Non-SFI mode: Add Linux asynchronous signal support

src/untrusted/pthread/nc_thread.c

Index: src/untrusted/pthread/nc_thread.c
diff --git a/src/untrusted/pthread/nc_thread.c b/src/untrusted/pthread/nc_thread.c
index 8b0fe042179cc507530b26668fc4ba32b2b256be..5346debd5ad0dd9f2fd36a086a7d1410ee090922 100644
--- a/src/untrusted/pthread/nc_thread.c
+++ b/src/untrusted/pthread/nc_thread.c
@@ -36,7 +36,7 @@
  * ABI tables for underyling NaCl thread interfaces. This is declared to be
  * global so that a user will be able to override it using the irt_ext API.
  */
-struct nacl_irt_thread __libnacl_irt_thread;
+struct nacl_irt_thread_v0_2 __libnacl_irt_thread;
 
 /*
  * These days, the thread_create() syscall/IRT call will align the
@@ -415,7 +415,10 @@ int pthread_create(pthread_t *thread_id,
   memset(esp, 0, kStackPadBelowAlign);
 
   /* Start the thread. */
-  retval = __libnacl_irt_thread.thread_create(nc_thread_starter, esp, new_tp);
+  nacl_irt_tid_t native_tid;
+  retval = __libnacl_irt_thread.thread_create(nc_thread_starter, esp, new_tp,

[Mark Seaborn, 2015/06/26 18:32:00]

This isn't going to work for SFI NaCl, because you're making this depend on
nacl_irt_thread_v0_2, which isn't defined for SFI NaCl.

I'm not sure if this change passes on the trybots?  (Can you figure out how to
get trybot access?)

I don't think it's really necessary to add support for TIDs to
src/untrusted/pthread/ (i.e. to nacl-newlib's libpthread), because in the NaCl
codebase we *should* only need access to them for testing.

See my other comment on the test code.

[Luis Héctor Chávez, 2015/07/06 23:45:00]

Done.

+                                              &native_tid);
+  (*thread_id)->native_tid = native_tid;

[Mark Seaborn, 2015/06/26 18:32:00]

BTW, assigning to (*thread_id)->native_tid as a separate step defeats the
purpose of having thread_create() set the TID rather than having a separate
gettid() syscall, which is to make both the assignment and the use of the TID
thread-safe.

Doing the assignment here means that it's not thread-safe for the child thread
to cause pthread_kill(child_thread, sig) to be called.

Also, the thread could have exited by this point, so (*thread_id)->native_tid
might point to deallocated memory.

The idea would have been to pass &(*thread_id)->native_tid to thread_create().

[Luis Héctor Chávez, 2015/07/06 23:45:00]

Done.

   if (0 != retval) {
     pthread_mutex_lock(&__nc_thread_management_lock);
     /* TODO(gregoryd) : replace with atomic decrement? */