Re: [PATCH 09/10] Change table chaining layout

[Matt Mackall]

On Mon, Oct 22, 2007 at 09:16:17PM +0100, Alan Cox wrote:
> On Mon, 22 Oct 2007 12:49:40 -0700 (PDT)
> Linus Torvalds <torvalds@xxxxxxxxxxxxxxxxxxxx> wrote:
> 
> > 
> > 
> > On Mon, 22 Oct 2007, Geert Uytterhoeven wrote:
> > >
> > > Better safe than sorry...
> > > 
> > > Is it possible that a chain entry pointer has bit 1 set on architectures
> > > (e.g. m68k) where the natural alignment of 32-bit quantities is _2_ bytes,
> > > not 4?
> > 
> > Better make sure that such alignment never happens... But no, I don't 
> > think it will, since these things would generally always have to be 
> > allocated with an allocator, and the *allocator* won't return 2-byte 
> > aligned data structures.
> 
> No - but a structure which has other objects in it before the object
> being written out may well be 2 byte aligned on M68K and some of the
> other externally 16bit platforms - ditto local dynamic objects.

Also, the current version of SLOB will return objects aligned at 2 bytes if the
architecture allows it.
 
> Why can't we just make the list one item longer than the entry count and
> stick a NULL on the end of it like normal people ? Then you need one bit
> which ought to be safe for everyone (and if the bit is a macro any CPU
> warped enough to have byte alignment is surely going to have top bits
> spare...)

I'm guessing the extra entry makes slab-like allocators unhappy.

-- 
Mathematics is the supreme nostalgia of our time.
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at  http://vger.kernel.org/majordomo-info.html
Please read the FAQ at  http://www.tux.org/lkml/