Re: spin_unlock optimization

Manfred Spraul (manfreds@colorfullife.com)
Fri, 31 Dec 1999 00:55:29 +0100

Steve Dodd wrote:
>
> On Thu, Dec 30, 1999 at 04:06:04AM +0100, Oliver Henning wrote:
>
> > well, not necessarily. If you would modify the spin_unlock asm sequence at
> > boot time, there would be no performance loss. But self-modifying code is
> > not very desirable, so...
>
> Erm, spin_unlock is inlined isn't it? So that won't work unless you want to
> grep for the code sequence all over the kernel and modules;

Lets ignore the modules, start with the main kernel. What about the
following idea:

* by default, "lock;btrl %0" is used.
* a new section for "self-modifying code pointer". The table could be
discarded after boot.
* each spin_unlock-macro adds an entry to the that section.

spin_unlock():
__asm__ __volatile__(

"1: lock; btrl %0"
".section .text.selfmodtable,ax"
".long &1" <<<<<<< pseudo code, but that should be possible
".previous"
...);

During early init, we walk through the .text.selfmodtable pointer table
and exchange all instructions with the optimal version. The only
limitation is that we cannot change the instruction length.

> also, the binary
> module people would be forced to brutally murder you :-)
>

Why? Their .text.selfmodtable section would be empty, i.e. no problems.

--
	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/