2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"

2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"

[Daniel Elstner]

Hi,

With kernel 2.4.4 SMP, I get some spurios errors from several
user-space programs. Unfortunately it's hard to reproduce, I had most
luck with the XFree86-4.0.3 build. When doing `make World', soon cpp0
(called by imake) dies with the following error message:

cpp0: : Value too large for defined data type

The message seems to correspond to EOVERFLOW in gcc's libiberty.
When calling imake directly, it fails 1 out of 10-20 times.
I couldn't reproduce this with calling cpp directly.

I also got a lot of that messages once at shutdown,
as init was trying to umount /proc.

The error occurs neither with 2.4.3 SMP nor with 2.4.4 UP.
(I'm using reiserfs, too.)

ABIT VP6
dual P3 866
gcc version 2.95.4 20010319 (prerelease)
binutils 2.11
glibc 2.2.3

Could you please give me further advice how to track this down?

Thanks,
-- Daniel

Re: 2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"

[Daniel Elstner]

Hi again,

the problem occurs only after setting up a parallel build tree with
lndir, removing the whole symlink tree, and running lndir again.
Maybe an reiserfs bug?

-- Daniel

> With kernel 2.4.4 SMP, I get some spurios errors from several
> user-space programs. Unfortunately it's hard to reproduce, I had most
> luck with the XFree86-4.0.3 build. When doing `make World', soon cpp0
> (called by imake) dies with the following error message:
> 
> cpp0: : Value too large for defined data type
> 
> The message seems to correspond to EOVERFLOW in gcc's libiberty.
> When calling imake directly, it fails 1 out of 10-20 times.
> I couldn't reproduce this with calling cpp directly.
> 
> I also got a lot of that messages once at shutdown,
> as init was trying to umount /proc.
> 
> The error occurs neither with 2.4.3 SMP nor with 2.4.4 UP.
> (I'm using reiserfs, too.)
> 
> ABIT VP6
> dual P3 866
> gcc version 2.95.4 20010319 (prerelease)
> binutils 2.11
> glibc 2.2.3
> 
> Could you please give me further advice how to track this down?

reiserfs+lndir problem [was: 2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"]

[Daniel Elstner]

Hi all,

unfortunately I have to correct me again.
The problem seems unrelated to the kernel version or SMP/UP
(though only 2.4.[34] tried yet).

Apparently it's a reiserfs/symlink problem.
I tried doing the lndir on an ext2 partition, sources still
on reiserfs. And it worked just fine!

Sorry for the rather large amount of noise, I hope it's
finally correct now :)

-- Daniel

> the problem occurs only after setting up a parallel build tree with
> lndir, removing the whole symlink tree, and running lndir again.
> Maybe an reiserfs bug?
> 
> -- Daniel
> 
> > With kernel 2.4.4 SMP, I get some spurios errors from several
> > user-space programs. Unfortunately it's hard to reproduce, I had most
> > luck with the XFree86-4.0.3 build. When doing `make World', soon cpp0
> > (called by imake) dies with the following error message:
> > 
> > cpp0: : Value too large for defined data type
> > 
> > The message seems to correspond to EOVERFLOW in gcc's libiberty.
> > When calling imake directly, it fails 1 out of 10-20 times.
> > I couldn't reproduce this with calling cpp directly.
> > 
> > I also got a lot of that messages once at shutdown,
> > as init was trying to umount /proc.
> > 
> > The error occurs neither with 2.4.3 SMP nor with 2.4.4 UP.
> > (I'm using reiserfs, too.)
> > 
> > ABIT VP6
> > dual P3 866
> > gcc version 2.95.4 20010319 (prerelease)
> > binutils 2.11
> > glibc 2.2.3
> > 
> > Could you please give me further advice how to track this down?

Re: reiserfs+lndir problem [was: 2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"]

[Chris Mason]

On Monday, April 30, 2001 10:55:57 PM +0200 Daniel Elstner
<daniel.elstner@gmx.net> wrote:

> Hi all,
> 
> unfortunately I have to correct me again.
> The problem seems unrelated to the kernel version or SMP/UP
> (though only 2.4.[34] tried yet).
> 
> Apparently it's a reiserfs/symlink problem.
> I tried doing the lndir on an ext2 partition, sources still
> on reiserfs. And it worked just fine!

Neat, thanks for the extra details.  Does that mean you can consistently
repeat on reiserfs now?  What happens when you do the lndir on reiserfs and
diff the directories?

Any useful messages in /var/log/messages?

Re: reiserfs+lndir problem [was: 2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"]

[Daniel Elstner]

Hi,

On Mon, 30 Apr 2001 21:03:47 -0400 Chris Mason <mason@suse.com> wrote:

> > Apparently it's a reiserfs/symlink problem.
> > I tried doing the lndir on an ext2 partition, sources still
> > on reiserfs. And it worked just fine!
> 
> Neat, thanks for the extra details.  Does that mean you can consistently
> repeat on reiserfs now?  What happens when you do the lndir on reiserfs and
> diff the directories?

I just played around a bit with the following results:

sources on reiserfs, lndir on reiserfs -> make fails, diff ok
sources on reiserfs, lndir on ext2     -> make ok
sources on ext2, lndir on reiserfs     -> make fails, diff ok

Doing the diff against a second copy of the tree shows no errors, too.
Always the same behaviour: You have to run lndir at least twice to
get the error. If the link tree was already set up after a boot, the
error occurs only after rm + lndir + rm + lndir.

There's a strange way to get things working just like after a reboot.
After diff'ing the link tree with the 2nd copy (both on reiserfs),
make World won't fail - at least once.

I also tried in the link tree:
    find ! -type d -exec cat '{}' \; >/dev/null
And it seems to have the same effect as the diff.

> Any useful messages in /var/log/messages?

Nope, not any single message (except the usual ones).

The error on shutdown (when umounting /proc) appeared once again,
with the the same error message. Though I have no clue how that
could be related to reiserfs.

Two missed details: the reiserfs partion exists on a software
raid0 device. I configured glibc-2.2.3 with --enable-kernel=2.4.1.

Shall I try moving back to glibc-2.2.2 / configuring for lower kernel?

-- Daniel

Re: reiserfs+lndir problem [was: 2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"]

[Chris Mason]

On Wednesday, May 02, 2001 12:41:52 AM +0200 Daniel Elstner
<daniel.elstner@gmx.net> wrote:

> Hi,
> 
> On Mon, 30 Apr 2001 21:03:47 -0400 Chris Mason <mason@suse.com> wrote:
> 
>> > Apparently it's a reiserfs/symlink problem.
>> > I tried doing the lndir on an ext2 partition, sources still
>> > on reiserfs. And it worked just fine!
>> 
>> Neat, thanks for the extra details.  Does that mean you can consistently
>> repeat on reiserfs now?  What happens when you do the lndir on reiserfs
>> and diff the directories?
> 
> I just played around a bit with the following results:
> 
> sources on reiserfs, lndir on reiserfs -> make fails, diff ok
> sources on reiserfs, lndir on ext2     -> make ok
> sources on ext2, lndir on reiserfs     -> make fails, diff ok
> 
> Doing the diff against a second copy of the tree shows no errors, too.
> Always the same behaviour: You have to run lndir at least twice to
> get the error. If the link tree was already set up after a boot, the
> error occurs only after rm + lndir + rm + lndir.
> 
> There's a strange way to get things working just like after a reboot.
> After diff'ing the link tree with the 2nd copy (both on reiserfs),
> make World won't fail - at least once.

Ok, can you reproduce with a set of sources other than X?  I would leave
glibc alone for now, unless you can reproduce on ext2.

-chris

Re: reiserfs+lndir problem [was: 2.4.4 SMP: spurious EOVERFLOW "Value too large for defined data type"]

[Daniel Elstner]

On Tue, 01 May 2001 18:48:51 -0400 Chris Mason <mason@suse.com> wrote:

> Ok, can you reproduce with a set of sources other than X?  I would leave
> glibc alone for now, unless you can reproduce on ext2.

No.
I tried building kernel 2.4.4 five times in a row - no errors.
Also some diff's after two `make dep' didn't show anything.

The xfree-dri sources from cvs failed, too.
Something special with imake?

-- Daniel


> > Hi,
> > 
> > On Mon, 30 Apr 2001 21:03:47 -0400 Chris Mason <mason@suse.com> wrote:
> > 
> >> > Apparently it's a reiserfs/symlink problem.
> >> > I tried doing the lndir on an ext2 partition, sources still
> >> > on reiserfs. And it worked just fine!
> >> 
> >> Neat, thanks for the extra details.  Does that mean you can consistently
> >> repeat on reiserfs now?  What happens when you do the lndir on reiserfs
> >> and diff the directories?
> > 
> > I just played around a bit with the following results:
> > 
> > sources on reiserfs, lndir on reiserfs -> make fails, diff ok
> > sources on reiserfs, lndir on ext2     -> make ok
> > sources on ext2, lndir on reiserfs     -> make fails, diff ok
> > 
> > Doing the diff against a second copy of the tree shows no errors, too.
> > Always the same behaviour: You have to run lndir at least twice to
> > get the error. If the link tree was already set up after a boot, the
> > error occurs only after rm + lndir + rm + lndir.
> > 
> > There's a strange way to get things working just like after a reboot.
> > After diff'ing the link tree with the 2nd copy (both on reiserfs),
> > make World won't fail - at least once.