Re: [PATCH 1/8] drivers/random: Cache align ip_random better

[Matt Mackall <mpm@selenic.com>]

On Wed, 2011-03-16 at 10:17 -0700, Hugh Dickins wrote:
> On Sun, 13 Mar 2011, George Spelvin wrote:
> 
> > Cache aligning the secret[] buffer makes copying from it infinitesimally
> > more efficient.
> > ---
> >  drivers/char/random.c |    2 +-
> >  1 files changed, 1 insertions(+), 1 deletions(-)
> > 
> > diff --git a/drivers/char/random.c b/drivers/char/random.c
> > index 72a4fcb..4bcc4f2 100644
> > --- a/drivers/char/random.c
> > +++ b/drivers/char/random.c
> > @@ -1417,8 +1417,8 @@ static __u32 twothirdsMD4Transform(__u32 const buf[4], __u32 const in[12])
> >  #define HASH_MASK ((1 << HASH_BITS) - 1)
> >  
> >  static struct keydata {
> > -	__u32 count; /* already shifted to the final position */
> >  	__u32 secret[12];
> > +	__u32 count; /* already shifted to the final position */
> >  } ____cacheline_aligned ip_keydata[2];
> >  
> >  static unsigned int ip_cnt;
> 
> I'm intrigued: please educate me.  On what architectures does cache-
> aligning a 48-byte buffer (previously offset by 4 bytes) speed up
> copying from it, and why?  Does the copying involve 8-byte or 16-byte
> instructions that benefit from that alignment, rather than cacheline
> alignment?

I think this alignment exists to minimize the number of cacheline
bounces on SMP as this can be a pretty hot structure in the network
stack. It could probably benefit from a per-cpu treatment.

-- 
Mathematics is the supreme nostalgia of our time.