DragonFly BSD
DragonFly kernel List (threaded) for 2005-05
[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]

Re: patch #2 to add pointer to errno in TCB (Re: The time has come for a kernel interfacing library layer)


From: YONETANI Tomokazu <qhwt+dfly@xxxxxxxxxx>
Date: Tue, 10 May 2005 19:32:37 +0900

On Tue, May 10, 2005 at 06:17:51PM +0800, Sepherosa Ziehau wrote:
> On 5/10/05, YONETANI Tomokazu <qhwt+dfly@xxxxxxxxxx> wrote:
> > Not essential for this patch, but before it looses its context,
> > 
> > On Tue, May 10, 2005 at 12:19:27AM -0700, Matthew Dillon wrote:
> > > @@ -78,14 +81,7 @@
> > >  {
> > >       void *self;
> > >
> > > -#if 0
> > > -     __asm __volatile ("movl %%gs:%1, %0"
> > > -         : "=r" (self)
> > > -         : "i" (__offsetof(struct tls_tcb, tcb_pthread)));
> > > -#else
> > >       __asm __volatile ("movl %%gs:8, %0" : "=r" (self));
> > > -#endif
> > > -
> > >       return(self);
> > >  }
> > >
> > 
> > this part(and other similar parts) could be written to avoid hard-coded
> > constants in __asm code by using an "m" constraint and a faked lvalue
> > for a member of the structure whose base address is zero:
> > 
> > __asm __volatile ("movl %%gs:%1, %0"
> >     : "=r" (self)
> >     : "m" (((struct tls_tcb *)0)->tcb_pthread));
> > 
> 
> Why not move __offsetof() from machine/stdint.h to sys/cdefs.h and
> resurrect `#if 0' code in tls.h

Because it gets assembled to something as below, and it's not something
expected here(that's why it's #ifdef'ed out):

   movl %gs:$8, %eax
   movl %eax, self

`%gs:$8' is an integer constant 8 prefixed by %gs, which is not a valid
operand for movl instruction.



[Date Prev][Date Next]  [Thread Prev][Thread Next]  [Date Index][Thread Index]