Linux Kernel Architecture

Chapter 6: Device Drivers

Configuration Information

In contrast to many of its predecessors, the PCI bus is a jumper-free system. In other words, expansion

devices can be fully configured by software means and without user intervention.^22 To support such

configuration, each PCI device has a 256-byte-long configuration space with information on the special

characteristics and requirements of the device. Even though 256 bytes may at first appear to be a paltry

figure given current memory configuration levels, a large amount of information can be stored, as shown

in Figure 6-23, which illustrates the layout of the configuration space as required by the PCI specification.

BIST

Base address 0 Base address 1 Base address 2 Base address 3

Base address 4 Base address 5 CardBus CISPointer SubsystemVendor ID SubsystemDevice ID

Expansion ROM

base address Reserved

IRQ

Line

IRQ

Pin

Min

Gnt

Max

Lat

Vendor

ID

Device

ID

Cmd

Reg

Status

Reg RevIDClass Code CacheLineLatencyTimerHeaderType

012 345 678 910111213141516

0–16

16–32

32– 45

48– 64

Figure 6-23: Layout of the PCI configuration space.

Although the structure must be 256 bytes in length, only the first 64 bytes are

standardized. The remainder are freely available and are typically used to exchange

additional information between the device and the driver. The structure of this

information is (or should be) defined in thehardware documentation. It should also

be noted that not all information in the first 64 bytes is mandatory; some items are

optional and may be filled with zeros if they are not required by a device. The

mandatory items are highlighted in darker gray in the figure.

Thevendor IDanddevice IDuniquely identify the vendor and device type. Whereas the former is assigned

by the PCI Special Interest Group (an industry consortium) to identify individual companies,^23 the latter

can be freely selected by the vendors — they alone are responsible for ensuring that there are no overlaps

in their address space. Taken together the two IDs are often referred to as thesignatureof a device. Two

additional fields with similar names —subsystem vendor IDandsubsystem device ID — may also be used

to more accurately describe generic interfaces. Therevision IDenables a distinction to be made between

various device revision levels. This helps users select device driver versions where known hardware

faults have been eliminated or where new features have been added.

Theclass codefield is used to assign devices to various function groups and is split into two parts. The

first 8 bits indicate the base class and the remaining 16 bits a subclass of the base class. Examples of

base classes and their subclasses are given below (I use the names of the corresponding constants in

<pci_ids.h>).

(^22) Some readers may well remember the ISA ‘‘game‘‘ where the cards, mostly miserably documented, were configured by manually
adjusting resources by means of jumpers.
(^23) The ID for Intel is0x8086...