On 8/2/05, Eric Dumazet <[EMAIL PROTECTED]> wrote: > Arnaldo Carvalho de Melo a écrit : > > Hi Eric, > > > > Interesting, I'm reading it now that I finished one more changeset on > > the > > generalisation of all the tcp_hashinfo routines, could you please take > > a look at: > > > > http://www.kernel.org/git/?p=linux/kernel/git/acme/timewait-2.6.14.git;a=summary > > > > ? > > > > yes :) Thanks for the link !
You're welcome :) > About > http://www.kernel.org/git/?p=linux/kernel/git/acme/timewait-2.6.14.git;a=commitdiff;h=7d3cecee43811f9493bff3fa794bf30b77e4e3cd > > Please move kmem_cache_t *bind_bucket_cachep; in a better place, not just > after spinlock_t portalloc_lock; > > bind_bucket_cachep is a read only pointer (only written at boot time), and if > you place it on a cache line that is heavily written (because > of atomic counters and spinlocks), processors wont be able to cache a local > copy of this pointer. > > About inet_lhashfn() and other 'hash functions', its better to make them > return 'unsigned int', so that 64 bits platforms dont have to sign > extend when using the result to index a table. The generated code would be > shorter. Thanks for the suggestions, I'm mostly moving, slowly, not changing the existing code too much in its basic functionality, to a new sock infrastructure based on the hard work Dave, Alexey and others did on TCP, there is for sure more work to do and improvements to be done, like a generic net_diag, based on today's tcp_diag, etc. I have lots of other patches in my experimental LLC tree, where I toyed with generalising tcp_sendmsg, tcp_recvmsg, tcp_sendpages, etc, and many more in my DCCP tree, that is mostly what I'm flushing now. Eventually I'll move on to the other legacy protocols making them use this infrastructure, for instance this file has one possible outcome: http://userweb.kernel.org/~acme/sock_hierarchy.ps Parts are already merged, parts are about to be submitted to David, like the partial timewait_sock hierarchy (for now restricted to INET transport protocols) and other bits I still have to dig out of my experimental trees, most probably after looking at what was done again and reworking some parts. > Thank you Thank you too! > > Good thing is that any optimization you do for TCP will be useful for > > DCCP as well, > > perhaps SCTP at some point :-) - 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
