On Sat, Aug 13, 2005 at 02:10:55PM -0700, David S. Miller wrote: > Yes, you need an atomic in the fast path, but it's only done > for fast clones and their parent SKBs, and such an atomic is > _MUCH_ cheaper than a kmem_cache_alloc especially since it > touches more cache local state (the atomic counter in the parent > SKB, whose cache line is hot for the first transmit of an SKB > usually).
Good point about the potential for cache misses and cache eviction with calling kmem_cache_alloc twice far apart from each other. But what about the following scheme for fast clones: 1) When you allocate for fast clones you do two kmem_cache_alloc's in a row. You store the second pointer in the shared skb data area. The fast path within kmem_cache_alloc is good enough that the second one should be free most of the time. 2) When you do a clone just use the stored pointer and forget about it since it can be freed independently of the first. Cheers, -- Visit Openswan at http://www.openswan.org/ Email: Herbert Xu ~{PmV>HI~} <[EMAIL PROTECTED]> Home Page: http://gondor.apana.org.au/~herbert/ PGP Key: http://gondor.apana.org.au/~herbert/pubkey.txt - To unsubscribe from this list: send the line "unsubscribe netdev" in the body of a message to [EMAIL PROTECTED] More majordomo info at http://vger.kernel.org/majordomo-info.html
