Re: [patch] remove the BKL (Big Kernel Lock), this time for real

[Manfred Spraul <manfred@colorfullife.com>]

Tony wrote:

>I coded a IPC system before use atomic add + share memory.  
>It works very well (fast) on 4 CPU SMP system, since it doesn't use 
>any locking API at all.    Very good for resource allocation for 
>SMP.     I implemented speciall malloc/free use by ISR, different
>prority process completely without any lock.
>
Without any lock or without any common cacheline that are accessed by 
atomic operations?
I usually consider the costs of
    atomic_inc(&global_atomic_var);
and
    spin_lock(&global_lock);
    global_var++;
    spin_unlock(&global_lock);
as nearly identical (assuming that global_var and global_lock are in the 
same cacheline): one cachline transfer per run. The 5 instructions under 
the spinlock and the theoretical chance that spin_lock() blocks are 
noise compared to the cost of the line transfer.
And that's without thinking about smb_mb__{before,after}_atomic_inc().

Btw, Ingo forgot to mention sequence locks and percpu_counter as two 
high-scalability locking primitives.

--
    Manfred