Re: [git pull] cpus4096 tree, part 3

[Andrew Morton <akpm@linux-foundation.org>]

On Mon, 26 Jan 2009 11:09:59 -0800 (PST)
Linus Torvalds <torvalds@linux-foundation.org> wrote:

> 
> 
> On Mon, 26 Jan 2009, Andrew Morton wrote:
> > > +	write = error_code & PF_WRITE;
> > 
> > What's going on here?  We set `error_code' to PF_WRITE, which is some
> > x86-specific thing.
> 
> No. We set "write" to non-zero if it was a write fault.
> 
> > >  	fault = handle_mm_fault(mm, vma, address, write);
> > 
> > and then pass it into handle_mm_fault(), which is expecting a bunch of
> > flags in the FAULT_FLAG_foo domain.
> 
> No. "handle_mm_fault()" takes an integer that is non-zero if it's a write, 
> zero if it's a read. That's how it has _always_ worked.
> 
> I don't see where you find that FAULT_FLAG_foo thing. That's much deeper 
> down, when people do things like
> 
>         unsigned int flags = FAULT_FLAG_NONLINEAR |
>                                 (write_access ? FAULT_FLAG_WRITE : 0);
> 
> based on that whole "write_access" flag.
> 

OK, thanks.  It's actually page_fault-retry-with-nopage_retry.patch
which got those things confused, and then confused me.  I'll go address
that in the other thread..