[parisc-linux] Re: Getting up and running

[parisc-linux] Re: Getting up and running

[John Marvin]

>
> How does this look?  I can't test it because I don't have a 720-era machine,
> but it ought to work, I think...
>

<Patch Deleted>

No, it needs to be a little more complicated than that. That patch doesn't
take into account byte and half-word accesses to a page. The real fix
needs to look at the instruction in the iir register in combination with
the ior value in order to determine whether the access is unaligned.

I thought for a second that the patch might work anyway, since the
unaligned fault handler has exception table support, so even though
the instruction might be unecessarily emulated, it would still
successfully bring the page in, but the unaligned fault handler
uses ldb, which would just cause an infinite loop ...

So, the correct fix should probably involve leveraging some of the
macros in unaligned.c and writing something like check_unaligned(regs)
that would be used by the trap 18 handler to determine whether or
not the access was unaligned or not.

John

Re: [parisc-linux] Re: Getting up and running

[Jochen Friedrich]

Hi John,

> I thought for a second that the patch might work anyway, since the
> unaligned fault handler has exception table support, so even though
> the instruction might be unecessarily emulated, it would still
> successfully bring the page in, but the unaligned fault handler
> uses ldb, which would just cause an infinite loop ...

init started:  BusyBox v0.51 (2001.06.27-21:43+0000) multi-call binary
rcS(14): unaligned access to 0x000411e5 at ip=0x0001a04f
Not-handled unaligned insn 0x60a0002a
Unaligned handler failed, ret = -1 #
# ps -ax
<4>sh(16): unaligned access to 0x000411e5 at ip=0x0001a04f
Not-handled unaligned insn 0x60a0002a
Unaligned handler failed, ret = -1 Bus error
# ls -la
<4>sh(17): unaligned access to 0x000411e5 at ip=0x0001a04f
Not-handled unaligned insn 0x60a0002a
Unaligned handler failed, ret = -1 Bus error

I had more success using the old suggested patch:

"Our" CPUs are very old. So there must be a line added in
the arch/parisc/kernel/trap.c file. In the handle_interruption()
you must add the line
        `case 18:'
where
        `case 15:'
        `case 26:'
is in the code, because these old CPUs use an older method of
"reporting" page faults, etc...

It looks like the expression "if (regs->ior & 3)" might be wrong here...

Cheers,
Jochen

[parisc-linux] installing a native built kernel with 2.4.0 (.92 dist).

[marka]

I've successfully installed and upgraded with the .92 distro, but I'm having
trouble getting a natively built kernel with this distro to load.

the load will either freeze at 'branching to kernel entry point' or
on the last time I rebuilt and copied, it complained:

Couldn't grok your kernel executable format
failed to load kernel

this is with kernel-latest and palo-latest from Sep 7 on a b180l

I select PA7300LC on 'make oldconfig' (this sets CONFIG_PA7100LC=y in
.config).

Any tips from the pros on building and installing -latest natively?  I'd
really like to get the fixes in 2.4.9, and there must be something more
interesting that needs to be done than copying/renaming lifimage over to /boot
and updating palo.conf and re-running palo.

-mark

Re: [parisc-linux] installing a native built kernel with 2.4.0 (.92 dist).

[Richard Hirst]

On Mon, Sep 17, 2001 at 03:13:40PM -0700, marka@sea.checkpoint.com wrote:
> 
> I've successfully installed and upgraded with the .92 distro, but I'm having
> trouble getting a natively built kernel with this distro to load.
> 
> the load will either freeze at 'branching to kernel entry point' or
> on the last time I rebuilt and copied, it complained:
> 
> Couldn't grok your kernel executable format
> failed to load kernel

If you are netbooting, you use a lifimage, if you are booting
from disk you use the vmlinux file produced from building the
kernel.

> this is with kernel-latest and palo-latest from Sep 7 on a b180l
> 
> I select PA7300LC on 'make oldconfig' (this sets CONFIG_PA7100LC=y in
> .config).

The config we use for the official kernels is in arch/parisc/debian-configs/32,
you can "make distclean; cp arch/parisc/debian-configs/32 .config;
make oldconfig; make; cp vmlinux /boot/vmtest", then reboot, interact
with IPL and change the vmlinux-whatever name to be vmtest.

> Any tips from the pros on building and installing -latest natively?  I'd
> really like to get the fixes in 2.4.9, and there must be something more
> interesting that needs to be done than copying/renaming lifimage over to /boot
> and updating palo.conf and re-running palo.

Don't copy lifimages to /boot, copy vmlinux.

Richard

Re: [parisc-linux] installing a native built kernel with 2.4.0 (.92 dist).

[marka]

Thanks for your help,

Ok, it turns out it is booting (the machine's so quiet it's hard to tell), and
not 'freezing' per-say at 'branching to kernel entry point 0x00100000'.  It's
simply not sending anything to ttyS0 any more after this point.  This seems to
be the 'swapped serial ports' issue mentioned earlier on this list.

--
Mark

Re: [parisc-linux] installing a native built kernel with 2.4.0 (.92 dist).

[Matthew Wilcox]

On Tue, Sep 18, 2001 at 10:10:43AM -0700, marka@sea.checkpoint.com wrote:
> Ok, it turns out it is booting (the machine's so quiet it's hard to tell), and
> not 'freezing' per-say at 'branching to kernel entry point 0x00100000'.  It's
> simply not sending anything to ttyS0 any more after this point.  This seems to
> be the 'swapped serial ports' issue mentioned earlier on this list.

That could be.  Can you boot the old kernel and report
what the serial ports are as Richard did in his message
http://lists.parisc-linux.org/pipermail/parisc-linux/2001-September/013670.html

Or if your machine boots as far as having a network, you could check in
dmesg to see what the inventory reported and paste it from there.

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] installing a native built kernel with 2.4.0 (.92 dist).

[Richard Hirst]

On Tue, Sep 18, 2001 at 10:10:43AM -0700, marka@sea.checkpoint.com wrote:
> Thanks for your help,
> 
> Ok, it turns out it is booting (the machine's so quiet it's hard to tell), and
> not 'freezing' per-say at 'branching to kernel entry point 0x00100000'.  It's
> simply not sending anything to ttyS0 any more after this point.  This seems to
> be the 'swapped serial ports' issue mentioned earlier on this list.

That should be fixed in the latest kernel source, please let us know
if it is still broken on your box.

Thanks,
  Richard

Re: [parisc-linux] installing a native built kernel with 2.4.0 (.92 dist).

[marka]

On Tue, 18 Sep 2001, Richard Hirst wrote:
> That should be fixed in the latest kernel source, please let us know
> if it is still broken on your box.

lo and behold it is fixed, thanks! broken in pa17 and fixed in pa22.

Serial driver version 5.05c (2001-07-08) with MANY_PORTS SHARE_IRQ SERIAL_PCI
enabled
ttyS00 at iomem 0xffd05800 (irq = 154) is a 16550A
ttyS01 at iomem 0xfff83800 (irq = 74) is a 16550A

--
Mark