Invert sense of SLB class bit

Sonny Rao sonny at burdell.org
Thu Sep 8 16:19:15 EST 2005


On Thu, Sep 08, 2005 at 08:45:17AM +1000, Paul Mackerras wrote:
> linas writes:
> 
> > Antonb had been talking about something like this more than a year 
> > ago ... as there was code added to not flush the 0xc segment.  
> > Did that code not work? Or did it get broken? Or was it never 
> > submitted upstream?
> 
> No, this is entirely different.  POWER5 creates ERAT entries for pages
> which are accessed in real mode (these are created automatically by
> the hardware as needed).  In other words, *all* accesses go through
> the ERAT, there isn't a bypass for real mode.  The real-mode ERAT
> entries are created with C=0, so by using C=0 for user mappings we
> were blowing away all the real-mode entries on every context switch.

So, when exactly do we go into real mode ?  I know the SLB handler is
now run in real mode, but is anything else? 

If we want to look at best-case improvements -- meaning a lot of SLB
reloading and userspace context-switching, I suppose I should select a
machine with a fair amount of memory.

Either way, I'm going to run SDET on a 2-way box w/ 8GB of RAM and see
if I can find a difference.  If necessary, I'll find a bigger box to
test on.

Sonny



More information about the Linuxppc64-dev mailing list