Alan Cox wrote:
>
> > distributions: we had to re-design LILO, because the Phoenix BIOS used
> > the upper end of base memory for storing the SMP config table (ask Zach
> > Brown), we had to work on the memory management initialization because
> > of the same problem, another bug in the mm initialization prevented the
> > system from using base memory (beats me why this hasn't surfaced
> > earlier). We had to build custom installation floppy images, kernel
>
> Basically they didnt show up because _only_ Fujitsu/Siemens use that paticular
> weird memory layout. You need a fairly specific setup where SMM mode
> code that uses EBDA space.
AFAIK Zach Brown had this problem too but not with a (then) Siemens
machine. And it wasn't related to SMM mode: the BIOS stored the SMP
config table in the EBDA at the end of base memory. When the second
stage LILO was loaded, it happily over-wrote part of the SMP config
table sending the kernel into nirwana during boot (BTW The fix, although
accepted by Werner, never made it into an official LILO release. SuSE
and RedHat both accepted my patches and put them onto their
distributions.)
Perhaps it was a generic Phoenix BIOS problem? At least Intel's SMP spec
(v1.4) specifically allowed this setup. BTW Werner Almesberger also
confirmed that he anticipated such a problem and that was the reason why
he left the top 4K unused. Howver, in certain configurations the EBDA
grew beyond proportion and mre tha 4K were allocated for it and the SMP
config table started below 636K.
> Free software is good at being reliable - but its a testing/review/fixing
> scheme that is nothing like as effective at finding situations where the code
> works for everyone currently but isnt quite to specification.
Please don't get me wrong. I wasn't complaining about the things that
happened in the past. I was merely trying to explain my (and our
managers') impatience:
Our customers want to run Linux. But at present, our management is
reluctant to invest into Linux. Therefore I have to try and minimize the
effort we have to spend to run Linux on our machines. Hence my desire to
get this bug (and Justin Gibbs confirmed that it _is_ a bug, although, I
admit, the fix he sent me obviously doesn't cover all instances) fixed
before SuSE and RedHat (the only two distributions we officially
support) start shipping 2.4-kernels which require a similar set of
patches as did the aforementioned problems. I already have to provide
boot floppy images, precompiled kernel-modules and kernel source patches
for SuSE 6.1, 6.2, 6.3, 6.4 and RedHat 6.1 and 6.2.
--
Josef M�llers
Fujitsu Siemens Computers
SHV Server DS 1
-
To unsubscribe from this list: send the line "unsubscribe linux-scsi" in
the body of a message to [EMAIL PROTECTED]
