blob: b643c668900f5d0849dbd63a3bd15248d365da70 (
plain)
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
|
This is a list of things that still need to be done:
* Fix the totally bogus _atomic_lock in arch/mips.
This is a nightmare because you really need to incur a ctxt switch
for pmaxen. R4000 has some instructions to do an atomic test-and-set
but not mips1. fuck fuck fuck. the ctx switch has to be sigsetmask
* Add thread stuff to the other archs in libc/arch/; ie change
some usages of ENTRY to SYSENTRY in some .S files and add
the new macros to their SYS.h. Only the i386 arch is working.
* Finish/verify the md stuff for setjmp and atomic_lock for sparc and
alpha. Boy do they look dodgy.
* Find a better way of doing the _sys_aliases stuff; ideally by
using the linker. jmp's (current implementation) are ugly, but
I don't want to have to hack ../libc/sys/Makefile.inc just to
partition the syscalls into wrapped and non-wrapped. The `.set'
assembler directive just doesn't work as I thought it would.
* Fix the bug where _thread_init() isn't called when libc_r.so is used
instead of libc_r.a ... some kind of shlib common problem? may
need to use crt0 instead? __CTOR_LIST__ is just not being updated by ld.
Frankly this one is driving me nuts. Maybe <bsd.lib.mk> is wrong.
This varies on different architectures and exec format.
* Test the objective-C compiler to see if it can use threads.
Really this shows up a nasty problem with needing to say -threads
whenever you want to use threads. Ideally, linking with -lc_r library
would override -lc; but the _POSIX_THREADS define would alter
some optimisations done in <stdio.h>... What to do!?! FreeBSD looks
like they are going to have an always-threaded libc with optimisations
for the 1-thread case and use weak symbols to override functions when
someone uses -lc_r. Do we have weak symbols everywhere? I wonder
what the performance hit is if #define LOCK() if(nthreads>1){lock();}
is used... need to profile this.
* Find all the static buffers in libc and make them per-thread. This
is not absolutely necessary but incurs no penalty for single-threaded
operation, and makes life easier & safer for when threads are used.
On the other hand, it adds a bit more code bloat when you use threads
and makes openbsd programs non-portable to other OSs with less-nicer
libraries. If we do the bare posix minimum, we promote portable code.
* Look into how netbsd are going with their kernel threads
* Look into how asynchronous I/O can help us. In particular, the
(unimplemented) aio*() functions.
* Update the libc manual pages to describe the posix re-entrant functions.
Although this is actually trivial to do, I have to decide on a
consistent way of adding them - maybe `.Sh THREAD-SAFE FUNCTIONS' ?
Should look into standards to see what they suggest/did..
May also need to document "This is not thread-safe" for some library
functions (yet to be identified).
* Find out where freebsd/netbsd use pread() and pwrite().. i think
its in the database routines mostly. We will need to rip their code.
* Figure out what to do with the configuration system variables (_SC_*)
$OpenBSD: TODO,v 1.3 1998/11/20 11:18:57 d Exp $
|
