Re: 2.1.48 problems (NFS, Oops and more)

Jan Kratochvil (short@k332.feld.cvut.cz)
Sat, 9 Aug 1997 12:17:16 +0200


Hi all

On Sat, Aug 09, 1997 at 11:10:46AM +0200, Viggo Norum wrote:
...
> uses Linux as it's only OS. I have tried a 2.1.48 kernel for a few
> hours, it works much better than the last 2.1.xx kernel I tried, but:
...
> Secondly (this may be a more serious problem),
> something odd happened to one of my data files:
>
> <snip>
> 0.10097E-10 0.23410E+01 0.12000E+02
> 0.72638E+00 0.23245E+01 0.12000E+02
> 0.14578E+01 0.22764E+01 0.12000E+02
> 0.22007E+01 0.22007E+01 0.12000E+02
> 0.29648E+01 0.21056E+01 0.12000E+02
> 0.37644E^@^@^@^@ 0.20038E+01 0.12000E+02
> -0.43301E+01 0.25000E+01 0.12000E+02
> -0.35523E+01 0.26547E+01 0.12000E+02
> <snip>
>
> What is the "^@^@^@^@" doing in there? It was not there 2 minutes ago
> when I used it as input file to a CFD post processor. No one has
> edited the file (confirmed by the time stamp). The file was on a 2MB
> e2fs partition on a Quantum Fireball.

I must confirm that something weird is in the actual 2.1.x releases. I had
several times page cache corruption by one '\0' character (in C files during
compile - fixing and recompiling didn't help as it occured again). Unfortunately
it's not repeatable for me now. And the machine runs stable otherwise so I
don't accept replies like setup/wait states.

Would it be possible to add checksumming of buffers/unshared page cache as
kernel compile option?

Jan Kratochvil