Messages in this thread | | | Date | Wed, 7 Oct 2009 12:25:03 -0700 (PDT) | From | Linus Torvalds <> | Subject | Re: [rfc][patch] store-free path walking |
| |
On Wed, 7 Oct 2009, Nick Piggin wrote: > > Oh I did that, used a "seqcount" which is the bare sequence counter > (and update it while holding d_lock).
Oh, I didn't notice, for a really silly reason: I looked at how you had changed DNAME_INLINE_LEN_MIN on 64-bit, and noticed that you were using 8 bytes for your sequence lock.
But that's only true due to the padding things, and the seqcount itself is indeed just 4 bytes.
Sad. But we do have the added requirement that we want to keep the commonly used fields together in the same cacheline, so maybe it's unavoidable.
Linus
| |