13d9383679
Group binding needs to be cleared before sched_releasetcb(), as otherwise group_leave() will be called and group->tg_nmembers decremented or group being released. group_leave() should be called only after group_join() is called, not after group_bind(). From Jussi Kivilinna.
2015-07-01 06:24:34 -06:00
342f5fe33d
Fix references to the no-longer-existent misc/ directory in comments, README files, and documentation
2015-06-28 08:08:57 -06:00
61af0d511f
Use up_timer_gettime for CLOCK_MONOTONIC in tickless mode. From Max Neklyudov.
2015-06-22 06:37:26 -06:00
e1846dff82
sched/pthread: Implement pthread_mutex_trylock() for recursive mutexes
2015-06-16 08:32:20 -06:00
d16053c33c
sched/pthread: Use -1 instead of 0 as PID for unclaimed mutexes
2015-06-16 08:27:38 -06:00
d8db596b3b
POSIX message queues: Move mq_setattr() and mq_getattr() from nuttx/libc/mqueue to nuttx/sched/mqueue. Also add sysyscall support for mq_setattr() and mq_getattr(). This is necessary in protected and kernel builds because in those cases the message queue structure is protect and cannot be accessed directly from user mode code. Noted by Jouko Holopainen.
2015-06-03 09:24:17 -06:00
84f0303fc0
mq_timedreceive() move the location where the errno value is set; the TIMEDOUT errno setting was being overwritten by subsequent actions before returning. Noted by Freddie Chopin.
2015-06-03 07:24:50 -06:00
4adda9d428
Fix numerous typos in configuration variable names. Tracked down by Alan Carvalho de Assis
2015-05-23 17:08:35 -06:00
3adcae8ffb
Update the type passed to watchdog timer handlers. Using uint32_t is a problem for 64-bit machines.
2015-05-18 08:53:42 -06:00
16276a4793
Change the order of some operations to prevent a debug assertion from firing.
2015-05-18 06:34:54 -06:00
cc77a28f40
If HAVE_GROUP_MEMBERS is not defined, then pthread_kill() really is the same as kill()
2015-05-13 20:34:21 -06:00
4417728955
Re-implemened pthread_kill(). It as just a simple wrapper around kill() but since the correct dispatching of singles for multi-threaded task groups has been implemented, calling kill() does no work. The corrected implementation of pthread_kill() will direct the signal specifically to the specific pthread and no other.
2015-05-13 14:43:43 -06:00
8cb4e78a1a
Complete implementation of the IPv6 routing logic. From Max Neklyudov.
2015-05-13 07:22:02 -06:00
071658bb6d
Another missing semicolon
2015-04-23 07:18:59 -06:00
093a144934
sem_wait(): Missing comma on debug statement caused compile error with DEBUG on. Why hasn't that been noted before?
2015-04-23 07:15:55 -06:00
f161a1d0f7
atexit: Fix reference to undefined variable in certain configurations. From Juha Niskanen
2015-04-14 07:31:35 -06:00
b4423c1eed
Make some file section headers more consistent with standard
2015-04-08 06:47:36 -06:00
af6d9e2684
sched: Do not use uninitialized boolean in waitpid(). From Juha Niskanen
2015-03-12 07:45:10 -06:00
caea44a624
Fix a case in mq_timedsend() where the return errno value was being overwritten
2015-03-10 12:05:33 -06:00
7d46801f46
Reorder some operations to minimize a race condition
2015-03-10 10:41:21 -06:00
4c6057eca1
mq_timedsend(): Do check for time errors if the message queue is not full. Noted by Freddie Chopin
2015-03-10 09:42:35 -06:00
12d61531aa
Move board_ prototypes from arch.h to board.h
2015-02-27 20:02:03 -06:00
127951e029
Fix places where the errno value was being overwritten by subsequent actions so that the returned errno value was incorrect. From Max Neklyudov.
2015-02-25 07:45:04 -06:00
c99d927527
When an RTC is used, clock_systimespec() must subtract the basetime from the RTC time
2015-02-25 07:24:03 -06:00
cded7ea682
Fix some time value changes; mostly changing greater than 1000000000 to greater than or equal to 1000000000. From Juha Niskanen
2015-02-20 07:07:36 -06:00
8840102554
Networking: Divide net_intiialize() into net_setup() and net_initialize() to solve a chicken-and-egg problem. net_setup() must be caleld before up_initialize() is called so that networking data structures are ready to register new network devices.
...
net_initialize() now does only timer related operations and is called AFTER up_initialize() where the timers are configured. This is really.
2015-02-14 06:36:53 -06:00
a1d3faaa8a
Update ChangeLog
2015-02-13 18:39:43 -06:00
bc273596bb
Back out commit d998cbdfa6c10dbd8074dd43c39c23a006505987
2015-02-13 18:32:15 -06:00
0fcd620404
Move the initialization of the networking subsystem to AFTER up_initialize is called. Fromo Macs Neklyudov.
2015-02-13 13:41:21 -06:00
3724a5e98e
RTC: Remove all backdoor interfaces from rtc.h
2015-02-13 08:41:34 -06:00
d9f960e97b
Fix a problem in clock_systimer64 that occurs when (1) the 64-bit system time is enabled, and (2) the value of CONFIG_USEC_PER_TICK is less than 1 millisconds (such as when using the tickless mode of operation). In that case, the convertion of time to 64-bit millisecond value in clock_systmer64() causes some bad times to be returned. Time was converted to milliseconds, then to configured ticks. Precision was lost in the millisecond convertion.
...
The fix is to first convert time to a 64-bit microsecond value, then to the configured tick value.
Noted by David Sidrane.
2015-02-13 06:13:47 -06:00
147042218b
gettimeofday() and settimeofday(): Move gittimeofdady() from sched/clock to libc/time. All remove gettimeofday() from NuttX system calls. It is only a wrapper around clock_settime() and does not need a trap. gettimeofday() is no longer tried as a core OS interface.
...
gettimeofday has been decremented in POSIX 2008. settimeofday() was never part of POSIX, but I decided to add it to libc as well just for symmetry.
2015-02-12 11:10:46 -06:00
1d534ff896
Convert the 64-bit usec limit to a 32-bit tick limit
2015-02-03 07:18:17 -06:00
c5ab7c39d3
Eliminates a warning. From Macs N
2015-02-03 06:44:59 -06:00
a448e0f69f
On my platform I have just 16-bit timer clocked at 32kHz. As result oneshot timer max delay is 2s. The patch adds limit of maximum ticks in sched_timer_start (nuttx/sched/sched/sched_timerexpiration.c). From Macs N.
2015-02-03 06:25:19 -06:00
e69003cbf7
clock_systimespec(): Fixes for compilation errors in certain configurations. From Macs N.
2015-01-30 08:19:40 -06:00
6bc9a0c9c6
Fix backward calculation in the work queue timing. From Liio Chen via the PX4 repository
2015-01-27 06:17:02 -06:00
826f5516ff
Semaphores: sem_waitirq.c must be built when signals are disabled. That is because not handles not only the case of semaphore wait being awakened by a signal, but also the case with sem_timedwait.c when the semaphore wait is awakened by a timeout.
2014-12-28 15:03:12 -06:00
78b755961f
Semahores: Remove a warning with SIGNALS are disabled
2014-12-28 14:54:40 -06:00
937f9f23f1
strncpy will not copy the terminating \0 into the destination if the source is larger than the size of the destination. Ensure that the last byte is always zero and let strncpy only copy CONFIG_TASK_NAME_SIZE bytes. The issue of unterminated names can be observed in ps when creating a pthread while CONFIG_TASK_NAME_SIZE is set to 8.
2014-12-17 12:24:02 -06:00
b3a1939020
Sempahores: Add logic to clean up after task_delete() or pthread_cancel() if the task happens to be waiting on a semaphore when it is cancelled
2014-12-13 12:02:25 -06:00
1003ee3e9a
Costmetic change to a comment
2014-12-13 10:41:57 -06:00
ff87e2e02a
In message queue created return ENOSPC error if size exceeds the configured size of pre-allocatd messages; Use ENOSPC vs ENOMEM per OpenGroup.org. From Pierre-Noel Bouteville
2014-12-06 07:18:48 -06:00
ad05793c0f
msg type should be char * not void * in mq_send, mq_timedsend, mq_receive, and mq_timedreceive. Noted by Pierre-Noel Bouteville
2014-12-05 19:16:14 -06:00
2ab604f74f
Fix more issues detected by cppcheck
2014-11-25 14:25:02 -06:00
1aa528a572
More fixes to problems noted by cppcheck. Some are kind of risky; some are real bugs.
2014-11-25 14:10:35 -06:00
6f94e5e4be
Cosmetic chnages to comments
2014-11-20 07:10:48 -06:00
8c28718bcb
alling mq_timedreceived with immediate timeout was getting stuck and not
...
timeout. Immediate timeout is achieved by setting absolute timeout value to
past time, for example abstime={ .tv_sec=0, .tv_nsec=0 }. However absolute
time was converted to relative time using unsigned integer arithmetic and
resulted large ticks count by clock_abstime2ticks, instead of expected negative
ticks value.
Patch corrects clock_abstime2ticks to return negative ticks, if absolute time
is in the past.
Signed-off-by: Jussi Kivilinna <jussi.kivilinna@haltian.com>
2014-11-19 09:25:00 -06:00
9e290c10b5
Fix some compile issues introduces with removal of CONFIG_MAX_TASK_ARGS
2014-11-14 09:53:11 -06:00
838f2053e4
Fix some compile errors introduce in last commits
2014-11-13 07:44:45 -06:00
Jussi Kivilinna