Inquiry about brl statement

Inquiry about brl statement

[Christian Kandeler]

Hello, 

in the definition of the BRL_COND_FSYS_BUBBLE_DOWN macro in 
arch/ia64/kernel/gate.S, shouldn't there be a stop bit after the brl.cond 
instruction? According to the Intel specs, the current definition (which has 
no stop bit) triggers undefined behavior.


Regards,
Christian Kandeler

Re: Inquiry about brl statement

[David Mosberger-Tang]

As I remember it, a taken branch acts as a stop whereas a non-taken
branch doesn't (so if no explicit stop bit is following a branch, then
it must be OK for the entire group to be executed in parallel).  I
suppose it's possible the definition changed or that my (admittedly
bad) memory is playing tricks on me. ;-)

  --david

On 4/27/07, Christian Kandeler <christian.kandeler@hob.de> wrote:
> Hello,
>
> in the definition of the BRL_COND_FSYS_BUBBLE_DOWN macro in
> arch/ia64/kernel/gate.S, shouldn't there be a stop bit after the brl.cond
> instruction? According to the Intel specs, the current definition (which has
> no stop bit) triggers undefined behavior.
>
>
> Regards,
> Christian Kandeler
> -
> To unsubscribe from this list: send the line "unsubscribe linux-ia64" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>


-- 
Mosberger Consulting LLC, http://www.mosberger-consulting.com/

RE: Inquiry about brl statement

[Luck, Tony]

> As I remember it, a taken branch acts as a stop whereas a non-taken
> branch doesn't (so if no explicit stop bit is following a branch, then
> it must be OK for the entire group to be executed in parallel).  I
> suppose it's possible the definition changed or that my (admittedly
> bad) memory is playing tricks on me. ;-)

The SDM page for "brl" doesn't look to have any get-out-of stop-bits-free
option.  It says:

 "This instruction must be immediately followed by a stop; otherwise
  its behaviour is undefined."

-Tony

Re: Inquiry about brl statement

[David Mosberger-Tang]

Tony,

I'm sure I missed that point about brl.  I probably just assumed it
would have the same behavior as regular branches.  Thanks for catching
that.

  --david

On 4/27/07, Luck, Tony <tony.luck@intel.com> wrote:
> > As I remember it, a taken branch acts as a stop whereas a non-taken
> > branch doesn't (so if no explicit stop bit is following a branch, then
> > it must be OK for the entire group to be executed in parallel).  I
> > suppose it's possible the definition changed or that my (admittedly
> > bad) memory is playing tricks on me. ;-)
>
> The SDM page for "brl" doesn't look to have any get-out-of stop-bits-free
> option.  It says:
>
>  "This instruction must be immediately followed by a stop; otherwise
>   its behaviour is undefined."
>
> -Tony
>


-- 
Mosberger Consulting LLC, http://www.mosberger-consulting.com/

RE: Inquiry about brl statement

[Jim Hull]

David:

If brl had existed from the beginning, then its stop-bit requirements would
have matched those of regular branches.  However, it was added after Merced
could no longer be changed, and because of where it is encoded, and the
desire to enable an OS to emulate it on Merced, it had to have the
must-always-be-followed-by-a-stop-bit rule.

 -- Jim 

> -----Original Message-----
> From: linux-ia64-owner@vger.kernel.org 
> [mailto:linux-ia64-owner@vger.kernel.org] On Behalf Of David 
> Mosberger-Tang
> Sent: Friday, April 27, 2007 1:01 PM
> To: Luck, Tony
> Cc: Christian Kandeler; linux-ia64@vger.kernel.org
> Subject: Re: Inquiry about brl statement
> 
> Tony,
> 
> I'm sure I missed that point about brl.  I probably just assumed it
> would have the same behavior as regular branches.  Thanks for catching
> that.
> 
>   --david
> 
> On 4/27/07, Luck, Tony <tony.luck@intel.com> wrote:
> > > As I remember it, a taken branch acts as a stop whereas a 
> non-taken
> > > branch doesn't (so if no explicit stop bit is following a 
> branch, then
> > > it must be OK for the entire group to be executed in parallel).  I
> > > suppose it's possible the definition changed or that my 
> (admittedly
> > > bad) memory is playing tricks on me. ;-)
> >
> > The SDM page for "brl" doesn't look to have any get-out-of 
> stop-bits-free
> > option.  It says:
> >
> >  "This instruction must be immediately followed by a stop; otherwise
> >   its behaviour is undefined."
> >
> > -Tony
> >
> 
> 
> -- 
> Mosberger Consulting LLC, http://www.mosberger-consulting.com/
> -
> To unsubscribe from this list: send the line "unsubscribe 
> linux-ia64" in
> the body of a message to majordomo@vger.kernel.org
> More majordomo info at  http://vger.kernel.org/majordomo-info.html
>