regen

unix v7是最后一个广泛发布的研究型UNIX版本

make unix
.DE
if the libraries /usr/sys/dev/LIB2 and /usr/sys/sys/LIB1,
and also c.o and l.o,
are correct.
Use
.DS
cd /usr/sys/conf
make all
.DE
to recompile everything
and recreate the libraries from scratch.
This is needed, for example, when a header included
in several source files is changed.
See `Setting Up UNIX' for other information about configuration
and such.
.PP
When the make is
done, the new system is present in the
current directory as `unix'.
It should be tested before destroying the currently
running `/unix', this is best done by doing something like
.DS
mv /unix /ounix
mv unix /unix
.DE
If the new system doesn't work, you can still boot `ounix'
and come up (see boot(8)).
When you have satisfied yourself that the new system works,
remove /ounix.
.PP
To install a new device driver,
compile it and
put it into its library.
The best way to put it into
the library is to use the command
.DS
ar uv LIB2 x.o
.DE
where x is the routine you just compiled.
(All the device drivers distributed with
the system are already in the library.)
.PP
Next, the device's interrupt vector must be entered in l.s.
This is probably already done by the routine mkconf(1), but if the
device is esoteric or nonstandard you will have to massage
l.s by hand.
This involves placing a pointer to a callout routine
and the device's priority level in the vector.
Use some other device (like the console) as a guide.
Notice that the entries in l.s must be in order
as the assembler does not permit moving the
location counter `.' backwards.
The assembler also does not permit assignation of
an absolute number to `.', which is the
reason for the `. = ZERO+100' subterfuge.
If a constant smaller than 16(10) is added to the
priority level,
this number will be available as the first argument of the interrupt routine.
This stratagem is used when
several similar devices share the same interrupt routine
(as in dl11's).
.PP
If you have to massage l.s, be sure to add the code
to actually transfer to the interrupt routine. Again use
the console as a guide. The apparent strangeness of this code
is due to running the kernel in separate I&D space.
The
.I call
routine
saves registers as required and prepares a C-style
call on the actual interrupt routine
named after the `jmp' instruction.
When the routine returns,
.I call
restores the registers and performs an
rti instruction.
As an aside, note that
external names in C programs have an
underscore (`_') prepended to them.
.PP
The second step which must be performed to add a 
device unknown to mkconf is
to add it to the configuration table
/usr/sys/conf/c.c.
This file contains two subtables,
one for block-type devices, and one for character-type devices.
Block devices include disks, DECtape, and magtape.
All other devices are character devices.
A line in each of these tables gives all the information
the system needs to know about the device handler;
the ordinal position of the line in the table implies
its major device number, starting at 0.
.PP
There are four subentries per line in the block device table,
which give its open routine, close routine, strategy routine, and
device table.
The open and close routines may be nonexistent,
in which case the name `nulldev' is given;
this routine merely returns.
The strategy routine is called to do any I/O,
and the device table contains status information for the device.
.PP
For character devices, each line in the table
specifies a routine for open,
close, read, and write, and one which sets and returns
device-specific status (used, for example, for stty and gtty
on typewriters).
If there is no open or close routine, `nulldev' may
be given; if there is no read, write, or status
routine, `nodev' may be given.
Nodev sets an error flag and returns.
.PP
The final step which must
be taken to install a device is to make a special file for it.
This is done by mknod(1), to which you must specify the
device class (block or character),
major device number (relative line in the configuration table)
and minor device number
(which is made available to the driver at appropriate times).
.PP
The documents
`Setting up Unix' and
`The Unix IO system'
may aid in comprehending these steps.
.SH
The Library libc.a
.PP
The library /lib/libc.a is where most of the subroutines
described in sections 2 and 3 of the manual are kept.
This library
can be remade using the following commands:
.DS
cd /usr/src/libc
sh compall
sh mklib
mv libc.a /lib/libc.a
.DE
If single routines need to be recompiled and replaced, use
.DS
cc \-c \-O x.c
ar vr /lib/libc.a x.o
rm x.o
.DE
The above can also be used to put new items into the library.
See ar(1), lorder(1), and tsort(1).
.PP
The routines in /usr/src/cmd/libc/csu (C start up) are not in
libc.a. These are separately assembled and put into
/lib. The commands to do this are
.DS
cd /usr/src/libc/csu
as \- x.s
mv a.out /lib/x
.DE
where x is the routine you want.
.SH
Other Libraries
.PP
Likewise,
the directories containing the source for the other libraries
have files compall (that recompiles everything)
and mklib (that recreates the library).
.SH
System Tuning
.PP
There are several tunable parameters in the system. These set
the size of various tables and limits. They are found in the
file /usr/sys/h/param.h as manifests (`#define's).
Their values are rather generous in the system as distributed.
Our typical maximum number of users is about 20, but there are
many daemon processes.
.PP
When any parameter is changed, it is prudent to recompile
the entire system, as discussed above.
A brief discussion of each follows:
.IP NBUF 12
This sets the size of the disk buffer cache. Each buffer is 512 bytes.
This number should be around 25 plus NMOUNT,
or as big as can be if the above number of
buffers cause the system to not fit in memory.
.IP NFILE 12
This sets the maximum number of open files. An entry is made in
this table every time a file is `opened' (see open(2), creat(2)).
Processes share these table entries across forks (fork(2)). This number
should be about the same size as NINODE below. (It can be a bit smaller.)
.IP NMOUNT 12
This indicates the maximum number of mounted file systems. Make it
big enough that you don't run out at inconvenient times.
.IP MAXMEM 12
This sets an administrative limit on the amount of memory
a process may have.
It is set automatically if the amount of physical memory is small,
and thus should not need to be changed.
.IP MAXUPRC 12
This sets the maximum number of processes that any one user can
be running at any one time. This should be set just large enough
that people can get work done but not so large that a user can
hog all the processes available (usually by accident!).
.IP NPROC 12
This sets the maximum number of processes that can be
active.
It depends on the demand pattern of the typical user;
we seem to need about 8 times the number
of terminals.
.DE
.IP NINODE 12
This sets the size of the inode table. There is one entry in the inode
table for every open device, current working directory,
sticky text segment,
open file, and mounted device.
Note that if two users have a file open there is still only one entry
in the inode table. A reasonable rule of thumb for the size of
this table is
.DS
NPROC + NMOUNT + (number of terminals)
.DE
.IP SSIZE 12
The initial size of a process stack. This may be made bigger
if commonly run processes have large data areas on the stack.
.IP SINCR 12
The size of the stack growth increment.
.IP NOFILE 12
This sets the maximum number of files that any one process can have
open.
20 is plenty.
.IP CANBSIZ 12
This is the size of the typewriter canonicalization buffer. It is
in this buffer that erase and kill processing is done. Thus this
is the maximum size of an input typewriter line. 256 is usually
plenty.
.IP CMAPSIZ 12
The number of fragments that memory can be broken into. This should
be big enough that it never runs out.
The theoretical maximum is twice the number of processes,
but this is a vast overestimate in practice.
50 seems enough.
.IP SMAPSIZ 12
Same as CMAPSIZ except for secondary (swap) memory.
.IP NCALL 12
This is the size of the callout table. Callouts are entered in this
table when some sort of internal system timing must be done, as in
carriage return delays for terminals. The number must be big enough
to handle all such requests.
.IP NTEXT 12
The maximum number of simultaneously executing pure programs. This
should be big enough so as to not run out of space under heavy load.
A reasonable rule of thumb is about
.DS
(number of terminals) + (number of sticky programs)
.DE
.IP NCLIST 12
The number of clist segments. A clist segment is 6 characters.
NCLIST should be big enough so that the list doesn't become exhausted
when the machine is busy. The characters that have arrived from a terminal
and are waiting to be given to a process live here. Thus enough space
should be left so that every terminal can have at least one average
line pending (about 30 or 40 characters).
.IP TIMEZONE 12
The number of minutes westward from Greenwich. See `Setting Up UNIX'.
.IP DSTFLAG 12
See `Setting Up UNIX' section on time conversion.
.IP MSGBUFS 12
The maximum number of characters of system error messages saved. This
is used as a circular buffer.
.IP NCARGS 12
The maximum number of characters in an exec(2) arglist. This
number controls how many arguments can be passed
into a process.
5120 is practically infinite.
.IP HZ 12
Set to the frequency of the system clock (e.g., 50 for
a 50 Hz. clock).