Re: [PATCH] Optimize away pte_chains for single mappings

[Daniel Phillips <phillips@arcor.de>]

On Monday 15 July 2002 16:02, Dave McCracken wrote:
> --On Saturday, July 13, 2002 03:13:35 PM +0200 Daniel Phillips
> <phillips@arcor.de> wrote:
> 
> > Why are we using up valuable real estate in page->flags when the low bit
> > of page->pte_chain is available?
> 
> Right now my flag is bit number 18 in page->flags out of 32.  Mechanisms
> already exist to manipulate this bit in a reasonable fashion.  I don't see
> any good reason for complicating things by putting a flag bit into a
> pointer, where we'd have to repeatedly check and clear it before we
> dereference the pointer.

Hi Dave,

It's not more complicated.  You have to check which type of pointer you
have anyway, and having to strip away the low bit on one of the two
paths is insignificant in terms of generated code.  The current patch
has to set and clear the flag bit separately.

> When I discussed this with Rik he said putting it
> in flags was reasonable.  We can always revisit it in the future if we run
> out of bits.

I prefer doing things the most efficient way in core code.

-- 
Daniel
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/