On Mon, Aug 16, 1999 at 08:33:15PM +0400, Vladimir A. Roganov wrote:
> We have some problems with modules loading in our Linux/R3k.
> The picture is following:
>
> 1. For modutils-2.1.121 rebuilt for linux-2.2.1/r3k any
> attempt to insmod something lead to insmod segfault.
> The line where it dies is obj/obj_mips.c:184:
>
> assert(v == l->value);
>
> The value of 'l' here is a (struct mips_hi16 *) 0x41
Not good :-)
> 2. For modutils placed in HardHat instimage the command
> 'insmod <object>' takes infinite time (locking kernel
> somewhere), but kernel looks alive (console input keys).
That one was definately buggy ...
> The modules we are testing is mtd physmem device: they works
> well for Intel and looks very platform-independent
> (I don't think that the problem located inside them).
>
> The behaviour is same for cross-compiled and native-compiled
> objects.
Yep. Btw, the modutils code has #ident ops which will be missompiled
by older compilers. Make shure you've got one of my patches for egcs 1.0.3.
They'll build this correctly. Alternatively just remove the #idents.
> Due it looks just like a some mismatch in our kernel, modutils
> and compiler versions, we are very interesting to get info
> about working combination for all above
> (or about new versions with more fresh and actual bugs :-)
Try to get the current modutils via anon-cvs from vger as described in the
FAQ, does it help?
Ralf
|