naught if there was no way to get information into and out of the computer system. This is the responsibility of the I/O subsystem. I/O at the machine level is considerably different than the interface high level languages and I/O subroutine libraries (like stdout.put) provide. At the machine level, I/O transfers consist of moving bytes (or other data units) between the CPU and device registers or memory.
The 80x86 family supports two types of programmed I/O: memory-mapped input/output and I/O-mapped I/O. PCs also provide a third form of I/O that is mostly independent of the CPU: direct memory access or DMA. Memory-mapped input/output uses standard instructions that access memory to move data between the system and the peripheral devices. I/O-mapped input/output uses special instructions, IN and OUT, to move data between the CPU and peripheral devices. I/O-mapped devices have the advantage that they do not consume memory addresses normally intended for system memory. However, the only access to devices using this scheme is through the IN and OUT instructions; you cannot use arbitrary instructions that manipulate memory to control such peripherals. Devices that use DMA have special hardware that let them transmit data to and from system memory without going through the CPU. Devices that use DMA tend to be very high performance, but this I/O mechanism is really only useful for devices that transmit large blocks of data at high speeds.
I/O devices have many different operating speeds. Some devices are far slower than the CPU while other devices can actually produce or consume data faster than the CPU. For devices that are slower than the CPU, some sort of handshaking mechanism is necessary in order to coordinate the data transfer between the CPU and the device. High-speed devices require a DMA controller or buffering since the CPU cannot handle the data rates of these devices. In all cases, some mechanism is necessary to tell the CPU that the I/O operation is complete so the CPU can go about other business.
In modern 32-bit operating systems like Windows and Linux, applications programs do not have direct access to the peripheral devices. The operating system coordinates all I/O via the use of device drivers. The good thing about device drivers is that you (usually) don't have to write them - the operating system provides them for you. The bad thing about writing device drivers is if you have to write one, they are very complex. A later volume in this text may discuss how to do this.
Because HLA programs usually run as applications under the OS, you will not be able to use most of the coding techniques this chapter discusses within your HLA applications. Nevertheless, understanding how device I/O works can help you write better applications. Of course, if you ever have to write a device driver for some device, then the basic knowledge this chapter presents is a good foundation for learning how to write such code.
|
| |
|
|
|