Willem Konynenberg wrote:
> > I'm a poor boy :-) , so I'm allowed to ipl Linux only on a LPAR'ed G3 box with
> > only a 3172 as network device (neither OSA, nor CTC (dedicated to production)).
>
> What kind of performance does that give you?
> (I want to compare to my Hercules box. ;-)
It depends. Linux Lpar have 10% of total CPU, dynamically assigned.
I see even 4~ BOGOMips, sometimes... (wow!), but that number varies depending on the
flavour of linux booting... (Redhat is the most generous...)
But the actual point is: "my" G3 is a "dead box walking".
In the last two years most of applications migrated from the blue iron box to plastic
boxes ;-) running unix (hpux and solaris) and in June 2002 the switch will be turned
off forever; my firm doesn't intend to invest on IBM mainframe hardware, of course.
Unless, meantime, something new happens, like e.g. a unix system running on s390,
with many many many tools _free_...
>
>
> > Question 1.
> > Suse and Turbolinux distributions are supported only on G5 or G6 machines
> > (unlike intel flavours, which should work also with ancient 386!).
>
> That's partly because the the Intel implementation of Linux was done by
> users of various kinds of old and new PCs. Had it been done by Intel,
> the result might have been different. ;-)
I agree...
>
>
> > Does Red Hat for s390 will have the same limitations?
>
> It's primarily a limitation of the GCC compiler tool chain, which expects
> certain recent additions to the instruction set.
So, if I get an older version of GCC, I'll be able to compile a 2.4.x kernel for G3
machines?
>
> The "hacked version" that you refer to ("the vintage patches") is basically
> a hacked version of the compiler toolchain, with a filter between the code
> generation and the assembler step to replace all the new instructions with
> equivalent older instructions.
> I think that deals with most of the issues, although there may be a
> few extra tweaks in the kernel source itself as well.
BTW: the address for vintage is: http://penguinvm.princeton.edu/programs/vintage/
>
> > Question 2.
> > lcs.o (the network driver for "lan channel stations" like 3172) is still
> > distributed as OCO (Object Code Only) also under SRPMS directory. Is it
> > permitted by GPL license?
>
> It's not permitted by the GPL, but it is permitted by the "Linus exception".
> I.e. Linus has explicitly stated that he consider binary modules legally
> acceptable, although undesirable enough that he won't make any effort
> whatsoever to ensure backward binary compatibility.
> If a vendor wants to distribute binary modules, they'll have to keep
> track of every single kernel upgrade to ensure that their module still
> works.
>
> Meanwhile, I have seen indications that IBM's policy on this is slowly
> changing, and a source release of an lcs driver might occur some time
> in the future.
> > (the true question is: there will be some way to be sure the instructions of
> > lcs.o are supported by pre G4 cpu?)
>
> Not easily if you only have the binary module.
> The vintage hack operates on assembler source, I believe.
>
>
> --
> Willem Konynenberg <wfk@xos.nl>
begin:vcard
n:Bellussi;Giorgio
tel;fax:++39-049-8288406
tel;work:++39-049-8288506
x-mozilla-html:TRUE
org:Produzione
version:2.1
email;internet:giorgio.bellussi@infocamere.it
adr;quoted-printable:;;InfoCamere S.C.p.A.=0D=0AC.so Stati Uniti 14;Padova;;35127;Italia
x-mozilla-cpt:;21536
fn:Giorgio Bellussi
end:vcard
