Tyan S4880

[YhLu]

Stefan,

I'm porting Tyan S4880 now. I found that the following building err, do you
meet it in AMD board?

Even I only enable the CPU0's memctrl in the auto.c

objcopy -O binary linuxbios linuxbios.strip
gcc -o buildrom /home/yhlu/xx/xx/freebios2/util/buildrom/buildrom.c
./buildrom linuxbios.strip linuxbios.rom ../../tg3.zelf 0x10000 0x18000
linuxbios image is 65854 bytes; only 65536 allowed
Linuxbios input file larger than allowed size!

Regards

YH.

[Stefan Reinauer]

* YhLu  [031125 03:37]:
> Stefan,
>
> I'm porting Tyan S4880 now. I found that the following building err, do you
> meet it in AMD board?

No, the AMD boards build fine for me. 
gcc-3.3.1-29
binutils-2.14.90.0.6-8
> Even I only enable the CPU0's memctrl in the auto.c

It seems the struct itself in auto.c is pretty uncritical for size.
Other parts like the log level need a lot more tuning if you want the
image to be as verbose as possible.

* Do you compile with romcc -O (not -O0 and not -O2)
* What's your max log level? The UP-build (Solo) allows me to 
  put MAXIMUM_CONSOLE_LOGLEVEL up to 10, other builds want less.
> objcopy -O binary linuxbios linuxbios.strip
> gcc -o buildrom /home/yhlu/xx/xx/freebios2/util/buildrom/buildrom.c
> ./buildrom linuxbios.strip linuxbios.rom ../../tg3.zelf 0x10000 0x18000
> linuxbios image is 65854 bytes; only 65536 allowed
> Linuxbios input file larger than allowed size!

Is this 64k including payload size? I use:
option ROM_SIZE=262144
option FALLBACK_SIZE=131072
option ROM_IMAGE_SIZE=0x10000

guess the size issue should be mentioned in the docs. Everybody
fiddling with LinuxBIOS on AMD64 _has_ seen it..

Stefan

Stefan Reinauer  writes:
> YhLu  [031125 03:37]:
> > Stefan,
> >
> > I'm porting Tyan S4880 now. I found that the following building err, do you
> > meet it in AMD board?
>
> No, the AMD boards build fine for me. 
> gcc-3.3.1-29
> binutils-2.14.90.0.6-8
> > Even I only enable the CPU0's memctrl in the auto.c
>
> It seems the struct itself in auto.c is pretty uncritical for size.
> Other parts like the log level need a lot more tuning if you want the
> image to be as verbose as possible.

I think that is more because changing the loglevel is something
we have real control over.  I think most of the bloat is in
auto.c but I have not gone back and looked for a while.

Before the memory controller setup was called in a loop it
made a real difference if you setup 1 memory controller or
2 because both calls were inlined.  With the loop the memory
controller code is only inlined once no matter how many memory
controllers you have.
> Do you compile with romcc -O (not -O0 and not -O2)

Hmm.  -O2 should work and better than -O.  I know it was
broken for a while...
> guess the size issue should be mentioned in the docs. Everybody
> fiddling with LinuxBIOS on AMD64 _has_ seen it..

Yes.  And this is why I keep working on making inlining optional
in romcc.

Eric