Re: ksyms_init considered too heavyweight for early bootstrap

[Michael <macallan%netbsd.org@localhost>]

Hello,

On Nov 17, 2008, at 5:29 AM, Martin Husemann wrote:

> I found out why a LOCKDEBUG sparc64 kernel crashes very early:
>
> "Traditionally" we have called ksyms_init() very early in MD  
> bootstrap(),
> even before running pmap_bootstrap(). Nowadays ksyms_init() does a  
> mutex_init(),
> which is, at least with LOCKDEBUG, a very heavy weight operation -  
> too much
> for the still very restricted runtime environment at this stage - on  
> sparc64
> neither curcpu nor curlwp are valid, and we have not yet setup our  
> trap
> handlers.
>
> Now I could solve this in a MD way (for example move the call to  
> cpu_startup),
> but I wonder if other archs run into the same problem (maybe with  
> less drastic
> failure mode) and the attached MI fix would be better?

I'm pretty sure arch/powerpc calls ksyms_init() either after or as the  
last thing in pmap_bootstrap().

have fun
Michael