CVS log for pegasus/src/Pegasus/Common/Attic/ThreadUnix_inline.h
	Help

Request diff between arbitrary revisions

 - The Solaris compiler moans that pthread_create is expecting a function with
   "C" linkage as the third parameter and we are passing one with "C++" linkage.
   Not only is this true it appears the compiler is correct to complain, even
   though "C" and "C++" linkage are the same on Solaris.

   Added a structure zosParmDef to hold the desired function and its parameter.
   The pass a wrapper function with "C" linkage to pthread_create that simply
   calls the function in zosParmDef with the included parameter.

   Note the structure is malloc'ed in Thread::run, but _must_ be free'd by the
   newly created thread. This is may not be pretty, but it does avoid some nasty
   race conditions.

   zosParmDef shamelessly "borrowed" from the IBM zOS port.

 - On Solaris 5.8 the IPC test (and some others) would deadlock with the
   process consuming large amounts of CPU time. It turns out the "dqe" thread
   in the IPC test has a "hard" loop checking for messages on the queue. If
   the main thread and one of the dqe threads ended up on the same LWP process
   then the dqe would spin and never get pre-empted. This locked out the main
   thread which never got a chance to stop the dqe thread.

   Changed the scheduling policy from SCHED_OTHER to SCHED_RR for Solaris. This
   seems to fix the IPC deadlock by pre-empting dqe after a set time.

 - All the above wrapped in an #ifdef SOLARIS so it doesn't affect other
   platforms.