diff options
| -rw-r--r-- | ChangeLog | 25 | ||||
| -rw-r--r-- | ChangeLogs/ChangeLog-02a | 25 | ||||
| -rw-r--r-- | ChangeLogs/ChangeLog-03a | 25 | ||||
| -rw-r--r-- | ace/WFMO_Reactor.cpp | 31 |
4 files changed, 91 insertions, 15 deletions
diff --git a/ChangeLog b/ChangeLog index 5c2fe396128..4bb21415806 100644 --- a/ChangeLog +++ b/ChangeLog @@ -1,3 +1,28 @@ +Thu May 09 00:38:22 2002 Irfan Pyarali <irfan@cs.wustl.edu> + + * ace/WFMO_Reactor.cpp (event_handling): Added a fix for bug + 1161. The problem was described aptly and the fix was applied + exactly as suggested by Lyn Headley <headley@wolve.com>: + + Problem: "The do-while loop in this function is problematic, + because active threads is incremented once prior to executing + the loop, and then decremented on every iteration of the loop + inside safe dispatch(). Thus on the second iteration of the + loop active threads is decremented to 2^32, which causes the + thread to block on waiting to change state inside update + state(), which is never signaled (because there are no other + active threads)." + + Solution: "Moving the do clause of the loop up a few statements + to include the increment of active threads (and the acquisition + of the lock) would probably fix the larger bug." + + Also, thanks to Lyn Headley <headley@wolve.com> and P Mason + <pmason@wolve.com> at Wolverine, and Kobi Cohen Arazi + <kobi@mivzak.com>, Mike Winter (mwinter@sonic.net), + <sturtesm@hotmail.com> who have all suggested fimilar problems + and fixes. + Thu May 09 00:06:47 2002 Irfan Pyarali <irfan@cs.wustl.edu> * examples/Reactor/WFMO_Reactor: Added the "Static Debug", "Static diff --git a/ChangeLogs/ChangeLog-02a b/ChangeLogs/ChangeLog-02a index 5c2fe396128..4bb21415806 100644 --- a/ChangeLogs/ChangeLog-02a +++ b/ChangeLogs/ChangeLog-02a @@ -1,3 +1,28 @@ +Thu May 09 00:38:22 2002 Irfan Pyarali <irfan@cs.wustl.edu> + + * ace/WFMO_Reactor.cpp (event_handling): Added a fix for bug + 1161. The problem was described aptly and the fix was applied + exactly as suggested by Lyn Headley <headley@wolve.com>: + + Problem: "The do-while loop in this function is problematic, + because active threads is incremented once prior to executing + the loop, and then decremented on every iteration of the loop + inside safe dispatch(). Thus on the second iteration of the + loop active threads is decremented to 2^32, which causes the + thread to block on waiting to change state inside update + state(), which is never signaled (because there are no other + active threads)." + + Solution: "Moving the do clause of the loop up a few statements + to include the increment of active threads (and the acquisition + of the lock) would probably fix the larger bug." + + Also, thanks to Lyn Headley <headley@wolve.com> and P Mason + <pmason@wolve.com> at Wolverine, and Kobi Cohen Arazi + <kobi@mivzak.com>, Mike Winter (mwinter@sonic.net), + <sturtesm@hotmail.com> who have all suggested fimilar problems + and fixes. + Thu May 09 00:06:47 2002 Irfan Pyarali <irfan@cs.wustl.edu> * examples/Reactor/WFMO_Reactor: Added the "Static Debug", "Static diff --git a/ChangeLogs/ChangeLog-03a b/ChangeLogs/ChangeLog-03a index 5c2fe396128..4bb21415806 100644 --- a/ChangeLogs/ChangeLog-03a +++ b/ChangeLogs/ChangeLog-03a @@ -1,3 +1,28 @@ +Thu May 09 00:38:22 2002 Irfan Pyarali <irfan@cs.wustl.edu> + + * ace/WFMO_Reactor.cpp (event_handling): Added a fix for bug + 1161. The problem was described aptly and the fix was applied + exactly as suggested by Lyn Headley <headley@wolve.com>: + + Problem: "The do-while loop in this function is problematic, + because active threads is incremented once prior to executing + the loop, and then decremented on every iteration of the loop + inside safe dispatch(). Thus on the second iteration of the + loop active threads is decremented to 2^32, which causes the + thread to block on waiting to change state inside update + state(), which is never signaled (because there are no other + active threads)." + + Solution: "Moving the do clause of the loop up a few statements + to include the increment of active threads (and the acquisition + of the lock) would probably fix the larger bug." + + Also, thanks to Lyn Headley <headley@wolve.com> and P Mason + <pmason@wolve.com> at Wolverine, and Kobi Cohen Arazi + <kobi@mivzak.com>, Mike Winter (mwinter@sonic.net), + <sturtesm@hotmail.com> who have all suggested fimilar problems + and fixes. + Thu May 09 00:06:47 2002 Irfan Pyarali <irfan@cs.wustl.edu> * examples/Reactor/WFMO_Reactor: Added the "Static Debug", "Static diff --git a/ace/WFMO_Reactor.cpp b/ace/WFMO_Reactor.cpp index cb674dd425c..0e8c79ebe8d 100644 --- a/ace/WFMO_Reactor.cpp +++ b/ace/WFMO_Reactor.cpp @@ -1603,25 +1603,26 @@ ACE_WFMO_Reactor::event_handling (ACE_Time_Value *max_wait_time, // called. ACE_Countdown_Time countdown (max_wait_time); - // Check to see if it is ok to enter ::WaitForMultipleObjects - // This will acquire <this->lock_> on success - // On failure, the lock will not be acquired - int result = this->ok_to_wait (max_wait_time, alertable); - if (result != 1) - return result; + int result; + do + { + // Check to see if it is ok to enter ::WaitForMultipleObjects + // This will acquire <this->lock_> on success On failure, the + // lock will not be acquired + result = this->ok_to_wait (max_wait_time, alertable); + if (result != 1) + return result; - // Increment the number of active threads - this->active_threads_++; + // Increment the number of active threads + this->active_threads_++; - // Release the <lock_> - this->lock_.release (); + // Release the <lock_> + this->lock_.release (); - // Update the countdown to reflect time waiting to play with the - // mut and event. - countdown.update (); + // Update the countdown to reflect time waiting to play with the + // mut and event. + countdown.update (); - do - { // Calculate timeout int timeout = this->calculate_timeout (max_wait_time); |