Re: [PATCH 1/5] pid: Implement task references.

[Eric Dumazet <dada1@cosmosbay.com>]

Eric W. Biederman a écrit :
> Eric Dumazet <dada1@cosmosbay.com> writes:
> 
>> Greg KH a écrit :
>>> On Mon, Jan 30, 2006 at 06:19:35AM +0100, Eric Dumazet wrote:
>>>> Example of improvement in kref_put() :
>>>>
>>>> [PATCH] kref : Avoid an atomic operation in kref_put() when the last
>>>> reference is dropped. On most platforms, atomic_read() is a plan read of the
>>>> counter and involves no atomic at all.
>>> No, we wat to decrement and test at the same time, to protect against
>>> any race where someone is incrementing right when we are dropping the
>>> last reference.
>> Sorry ? Me confused !
> 
> Largely I think you have the right of it, that the optimization is
> correct.  My biggest complaint is that the common case is going to be
> several references to the data structure.  Releasing the references
> will always be slow.  To do the read you need to get the value into
> the cache line.
> 
> So it looks to me like you are optimizing the wrong case and
> bloating the icache with unnecessary code.

This function is not inlined.

Adding a test and a branch is a matter of 7 bytes.

  'Bloating the icache' is a litle bit off :)

Avoiding an atomic is important. This is already done elsewhere in the kernel, 
in a inlined function with *many* call sites :

(See kfree_skb() in include/linux//skbuff.h )

/*
  * If users == 1, we are the only owner and are can avoid redundant
  * atomic change.
  */

/**
  *      kfree_skb - free an sk_buff
  *      @skb: buffer to free
  *
  *      Drop a reference to the buffer and free it if the usage count has
  *      hit zero.
  */
static inline void kfree_skb(struct sk_buff *skb)
{
         if (likely(atomic_read(&skb->users) == 1))
                 smp_rmb();
         else if (likely(!atomic_dec_and_test(&skb->users)))
                 return;
         __kfree_skb(skb);
}


This is a valid optimization : an atomic_dec_and_test() is very expensive.

Eric