| Age | Commit message (Collapse) | Author |
|---|
|
- better function descriptions
|
|
Also, for safety lock the _thread_fd_table when removing entries.
|
|
|
|
|
|
_thread_fd_table_dup
|
|
|
|
to notice that the file exists. This fixes a problem where an
application may think a file was in non-block mode because the
threads kernel played with the flags. Also fix a stupid error
introduced in the last commit -- the threaded version of dup and
dup2 were foobared. Bad marc.
|
|
for dup-ed fds are shared to ensure proper flag handling. A refcnt
was added to control when entries should be freed. Specific changes:
close: don't free entry unless refcnt is zero
dup: rewrite to use new function _thread_fd_table_dup
dup2: rewrite to use new function _thread_fd_table_dup
fcntl: use _thread_fd_table_dup
uthread_fd: initialize thread fd table, searching for dup-ed fds. Add
function to share _thread_fd_table entries when an fd is dup-ed.
uthread_init: make it readable. Call fd init functions.
All current regression tests plus the mysql torture test pass. The
new stdfiles regression test fails (I/O redirection problem). Part
2 is intended to fix that problem
|
|
|
|
|
|
|
|
Clean up (compiler warning elimination). Compile check options added
but commented out as they have not been checked on all architectures, yet.
|
|
of the thread regressions tests pass on i386, sparc, sparc64
(save the siginfo test on sparc64 due to a kernel issue) and alpha.
The tests should also pass on ppc.
In addition, it fixes the problems with the "mysql torture test"
provided by one of our users. The python port also appears to
work correctly with these changes.
Summary of changes:
* check_pending removed from thread structure, no longer used.
* unused elements of sigstatus structure removed. The sigstatus
structure is now used to keep track of siginfo data instead of
defining a local array in uthread_sig.c.
* _thread_kern_sched_sig removed
* _thread_sig_process goes away -- can't have a lock active when
signal handlers are called. Functions now call _thread_sig_handle
directly.
* _thread_clear_pending now used lib wide to clear pending flags.
It was named _clean_pending_flag and only used in uthread_sig.c.
The function clears both per thread signals, and per process signals.
* _thread_sig_handle now returns a value.
* unused debugging macros removed from the thread kernel
* dispatch pending signals after switching to a thread providing
that signal handling hasn't been deferred by the thread.
* process thread switchhooks before dispatching pending sigs
* check for thread cancellation before dispatching pending sigs
* re-wrote pthread-kill to do the correct thing. It now does
minimal thread-kill-specific processing and then calls the
existing code in uthread_sig to process the generated signal.
* shut the compiler up when compiling uthread_mutex.c
* no more "signal_lock". It does more harm than good.
* keep track of "per-process" signals.
* don't bother saving siginfo_t data for the scheduling signal.
* per posix: SIGSTOP cleared when SIGCONT received and vice versa.
* add _dispatch_signal to properly dispatch a signal to a thread.
It sets the appropriate signal mask, something that was missing
in the previous implementation. This fixes several bugs.
The previous method held a lock. If the signal handler longjmp-ed
the lock was never cleared and no more signals were processed.
|
|
half-sized areas. fixes preemption_float on powerpc.
|
|
|
|
If the fpu isn't enabled, don't save state, otherwise, save the state,
mark both sides clean and disable the fpu.
On restore:
If the saved fpu state has the fpu disabled, do nothing, otherwise, restore
everything.
|
|
|
|
an arch/i386 patch previously commited and arch/sparc64 patches
from jason@ make the preemption_float test pass on those two
architectures.
Do not run signal handlers for a thread until the thread has
been made current, ensuring the proper context. Solves several
(if not all) of the '_pq_insert_tail: Already in priority queue'
problems.
|
|
|
|
|
|
|
|
This is necessary for alpha setjmp. The alpha setjmp/longjmp
regression tests pass with -pthread with this change
|
|
|
|
are both void.
The select call is a cancellation point per IEEE Std 1003.1-2001.
This should fix a problem espie@ found in kde.
|
|
|
|
takes their place. The -pthread option to gcc uses libpthread.
ppc portion of patch from drahn@ (thanks). gcc portions of patch
OK espie@. Ports testing (no bad effects!) by naddy@
best build sequence is: build gcc, build system, build ports
remove /usr/lib/lib{c_r,npthread}.*. Note, once libc_r and
libnpthread is removed threaded ports will stop working until
they have been re-built using the libpthread and the updated gcc.
That's why the above build sequence is recommended.
|
|
is defined as an int. This solves several alpha/pthread problems
as repoted by various ports people.
|
|
|
|
|
|
now build libraries with propolice enabled. Without this, existing
binaries (such as ports/packages) that link with any system library
other than libc will fail with an undefined symbol of "___guard"
(__guard on ELF).
Pointed out by markus@ and discussed with deraadt@
|
|
|
|
putchar_unlocked. Crank the minor on related libs. OK fgs@, deraadt@
|
|
The library is installed as libnpthread, not libpthread. This gets
around broken autoconf ports that attempt to link using both
-lpthread and -pthread.
The next step will be to change the gcc -pthread flag to use this lib
instead of libc_r.
|
|
|
|
always build with -Wall
bump version to 0.7 (matches libc_r x.7)
|
|
registers. ok marc@
|
|
|
|
|
|
is created.
|
|
in libc so static linking works with -lpthreads. There is a
(linker?) problem that still shows up with static -pthread, though.
Solution to problem from freebsd.
|
|
|
|
Thanks to miod@ for m68k and vax fixes
|
|
not zero on all arches
|
|
|
|
alpha (millert@), i386 (marc@), m68k (millert@ and miod@),
powerpc (drahn@ and dhartmei@), sparc (millert@ and marc@),
sparc64 (marc@), and vax (millert@ and miod@).
Thanks to millert@, miod@, and mickey@ for fixes along the way.
|
|
|
|
ok marc@
|
|
|
|
missing) _thread_fd_unlock_owned function can be added with minimal pain.
The incorrect special handling of the stdio fds was also removed.
Tested with the libc_r regression tests and the mysql regression tests.
No complaints from any developers
|
|
handling errors in the threads library. Most of the libc_r regression
tests are now ok. thread specific kill semantics are still not correct.
No negative comments after posting to tech@ a week or so ago.
siginfo test fails on sparc64 due to sparc64 oddity.
|
