[Sussex] Kernel 2.5.47 Problems

Steve Dobson SDobson at manh.com
Mon Nov 18 11:03:00 UTC 2002


On 11/18/2002 Geoff Teale wrote:
> Morning Steve :)

Morning Geoff;

> Steve wrote:
<snip> 
> > Did you get a core dump from tuxracer?  If so did you check
> > it?  My guess is that you were out of memory at the time and
> > malloc() returned a zero which was not checked for my the
> > tuxracer developer.  If my guess is correct: "Bad texracer
> > developer!!!"
>
> Actually, you have the right problem but the wrong developer.
> It was libPNG that was to blame.
>
> Frankly, I wasn't that fussed (although I send a bug report in)
> - I wouldn't have even attempted to play a fast-3d game along
> side a large compilation and a mail client on Windows..

I glad you sent in the bug report.  In my view it is a capital 
crime not to check the return code of a route.  Bad when done
in an application but very, very poor to be in a library.

> > Hold on a momement - You're compiling a development kernel 
> > with GCC 3.2!!!!!
>
> Yup.. I'm a speed freak ;)
>
> > Major risk. The kernel team don't support 3.2 yet.  In fact
> > the kernel build (so I've been told) does build/work if you
> > turn off or change some of the optomisation options from
> > 2.95.4.  Have you ruled out the compiler as being the
> > problem here?  If you haven't I suggest to try compiling
> > against GCC 2.95.4 and report back.
>
> Answer.  
> 
> Already tried that, first thing I did was switch the link in
> /usr/bin to point to the 2.95 binary (this ultimately would
> cause other problems but I wanted to eliminate it), the result
> is the same.

Hmmmmm....

> I agree GCC 3.2 makes life a little more "fun" but shouldn't
> be the problem.  I have succesfully earlier 2.5 kernels with
> 3.2 and I know several gentoo people have compiled and run
> 2.5.47 under 3.2.

I agree it shouldn't be an issue, but some times it is, esp. in
kernel code.  Some of the kernel hackers are bright little sparks
and exploit "features" of the compiler.

> Really what I need to know is what happens with loading the
> kernel prior to it producing any output - something is hanging
> it very quickly.

As I recall Linux uses a two stange load processes (FreeBSD uses
three).  The boot loader (lilo/grub) is configured with the
physical location on disk of the kernel image.  It loads this
into memeory and then jumps to it start address.

The first think the kernel does is report it's version and then
start de-compressing itself into kernal land.  Do you see the
      Linux 2.5.47.........................
I've always thought that the dots are the de-compressiong running.
The next line after than it the kernel starting up and finding and
configuing the hardware.

> It's not really that important.. ...I only pulled it because
> I'd read it was up to 3 times faster in some core functions,
> and like I said, I'm a speed junkie.. ..I need to be to keep
> my old P3 performing OK in the face of ever expanding software.

So all this is because "you feel the need, the need for speed"!
So when are you going to but that Cray and have it runing Linux?

Steve




More information about the Sussex mailing list