[
https://issues.apache.org/jira/browse/DISPATCH-1956?page=com.atlassian.jira.plugin.system.issuetabpanels:comment-tabpanel&focusedCommentId=17357509#comment-17357509
]
ASF GitHub Bot commented on DISPATCH-1956:
------------------------------------------
mgoulish opened a new pull request #1243:
URL: https://github.com/apache/qpid-dispatch/pull/1243
I noticed that almost all calls to qd_log_source_lh() were very simple:
take the lock, call the function, release the lock.
But that means that we could move the lock take-and-release inside of that
function, with no change in behavior.
And then that means I don't have to ask the caller to add my new log_source
(if it is new) to the list -- I can do that myself from within the function and
simplify everything -- but do it after I drop the log_source_lock, so we don't
get the lock inversion.
SO I did that, and removed the _lh from the end of the function name -- but
that made it collide with the name of another function. So I changed it to
qd_log_source_find_or_make().
I.e. "if you don't find one of this name, then make it."
The only slight disturbance was the function qd_log_entity() which took the
lock and used it to cover this function and a couple other calls to a different
_lh() function. SO for that one I just inserted new take-and-releases of
the lock, right around the other function calls. Now there is only one line
of code between each take-and-release, which seems much nicer.
Also qd_log_source_reset() bugged me because it called my new function but
then immediately had to take the lock just so it could set the log_source to
default values. SO I just added a flag to my new fn so that
qd_log_source_reset() could tell me to do that for him right after I retrieve
it.
--
This is an automated message from the Apache Git Service.
To respond to the message, please log on to GitHub and use the
URL above to go to the specific comment.
For queries about this service, please contact Infrastructure at:
[email protected]
> Potential deadlock: logging lock vs entity cache lock
> -----------------------------------------------------
>
> Key: DISPATCH-1956
> URL: https://issues.apache.org/jira/browse/DISPATCH-1956
> Project: Qpid Dispatch
> Issue Type: Bug
> Components: Router Node
> Affects Versions: 1.15.0
> Reporter: Ken Giusti
> Assignee: Michael Goulish
> Priority: Major
> Labels: deadlock, tsan
> Fix For: 1.17.0
>
> Attachments: tsan.supp
>
>
> {noformat}
> WARNING: ThreadSanitizer: lock-order-inversion (potential deadlock)
> (pid=1474955)
> Cycle in lock order graph: M11 (0x7b10000002c0) => M9 (0x7b1000000240) =>
> M11
>
> Mutex M9 acquired here while holding mutex M11 in main thread:
> #0 pthread_mutex_lock <null> (libtsan.so.0+0x528ac)
> #1 sys_mutex_lock
> /home/kgiusti/work/dispatch/qpid-dispatch/src/posix/threading.c:57
> (libqpid-dispatch.so+0x8cb7d)
> #2 push_event
> /home/kgiusti/work/dispatch/qpid-dispatch/src/entity_cache.c:63
> (libqpid-dispatch.so+0x6fa13)
> #3 qd_entity_cache_add
> /home/kgiusti/work/dispatch/qpid-dispatch/src/entity_cache.c:69
> (libqpid-dispatch.so+0x6fc26)
> #4 qd_alloc_init
> /home/kgiusti/work/dispatch/qpid-dispatch/src/alloc_pool.c:302
> (libqpid-dispatch.so+0x5878b)
> #5 qd_alloc /home/kgiusti/work/dispatch/qpid-dispatch/src/alloc_pool.c:318
> (libqpid-dispatch.so+0x5878b)
> #6 new_qd_log_entry_t /home/kgiusti/work/dispatch/qpid-dispatch/src/log.c:61
> (libqpid-dispatch.so+0x75891)
> #7 qd_vlog_impl /home/kgiusti/work/dispatch/qpid-dispatch/src/log.c:426
> (libqpid-dispatch.so+0x76205)
> #8 qd_log_impl /home/kgiusti/work/dispatch/qpid-dispatch/src/log.c:453
> (libqpid-dispatch.so+0x76580)
> #9 qd_python_log
> /home/kgiusti/work/dispatch/qpid-dispatch/src/python_embedded.c:547
> (libqpid-dispatch.so+0x8d1cb)
> #10 <null> <null> (libpython3.8.so.1.0+0x12a23b)
> #11 main_process
> /home/kgiusti/work/dispatch/qpid-dispatch/router/src/main.c:95
> (qdrouterd+0x40281c)
> #12 main /home/kgiusti/work/dispatch/qpid-dispatch/router/src/main.c:367
> (qdrouterd+0x4024fc)
>
> Hint: use TSAN_OPTIONS=second_deadlock_stack=1 to get more informative
> warning message
>
> Mutex M11 acquired here while holding mutex M9 in main thread:
> #0 pthread_mutex_lock <null> (libtsan.so.0+0x528ac)
> #1 sys_mutex_lock
> /home/kgiusti/work/dispatch/qpid-dispatch/src/posix/threading.c:57
> (libqpid-dispatch.so+0x8cb7d)
> #2 qd_vlog_impl /home/kgiusti/work/dispatch/qpid-dispatch/src/log.c:425
> (libqpid-dispatch.so+0x76200)
> #3 qd_log_impl /home/kgiusti/work/dispatch/qpid-dispatch/src/log.c:453
> (libqpid-dispatch.so+0x76580)
> #4 qd_python_log
> /home/kgiusti/work/dispatch/qpid-dispatch/src/python_embedded.c:547
> (libqpid-dispatch.so+0x8d1cb)
> #5 <null> <null> (libpython3.8.so.1.0+0x12a23b)
> #6 main_process
> /home/kgiusti/work/dispatch/qpid-dispatch/router/src/main.c:95
> (qdrouterd+0x40281c)
> #7 main /home/kgiusti/work/dispatch/qpid-dispatch/router/src/main.c:367
> (qdrouterd+0x4024fc)
>
> SUMMARY: ThreadSanitizer: lock-order-inversion (potential deadlock)
> (/lib64/libtsan.so.0+0x528ac) in __interceptor_pthread_mutex_lock
> {noformat}
--
This message was sent by Atlassian Jira
(v8.3.4#803005)
---------------------------------------------------------------------
To unsubscribe, e-mail: [email protected]
For additional commands, e-mail: [email protected]
