I've been out of town so my mail is piling up horribly, but I'll
take a minute for this:
> I suspect that the real fix is in fact to rethink the module
> versioning.
I 100% agree.
For instance, here is a problem: kernel/ksyms.c has a definition of
__export_Using_Versions conditionalized on MODVERSIONS. This symbols is
not defined for resident files such as kernel/ksyms.c. So the resident
kernel never defines __export_Using_Versions and insmod never sees it.
I believe this bug is easily fixed by changing the MODVERSIONS test
to CONFIG_MODVERSIONS. I don't think module versioning is broken
because it has a hard-to-spot, easy-to-fix bug. I think it's broken
because it's been like this since about 2.1.20-something and no one
has noticed that the symbol versions are not working.
Or, of course, an alternative explanation is that my analysis is flawed.
I've been wrong in public before. :) I would welcome someone checking
my logic and pointing that out to me.
> However, I prefer to follow up my suggestions with at least a snippet
> of code, and have no clue how the version system works ATM, (or doesn't
> work, as the case may be).
For a short-term fix, consider using of the genksyms -p flag.
Generate a prefix that encodes the SMP state, the CONFIG_MAX_MEMSIZE state,
and any other state that is a tacit part of the kernel/module interface.
Regards,
Michael Chastain
<mailto:mec@shout.net>
"love without fear"
-
To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
the body of a message to majordomo@vger.rutgers.edu