adding logging for thread-related stuff (Unified Logging)

[Thomas Stüfe]

Hi all,

I would like to add tracing to trace things around threads, from the OS
point of view:
- trace thread starts and finish, incl. pthread id, kernel thread id...
- trace stack locations
- trace guard page locations and when they are protected, unprotected
- ...

In the future there may be more OS related stuff to trace, like signal
handling etc.

This is mainly motivated by the AIX port (see e.g.
https://bugs.openjdk.java.net/browse/JDK-8144343?filter=22798), but I would
like to add this for all platforms. So, I have some questions:

1) are there any objections?

2) if not, how should I name the tag? "os"? "os-threads"? "threads"? I
would like to use "os" as "anything at the VM-OS border".

3) for errno values: I have never been happy with strerror(). I propose to
add a function which returns the name of the errno enum for the value (e.g.
"EINVAL" for EINVAL). This would have the following advantages compared
with strerror():
- short, precise output in logs (every programmer knows "EINVAL" or
"EACCES"). Easy to grep for.
- no trouble with localizations. strerror output is localized depending on
the customers LC_MESSAGES. So, you may get "Illegal Seek" or "Nicht
erlaubter Seek" or "Repérage non permis" which is even harder to grep. You
also may just get "????" if the code conversion is not supported by the
C-Runtime.
- making this function threadsafe is trivial. strerror() is not threadsafe
and strerror_r() requires error checking.

Does this make sense? What do you think?

Thanks & Regards, Thomas