lkml.org 
[lkml]   [1999]   [Nov]   [26]   [last100]   RSS Feed
Views: [wrap][no wrap]   [headers]  [forward] 
 
Messages in this thread
/
Date
From
SubjectRe: spin_unlock optimization(i386)
Erich Boleyn wrote:
>
> But, it is not needed to avoid any kind of re-ordering that IA32 will
> do. I still maintain that.
>
> IA32 just doesn't do program visible reordering. Any reordering is
> done only as a performance enhancement and with temporary results that
> are checked prior to becoming committed state.
>
It's invisible on single CPU system, but clearly visible on SMP:

* you (speculatively) execute read instructions ahead of write
instructions.
* you do not send write instructions immediately out of the CPU core,
they sit in a write buffer.
* the write buffer is private to each CPU.
* on _single_ cpu, you correct the effects of the reordering (ie if a
write sits in the write buffer, and I reread that memory address, the
the CPU notices that, and uses the new value from the write
buffer/discards the temporary result.)

BUT: since the write buffer is private to each CPU, you cannot correct
the effects of the reordering on SMP.

The program that I sent you yesterday exploits this problem, you could
run it with a suitable tracer:
* with "lock;bts", it doesn't hang.
* with "mov", it hangs.


--
Manfred


-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu
Please read the FAQ at http://www.tux.org/lkml/

\
 
 \ /
  Last update: 2005-03-22 13:55    [from the cache]
©2003-2011 Jasper Spaans