Valgrind warnings "Syscall param epoll_ctl(event) points to uninitialised byte(s)unitialized bytes" at threadpool_unix.cc

[Elena Stepanova]

==28641== Thread 5:
==28641== Syscall param epoll_ctl(event) points to uninitialised byte(s)
==28641== at 0x42C7C1E: epoll_ctl (in /lib/libc-2.11.2.so)
==28641== by 0x83B22D4: io_poll_associate_fd(int, int, void*) (threadpool_unix.cc:208)
==28641== by 0x83B4512: start_io(connection_t*) (threadpool_unix.cc:1192)
==28641== by 0x83B45C6: handle_event(pool_event_t*) (threadpool_unix.cc:1221)
==28641== by 0x83B48FC: worker_main(void*) (threadpool_unix.cc:1269)
==28641== by 0x4049B24: start_thread (in /lib/libpthread-2.11.2.so)
==28641== by 0x42C734D: clone (in /lib/libc-2.11.2.so)
==28641== Address 0x7b7023c is on thread 5's stack
==28641==
==28641== Syscall param epoll_ctl(event) points to uninitialised byte(s)
==28641== at 0x42C7C1E: epoll_ctl (in /lib/libc-2.11.2.so)
==28641== by 0x83B231A: io_poll_start_read(int, int, void*) (threadpool_unix.cc:218)
==28641== by 0x83B4533: start_io(connection_t*) (threadpool_unix.cc:1195)
==28641== by 0x83B45C6: handle_event(pool_event_t*) (threadpool_unix.cc:1221)
==28641== by 0x83B48FC: worker_main(void*) (threadpool_unix.cc:1269)
==28641== by 0x4049B24: start_thread (in /lib/libpthread-2.11.2.so)
==28641== by 0x42C734D: clone (in /lib/libc-2.11.2.so)
==28641== Address 0x7b7023c is on thread 5's stack
==28641==
111228 19:53:26 [Note] /home/elenst/lp-5.5-threadpool/debug/sql/mysqld: Normal shutdown

111228 19:53:26 [Note] Event Scheduler: Purging the queue. 0 events
111228 19:53:31 [Note] Debug sync points hit: 779
111228 19:53:31 [Note] Debug sync points executed: 0
111228 19:53:31 [Note] Debug sync points max active per thread: 0
111228 19:53:31 [Note] /home/elenst/lp-5.5-threadpool/debug/sql/mysqld: Shutdown complete

==28641== Thread 7:
==28641== Syscall param epoll_ctl(event) points to uninitialised byte(s)
==28641== at 0x42C7C1E: epoll_ctl (in /lib/libc-2.11.2.so)
==28641== by 0x83B22D4: io_poll_associate_fd(int, int, void*) (threadpool_unix.cc:208)
==28641== by 0x83B359E: thread_group_close(thread_group_t*) (threadpool_unix.cc:791)
==28641== by 0x83B4B32: tp_end() (threadpool_unix.cc:1329)
==28641== by 0x81B893F: clean_up(bool) (mysqld.cc:1827)
==28641== by 0x81B852D: unireg_end() (mysqld.cc:1695)
==28641== by 0x81B83F4: kill_server(void*) (mysqld.cc:1623)
==28641== by 0x81B8417: kill_server_thread (mysqld.cc:1646)
==28641== by 0x855F5BC: pfs_spawn_thread (pfs.cc:1015)
==28641== by 0x4049B24: start_thread (in /lib/libpthread-2.11.2.so)
==28641== by 0x42C734D: clone (in /lib/libc-2.11.2.so)
==28641== Address 0x80000ec is on thread 7's stack
==28641==

Importantly, I'm getting it only on a 1-core 32-bit machine, but not on a 4-core 64-bit machine. I'm not sure what makes the difference, the number of CPU or the architecture, or something else. Both are openSUSE 11.3.

How to repeat:

I'm getting it while running the MTR test 1st from the main suite, with all default settings and valgrind-mysqld.

Not reproducible with thread_handling=one-thread-per-connection or no-threads.

bzr version-info
revision-id: <email address hidden>
date: 2011-12-27 17:54:04 +0100
build-date: 2011-12-28 18:49:26 +0400
revno: 3181
branch-nick: lp-5.5-threadpool