> Richard B. Johnson wrote:
> >
> > I just replaced the root file-system, utilities, and kernel in an old
> > '386SX (16MHz). Compiled with math emulation, etc.
> >
> > . . .
> >
> > The machine will hang. No interrupt activity whatsoever. Looking at
> > IRQ0 (timer-tick) on the bus shows it permanently high. Perhaps these
> > old machines can't use level-triggered interrupts. It worked with
> > an old version of Linux for several years. It still worked before I
> > installed 2.1.42. The old version was 0.99 (Yaggdrasl). Now I am
> > gonna be in trouble with the owner (an executive).
> >
> > I will probably have to give him a "new" motherboard with a '486-DX/66
> > to get him off my back (I have a old spare).
> >
> > I would suggest that support for '386s be disabled sometime soon,
>
> What I don't understand about this whole thread is why someone would
> blindly assume that a development kernel would work on anything
> without testing it first, especially to the exclusion of being able
> to go backwards.
The kernel installation process is something that __I__ do, not something
that gets done by magic. As I mentioned, I made a new system disk with
all new ELF binaries, etc. The present kernel, and its bug-fixes has
been in use here on over 20 machines.
> kernel so you can have it boot if the new one doesn't work, and if
> you lost that possibility, you could boot off the Yaggdrasl boot disk
> and re-install the old working kernel to salvage your mistakes rather
> than giving hardware away.
These are not my __MISTAKES__. The 0.99 kernel does not even like the
present version of the ext2 file-system. It is not supposed to be. There
have been many improvements over the past 6 years.
> I also think that you can't find any
> professional who wouldn't suggest doing a complete backup before any
> major changes to any system.
Not only was all the user's stuff backed up to a SCSI tape, but I also
have his original root file-system disk, i.e., a 330B slow-speed
Seagate. The backup was done first because I know that the old SCSI
disk has a different sector translation scheme than the new one so
I would not be able to retreive the old stuff except from tape.
I can always hand this VP back is favorite machine, saying; "I can't
make it be able to play the ELF binaries you get off the Internet".
I can also try to make the latest working stuff work on his machine.
Or, I can secretly replace his motherboard -- but his RS-232C lights
won't flicker on the front panel so he'll know I cheated. This
is an AST 386SX/16 that his son have him just before his son was
killed. You don't just throw these things away.
The whole idea was to use a complete __WORKING__ system disk with all
the latest tools, with all the latest networking improvements, and
all the latest X-Windows "User stuff", on the old machine. To assure
compatibility, the old hard-disk controller, and the old network card
were used and tested, complete --everything on a more modern motherboard.
The newer kernels simply do not work properly with a '386SX/16.. And
the old 0.99 kernel will not work with the new ELF binaries and the
improvements made in the file-systems (of course).
I would like to get the code to work on the old junker, in fact if
I leave out the Ethernet card (so it doesn't get pranged with a
lot of interrupts), change only the cascade interrupt to a specific
EIO in the kernel, I can keep the machine up long enough to rebuild the
kernel (six and a half hours!!!). However, If I were to make any
typing mistake so the "bell" rings, the machine halts.
Cheers,
DJ
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-
Richard B. Johnson
Analogic Corporation
Email : rjohnson@analogic.com, johnson@analogic.com
Penguin : Linux version 2.1.42 on an i586 machine (66.15 BogoMips).
Warning : It's hard to stay on the trailing edge of technology.
-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-=-