Review for libpthread support

[Daniel Eischen]

On Sun, 5 Oct 2003, Jake Burkholder wrote:

> On Sunday 05 October 2003 13:46, Daniel Eischen wrote:
> > On Sun, 5 Oct 2003, Jake Burkholder wrote:
> > > On Sunday 05 October 2003 12:21, Daniel Eischen wrote:
> > > > Hi guys,
> > > >
> > > > I have a patch for adding libpthread support for sparc64.  It's jake's
> > > > patch, with a few mods by myself.  I'd like for some folks to look it
> > > > over.  I haven't tested it; I need a couple of questions answered
> > > > first so I can fill in a missing piece.
> > > >
> > > >   o Can someone look at the end of __thr_getcontext (line 524 of
> > > >     the patch)?  I need to test %i2 for NULL and branch to 1 if true.
> > > >
> > > > 	+/*	bz	%i2, 1f */
> > > > 	+	stx	%i1, [%i2]
> > > > 	+1:	ldx	[%sp], %i1			/* restore %i1 */
> > > > 	+	ldx	[%sp + 8], %i2			/* restore %i2 */
> > >
> > > brz.  Beware of delay slots, the instruction after a branch is executed
> > > before the branch is taken.
> >
> > OK.
> >
> > > Use the locals as temporaries instead of using the stack, eg
> >
> > What if a thread gets interrupted while using locals?  _thr_setcontext()
> > is suppose to be able to resume a thread that was interrupted anywhere.
> > What if some library function used locals, thread T1 had a page
> > fault, and it's context was exported back to libpthread?  The
> > library may run other threads before resuming T1 again.
> 
> They're part of T1's context.  They'll get saved on T1's stack on entry to the 
> kernel to handle the page fault.

Oh, so they're not stored in the mcontext, but when resumed
they'll be popped off the stack.

Thanks, I'll use the local's then.

-- 
Dan Eischen