Status: - working fallback linuxbios that was built on redhat 7.3 w/ gcc 2.96 (bogus redhat version :-) - trying to get a supermicro p4dpe to work on redhat 8.0 - kernel+initrd is loaded from IDE-flash via etherboot. This is a bproc "phase 1" kernel, and will download and boot a "phase 2" kernel. The phase 1 kernel is non-smp, and should use PIRQ tables or similar to set up IRQs, it turns out. It appears not to be MP-table aware. The phase 2 kernel is SMP- and MP-table aware.
If I build linuxbios with gcc 3.2, it loads the phase 1 kernel but ... that kernel can't find any interrupts for the myrinet card. If I build linuxbios with gcc 2.96, that kernel can find interrupts and download the phase 2 kernel. The phase 2 kernel finds the MP table and gets everything ready to go, but panics when trying to set up the initrd for reasons that are not clear. It seems to think it has 4 GB memory, when it fact it has 1 GB memory with a big memory hole. Can Linux 2.4.19 handle this kind of sparse memory map? right now, experience is saying "no", although I thought this was working.
Now I think I need to look at the PIRQ table as compiled by gcc 3.2. The MP table seems to be created just fine. Interesting.
ron
I had similar trouble with 2.4.19 on a Tyan E7500 motherboard and RH80. The stock RH 2.4.18 wouldn't work either and none of the pre 2.4.20's I tried would work. I could only get 2.4.18 to run stable. I used the gcc that came with RH80 kernel builds I think, though I've since downgraded to the previous clean release of gcc. I also had to get the latest unrelease BIOS from Tyan so the BIOS was at least part of the problem. Its not a machine I can play with to see if 2.4.20 works or if building 2.4.19 with different version of gcc would have worked.
GO
On Tuesday 03 December 2002 00:12, Ronald G. Minnich wrote:
Status:
- working fallback linuxbios that was built on redhat 7.3 w/ gcc 2.96 (bogus redhat version :-)
- trying to get a supermicro p4dpe to work on redhat 8.0
- kernel+initrd is loaded from IDE-flash via etherboot. This is a bproc "phase 1" kernel, and will download and boot a "phase 2" kernel. The phase 1 kernel is non-smp, and should use PIRQ tables or similar to set up IRQs, it turns out. It appears not to be MP-table aware. The phase 2 kernel is SMP- and MP-table aware.
If I build linuxbios with gcc 3.2, it loads the phase 1 kernel but ... that kernel can't find any interrupts for the myrinet card. If I build linuxbios with gcc 2.96, that kernel can find interrupts and download the phase 2 kernel. The phase 2 kernel finds the MP table and gets everything ready to go, but panics when trying to set up the initrd for reasons that are not clear. It seems to think it has 4 GB memory, when it fact it has 1 GB memory with a big memory hole. Can Linux 2.4.19 handle this kind of sparse memory map? right now, experience is saying "no", although I thought this was working.
Now I think I need to look at the PIRQ table as compiled by gcc 3.2. The MP table seems to be created just fine. Interesting.
ron
Linuxbios mailing list Linuxbios@clustermatic.org http://www.clustermatic.org/mailman/listinfo/linuxbios
-
GNUOrder -
Ronald G. Minnich