Re: Potentially serious problem with kernel-headers...
Monty <xiphmont@MIT.EDU> writes:
> >Experimentally: does the new binary work on the old kernels?
>
> Not tried this. I can't reproduce the problem locally. zif? Dale?
It appears to, actually. I haven't tried anything extensive but it doesn't
blow up immediately like it does the other way around.
> It's clean in CodeCenter and Electric Fence. Note that this problem
> happens *very* early on (first malloc).
The other type of bug it could be is that it's "#include"ing a "private"
header in the kernel. Those are allowed to change between minor releases,
I presume? I haven't looked for this - nor do I have the expertise to find
this kind of bug. I'm not a kernel wonk :-)
> >If not, and if the version compiled on kernel 2.0.34 includes works
> >on version 2.0.32 I'm inclined to suspect a libc bug.
>
> The experiment is worth running. I'll bet the package system is going
> to bitch though :-) I will stake a sizable chunk of my reputation on
> this one not being libc's fault. I have maligned libc quality control
> mightily in the past, but not this time, not yet.
:-)
Like I said, it looks like the binary compiled with kernel headers from
2.0.34 will work under 2.0.31. I've successfully sampled a song this way
(it did sample REALLY slow, though. Sampling a 5 minute song took something
like 15 minutes, with no indication of a problem with the CD istelf.) If I
use headers from 2.0.32 with kernel 2.0.34 running, it blows up immediately
while trying to query the drive.
It seems to me that we might just be really lucky that new binary works on
old kernels, and that this could be evidence of a much larger problem.
Later,
Dale
--
+-------------------- finger for pgp public key ---------------------+
| Dale E. Martin | Clifton Labs, Inc. | Senior Computer Engineer |
| dmartin@clifton-labs.com | http://www.clifton-labs.com |
+----------------------------------------------------------------------+
--
To UNSUBSCRIBE, email to debian-devel-request@lists.debian.org
with a subject of "unsubscribe". Trouble? Contact listmaster@lists.debian.org
Reply to: