[ltp] Linux kernel instability? (Rant/Panic/Cry-for-help!)

[Richard Neill]

Henrique de Moraes Holschuh wrote:
> If you guys want any of us people closer to the kernel to even participate
> on threads as this, there is an EXTREMELY important information you must
> give:  whether you are using any binary blobs.  Also which kernel, but most
> of you provide that information already.

Dear Henrique,

Thank you for a very helpful and informative reply. I think there are 
several key points:

1. Binary Blobs are bad, and taint the kernel
Yes, on my desktop, I do use the nvidia driver, as there is no other way 
to get dual-head working. However, on the majority of the systems I run, 
I use the vesa (or sometimes nv) driver, because the nvidia one is such 
a pain.  My T60 thinkpad uses Vesa rather than fglrx. We aren't using 
WiFi anywhere. The server uses the intel onboard graphics driver. System 
reliability doesn't seem to be correlated to the binary blob use.

As far as I can tell, most of the bugs I've recently hit seem to be 
related to core parts of the kernel, eg SMP, NFS4, UnionFS, networking 
rather than to the fancy stuff (suspend, 3d, wifi).

2. Debugging.
I do try to submit bug reports when I can. However, when the system 
locks up, what can we do? The reset button is the only interface that 
still works!


3. Kernel versions.
Im using pretty well standard distro-released kernels (mostly 2.6.24 at 
the moment) for Ubuntu and Mandriva. Is it even possible to get a 2.6.16 
kernel on Ubuntu Hardy?  (from an official repository).


4. Production systems.
Most of these systems are "production" machines, eg office desktops. I'm 
a reasonably good sysadmin, but not a kernel programmer. I really do 
need a system that "just works", and "doesn't crash". It has to be 
distribution-supported in some way too.



> I don't use Ubuntu's patched kernel (I'd use either Fedora's or SuSE's, they
> DO have a team large enough to take care of it.  Ubuntu doesn't). 
> 
> This doesn't mean the quality of mainline's core is not changing for the
> worst, it really might be.  But it DOES mean that there is such a thing as
> "playing it safer"...
> 

Is there any way to find that guaranteed subset of "playing safe" 
features? I love being able to write shell scripts, customise the 
desktop, etc, but is it too much to ask that I could just be an 
"end-user" of the kernel, and not have to worry about it?

Is, for example, Fedora really more stable than Ubuntu? They use the 
same kernel, after all. Are the various "enterprise" distros really 
better tested and debugged? I don't care about (most) features not being 
implemented; what I do care about is that the implemented features work 
and don't crash.

Richard



P.S. Anyone folloing one of the earlier points I raised (about BSD kerne 
  + Linux userland)  might be interested to try this:
http://glibc-bsd.alioth.debian.org/ging/