Re: [PATCH]The idea about multi-threaded initcall

[Greg KH <greg@xxxxxxxxx>]

On Thu, May 03, 2007 at 02:02:32PM +0800, Yang Sheng wrote:
>  Hi,
> 
>  Thanks for the notice!
> 
>  I am a new kernel developer(also a kernel newbie :-) ) and working on a
>  multi-threaded initcall patch in order to speed up the boot stage now. This
>  idea provided by my mentor.

I don't think it will really speed up things, boot is still limited by
I/O speed, not kernel startup execution speed.  I implemented
multi-threaded device probing (which caused all sorts of problems and
had to be ripped out in the end), and it only caused a very small
increase in speed.

To speed up boot time, you need to attack the issues _after_ init starts
up, not before.

>  The basic idea is using kthread with initcall and register a
>  wait_for_initcall() with xxx_initcall_sync() which introduced by Andrew
>  Morton for multi-threaded probing <http://lkml.org/lkml/2006/10/27/157> used
>  as barrier for each level.
> 
>  I supposed the initcalls in same level can be run in parallel. I know this
>  maybe not true, but I wrote this patch and tried.

It is not true, there are a lot of dependencies on the link order of
these initcalls.  You will run into troubles.

Good luck,

greg k-h

--
To unsubscribe from this list: send an email with
"unsubscribe kernelnewbies" to ecartis@xxxxxxxxxxxx
Please read the FAQ at http://kernelnewbies.org/FAQ