Lost memory reported in valgrind in os_aio_simulated_handle

[Michael Widenius]

When running the test innodb_plugin.innodb-use-sys-malloc (not support test), xtradb doesn't properly free memory when test is aborted.

How to repeat:

mysql-test-run --valgrind innodb_plugin.innodb-use-sys-malloc

Error output:

innodb_plugin.innodb_mysql w3 [ skipped ] Test requires: 'true'
***Warnings generated in error logs during shutdown after running tests: innodb_plugin.innodb-use-sys-malloc

==8801== 1,310,720 bytes in 3 blocks are still reachable in loss record 7 of 7
==8801== at 0x4C259AE: malloc (vg_replace_malloc.c:207)
==8801== by 0xA056A2: ut_malloc_low (ut0mem.c:106)
==8801== by 0x9B8896: os_aio_simulated_handle (os0file.c:4191)
==8801== by 0xA3F027: fil_aio_wait (fil0fil.c:4892)
==8801== by 0x9F07B7: io_handler_thread (srv0start.c:492)
==8801== by 0x504906F: start_thread (in /lib64/libpthread-2.9.so)
==8801== by 0x5FFE11C: clone (in /lib64/libc-2.9.so)

This affects MariaDB 5.1. MySQL 5.1 doesn't have this issue

[Christoffer Hall (christoffer-hall-percona)]

It seems the os_aio_thread_buffer for read/write combining gets allocated on the fly but never deallocated at the end. This simple patch fixes it.

[Christoffer Hall (christoffer-hall-percona)]

Verified that it works on percona-xtradb