On Mon, Jan 19, 1998 at 10:35:16AM +0100, Harald Koerfgen wrote:
> Can it be that this affects the stack manipulation in do_sys(...)?
> The DECstation kernel is now dying in the return from the first syscall
> in kernel_thread(...).
If you're getting that far without kludgology for the rest of the kernel
then things are actually looking quite good.
> From what I have been able to find out so far
> (with the help of two good old friends named printk and PRINT), the
> regs->cp0_epc += 4;
> in do_sys(...) is not visible after RESTORE_ALL in ret_from_syscall.
Can you give me some disassembler listing and point right to the program
counter where the epc seems to be unchanged again?
> > Suggested alternative strategy: build an ELF kernel
> > like all other systems do and convert it into ECOFF using the
> > ELF -> ECOFF converter program from the Milo sources. That's will be
> > a quite robust solution.
> I have done some experiments with that, but I am sorry to tell that this
> does not work :-(. It looks like the DECstation boot prom is very picky
> about the layout of kernel images. Maybe it's time to put some effort in
> a bootloader which is able to boot ELF kernels.
I should work, the converter program is based on Ted Lemon's from the
NetBSD sources for the pmax. You didn't use the ELF -> a.out converter
from the Milo sources by accident?