Austin Group Defect Tracker

Aardvark Mark IV


Viewing Issue Simple Details Jump to Notes ] Issue History ] Print ]
ID Category Severity Type Date Submitted Last Update
0001216 [1003.1(2016/18)/Issue7+TC2] System Interfaces Comment Enhancement Request 2018-11-26 18:53 2021-05-10 13:26
Reporter mikecrowe View Status public  
Assigned To ajosey
Priority normal Resolution Accepted As Marked  
Status Applied  
Name Mike Crowe
Organization
User Reference
Section pthread
Page Number 0
Line Number 0
Interp Status ---
Final Accepted Text Note: 0005339
Summary 0001216: Adding clockid parameter to functions that accept absolute struct timespec timeouts
Description POSIX contains several functions that support waiting with an absolute
timeout passed as a struct timespec. This time must almost always be
measured against CLOCK_REALTIME. (pthread_cond_timedwait also supports a
single alternative clock specified at construction time via
pthread_condattr_setclock.)

Embedded systems and desktop computers may not have a good source of
accurate time, particularly at boot. This can result in CLOCK_REALTIME
warping by a large amount when the real time is known. In such situations,
CLOCK_REALTIME is not a good choice for expressing timeouts. A member of
the Android libc team has reported[1] that this has been the cause of real
world bugs in Android applications. I've worked on software at different
companies where we had to work around this problem.

The C++ standard provides std::condition_variable::wait_until and
std::timed_mutex::try_lock_until methods which support arbitrary clocks.
Current implementations that build upon POSIX convert these clocks to
CLOCK_REALTIME, which can cause race conditions when CLOCK_REALTIME is
warped. The C++ standard requires the clock to be specified at the time of
the wait, which means that pthread_condattr_setclock isn't useful.

The above problems can be solved by adding variants of the affected
functions that take an extra clockid_t parameter to indicate the clock that
should be used. Initially, implementations would be required to only
support passing CLOCK_REALTIME which would make adding support
straightforward. Support for CLOCK_MONOTONIC would be suggested, and
implementations would be free to support other clocks if they wished.

This proposal is the result of a thread[2] on the mailing list and my
original defect report[3] only covering pthread_cond_timedwait.

Various naming options for the new functions were discussed[4] and the
following names are based on one of the more popular options. In all cases
the clock immediately precedes the timespec timeout.

int
pthread_mutex_clocklock(
    pthread_mutex_t *restrict mutex,
    clockid_t clock,
    const struct timespec *restrict abstime)

int
pthread_rwlock_clockrdlock(
    pthread_rwlock_t *restrict rwlock,
    clockid_t clock,
    const struct timespec *restrict abstime)

int
pthread_rwlock_clockwrlock(
    pthread_rwlock_t *restrict rwlock,
    clockid_t clock,
    const struct timespec *restrict abstime)

int
pthread_cond_clockwait(
    pthread_cond_t *restrict cond,
    pthread_mutex_t *restrict mutex,
    clockid_t clock,
    const struct timespec *restrict abstime)

int
sem_clockwait(
    sem_t *restrict sem,
    clockid_t clock,
    const struct timespec *restrict abstime)

ssize_t
mq_clockreceive(
    mqd_t mqdes, char *restrict msg_ptr,
    size_t msg_len,
    unsigned int *restrict msg_prio,
    clockid_t clock,
    const struct timespec *restrict abs_timeout)

int
mq_clocksend(
    mqd_t mqdes, const char *restrict msg_ptr,
    size_t msg_len, unsigned int msg_prio,
    clockid_t clock,
    const struct timespec *restrict abs_timeout)

These functions all behave the same as their "timed" equivalents, but
measure the timeout against the specified clock rather than CLOCK_REALTIME.

If passed an unsupported clock, these functions indicate failure in the
same way as their "timed" equivalents and return/set errno to ENOTSUP as
required.

Support for a clock by one function does not require that the clock be
supported by any of the others.

[1] https://www.mail-archive.com/austin-group-l@opengroup.org/msg02902.html [^]
[2] https://www.mail-archive.com/austin-group-l@opengroup.org/msg02813.html [^]
[3] http://austingroupbugs.net/view.php?id=1164 [^]
[4] https://www.mail-archive.com/austin-group-l@opengroup.org/msg03034.html [^]
Desired Action The addition of the above functions, or ones that provide equivalent functionality.
Tags issue8
Attached Files

- Relationships
related to 0001164Closed Correct C++11 std::condition_variable requires a version of pthread_cond_timedwait that supports specifying the clock 
related to 0001426Applied Clarifications regarding functions that synchronize memory 

-  Notes
(0004171)
nick (manager)
2018-11-27 16:02

As a reminder, the Austin Group procedures (see https://opengroup.org/austin/docs/austin_sd6.txt) [^] state:

New Work Items


From time to time, a defect report may propose new work
items that are outside the scope of maintenance of the Austin Group
specifications. This section addresses how these are handled.

The Austin Group is not a development body for new material apart from
integration issues arising from the merger of the approved standards
that were the Base documents into the revision.

The Austin Group expects to take a similar approach for a future revision.
Thus if a defect report raises the possibility of new interfaces
for inclusion, the standard response will be that it is out of scope
for either a TC or Interpretation, and that if the new material were
to meet the some criteria it may be considered for inclusion in a
future revision subject to the agreed scope determined at that time,
although there is no guarantee.

The recommended criteria for development of new interfaces to enable
them to be considered for inclusion in a future revision are as follows:

1.There must be a written specification that has undergone a formal
consensus based approval process and is suitable for inclusion.

Parties interested in submitting new work items through one of the
three organizations within the Austin Group (The Open Group, IEEE, ISO/IEC)
should contact the appropriate Organizational Representative for further
information and advice on how each organization handles new work items.
Submissions from other organizations will also be considered.
Items 2 through 4 below apply to all submissions regardless of
origin.

2.There must be an implementation, preferably a reference implementation.

3.The specification must be "sponsored" by one of three organizations
(The Open Group, IEEE, ISO/IEC) within the Austin Group, i.e. they would
support and champion its inclusion.

4.Submitters must provide an outline plan of the editing instructions to
merge the document with the Austin Group specifications, and assistance
to the Austin Group editors as required to complete the merger.
(0004196)
mikecrowe (reporter)
2019-01-07 21:26

This issue was discussed in the 2018-11-29 teleconference[1] and I received
advice on how to proceed.

In particular, during the call we discussed making the new functions
distinguish between valid-but-unsupported clocks (such as
CLOCK_PROCESS_CPUTIME_ID) and an entirely invalid clock. After further
thought, I've realised that there are actually three separate error cases:

1. Valid, but nonsensical clock (such as CLOCK_PROCESS_CPUTIME_ID.)

2. Valid, but (currently) unsupported clock (such as CLOCK_MONOTONIC in an
   implementation that doesn't support it yet.)

3. Invalid clock ID.

Whilst implementing the new functions within glibc I realised that
returning different errors for these situations is error-prone because a
complete list of valid clocks is required for the target platform. There is
a high risk that a new clock is added, but not all the functions are
updated to return an appropriate error.

Therefore, I propose that all three cases return/set errno to EINVAL.

[1] https://www.mail-archive.com/austin-group-l@opengroup.org/msg03074.html [^]
(0004215)
mikecrowe (reporter)
2019-01-19 04:08

Here's my suggested wording for pthread_cond_clockwait (with a single hint
to the other new functions too.) It would be great to receive feedback on
whether the changes are broadly appropriate, and whether there is a better
way to describe the changes. Once I know whether I'm on the right track, I
can add the remaining functions (which will hopefully require fewer
changes.)

(I have glibc implementations of sem_clockwait, pthread_mutex_clockwait,
pthread_cond_clockwait, pthread_rwlock_clockrdlock and
pthread_rwlock_clockwrlock working. I just need to get the commit messages
into shape before posting them to the glibc list.)

Thanks.

Mike.

*** Add to lists at lines 3070, 10833, 18119, 52281, 52599, 52710, 55292, 124395, 130171:
pthread_cond_clockwait()

*** After line 10727 insert:
int pthread_cond_clockwait(pthread_cond_t *restrict,
    pthread_mutex_t *restrict, clockid_t,
    const struct timespec *restrict);

*** After line 10767 insert:
It calls pthread_cond_clockwait() with m as the mutex argument and the call returns zero or
certain error numbers (see pthread_cond_clockwait()).

*** After line 17972 insert:
It blocks in a call to pthread_cond_clockwait() with m as the mutex
argument.

*** Line 30449
Change "and pthread_cond_timedwait" to ", pthread_cond_timedwait and
pthread_cond_clockwait".

*** Lines 52513, 52515, 52519, 52582, 124488
Change "or pthread_cond_timedwait" to ", pthread_cond_timedwait or
pthread_cond_clockwait".

*** After line 52743 insert:
int pthread_cond_clockwait(pthread_cond_t *restrict cond,
    pthread_mutex_t *restrict mutex,
    clockid_t clock_id,
    const struct timespec *restrict abstime);

*** Lines 52747, 52762, 52763, 52766, 52781, 52783, 52786, 52848, 52882
Add ", pthread_cond_clockwait()" after "pthread_cond_timedwait()"

*** After line 52797 insert:
The pthread_cond_clockwait() function shall be equivalent to
pthread_cond_timedwait(), except that the absolute time specified by
abstime is measured against the clock indicated by clock_id rather than the
clock specified in the condition variable's clock attribute. All
implementations shall support passing CLOCK_REALTIME to
pthread_cond_clockwait() as the clock_id parameter. If the Monotonic Clock
option is supported, all implementations shall support a clock_id of
CLOCK_MONOTONIC.

*** Line 52822:
Change to "The pthread_cond_timedwait() and pthread_cond_clockwait()
functions shall fail if:"

*** Line 52823, 52850:
Change "pthread_cond_timedwait()" to "pthread_cond_timedwait() or
pthread_cond_clockwait()"

*** Line 52825:
Add "or the clock_id parameter passed to pthread_cond_clockwait is invalid
or not supported."

*** Insert after line 52879:
Choice of clock

Care should be taken to decide which clock is most appropriate when waiting
with a timeout. The system clock, CLOCK_REALTIME, as used by default with
pthread_cond_timedwait() may be subject to jumps forwards and backwards in
order to correct it against actual time. CLOCK_MONOTONIC is guaranteed not
to jump backwards and must also advance in real time, so using it via
pthread_cond_clockwait() or pthread_condattr_setclock() may be
more appropriate.

*** Line 52943:
Replace CLOCK_REALTIME with CLOCK_MONOTONIC

*** Line 52947:
Replace line with:
 rc = pthread_cond_clockwait(&t.cond, &t.mn, CLOCK_MONOTONIC, &ts);

*** After line line 52956 insert:
Using CLOCK_MONOTONIC rather than CLOCK_REALTIME means that the timeout is
not influenced by the system clock being changed.

*** After line 53084 insert:
Note that the /clock/ attribute shall have no effect on the
pthread_cond_clockwait() function.

*** Lines 54147, 54186, 54321, 55282:
After "pthread_cond_timedwait()" insert ", pthread_cond_clockwait()".

*** Line 123502:
Delete remainder of paragraph starting with "This capability has not been
added to other functions...". Insert in its place:

For condition variables, this capability is also available by passing
CLOCK_MONOTONIC to the pthread_cond_clockwait() function. Similarly,
CLOCK_MONOTONIC can be specified when calling sem_clockwait,
pthread_mutex_clocklock, pthread_rwlock_clockrdlock and
pthread_rwlock_clockwrlock.

*** After line 123535:
Subsequently, it was discovered that using a particular clock for the
timeout was desired by some users. In particular, the ISO C++11 standard
requires the clock to be a property of the wait for its
std::condition_variable and std::timed_mutex. This led to the addition of
sem_clockwait(), pthread_mutex_clockwait(), pthread_cond_clockwait(),
pthread_rwlock_clockrdlock() and pthread_rwlock_clockwrlock() functions
that accepted a clock parameter to measure their absolute timeout against.
The addition of the corresponding mq_clockreceive and mq_clocksend may be
considered in the future.
(0004276)
mikecrowe (reporter)
2019-03-06 13:17

I've posted patches for pthread_cond_clockwait, sem_clockwait, pthread_rwlock_clockrdlock and pthread_rwlock_clockwrlock to the glibc mailing list. I received only one reply for a straightforward-to-fix problem.

https://sourceware.org/ml/libc-alpha/2019-02/msg00637.html [^]

pthread_mutex_clocklock requires more work.
(0004387)
mikecrowe (reporter)
2019-05-07 12:09

Suggested wording updates for all new functions:

** General
*** Insert after p3622:123506
It was later found necessary to add variants of almost all interfaces that
accept absolute timeouts that allow the clock to be specified. This is
because, despite the claim in the previous paragraph, it is not possible to
safely use a CLOCK_REALTIME absolute timeout even to prevent errors when
the system clock is warped by a potentially large amount. A "safety
timeout" of a minute on a call pthread_mutex_timedlock() could actually
mean no timeout at all if the system clock is warped forwards immediately
prior to the call. This functionality is provided by the new "clockwait"
variants of the existing "timedwait" functions that accept an extra
clockid_t parameter to indicate the clock to be used for the wait. The
clock parameter is passed rather than using attributes as previously for
pthread_cond_timedwait in order to allow ISOXXX C++11 and later to be
implemented correctly. C++ requires that the clock to use for the wait is
not known until the time of the wait call, so it cannot be supplied during
creation. The new functions are sem_clockwait(), pthread_cond_clockwait(),
pthread_mutex_clocklock(), pthread_mutex_clockrdlock() and
pthread_mutex_clockwrlock(). It is expected that mq_clockreceive() and
mq_clocksend() functions will be added in a future version of this standard
too.
** pthread_cond_clockwait
*** Add to lists at lines
- p111:3070
- p318:10833
- p518:18119
- p1604:52281
- p1612:52599
- p1615:52710
- p1695:55292
- p3641:124395
- p3793:130171
pthread_cond_clockwait()
*** After line p316:10727 insert:
int pthread_cond_clockwait(pthread_cond_t *restrict,
    pthread_mutex_t *restrict, clockid_t,
    const struct timespec *restrict);
*** After line p514:17967 insert:
It calls pthread_cond_clockwait() with m as the mutex argument and the call returns zero or
certain error numbers (see pthread_cond_clockwait()).
*** After line p514:17972 insert:
It blocks in a call to pthread_cond_clockwait() with m as the mutex
argument.
*** Line p900:30449
Change "and pthread_cond_timedwait" to ", pthread_cond_timedwait and
pthread_cond_clockwait".
*** Lines p1610:52513,52515,52519, p1611:52582, p3643:124488
Change "or pthread_cond_timedwait" to ", pthread_cond_timedwait or
pthread_cond_clockwait".
*** After line p1617:52743 insert:
int pthread_cond_clockwait(pthread_cond_t *restrict cond,
    pthread_mutex_t *restrict mutex,
    clockid_t clock_id,
    const struct timespec *restrict abstime);
*** Lines p1617:52747,52762,52763,52766,52781,52783, p1618:52786, p1619:52848, p1620:52882
Add ", pthread_cond_clockwait()" after "pthread_cond_timedwait()"

*** After line p1618:52797 insert:
The pthread_cond_clockwait() function shall be equivalent to
pthread_cond_timedwait(), except that the absolute time specified by
abstime is measured against the clock indicated by clock_id rather than the
clock specified in the condition variable's clock attribute. All
implementations shall support passing CLOCK_REALTIME to
pthread_cond_clockwait() as the clock_id parameter. If the Monotonic Clock
option is supported, all implementations shall support a clock_id of
CLOCK_MONOTONIC.
*** Line p1618:52822:
Change to "The pthread_cond_timedwait() and pthread_cond_clockwait()
functions shall fail if:"

*** Line p1618:52823, p1619:52850:
Change "pthread_cond_timedwait()" to "pthread_cond_timedwait() or
pthread_cond_clockwait()"

*** Line p1618:52825:
Add "or the clock_id parameter passed to pthread_cond_clockwait is invalid
or not supported."

*** Insert after line p1620:52879:
Choice of clock

Care should be taken to decide which clock is most appropriate when waiting
with a timeout. The system clock, CLOCK_REALTIME, as used by default with
pthread_cond_timedwait() may be subject to jumps forwards and backwards in
order to correct it against actual time. CLOCK_MONOTONIC is guaranteed not
to jump backwards and must also advance in real time, so using it via
pthread_cond_clockwait() or pthread_condattr_setclock() may be
more appropriate.

*** Line p1621:52943:
Replace CLOCK_REALTIME with CLOCK_MONOTONIC
*** Line p1621:52947:
Replace line with:
 rc = pthread_cond_clockwait(&t.cond, &t.mn, CLOCK_MONOTONIC, &ts);
*** After line line p1622:52956 insert:
Using CLOCK_MONOTONIC rather than CLOCK_REALTIME means that the timeout is
not influenced by the system clock being changed.
*** After line p1626:53084 insert:
Note that the /clock/ attribute shall have no effect on the
pthread_cond_clockwait() function.
*** Lines p1661:54147, p1662:54186, p1665:54321, p1694:55282:
After "pthread_cond_timedwait()" insert ", pthread_cond_clockwait()".
*** Line p3621:123502:
Delete remainder of paragraph starting with "This capability has not been
added to other functions...". Insert in its place:

For condition variables, this capability is also available by passing
CLOCK_MONOTONIC to the pthread_cond_clockwait() function. Similarly,
CLOCK_MONOTONIC can be specified when calling sem_clockwait,
pthread_mutex_clocklock, pthread_rwlock_clockrdlock and
pthread_rwlock_clockwrlock.

*** After line p3622:123535:
Subsequently, it was discovered that using a particular clock for the
timeout was desired by some users. In particular, the ISO C++11 standard
requires the clock to be a property of the wait for its
std::condtion_variable and std::timed_mutex. This led to the addition of
sem_clockwait(), pthread_mutex_clockwait(), pthread_cond_clockwait(),
pthread_rwlock_clockrdlock() and pthread_rwlock_clockwrlock() functions
that accepted a clock parameter to measure their absolute timeout against.
The addition of the corresponding mq_clockreceive and mq_clocksend may be
considered in the future.
** sem_clockwait
*** Add to list at p111:3069
sem_clockwait
*** After line p329:11164 insert
int sem_clockwait(sem_t *restrict, clockid_t clock_id, const struct
                  timespec *restrict)
*** Add to list at p427:14548
sem_clockwait
*** Add to list at p517:18120
sem_clockwait
*** Add to lists at p681:23337,23362
sem_clockwait
*** Add to lists at p1849:59892:
sem_clockwait
*** Add to lists at p1851:59918,59922,59924,59946
sem_clockwait
*** Add to list at p1853:59972
sem_clockwait
*** Add to list at p1855:60057
sem_clockwait
*** Add to lists at p1856:60113,60121
sem_clockwait
*** p1858:60140
Add sem_clockwait to NAME
*** After p1858:60145 insert
int sem_clockwait(sem_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict abstime)
*** After p1858:60147 "sem_timedwait", insert
and sem_clockwait
*** Before p1858:60155 insert:
For sem_timedwait,
*** On p1858:60156 insert new sentence after sentence end:
For sem_clockwait, the timeout shall be measured against the clock
represented by the clock_id parameter.
*** p1858:60162, p1858:60167 replace "function" with
and sem_clockwait() functions
*** Append to p1858:p60170
Alternatively the sem_clockwait() function was passed an invalid or
unsupported clock_id.
*** p1860:60230
Replace "CLOCK_REALTIME" with "CLOCK_MONOTONIC".
*** p1860:60235
Replace "sem_timedwait" with "sem_clockwait".
*** p1860:60236
Replace "sem_timedwait(&sem, &ts)" with "sem_clockwait(&sem,
CLOCK_MONOTONIC, &ts)"
*** p1860:60241,60243,60245:
Replace "sem_timedwait" with "sem_clockwait".
*** Add to list p1862:60307
sem_clockwait
*** Add to list p3563:120877
sem_clockwait
*** Add to list p3770:129389,129400
sem_clockwait
*** Add to list p3792:130143
sem_clockwait
** pthread_mutex_clocklock
*** Add to list p111:3075
pthread_mutex_clocklock
*** Insert after p317:10761
int pthread_mutex_clocklock(pthread_mutex_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Add to list p318:10838
pthread_mutex_clocklock
*** Add to list p427:14547:
pthread_mutex_clocklock
*** Insert after p514:17962:
It calls pthread_mutex_clocklock( ) with m as the mutex argument and the
call returns zero or [EOWNERDEAD].
*** Add to list p681:23337:
pthread_mutex_clocklock
*** Add to list p1665:54333:
pthread_mutex_clocklock
*** Add to list p1668:54417:
pthread_mutex_clocklock
*** Add to list p1673:54574:
pthread_mutex_clocklock
*** In line p1676:54616, insert
pthread_mutex_clocklock
*** After line p1676:54621, insert
int pthread_mutex_clocklock(pthread_mutex_t *restrict mutex,
    clockid_t clock_id,
    const struct timespec *restrict abstime);
*** Begin line p1676:54623 with
The pthread_mutex_timedlock() and pthread_mutex_clocklock() functions shall
lock...
*** Begin line p1676:54631 with
For pthread_mutex_timedlock, the timeout shall...
*** On line p1676:54632, insert new sentence after sentence end:
For pthread_mutex_clocklock, the timeout shall be measured against the
clock represented by the clock_id parameter.
*** On line p1676:54642 and p1676:54644 replace "pthread_mutex_timedlock()" with
pthread_mutex_timedlock() or pthread_mutex_clocklock()
*** On line p1676:54653 and p1676:54656 replace "function" with
and pthread_mutex_clocklock() functions
*** Insert after p1677:54666
Alternatively the pthread_mutex_clocklock() function was passed an invalid
or unsupported clock_id.
*** On line p1677:54675 replace "function" with
and pthread_mutex_clocklock() functions
*** Add to list at p3563:120888
pthread_mutex_clocklock
*** Add to list at p3770:129399
pthread_mutex_clocklock
*** Add to list at p3793:130175
pthread_mutex_clocklock
** pthread_rwlock_clockrdlock & pthread_rwlock_clockwrlock
*** Add to list at p111:3072
pthread_rwlock_clockwrlock, pthread_rwlock_clockrdlock
*** Insert after line p317:10787
int pthread_rwlock_clockrdlock(pthread_rwlock_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Insert after line p317:10789
int pthread_rwlock_clockwrlock(pthread_rwlock_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Add to list at p318:10841
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p319:10842
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p427:14548
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p519:18175
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p1705:55499
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p1708:55600
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to line p1710:55625
pthread_rwlock_clockrdlock
*** Insert after line p1710:55630
int pthread_rwlock_clockrdlock(pthread_rwlock_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Replace p1710:55632 "function" with
and pthread_rwlock_clockrdlock() functions
*** Insert at beginning of p1710:55639
For pthread_rwlock_timedrdlock(),
*** On line p1710:55640, insert new sentence after sentence end
For pthread_rwlock_clockrdclock(), the timeout shall be measured against
the clock represented by the clock_id parameter.
*** On line p1710:55645, after "pthread_rwlock_timedrdlock" insert
or pthread_rwlock_clockrdlock
*** On line p1710:55650, replace "function" with
and pthread_rwlock_clockrdlock functions
*** On line p1710:55654, replace "function" with
and pthread_rwlock_clockrdlock functions
*** On line p1710:55656, replace "function" with
and pthread_rwlock_clockrdlock functions
*** On line p1710:55661 add
Alternatively, the pthread_rwlock_clockrdlock() function was passed an
invalid or unsupported clock_id.
*** On line p1711:55671, after "pthread_rwlock_timedrdlock()" add
or pthread_rwlock_clockrdlock()
*** Add to list at p1711:55676:
pthread_rwlock_clockwrlock()
*** Add to line p1712:55688
pthread_rwlock_clockwrlock
*** Insert after line p1712:55693
int pthread_rwlock_clockwrlock(pthread_rwlock_t *restrict rwlock,
    clockid_t clock_id,
    const struct timespec *restrict abstime);
*** Replace p1712:55695 "function" with
and pthread_rwlock_clockwrlock() functions
*** Insert at beginning of p1712:55702
For pthread_rwlock_timedrdlock,
*** On line p1712:55703, insert new sentence after sentence end
For pthread_rwlock_clockwrclock(), the timeout shall be measured against
the clock represented by the clock_id parameter.
*** On line p1712:55708, after "pthread_rwlock_timedwrlock" insert
or pthread_rwlock_clockwrlock
*** On line p1712:55713 replace function with
and pthread_rwlock_clockwrlock functions
*** On line p1712:55717 replace function with
and pthread_rwlock_clockwrlock functions

*** On line p1712:55719 replace function with
and pthread_rwlock_clockwrlock functions

*** On line p1712:55723 add
Alternatively, the pthread_rwlock_clockwrlock() function was passed an
invalid or unsupported clock_id.
*** On line p1713:55732, after "pthread_rwlock_timedwrlock()" add
or pthread_rwlock_clockwrlock()
*** Add to line p1713:55737
pthread_rwlock_clockrdlock
*** Add to list on p1716:55801
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p1718:55863
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p3563:120899
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p3770:129399
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p3792:130137
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
(0004391)
enh (reporter)
2019-05-09 18:03

tomcherry from Android's C/C++ team has implemented this for Android too (https://android-review.googlesource.com/c/platform/bionic/+/958058) [^] and
we're happy to submit that as soon this gets accepted into POSIX.

He had two quick comments about the wording (but can't log in for some reason):

The first is on the phrasing of 'A "safety timeout" of a minute on a
call pthread_mutex_timedlock() could actually mean no timeout at all
if the system clock is warped forwards immediately prior to the call.'

I'm used to the term 'no timeout' meaning that the call will block
indefinitely, but in this case, pthread_mutex_timedlock() will
immediately return, right? Would it be more clear if 'could actually
mean no timeout at all' were replaced with 'would return immediately
regardless of if the lock was acquired' ?

The second is that it may be worth commenting on the other situation,
where time is warped backwards immediately prior to the call and the
time out happens much later than originally expected. It's actually
that issue that has caused more issues for Android than the former.
(0004478)
mikecrowe (reporter)
2019-07-09 17:31

Updated wording hopefully addressing Tom Cherry's comments and improving the description of EINVAL returns.

** General
*** Insert after p3622:123506
It was later found necessary to add variants of almost all interfaces that
accept absolute timeouts that allow the clock to be specified. This is
because, despite the claim in the previous paragraph, it is not possible to
safely use a CLOCK_REALTIME absolute timeout even to prevent errors when
the system clock is warped by a potentially large amount. A "safety
timeout" of a minute on a call pthread_mutex_timedlock() could actually
mean that the call would return ETIMEDOUT early without acquiring the lock
if the system clock is warped forwards immediately prior to or during the
call. On the other hand, a short timeout could end up being arbitrarily
long if the system clock is warped backwards immediately prior to or during
the call. These problems are solved by the new "clockwait" variants of the
existing "timedwait" functions. These variants accept an extra clockid_t
parameter to indicate the clock to be used for the wait. The clock
parameter is passed rather than using attributes as previously for
pthread_cond_timedwait in order to allow ISO/IEC 14882:2011 (C++11) and
later to be implemented correctly. C++ requires that the clock to use for
the wait is not known until the time of the wait call, so it cannot be
supplied during creation. The new functions are sem_clockwait(),
pthread_cond_clockwait(), pthread_mutex_clocklock(),
pthread_mutex_clockrdlock() and pthread_mutex_clockwrlock(). It is expected
that mq_clockreceive() and mq_clocksend() functions will be added in a
future version of this standard too.
** pthread_cond_clockwait
*** Add to lists at lines
- p111:3070
- p318:10833
- p518:18119
- p1604:52281
- p1612:52599
- p1615:52710
- p1695:55292
- p3641:124395
- p3793:130171
pthread_cond_clockwait()
*** After line p316:10727 insert:
int pthread_cond_clockwait(pthread_cond_t *restrict,
    pthread_mutex_t *restrict, clockid_t,
    const struct timespec *restrict);
*** After line p514:17967 insert:
It calls pthread_cond_clockwait() with m as the mutex argument and the call returns zero or
certain error numbers (see pthread_cond_clockwait()).
*** After line p514:17972 insert:
It blocks in a call to pthread_cond_clockwait() with m as the mutex
argument.
*** Line p900:30449
Change "and pthread_cond_timedwait" to ", pthread_cond_timedwait and
pthread_cond_clockwait".
*** Lines p1610:52513,52515,52519, p1611:52582, p3643:124488
Change "or pthread_cond_timedwait" to ", pthread_cond_timedwait or
pthread_cond_clockwait".
*** After line p1617:52743 insert:
int pthread_cond_clockwait(pthread_cond_t *restrict cond,
    pthread_mutex_t *restrict mutex,
    clockid_t clock_id,
    const struct timespec *restrict abstime);
*** Lines p1617:52747,52762,52763,52766,52781,52783, p1618:52786, p1619:52848, p1620:52882
Add ", pthread_cond_clockwait()" after "pthread_cond_timedwait()"

*** After line p1618:52797 insert:
The pthread_cond_clockwait() function shall be equivalent to
pthread_cond_timedwait(), except that the absolute time specified by
abstime is measured against the clock indicated by clock_id rather than the
clock specified in the condition variable's clock attribute. All
implementations shall support passing CLOCK_REALTIME to
pthread_cond_clockwait() as the clock_id parameter. If the Monotonic Clock
option is supported, all implementations shall support a clock_id of
CLOCK_MONOTONIC.
*** Line p1618:52822:
Change to "The pthread_cond_timedwait() and pthread_cond_clockwait()
functions shall fail if:"

*** Line p1618:52823, p1619:52850:
Change "pthread_cond_timedwait()" to "pthread_cond_timedwait() or
pthread_cond_clockwait()"

*** Line p1618:52825:
Add "or the clock_id parameter passed to pthread_cond_clockwait is invalid
or not supported."

*** Insert after line p1620:52879:
Choice of clock

Care should be taken to decide which clock is most appropriate when waiting
with a timeout. The system clock, CLOCK_REALTIME, as used by default with
pthread_cond_timedwait() may be subject to jumps forwards and backwards in
order to correct it against actual time. CLOCK_MONOTONIC is guaranteed not
to jump backwards and must also advance in real time, so using it via
pthread_cond_clockwait() or pthread_condattr_setclock() may be
more appropriate.

*** Line p1621:52943:
Replace CLOCK_REALTIME with CLOCK_MONOTONIC
*** Line p1621:52947:
Replace line with:
 rc = pthread_cond_clockwait(&t.cond, &t.mn, CLOCK_MONOTONIC, &ts);
*** After line line p1622:52956 insert:
Using CLOCK_MONOTONIC rather than CLOCK_REALTIME means that the timeout is
not influenced by the system clock being changed.
*** After line p1626:53084 insert:
Note that the /clock/ attribute shall have no effect on the
pthread_cond_clockwait() function.
*** Lines p1661:54147, p1662:54186, p1665:54321, p1694:55282:
After "pthread_cond_timedwait()" insert ", pthread_cond_clockwait()".
*** Line p3621:123502:
Delete remainder of paragraph starting with "This capability has not been
added to other functions...". Insert in its place:

For condition variables, this capability is also available by passing
CLOCK_MONOTONIC to the pthread_cond_clockwait() function. Similarly,
CLOCK_MONOTONIC can be specified when calling sem_clockwait,
pthread_mutex_clocklock, pthread_rwlock_clockrdlock and
pthread_rwlock_clockwrlock.

*** After line p3622:123535:
Subsequently, it was discovered that using a particular clock for the
timeout was desired by some users. In particular, the ISO C++11 standard
requires the clock to be a property of the wait for its
std::condtion_variable and std::timed_mutex. This led to the addition of
sem_clockwait(), pthread_mutex_clockwait(), pthread_cond_clockwait(),
pthread_rwlock_clockrdlock() and pthread_rwlock_clockwrlock() functions
that accepted a clock parameter to measure their absolute timeout against.
The addition of the corresponding mq_clockreceive and mq_clocksend may be
considered in the future.
** sem_clockwait
*** Add to list at p111:3069
sem_clockwait
*** After line p329:11164 insert
int sem_clockwait(sem_t *restrict, clockid_t clock_id, const struct
                  timespec *restrict)
*** Add to list at p427:14548
sem_clockwait
*** Add to list at p517:18120
sem_clockwait
*** Add to lists at p681:23337,23362
sem_clockwait
*** Add to lists at p1849:59892:
sem_clockwait
*** Add to lists at p1851:59918,59922,59924,59946
sem_clockwait
*** Add to list at p1853:59972
sem_clockwait
*** Add to list at p1855:60057
sem_clockwait
*** Add to lists at p1856:60113,60121
sem_clockwait
*** p1858:60140
Add sem_clockwait to NAME
*** After p1858:60145 insert
int sem_clockwait(sem_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict abstime)
*** After p1858:60147 "sem_timedwait", insert
and sem_clockwait
*** Before p1858:60155 insert:
For sem_timedwait,
*** On p1858:60156 insert new sentence after sentence end:
For sem_clockwait, the timeout shall be measured against the clock
represented by the clock_id parameter.
*** p1858:60162, p1858:60167 replace "function" with
and sem_clockwait() functions
*** Append to p1858:p60170
Alternatively the sem_clockwait() function was passed an invalid or
unsupported clock_id.
*** p1860:60230
Replace "CLOCK_REALTIME" with "CLOCK_MONOTONIC".
*** p1860:60235
Replace "sem_timedwait" with "sem_clockwait".
*** p1860:60236
Replace "sem_timedwait(&sem, &ts)" with "sem_clockwait(&sem,
CLOCK_MONOTONIC, &ts)"
*** p1860:60241,60243,60245:
Replace "sem_timedwait" with "sem_clockwait".
*** Add to list p1862:60307
sem_clockwait
*** Add to list p3563:120877
sem_clockwait
*** Add to list p3770:129389,129400
sem_clockwait
*** Add to list p3792:130143
sem_clockwait
** pthread_mutex_clocklock
*** Add to list p111:3075
pthread_mutex_clocklock
*** Insert after p317:10761
int pthread_mutex_clocklock(pthread_mutex_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Add to list p318:10838
pthread_mutex_clocklock
*** Add to list p427:14547:
pthread_mutex_clocklock
*** Insert after p514:17962:
It calls pthread_mutex_clocklock( ) with m as the mutex argument and the
call returns zero or [EOWNERDEAD].
*** Add to list p681:23337:
pthread_mutex_clocklock
*** Add to list p1665:54333:
pthread_mutex_clocklock
*** Add to list p1668:54417:
pthread_mutex_clocklock
*** Add to list p1673:54574:
pthread_mutex_clocklock
*** In line p1676:54616, insert
pthread_mutex_clocklock
*** After line p1676:54621, insert
int pthread_mutex_clocklock(pthread_mutex_t *restrict mutex,
    clockid_t clock_id,
    const struct timespec *restrict abstime);
*** Begin line p1676:54623 with
The pthread_mutex_timedlock() and pthread_mutex_clocklock() functions shall
lock...
*** Begin line p1676:54631 with
For pthread_mutex_timedlock, the timeout shall...
*** On line p1676:54632, insert new sentence after sentence end:
For pthread_mutex_clocklock, the timeout shall be measured against the
clock represented by the clock_id parameter.
*** On line p1676:54642 and p1676:54644 replace "pthread_mutex_timedlock()" with
pthread_mutex_timedlock() or pthread_mutex_clocklock()
*** On line p1676:54653 and p1676:54656 replace "function" with
and pthread_mutex_clocklock() functions
*** Insert after p1677:54666
Alternatively the pthread_mutex_clocklock() function was passed an invalid
or unsupported clock_id.
*** On line p1677:54675 replace "function" with
and pthread_mutex_clocklock() functions
*** Add to list at p3563:120888
pthread_mutex_clocklock
*** Add to list at p3770:129399
pthread_mutex_clocklock
*** Add to list at p3793:130175
pthread_mutex_clocklock
** pthread_rwlock_clockrdlock & pthread_rwlock_clockwrlock
*** Add to list at p111:3072
pthread_rwlock_clockwrlock, pthread_rwlock_clockrdlock
*** Insert after line p317:10787
int pthread_rwlock_clockrdlock(pthread_rwlock_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Insert after line p317:10789
int pthread_rwlock_clockwrlock(pthread_rwlock_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Add to list at p318:10841
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p319:10842
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p427:14548
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p519:18175
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p1705:55499
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list at p1708:55600
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to line p1710:55625
pthread_rwlock_clockrdlock
*** Insert after line p1710:55630
int pthread_rwlock_clockrdlock(pthread_rwlock_t *restrict,
    clockid_t clock_id,
    const struct timespec *restrict);
*** Replace p1710:55632 "function" with
and pthread_rwlock_clockrdlock() functions
*** Insert at beginning of p1710:55639
For pthread_rwlock_timedrdlock(),
*** On line p1710:55640, insert new sentence after sentence end
For pthread_rwlock_clockrdclock(), the timeout shall be measured against
the clock represented by the clock_id parameter.
*** On line p1710:55645, after "pthread_rwlock_timedrdlock" insert
or pthread_rwlock_clockrdlock
*** On line p1710:55650, replace "function" with
and pthread_rwlock_clockrdlock functions
*** On line p1710:55654, replace "function" with
and pthread_rwlock_clockrdlock functions
*** On line p1710:55656, replace "function" with
and pthread_rwlock_clockrdlock functions
*** On line p1710:55661 add
Alternatively, the pthread_rwlock_clockrdlock() function was passed an
invalid or unsupported clock_id.
*** On line p1711:55671, after "pthread_rwlock_timedrdlock()" add
or pthread_rwlock_clockrdlock()
*** Add to list at p1711:55676:
pthread_rwlock_clockwrlock()
*** Add to line p1712:55688
pthread_rwlock_clockwrlock
*** Insert after line p1712:55693
int pthread_rwlock_clockwrlock(pthread_rwlock_t *restrict rwlock,
    clockid_t clock_id,
    const struct timespec *restrict abstime);
*** Replace p1712:55695 "function" with
and pthread_rwlock_clockwrlock() functions
*** Insert at beginning of p1712:55702
For pthread_rwlock_timedrdlock,
*** On line p1712:55703, insert new sentence after sentence end
For pthread_rwlock_clockwrclock(), the timeout shall be measured against
the clock represented by the clock_id parameter.
*** On line p1712:55708, after "pthread_rwlock_timedwrlock" insert
or pthread_rwlock_clockwrlock
*** On line p1712:55713 replace function with
and pthread_rwlock_clockwrlock functions
*** On line p1712:55717 replace function with
and pthread_rwlock_clockwrlock functions

*** On line p1712:55719 replace function with
and pthread_rwlock_clockwrlock functions

*** On line p1712:55723 add
Alternatively, the pthread_rwlock_clockwrlock() function was passed an
invalid or unsupported clock_id.
*** On line p1713:55732, after "pthread_rwlock_timedwrlock()" add
or pthread_rwlock_clockwrlock()
*** Add to line p1713:55737
pthread_rwlock_clockrdlock
*** Add to list on p1716:55801
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p1718:55863
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p3563:120899
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p3770:129399
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
*** Add to list on p3792:130137
pthread_rwlock_clockwrlock(), pthread_rwlock_clockrdlock()
(0004481)
mikecrowe (reporter)
2019-07-14 18:08

Implementations of these functions have landed for glibc 2.30. See https://sourceware.org/git/?p=glibc.git;a=log;h=65dd7e9ce36ca8485a68ed41ac731188b6473dd2 [^]
(0004485)
enh (reporter)
2019-07-16 02:46

great to hear! we'd just been talking about whether to merge the bionic change first, but now we can just be followers :-)

this will be in Android R. see https://android-review.googlesource.com/c/platform/bionic/+/958058. [^]
(0005057)
geoffclare (manager)
2020-10-21 15:13

The *_clockwait() and *_clocklock() additions have been made in the Issue8NewAPIs branch in gitlab, based on Note: 0004478.

Some of the suggested changes were to SEE ALSO sections and other cross-references, which were not needed as those are references to pages, not individual functions. There were also some that would have updated old change history.

Finally, some pointer pages are needed but I have deferred adding them as I think it would be better to rename the *_timed*() pages to *_clock*() and add pointer pages for the *_timed*() functions. This could be handled by making the eventual resolution of this bug something like, "Make the changes from The Open Group document Cxxx and also rename the affected *_timed*() pages to *_clock*() and add pointer pages for the *_timed*() functions."
(0005122)
geoffclare (manager)
2020-11-26 10:57

The changes to add new functions in XBD 4.12 Memory Synchronization overlap with bug 0001426. This should be resolved by applying 1426 first and then, instead of just adding the new functions to the table, treat each *_clock*() function the same as the corresponding *_timed*() function in the updated XBD 4.12.
(0005339)
geoffclare (manager)
2021-04-29 15:40

Make the changes from "Additional APIs for Issue 8, Part 1" (Austin/1110). Resolve the overlap with bug 1426 by applying 1426 first and then, instead of just adding the new functions to the table, treat each *_clock*() function the same as the corresponding *_timed*() function in the updated XBD 4.12. In addition, rename the *_timed*() pages to *_clock*() and add pointer pages for the *_timed*() functions.

- Issue History
Date Modified Username Field Change
2018-11-26 18:53 mikecrowe New Issue
2018-11-26 18:53 mikecrowe Status New => Under Review
2018-11-26 18:53 mikecrowe Assigned To => ajosey
2018-11-26 18:53 mikecrowe Name => Mike Crowe
2018-11-26 18:53 mikecrowe Section => pthread
2018-11-26 18:53 mikecrowe Page Number => 0
2018-11-26 18:53 mikecrowe Line Number => 0
2018-11-27 09:23 geoffclare Project 1003.1(2008)/Issue 7 => 1003.1(2016/18)/Issue7+TC2
2018-11-27 09:23 geoffclare Relationship added related to 0001164
2018-11-27 16:02 nick Note Added: 0004171
2019-01-07 21:26 mikecrowe Note Added: 0004196
2019-01-19 04:08 mikecrowe Note Added: 0004215
2019-03-06 13:17 mikecrowe Note Added: 0004276
2019-05-07 12:09 mikecrowe Note Added: 0004387
2019-05-09 18:03 enh Note Added: 0004391
2019-07-09 17:31 mikecrowe Note Added: 0004478
2019-07-14 18:08 mikecrowe Note Added: 0004481
2019-07-16 02:46 enh Note Added: 0004485
2020-10-21 15:13 geoffclare Note Added: 0005057
2020-11-26 10:52 geoffclare Relationship added related to 0001426
2020-11-26 10:57 geoffclare Note Added: 0005122
2021-04-29 15:40 geoffclare Note Added: 0005339
2021-04-29 15:41 geoffclare Interp Status => ---
2021-04-29 15:41 geoffclare Final Accepted Text => Note: 0005339
2021-04-29 15:41 geoffclare Status Under Review => Resolved
2021-04-29 15:41 geoffclare Resolution Open => Accepted As Marked
2021-04-29 15:41 geoffclare Tag Attached: issue8
2021-05-10 13:26 geoffclare Status Resolved => Applied


Mantis 1.1.6[^]
Copyright © 2000 - 2008 Mantis Group
Powered by Mantis Bugtracker