1 Linuxthreads - POSIX 1003.1c kernel threads for Linux
3 Copyright 1996, 1997 Xavier Leroy (Xavier.Leroy@inria.fr)
8 This is release 0.7 (late beta) of LinuxThreads, a BiCapitalized
9 implementation of the Posix 1003.1c "pthread" interface for Linux.
11 LinuxThreads provides kernel-level threads: each thread is a separate
12 Unix process, sharing its address space with the other threads through
13 the new system call clone(). Scheduling between threads is handled by
14 the kernel scheduler, just like scheduling between Unix processes.
19 - Linux version 2.0 and up (requires the new clone() system call
20 and the new realtime scheduler).
22 - For Intel platforms: libc 5.2.18 or later is required.
23 5.2.18 or 5.4.12 or later are recommended;
24 5.3.12 and 5.4.7 have problems (see the FAQ.html file for more info).
26 - Also supports glibc 2 (a.k.a. libc 6), which actually comes with
27 a specially-adapted version of this library.
29 - Currently supports Intel, Alpha, Sparc, Motorola 68k, ARM and MIPS
32 - Multiprocessors are supported.
37 - Edit the Makefile, set the variables in the "Configuration" section.
46 gcc -D_REENTRANT ... -lpthread
48 A complete set of manual pages is included. Also see the subdirectory
49 Examples/ for some sample programs.
54 - All functions in the Posix 1003.1c base interface implemented.
55 Also supports priority scheduling.
57 - For users of libc 5 (H.J.Lu's libc), a number of C library functions
58 are reimplemented or wrapped to make them thread-safe, including:
60 * stdio functions (define _REENTRANT before including <stdio.h>)
61 * per-thread errno variable (define _REENTRANT before including <errno.h>)
62 * directory reading functions (opendir(), etc)
64 * gmtime(), localtime()
66 New library functions provided:
67 * flockfile(), funlockfile(), ftrylockfile()
68 * reentrant versions of network database functions (gethostbyname_r(), etc)
69 and password functions (getpwnam_r(), etc).
71 - libc 6 (glibc 2) provides much better thread support than libc 5,
72 and comes with a specially-adapted version of LinuxThreads.
73 For serious multithreaded programming, you should consider switching
74 to glibc 2. It is available from ftp.gnu.org:/pub/gnu and its mirrors.
79 Many existing libraries are not compatible with LinuxThreads,
80 either because they are not inherently thread-safe, or because they
81 have not been compiled with the -D_REENTRANT. For more info, see the
82 FAQ.html file in this directory.
84 A prime example of the latter is Xlib. If you link it with
85 LinuxThreads, you'll probably get an "unknown 0 error" very
86 early. This is just a consequence of the Xlib binaries using the
87 global variable "errno" to fetch error codes, while LinuxThreads and
88 the C library use the per-thread "errno" location.
90 See the file README.Xfree3.3 for info on how to compile the Xfree 3.3
91 libraries to make them compatible with LinuxThreads.
94 KNOWN BUGS AND LIMITATIONS:
96 - Threads share pretty much everything they should share according
97 to the standard: memory space, file descriptors, signal handlers,
98 current working directory, etc. One thing that they do not share
99 is their pid's and parent pid's. According to the standard, they
100 should have the same, but that's one thing we cannot achieve
101 in this implementation (until the CLONE_PID flag to clone() becomes
104 - The current implementation uses the two signals SIGUSR1 and SIGUSR2,
105 so user-level code cannot employ them. Ideally, there should be two
106 signals reserved for this library. One signal is used for restarting
107 threads blocked on mutexes or conditions; the other is for thread
110 *** This is not anymore true when the application runs on a kernel
111 newer than approximately 2.1.60.
113 - The stacks for the threads are allocated high in the memory space,
114 below the stack of the initial process, and spaced 2M apart.
115 Stacks are allocated with the "grow on demand" flag, so they don't
116 use much virtual space initially (4k, currently), but can grow
119 Reserving such a large address space for each thread means that,
120 on a 32-bit architecture, no more than about 1000 threads can
121 coexist (assuming a 2Gb address space for user processes),
122 but this is reasonable, since each thread uses up one entry in the
123 kernel's process table, which is usually limited to 512 processes.
125 Another potential problem of the "grow on demand" scheme is that
126 nothing prevents the user from mmap'ing something in the 2M address
127 window reserved for a thread stack, possibly causing later extensions of
128 that stack to fail. Mapping at fixed addresses should be avoided
129 when using this library.
131 - Signal handling does not fully conform to the Posix standard,
132 due to the fact that threads are here distinct processes that can be
133 sent signals individually, so there's no notion of sending a signal
134 to "the" process (the collection of all threads).
135 More precisely, here is a summary of the standard requirements
136 and how they are met by the implementation:
138 1- Synchronous signals (generated by the thread execution, e.g. SIGFPE)
139 are delivered to the thread that raised them.
142 2- A fatal asynchronous signal terminates all threads in the process.
143 (OK. The thread manager notices when a thread dies on a signal
144 and kills all other threads with the same signal.)
146 3- An asynchronous signal will be delivered to one of the threads
147 of the program which does not block the signal (it is unspecified
149 (No, the signal is delivered to the thread it's been sent to,
150 based on the pid of the thread. If that thread is currently
151 blocking the signal, the signal remains pending.)
153 4- The signal will be delivered to at most one thread.
154 (OK, except for signals generated from the terminal or sent to
155 the process group, which will be delivered to all threads.)
157 - The current implementation of the MIPS support assumes a MIPS ISA II
158 processor or better. These processors support atomic operations by
159 ll/sc instructions. Older R2000/R3000 series processors are not
160 supported yet; support for these will have higher overhead.
162 - The current implementation of the ARM support assumes that the SWP
163 (atomic swap register with memory) instruction is available. This is
164 the case for all processors except for the ARM1 and ARM2. On StrongARM,
165 the SWP instruction does not bypass the cache, so multi-processor support
166 will be more troublesome.