Re: [Valgrind-developers] More updates

[Julian S. <js...@ac...>]

[Nick -- q for you below]

Merging more stuff in.  I just did 08-skin-clientreq.

Also partially did 13-track-condvar-mutex.  I agree, the tracking 
was wrong.  However, I couldn't see the purpose of the part of
the patch dealing with the value of vg_tid_currently_in_baseBlock
when baseBlock is empty and so didn't merge that bit (shown
below).  It seems like, with that change, it will be impossible to
know from the value of vg_tid_currently_in_baseBlock whether or 
not baseBlock is "full".

Nick: I spotted the infamous (?) VG_(get_current_tid_1_if_root)
and specifically this:

   if (0 == vg_tid_currently_in_baseBlock)
      return 1;     /* root thread */

What's the meaning of the 0 here?  Where is it set?  AFAICS the only
valid values of vg_tid_currently_in_baseBlock are either 1 .. VG_N_THREADS
or VG_INVALID_THREADID, so I guess I'm missing something here?

I don't feel like I'm seeing a consistent story I'm happy with re
vg_tid_currently_in_baseBlock -- can you two clarify?

J




@@ -468,7 +468,7 @@ void VG_(save_thread_state) ( ThreadId t
    Int i;
    const UInt junk = 0xDEADBEEF;
 
-   vg_assert(vg_tid_currently_in_baseBlock != VG_INVALID_THREADID);
+   vg_assert(vg_tid_currently_in_baseBlock == tid);
 
 
    /* We don't copy out the LDT entry, because it can never be changed
@@ -2168,6 +2168,7 @@ void do__apply_in_new_thread ( ThreadId 
    /* Copy the parent's CPU state into the child's, in a roundabout
       way (via baseBlock). */
    VG_(load_thread_state)(parent_tid);
+   vg_tid_currently_in_baseBlock = tid;
    VG_(save_thread_state)(tid);
 
    /* Consider allocating the child a stack, if the one it already has