![[BACK]](/cvsweb/icons/back.gif){kind=icon}
Return to sys.t CVS log ![[TXT]](/cvsweb/icons/text.gif){kind=icon}
![[DIR]](/cvsweb/icons/dir.gif){kind=icon}
Up to [CSRG BSD Unix] / 43BSDReno / share / doc / smm / 13.kchanges|
Return to sys.t CVS log | Up to [CSRG BSD Unix] / 43BSDReno / share / doc / smm / 13.kchanges |
1.1 root 1: .\" Copyright (c) 1986 Regents of the University of California.
2: .\" All rights reserved. The Berkeley software License Agreement
3: .\" specifies the terms and conditions for redistribution.
4: .\"
5: .\" @(#)sys.t 1.7 (Berkeley) 4/11/86
6: .\"
7: .NH
8: Changes in the kernel proper
9: .PP
10: The next several sections describe changes in the parts of the kernel
11: that reside in \fB/sys/sys\fP.
12: This section summarizes several of the changes that impact
13: several different areas.
14: .NH 2
15: Process table management
16: .PP
17: Although the process table has grown considerably since its original
18: design, its use was largely the same as in its first incarnation.
19: Several parts of the system used a linear search of the entire
20: table to locate a process, a group of processes, or group of processes
21: in a certain state.
22: 4.2BSD maintained linkages between the children of each parent process,
23: but made no use of these pointers.
24: In order to reduce the time spent examining the process table,
25: several changes have been made.
26: The first is to place all process table entries onto one of three
27: doubly-linked lists,
28: one each for entries in use by existing processes (\fIallproc\fP),
29: entries for zombie processes (\fIzombproc\fP), and free entries
30: (\fIfreeproc\fP).
31: This allows the scheduler and other facilities that must examine all
32: existing processes to limit their search to those entries actually in use.
33: Other searches are avoided by using the linkage among the children
34: of each process and by noting a range of usable process ID's when searching
35: for a new unique ID.
36: .NH 2
37: Signals
38: .PP
39: One of the major incompatibilities introduced in 4.2BSD
40: was that system calls interrupted by a caught signal were restarted.
41: This facility, while necessary for many programs that use signals to drive
42: background activities without disrupting the foreground processing,
43: caused problems for other, more naive, programs.
44: In order to resolve this difficulty, the 4.2BSD signal model has been
45: extended to allow signal handlers to specify whether or not the signal
46: is to abort or to resume interrupted system calls.
47: This option is specified with the \fIsigvec\fP call used to specify the handler.
48: The \fIsv_onstack\fP field has been usurped for a flag field,
49: with flags available to indicate whether the handler should be invoked
50: on the signal stack and whether it should interrupt pending system calls
51: on its return.
52: As a result of this change, those system calls that may be restarted
53: and that therefore take control over system call interruptions
54: must be modified to support this new behavior.
55: The calls affected in 4.3BSD are \fIopen\fP, \fIread\fP/\fIwrite\fP,
56: \fIioctl\fP, \fIflock\fP and \fIwait\fP.
57: .PP
58: Another change in signal usage in 4.3BSD affects fewer programs
59: and less kernel code.
60: In 4.2BSD, invocation of a signal handler on the signal stack caused
61: some of the saved status to be pushed onto the normal stack before
62: switching to the signal stack to build the call frame.
63: The status information on the normal stack included the saved PC and PSL;
64: this allowed a user-mode \fIrei\fP instruction to be used in implementing
65: the return to the interrupted context.
66: In order to avoid changes to the normal runtime stack when switching to
67: the signal stack, the return procedure has been changed.
68: As the return mechanism requires a special system call for restoring
69: the signal state, that system call was replaced with a new call,
70: \fIsigreturn\fP, that implements the complete return to the previous context.
71: The old call, number 139, remains in 4.3BSD for binary compatibility with
72: the 4.2BSD version of \fIlongjmp\fP.
73: .NH 2
74: Open file handling
75: .PP
76: Previous versions of
77: .UX
78: have traditionally limited each process
79: to at most 20 files open simultaneously.
80: In 4.2BSD, that limit could not be increased past 30,
81: as a 5-bit field in the page table entry was used to specify
82: either a file number or the reserved values PGTEXT or PGZERO
83: (fill from text file or zero fill).
84: However, the file mapping facility that previously used this field
85: no longer existed, and its replacement is unlikely to require this low
86: limit.
87: Accordingly, the internal virtual memory system support for mapped files
88: has been removed and the number of open files increased.
89: The standard limit is 64, but this may easily be increased if sufficient
90: memory for the user structure is provided.
91: In order to avoid searching through this longer list of open files
92: when the actual number in use is small,
93: the index of the last used open file slot is maintained in the field
94: \fIu.u_lastfile\fP.
95: The routines that implement open and close or implicit close (\fIexit\fP
96: and \fIexec\fP) maintain this field, and it is used whenever
97: the open file array \fIu.u_ofile\fP is scanned.
98: .NH 2
99: Niceness
100: .PP
101: The values for \fInice\fP used in 4.2BSD and previous systems
102: ranged from 0 though 39.
103: Each use of this scheduling parameter offset the actual value by the default,
104: NZERO (20).
105: This has been changed in 4.3BSD to use a range of -20 to 20,
106: with NZERO redefined as zero.
107: .NH 2
108: Software interrupts and terminal multiplexors
109: .PP
110: The DH11 and DZ11 terminal multiplexor handlers had been modified
111: to use the hardware's received-character silo
112: when those devices were used by the Berknet network.
113: In order to avoid stagnation of input characters and slow response to input
114: during periods of reduced input,
115: the low-level software clock interrupt handler had been made to call
116: the terminal drivers to drain input.
117: When the clock rate was increased in 4.2BSD, the overhead of checking
118: the input silos with each clock tick was increased,
119: and the use of specialized network hardware reduced the need for this
120: optimization.
121: Therefore, the terminal multiplexors in 4.3BSD use per-character
122: interrupts during periods of low input rate, and enable the silos
123: only during periods of high-speed input.
124: While the silo is enabled, the routine to drain it runs less
125: frequently than every clock tick; it is scheduled
126: using the standard timeout mechanism.
127: As a result, the software clock service routine
128: need not to be invoked on every clock tick, but only when timeouts or
129: profiling require service.
This archive runs on limited infrastructure. Preserving old code on modern bandwidth. Automated agents are requested to crawl responsibly.