Kernel Configuration

This contains links to information used in the lectures.

The Unix Kernel provides interfaces to the following

• Processes (time-sharing, protected address space)
• Signals and semaphores
• Virtual memory (swapping, paging, mapping)
• Filesystem
• Interprocess communication (pipes and network connections)

ATT v BSD

• Can provide both with explicit information on hardware
• BSD kernels ask for guidance in advance
• ATT kernels try to auto-configure
• HP-UX and SunOS are both BSD like

When to Configure

• Installing a new system
  Most systems come with generic kernel and configuration file. Reconfigure to make kernel smaller.
• Adding device drivers
  Make need to configure from scratch. Sometimes can add using program, or by a "loadable module"
• Tuning Table Sizes
  You may need to change table size to increase or decrease
  • maximum number of open files
  • maximum number of processes

Building A BSD kernel - SunOS

conf.common/    net/            sbusdev/        sun4m/          sys/
debug/          netinet/        scsi/           sundev/         ufs/
krpc/           nfs/            sparc/          sunif/          vm/
lofs/           os/             specfs/         sunwindow/
mon/            rpc/            sun/            sunwindowdev/

NIMITZ/         cpu.h           intreg.h        openprom_xxx.c  seg_kmem.h
OBJ/            dbx_machdep.c   iocache.h       param.h         setjmp.h
a.out.h         debug/          iommu.h         pcb.h           trap.h
asm_linkage.h   devaddr.h       libprom.a*      psl.h           vm_hat.h
async.h         devr.h          llib-lprom.ln*  pte.h           vmparam.h
auxio.h         eeprom.h        machine@        reg.h
buserr.h        enable.h        memerr.h        romvec.h
clock.h         fpu@            mmu.h           sas_simdef.h
conf/           frame.h         module.h        scb.h

• directories with uppercase names are compilation areas
• each kernel configuration has a name - normally machine name
• the names of the kernel configuration files and compilation areas are same
• conf holds the text files that describe a configuration
• the config command reads a file from /usr/sys/conf and generates files in the appropriate compilation directory
• the rest of the directories contain parts that are assembled to create executable image of kernel

• Identify the system's hardware
• Create and edit the kernel's configuration file in /usr/sys/conf
• Create the kernel's compilation directory in /usr/sys
• Run the config program from conf directory
• Run make dependto build dependencies in Makefile
• Build the kernel with make
• Archive the old kernel and install thenew one
• Test and debug new kernel
• Document the new kernel

• Identify the system's hardware
  Include:
  • Disk drives and controllers
  • Tape drives and controllers
  • Network Interfaces
  • Terminal interfaces
  • Coprocessors (encryption, graphics, video, etc)
  • Frame buffers
  • Keyboard and Mouse
• Create and edit the kernel's configuration file in /usr/sys/conf
  This should have the same name as the kernel. This is normally done by copying the GENERIC one and deleting lines that are not needed. Example of GENERIC and NIMITZ configuration files.
• Create the kernel's compilation directory in /usr/sys
  Some versions of config do this automatically.
• Run the config program from conf directory
  cd to /usr/sys/conf and run config.

       /usr/etc/config [ -fgnp ] [ -o obj_dir ] config_file
  

  config does the preparation necessary for building a new system kernel with make(1). The config_file named on the command line describes the kernel to be made in terms of options you want in your system, size of tables, and device drivers to be included. When you run config, it uses several input files located in the current directory (typically the conf subdirectory of the system source including your config_file). If the directory named ../config_file does not exist, config will create one.
• Run make dependto build dependencies in Makefile
  cd to the configuration directory for new kernel and run make depend.
• Build the kernel with make
  In the compilation directory, make kernelname, on SunOS and HP-UX 10.XX this is usually make vmunix, on HP-UX 9.XX it is make hp-ux.
  It is a good idea to keep a copy of all make errors using for example

  make vmunix |& tee ERRS.LOG
  

• Archive the old kernel and install the new one
  Always keep the old kernel until the new one is thouroughly debugged. Check carefully in the system documentation how to boot an alternative (non-default) kernel. If the system supports booting arbitrary kernels copy the kernel to vmunix.bak or vmunix.date, where date is the date of its creation.
• Test and debug new kernel
  • Bring system to single-user mode and run fsck -p
  • Bring system up with new kernel
  • Run ps. If get "no namelist", then either the kernel name is wrong, the permission on it are incorrect, or something important did not get compiled in.
  • Run ls on some directory. It will fail if the partition tables in the kernel are incorrect or the filesystem is not functioning corrcectly.
• Document the new kernel
  Go back to the configuration file - /usr/sys/conf/NIMITZ - in our case, and add comments indicating what was done.

Creating a BSD Configuration file - SunOS

machine "type"

This is system is to run on the machine type specified. Only one machine type can appear in the config file. The legal types for a Sun system are sun3, sun3x, sun4, sun4c, sun4m, and sun386. Note: the double quotes around type are part of t he syntax, and must be included.

cpu "type"

This system is to run on the CPU type specified. More than one CPU type can appear in the config file.

ident name

Give the system identifier - a name for the machine or machines that run this kernel. Note: name must be enclosed in double quotes if it contains both letters and digits. Also, note that if name is GENERIC, you need not include the `options GENERIC' clause in order to specify `swap generic'.

maxusers number
The maximum expected number of simultaneously active user on this system is number. This number is used to size several system data structures.

options optlist

Compile the listed options into the system. Options in this list are separated by commas. A line of the form:

options FUNNY,HAHA

yields

-DFUNNY -DHAHA

to the C compiler. An option may be given a value, by following its name with = (equal sign) then the value enclosed in (double) quotes. None of the standard options use such a value. In addition, options can be used to bring in additional files if the option is listed in the files files. All options should be listed in upper case. In this case, no corresponding option.h will be created as it would be using the corresponding pseudo-device method.

config sysname config_clauses...

Generate a system with name sysname and configuration as specified in config-clauses. The sysname is used to name the resultant binary image and per-system swap configuration files. The config_clauses indicate th e location for the root file system, one or more disk partitions for swapping and paging, and a disk partition to which system dumps should be made. All but the root device specification may be omitted; config will assign default values as described below.

root

A root device specification is of the form:

root on xy0d

If a specific partition is omitted - for example, if only root on xy0 is specified - the `a' partition is assumed. When a generic system is being built, no root specification should be given; the root device will be defined at boot time by prompting the console.

swap

To specify a primary swap partition, use a clause of the form:

swap on partition

Swapping areas may be almost any size. Partitions used for swapping are sized at boot time by the system; to override dynamic sizing of a swap area the number of sectors in the swap area can be specified in the config file. For example, swap on xy0b size 99999 would configure a swap partition with 99999 sectors. If swap generic or no partition is specified with on, partition b on the root device is used. For dataless clients, use:

swap on type nfs

dumps

The location to which system dumps are sent may be specified with a clause of the form:

dumps on xy1

If no dump device is specified, the first swap partition specified is used. If a device is specified without a particular partition, the `b' partition is assumed. If a generic configuration is to be built, no dump device should be specified; the dump devi ce will be assigned to the swap device dynamically configured at boot time. Dumps are placed at the end of the partition specified. Their size and location is recorded in global kernel variables dumpsize and dumplo, respectively, for use by savecore(8).

Device names specified in configuration clauses are mapped to block device major numbers with the file devices.machine, where machine is the machine type previously specified in the configuration file. If a device name to block device major number mapping must be overridden, a device specification may be given in the form:

major x minor y

dev_type is either controller, disk, tape, device, devicedriver, or pseudo-device. These types have the following meanings:

controller

A disk or tape controller.

disk or tape

Devices connected to a controller.

device

Something "attached" to the main system bus, like a cartridge tape interface.

device-driver

This declares support for a device of name dev_name. For most devices on desktop SPARCsystems, this is all that is required.

pseudo-device

A software subsystem or driver treated like a device driver, but without any associated hardware. Current examples are the pseudo-tty driver and various network subsystems. For pseudo-devices, more_info may be specified as an integer, that gives the value of the symbol defined in the header file created for that device, and is generally used to indicate the number of instances of the pseudo-device to create.

con_dev is what the device you are specifying is connected to. It is either nexus?, a bus type, or a controller. There are several bus types which are used by config and the kernel.

The possible bus types are:

obmem

On board memory

obio

On board io vme16d16 (vme16) 16 bit VMEbus/ 16 bit data vme24d16 (vme24) 24 bit VMEbus/ 16 bit data

vme32d16

32 bit VMEbus/ 16 bit data

vme16d32

16 bit VMEbus/ 32 bit data

vme24d32

24 bit VMEbus/ 32 bit data vme32d32 (vme32) 32 bit VMEbus/ 32 bit data

ipi

IPI pseudo bus (sun4 system only)

more_info is a sequence of the following:

csr address

Specify the address of the csr (command and status registers) for a device. The csr addresses specified for the device are the addresses within the bus type specified.

The csr address must be specified for all controllers, and for all devices connected to a main system bus.

drive number

For a disk or tape, specify which drive this is.

flags number

These flags are made available to the device driver, and are usually read at system initialization time.

priority level

For devices which interrupt, specify the interrupt level at which the device operates.

vector intr number [ intr number . . . ]

For devices which use vectored interrupts on VMEbus systems, intr specify the vectored interrupt routine and number the corresponding vector to be used (0x40-0xFF). A ? may be substituted for a number in two places and the system will figure out what to fill in for the ? when it boots. You can put question marks on a con_dev (for example, at virtual `?'), or on a drive number (for example, drive `?'). T his allows redundancy, as a single system can be built which will boot on different hardware configurations.