Re: kern/52796: repeated modload panics netbsd-7

[Paul Goyette <paul%whooppee.com@localhost>]

On Sat, 9 Dec 2017, Valery Ushakov wrote:

> >  The "symbol redefined" error messages you see on NetBSD-8 are
> >  actually a result of the fix for this PR!
>
> I suspected as much, but there's the second (first) part to this
> problem too.
>
> On NetBSD-6 modload name and modload path behave the same:
>
>   nb6# modload nand
>   modload: File exists
>   nb6# modload /stand/i386/6.1/modules/nand/nand.kmod
>   modload: File exists
>
> On 7 and later modload name fails with EEXIST right away without
> actually trying to load the module.  But modload path does try to load
> the module (which causes bad things on 7 and is caught by the kernel
> linker on 8).  Shouldn't modload path detect EEXIST without trying to
> load the module too, like it did on 6?

Unfortunately, the kobj magic (in src/sys/kern/subr_kobj.c) that handles 
symbol relocation runs as soon as the module file is read into memory, 
and that is where the linkset information (required to obtain the name 
and other stuff for the modinfo_t structure) gets identified.  It would 
require some surgery on the kobj stuff to separate section processing 
from symbol/relocation processing.

I'm not quite sure how/why this part worked on netbsd-6


+------------------+--------------------------+----------------------------+
| Paul Goyette     | PGP Key fingerprint:     | E-mail addresses:          |
| (Retired)        | FA29 0E3B 35AF E8AE 6651 | paul at whooppee dot com   |
| Kernel Developer | 0786 F758 55DE 53BA 7731 | pgoyette at netbsd dot org |
+------------------+--------------------------+----------------------------+