Re: 2.0.20: async io complete on unlocked page

Eric Smith (eric@goonsquad.spies.com)
Mon, 23 Sep 1996 11:17:55 -0700 (PDT)

Five days ago I send email to the list about problems upgrading my UDB
running Red Hat 3.0.3 from a 2.0.0 kernel to 2.0.20. The machine would boot,
but every time I would try something that did a lot of disk activity, I
would get eight messages of the form:

> Whoops: unlock_buffer: async io complete on unlocked page

And the process would hang. And the file systems couldn't be unmounted. And
when I switched back to 2.0.0 it crashed every time it tried to fsck them.

So I switched back to 1.3.90, and everything works again. The file system
which couldn't be fsck'd under 2.0.0 was not even corrupted. It seems to me
that something must have gotten broken since 1.3.90.

I also mentioned that I've had lots of problems with the 53c7,8xx driver in
general, on both Alpha and Intel boxes. Various people have suggested that
I may have bad cabling, bad termination, or the like. I have very carefully
checked all of it, to the point of using an oscilliscope to verify the signal
integrity on many of the lines. I found no problems there, but I didn't really
expect to, as the same setups work quite well with Windows NT (even under very
heavy disk workloads).

Cheers,
Eric