[Mac] Release the mach_host_self() Mach port when done with it

[Mac] Release the mach_host_self() Mach port when done with it.

mach_host_self() and mach_thread_self() return send rights to their
respective Mach ports, which are finite system-wide resources. The port
returned by these calls must be released when the caller finishes using
the port. Failure to do so can exhaust the supply of ports system-wide.
For an example of the type of problem that this can cause, see
http://crbug.com/105513.

I don't believe that PR_GetPhysicalMemorySize is used by Chrome
anywhere, but Chrome does build this code on Mac, and this
mach_host_self() call came up in an audit of un-deallocated ports. This
patch should also be sent upstream.

Note that while you must mach_port_deallocate() the ports you get from
mach_host_self() and mach_thread_self(), you must not normally
deallocate the one from mach_task_self(), because mach_task_self() is
actually just a macro that references a global variable. It doesn’t add
any port rights at all. See <mach/mach_init.h>. If you bypass the macro
and call the real mach_task_self() system call, you do need to call
mach_port_deallocate(), but this situation is incredibly rare.

R=wtc

[wtc, 2014-05-12 21:02:48]

Patch set 1 LGTM. Please don't commit this CL yet.

We need to add a patch file and document the patch in README.chromium. If this
CL is not urgent, we can also just check it in to the NSPR upstream, and wait
until the next time I update NSPR to pick this up.

What would you like to do?

[Mark Mentovai, 2014-05-12 21:13:44]

It’s not urgent. We can prepare a patch to the upstream and import it normally.

[wtc, 2014-05-12 23:30:34]

Mark:

I checked in your patch to the NSPR upstream:
https://hg.mozilla.org/projects/nspr/rev/f4f1851bb7a6

You can close this CL now. Thanks.

[Mark Mentovai, 2014-05-13 14:14:52]

Excellent, thanks!

https://hg.mozilla.org/projects/nspr/rev/f4f1851bb7a6
https://bugzilla.mozilla.org/show_bug.cgi?id=1009270