cancellation and unwind

[Petr Salinger]

Hello,

I am trying to understand how NPTL works wrt cancellation.
Please, could someone provide brief explanation here ?

So far I hit into this:

a)
nptl/forward.c
FORWARD_NORETURN(__pthread_unwind,
...
                        INTERNAL_SYSCALL_DECL (err);
                        INTERNAL_SYSCALL (kill, err, 1, SIGKILL);

But usually  "int kill(pid_t pid, int sig)" holds.

It looks like pid 9 is killed by random signal ...

b)
nptl/cancellation.c: __pthread_disable_asynccancel (int oldtype)


   /* We cannot return when we are being canceled.  Upon return the
      thread might be things which would have to be undone.  The
      following loop should loop until the cancellation signal is
      delivered.  */
   while (__builtin_expect ((newval & (CANCELING_BITMASK | CANCELED_BITMASK))
                            == CANCELING_BITMASK, 0))
     {
       lll_futex_wait (&self->cancelhandling, newval, LLL_PRIVATE);
       newval = THREAD_GETMEM (self, cancelhandling);
     }

Can this ever return, or will it only again and again call kernel,
until thread is somehow terminated ?


c)
nptl x86_64 specific cancellation.S

The comment says:
/* It is crucial that the functions in this file don't modify registers
    other than %rax and %r11.  The syscall wrapper code depends on this
    because it doesn't explicitly save the other registers which hold
    relevant values.  */

But later on:

3:      andl    $(TCB_CANCELING_BITMASK|TCB_CANCELED_BITMASK), %eax
         cmpl    $TCB_CANCELING_BITMASK, %eax
         je      4f
1:      ret

         /* Performance doesn't matter in this loop.  We will
            delay until the thread is canceled.  And we will unlikely
            enter the loop twice.  */
4:      mov     %fs:0, %RDI_LP
         movl    $__NR_futex, %eax
         xorq    %r10, %r10
         addq    $CANCELHANDLING, %rdi
         LOAD_PRIVATE_FUTEX_WAIT (%esi)
         syscall
         movl    %fs:CANCELHANDLING, %eax
         jmp     3b

Again, can this ever return ?
Or the modification of registers is not problem any more ?

And moreover, the third agument of futex (rdx, old value) is not set 
before syscall at all ...

Am I mising some obvious ?

Petr