Re: [PATCH 1/9] Add the low level infrastructure for pthreads lockelision with TSX

[Torvald Riegel <triegel at redhat dot com>]

On Wed, 2013-01-23 at 09:16 +0100, Florian Weimer wrote:
> On 01/22/2013 10:39 PM, Andi Kleen wrote:
> 
> > +      /* When this could be a nested trylock that is not explicitely
> > +	 declared an elided lock abort. This makes us follow POSIX
> > +	 paper semantics. */
> > +      if (upgraded)
> > +        _xabort (0xfd);
> > +
> > +      if ((status = _xbegin()) == _XBEGIN_STARTED)
> > +	{
> > +	  if (*futex == 0)
> > +	    return 0;
> > +
> > +	  /* Lock was busy. Fall back to normal locking.
> > +	     Could also _xend here but xabort with 0xff code
> > +	     is more visible in the profiler. */
> > +	  _xabort (0xff);
> > +	}
> 
> Shouldn't 0xfd, 0xff be #defines or enum constants?

Agreed.

Andi, you've previously said that those constants would become part of
an ABI.  Is there any common header / specification / ... for this ABI
yet?  Where do you think will this get specified properly?  Or will this
remain a convention?

This is also somewhat related to the approach you took for txn-assert.
Those constants (eg, 0xff) are already used in GCC libitm too, so it
might make sense to find a common place for them at some point in time..


Torvald