[parisc-linux] JFS on PA-RISC-Linux

[parisc-linux] JFS on PA-RISC-Linux

[Peter Weilbacher]

Hi!

Did anyone try JFS on other architectures than iX86 (the 1.0.0
announcement did only restrict it to 2.4.X :-) )? I wanted to test
on PA-RISC-Linux today, and the patched kernel and the utilities
compiled OK (Kernel calls itself 2.4.0-pa49 #1), but when trying to
create a JFS partition with mkfs.jfs I either get

	mkfs.jfs development version: $Name: v1_0_0 $

	Warning!  All data on device /dev/sdb1 will be lost!

	Continue? (Y/N) Y
	  |
	Bus error

or

        mkfs.jfs development version: $Name: v1_0_0 $

        Warning!  All data on device /dev/sdb1 will be lost!

        Continue? (Y/N) Y
	<4>mkfs.jfs(379): unaligned access to 0x00035245 at ip=0x000137af
	Not-handled unaligned insn 0x4a740108
	Unaligned handler failed, ret = -1

(optionally without <4> or with Bus error)

Is it JFS's fault or that of the PA-RISC-Linux port? The messages do
not seem to be from mkfs.jfs, as I did not find anything like that
in its source. Is it too early to try things like this?

Peter.

Re: [parisc-linux] JFS on PA-RISC-Linux

[Matthew Wilcox]

On Tue, Jul 03, 2001 at 12:51:27PM +0200, Peter Weilbacher wrote:
> 	<4>mkfs.jfs(379): unaligned access to 0x00035245 at ip=0x000137af
> 	Not-handled unaligned insn 0x4a740108
> 	Unaligned handler failed, ret = -1
> 
> (optionally without <4> or with Bus error)
> 
> Is it JFS's fault or that of the PA-RISC-Linux port? The messages do
> not seem to be from mkfs.jfs, as I did not find anything like that
> in its source. Is it too early to try things like this?

It's the fault of both, really.  The userspace program is attempting to do
an unaligned access.  say, loading an int from a non-word-aligned address.
Most CPUs can't handle this and take a fault.  We haven't yet implemented
a trap handler for that fault.

Or rather, tausq has implemented one, but hasn't activated it yet,
I believe.... tausq?

-- 
Revolutions do not require corporate support.

Re: [parisc-linux] JFS on PA-RISC-Linux

[Randolph Chung]

> It's the fault of both, really.  The userspace program is attempting to do
> an unaligned access.  say, loading an int from a non-word-aligned address.
> Most CPUs can't handle this and take a fault.  We haven't yet implemented
> a trap handler for that fault.
> 
> Or rather, tausq has implemented one, but hasn't activated it yet,
> I believe.... tausq?

The trap handler is activated but doesn't handle this particular load
instruction yet. I'll fix this tomorrow.

randolph
-- 
Debian Developer <tausq@debian.org>
http://www.TauSq.org/

Re: [parisc-linux] JFS on PA-RISC-Linux

[Grant Grundler]

Peter Weilbacher wrote:
...
I can't comment on the bus error.

>         mkfs.jfs development version: $Name: v1_0_0 $
> 
>         Warning!  All data on device /dev/sdb1 will be lost!
> 
>         Continue? (Y/N) Y
> 	<4>mkfs.jfs(379): unaligned access to 0x00035245 at ip=0x000137af
> 	Not-handled unaligned insn 0x4a740108
> 	Unaligned handler failed, ret = -1
> 
> (optionally without <4> or with Bus error)
> 
> Is it JFS's fault or that of the PA-RISC-Linux port?

The parisc-linux port. Randolph Chung recently added support for unaligned
access emulation.  Randolph needs more info to fix this problem.
See http://lists.parisc-linux.org/pipermail/parisc-linux/2001-June/012946.html

> The messages do
> not seem to be from mkfs.jfs, as I did not find anything like that
> in its source. Is it too early to try things like this?

No. If no one tries, we won't know what's broken. The basic stuff works.

thanks,
grant

Grant Grundler
parisc-linux {PCI|IOMMU|SMP} hacker
+1.408.447.7253

Re: [parisc-linux] JFS on PA-RISC-Linux

[Matthew Wilcox]

Please do not cc closed (moderated) lists and lists at parisc-linux.org.
Failure to follow this friendly warning may result in offenders being
barred from lists on this server.  Thanks.

-- 
Revolutions do not require corporate support.

[parisc-linux] Once more: JFS

[Peter Weilbacher]

Hi again,

I gave it a new try with JFS on PA-RISC-Linux. I got kernel
2.4.7-pa2 (linux-20010810.tar.gz) and patched it with JFS 1.0.2.
Cross-compiling the kernel went well, but I had to use the native
PA-RISC-gcc to compile the utilities.

mkfs.jfs now works ok, as opposed to kernel 2.4.0-pa49, and I can
copy files to that partition, but fsck.jfs seems to have problems.
When checking on boot the following messages are written to dmesg:

fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e50b
fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e517
fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e533
fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e443
fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e50b
fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e517
fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e533
fsck.jfs(34): unaligned access to 0xfaf0050f at ip=0x0002e443
fsck.jfs(34): unaligned access to 0xfaf0058f at ip=0x0002e50b
fsck.jfs(34): unaligned access to 0xfaf0058f at ip=0x0002e517
fsck.jfs(34): unaligned access to 0xfaf0058f at ip=0x0002e533
fsck.jfs(34): unaligned access to 0xfaf0058f at ip=0x0002e443
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e50b
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e517
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e533
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e443
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e50b
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e517
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e533
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e443
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e50b
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e517
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e533
fsck.jfs(34): unaligned access to 0xfaf004cf at ip=0x0002e443

And hundreds more similar errors when doing a manual fsck.jfs are
written to kern.log, and this check never finishes...

This time it seems to be a PA-RISC-Linux problem, as I recall
earlier problems with "unaligned access" messages.

Greetings,
	Peter.

Re: [parisc-linux] Once more: JFS

[Randolph Chung]

> And hundreds more similar errors when doing a manual fsck.jfs are
> written to kern.log, and this check never finishes...
> 
> This time it seems to be a PA-RISC-Linux problem, as I recall
> earlier problems with "unaligned access" messages.

The unaligned messages by themselves indicate a *user-space* problem. We can
look at the fsck.jfs code and see what it's doing.... my guess is that
there's some casting going on that is causing the unaligned accesses.
For example, if you had something like this:

char foo[10];
long *bar = (long *)(foo+2);

*bar = 10;

then it will trigger an unaligned access. (usually it's not quite that
obvious though)

If you see a "Not-handled unaligned access" message then there's a
kernel issue that we need to address, but doesn't seem to be the case
here.

Please visit http://bugs.parisc-linux.org/ and follow the instructions
to file a bug, and we will look into this.

randolph
-- 
   @..@                                         http://www.TauSq.org/
  (----)
 ( >__< )
 ^^ ~~ ^^

Re: [parisc-linux] Once more: JFS

[Peter Weilbacher]

On Tue, 14 Aug 2001, Randolph Chung wrote:

> The unaligned messages by themselves indicate a *user-space* problem. We can
> look at the fsck.jfs code and see what it's doing.... my guess is that
> there's some casting going on that is causing the unaligned accesses.
[...]
> then it will trigger an unaligned access. (usually it's not quite that
> obvious though)

I am a bit out of programming, so I reported it to the JFS people,
they should be faster at finding this than me.

> If you see a "Not-handled unaligned access" message then there's a
> kernel issue that we need to address, but doesn't seem to be the case
> here.

No, none like that. Thanks for the explenation!

	Peter.