Re: [patchset] Lockfree fd lookup 0 of 5

[Manfred Spraul <manfred@colorfullife.com>]

viro wrote:

>How about this for comparison?  That's just a dumb "convert to rwlock"
>patch; we can be smarter in e.g. close_on_exec handling, but that's a
>separate story.
>  
>
That won't help:
The problem is the cache line trashing from fget() and fget_light() with 
multithreaded apps. A sequence lock might help, but an rw lock is not a 
solution.
Actually: 2.4 had an rwlock, it was converted to a spinlock because 
spinlocks are faster on i386:

    read_lock();
    short operation();
    read_unlock();

is a bit slower than

    spin_lock();
    short operation();
    spin_unlock();

because read_unlock() is a full memory barrier and spin_unlock is not a 
memory barrier (not necessary because writes are ordered)

--
    Manfred