Skip to contentSkip to conversation

a pretty big change

parisc-linux Paul Bame 3 participants 7 messages Dec 4, 1999 – Dec 12, 1999

lore
Share & tools
lore

Paul Bame

author
#0 lore raw

I just committed the changes to embed real-mode code in a separate executable inside vmlinux. It should cause no functional changes in theory. Read all about it in arch/parisc/real/README and ee.html. If you have terrible new problems,
I've attached a list of CVS commands to check out the files before I started hacking on them. I haven't seen a shell prompt from this set of hacks so beware...

-Paul Bame

prumpf- I moved kernel_thread and friends from
real/<something> to kernel/head.S I also re-enabled BTLB in real/setup.c so I could get further in the boot. Left some debugging printks in the handle_real_interrupt handler too.

cvs update -r1.25 arch/parisc/Makefile
cvs update -A arch/parisc/Makefile
cvs update -r1.28 arch/parisc/kernel/Makefile
cvs update -A arch/parisc/kernel/Makefile
cvs update -r1.35 arch/parisc/kernel/head.S
cvs update -A arch/parisc/kernel/head.S
cvs update -r1.39 arch/parisc/kernel/irq.c
cvs update -A arch/parisc/kernel/irq.c
cvs update -r1.14 arch/parisc/kernel/pdc.c
cvs update -A arch/parisc/kernel/pdc.c
cvs update -r1.25 arch/parisc/kernel/pdc_cons.c
cvs update -A arch/parisc/kernel/pdc_cons.c
cvs update -r1.73 arch/parisc/kernel/setup.c
cvs update -A arch/parisc/kernel/setup.c
cvs update -r1.11 arch/parisc/kernel/traps.c
cvs update -A arch/parisc/kernel/traps.c
cvs update -r1.2 arch/parisc/real/Makefile
cvs update -A arch/parisc/real/Makefile
cvs update -r1.2 arch/parisc/real/head.S
cvs update -A arch/parisc/real/head.S
cvs update -r1.2 arch/parisc/real/interruption.S
cvs update -A arch/parisc/real/interruption.S
cvs update -r1.2 arch/parisc/real/lib.c
cvs update -A arch/parisc/real/lib.c
cvs update -r1.2 arch/parisc/real/pdc.c
cvs update -A arch/parisc/real/pdc.c
cvs update -r1.2 arch/parisc/real/real.h
cvs update -A arch/parisc/real/real.h
cvs update -r1.2 arch/parisc/real/real_s.S
cvs update -A arch/parisc/real/real_s.S
cvs update -r1.2 arch/parisc/real/setup.c
cvs update -A arch/parisc/real/setup.c
cvs update -r1.1 arch/parisc/real/traps.c
cvs update -A arch/parisc/real/traps.c
cvs update -r1.1 include/asm-parisc/real.h
cvs update -A include/asm-parisc/real.h

author
#1 lore raw

Bad failures occur (for me) with the hpux boot loader. I have no idea whether I introduced them, but I might have. It looks like the first instruction of vmlinux doesn't even get executed, so probably something in the SOM header is telling the hpux boot loader to do something stupid. som_relocate may need an upgrade, but I'm stumped about what's going wrong.

	-P
#6 lore raw
In reply to Paul Bame

I read ee.html this evening. Having already tracked down one problem where the real-mode code went off into never-never land when I tried to print out a string, I agree that separate comilation, linking and conversion is a good idea. The size of real-mode code is limited with our current scheme. Thus, we may reach the point where it is impossible to link all real-mode code.

Regarding your example, the problem is that a function address is a PLABEL. PLABELs are stored in the text space. I think the problem can be avoided if the function address is put into a structure which is stored in the data space.

Quoted 57 lines

I just committed the changes to embed real-mode code in a separate executable inside vmlinux. It should cause no functional changes in theory. Read all about it in arch/parisc/real/README and ee.html. If you have terrible new problems,
I've attached a list of CVS commands to check out the files before I started hacking on them. I haven't seen a shell prompt from this set of hacks so beware...

-Paul Bame

prumpf- I moved kernel_thread and friends from
real/<something> to kernel/head.S I also re-enabled BTLB in real/setup.c so I could get further in the boot. Left some debugging printks in the handle_real_interrupt handler too.

cvs update -r1.25 arch/parisc/Makefile
cvs update -A arch/parisc/Makefile
cvs update -r1.28 arch/parisc/kernel/Makefile
cvs update -A arch/parisc/kernel/Makefile
cvs update -r1.35 arch/parisc/kernel/head.S
cvs update -A arch/parisc/kernel/head.S
cvs update -r1.39 arch/parisc/kernel/irq.c
cvs update -A arch/parisc/kernel/irq.c
cvs update -r1.14 arch/parisc/kernel/pdc.c
cvs update -A arch/parisc/kernel/pdc.c
cvs update -r1.25 arch/parisc/kernel/pdc_cons.c
cvs update -A arch/parisc/kernel/pdc_cons.c
cvs update -r1.73 arch/parisc/kernel/setup.c
cvs update -A arch/parisc/kernel/setup.c
cvs update -r1.11 arch/parisc/kernel/traps.c
cvs update -A arch/parisc/kernel/traps.c
cvs update -r1.2 arch/parisc/real/Makefile
cvs update -A arch/parisc/real/Makefile
cvs update -r1.2 arch/parisc/real/head.S
cvs update -A arch/parisc/real/head.S
cvs update -r1.2 arch/parisc/real/interruption.S
cvs update -A arch/parisc/real/interruption.S
cvs update -r1.2 arch/parisc/real/lib.c
cvs update -A arch/parisc/real/lib.c
cvs update -r1.2 arch/parisc/real/pdc.c
cvs update -A arch/parisc/real/pdc.c
cvs update -r1.2 arch/parisc/real/real.h
cvs update -A arch/parisc/real/real.h
cvs update -r1.2 arch/parisc/real/real_s.S
cvs update -A arch/parisc/real/real_s.S
cvs update -r1.2 arch/parisc/real/setup.c
cvs update -A arch/parisc/real/setup.c
cvs update -r1.1 arch/parisc/real/traps.c
cvs update -A arch/parisc/real/traps.c
cvs update -r1.1 include/asm-parisc/real.h
cvs update -A include/asm-parisc/real.h

---------------------------------------------------------------------------
To unsubscribe: send e-mail to [email protected] with `unsubscribe' as the subject.

Signature · 2 lines
J. David Anglin                                  [email protected]
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)
#2 lore raw
In reply to Paul Bame

After a day of hacking, I was finally able to link the kernel successfully. It booted through to the end of realmode_setup. Probably, further work is needed on the TLB handler. I wonder if calls between real and virt, and visa versa are working. No change to som_relocate was necessary.

There is one problem. I now have to turn off the system power to get the system to reboot. Formerly, I could just press the TOC button.

It also looks like the "real_" defines and protos can be eliminated.

Dave

Quoted 16 lines

Bad failures occur (for me) with the hpux boot loader. I have no idea whether I introduced them, but I might have. It looks like the first instruction of vmlinux doesn't even get executed, so probably something in the SOM header is telling the hpux boot loader to do something stupid. som_relocate may need an upgrade, but I'm stumped about what's going wrong.

	-P

---------------------------------------------------------------------------
To unsubscribe: send e-mail to [email protected] with `unsubscribe' as the subject.

Signature · 2 lines
J. David Anglin                                  [email protected]
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)
author
#3 lore raw

= After a day of hacking, I was finally able to link the kernel successfully. = It booted through to the end of realmode_setup. Probably, further work is = needed on the TLB handler. I wonder if calls between real and virt, and = visa versa are working. No change to som_relocate was necessary.

Glad you had better luck with the HPUX boot loader than I had.

= There is one problem. I now have to turn off the system power to get the = system to reboot. Formerly, I could just press the TOC button.

That happens pretty frequently to me but I've not tracked down why it happens. Did you try holding the TOC button in for several seconds? On some workstations I've been told that generates a hard reset.

= It also looks like the "real_" defines and protos can be eliminated.

As I merge in your changes I'll look into that.

	-P
#4 lore raw
Quoted 3 lines

That happens pretty frequently to me but I've not tracked down why it happens. Did you try holding the TOC button in for several seconds? On some workstations I've been told that generates a hard reset.

TOC results in (providing that the checksum passes) transfer of control to a routine pointed at from page 0. If something is trashing that, and the checksum'ed length is set to 0, then you'll branch off into the ozone.

If the checksum fails, then you get the PDC default of rebooting.

Don't know if that's your problem, but it's probably worth checking into whether something is trashing physical address 8 (TOC vector - 0xc is the length of the checksum.)

lamont

#5 lore raw
Quoted 9 lines
Quoted 4 lines

That happens pretty frequently to me but I've not tracked down why it happens. Did you try holding the TOC button in for several seconds? On some workstations I've been told that generates a hard reset.

TOC results in (providing that the checksum passes) transfer of control to a routine pointed at from page 0. If something is trashing that, and the checksum'ed length is set to 0, then you'll branch off into the ozone.

Quoted 5 lines

If the checksum fails, then you get the PDC default of rebooting.

Don't know if that's your problem, but it's probably worth checking into whether something is trashing physical address 8 (TOC vector - 0xc is the length of the checksum.)

I believe that I have located the cause of the problem. The real mode code in the kernel startup used the stack pointer passed from the system loader. The initial stack pointer passed by the hpux loader is 0. This caused some items in PAGE0 to be trashed, although it is not clear that the TOC vector itself was being trashed.

Dave

Signature · 2 lines
J. David Anglin                                  [email protected]
National Research Council of Canada              (613) 990-0752 (FAX: 952-6605)