ostimer.c functions not using semaphore

[skliper]

The VxWorks ostimer.c has a static table and a semaphore to protect it, as in:
{{{
static OS_timer_record_t OS_timer_table[OS_MAX_TIMERS];
static uint32 os_clock_accuracy;

/*
** The Semaphore for protecting the above table
*/
static SEM_ID OS_timer_table_sem;
}}}

Unfortunately, only OS_TimerCreate(), and OS_TimerGetInfo() use that semaphore.

'''OS_TimerCreate(), OS_TimerSet(), OS_TimerDelete() do not''' and they '''modify''' the table. '''OS_TimerGetIdByName()''' and the internal '''OS_TimerSignalHandler()''' also access (read) the table without using the semaphore.

The OS_TimerAPIInit() creates the semaphore, the time conversion functions don't access the table.

Recommend fixing all the functions that access or modify the table to acquire the semaphore after passing input argument checks.

[skliper]

Imported from trac issue 69. Created by abrown4 on 2015-07-01T16:43:49, last modified: 2015-12-01T14:13:03

[skliper]

Trac comment by abrown4 on 2015-07-08 13:11:06:

commit: [changeset:1a6e450] Trac #92, Fixed VxWorks ostimer.c to properly use semaphore.
Built upon ostimer.c coverage tests on Track #45. Made unit test changes to differentiate between failures due to Trac #92 and Trac #88.

[skliper]

Trac comment by abrown4 on 2015-07-08 13:22:36:

Opened Trac #94 for the posix version, which has similar problems.
Not all methods in the RTEMS version are protected... that may need its own issue as well. Created #95.

[skliper]

Trac comment by abrown4 on 2015-07-28 12:36:33:

Reviewed the above commit with #103 in mind.

If the functions that access the table, OS_TimerGetIdByName() and the internal OS_TimerSignalHandler(), follow the same paradigm then they don't need modification. Similarly, OS_TimerSet() modifies the table entry but the access pattern might not collide with other tasks given the expected usage patterns. Discuss.

MUST CHANGE: OS_TimerSignalHandler() invokes a callback. This should not occur while the table is locked.

[skliper]

Trac comment by abrown4 on 2015-07-28 14:58:16:

commit: [changeset:270c2dc] Trac #92, Removed table locks in ostimer.c OS_TimerSignalHandler().
Locking a global table while calling a non-osal callback isn't a good idea.

[skliper]

Trac comment by abrown4 on 2015-08-20 18:24:31:

From the above commits, as of #45 [changeset:6a9db70], the white-box unit test still shows that OS_TimerSignalHandler() doesn't safely access the OS_timer_table (3 test failures from ostimer_wb test).

[skliper]

Trac comment by abrown4 on 2015-11-25 12:24:13:

Traceability: The vxworks code changes from #92, [changeset:1a6e450] and [changeset:270c2dc] have been reflected in #138 [changeset:ab2d6d7]. Recommend closing #92 in favor of #138.