mm and Oops

mm and Oops

[Andries.Brouwer]

After sending util-linux out, I booted a kernel that had kdev_t
a pointer type, to see whether that still works.
And all (minus md/lvm/nfs that didnt compile) was fine
except that kswapd produced an oops. Investigation shows
that it was caused by the combination of what <linux/swap.h> calls
"Ugly ugly ugly HACK" and what <linux/kdev_t.h> judges with "yuk".

So, maybe we must remove one or both of these sore spots.
The second one is the use of a special constant B_FREE
as device value to indicate that the buffer is free.
I'll look at this tomorrow but perhaps someone knows:
must the constant B_FREE (used only in fs/buffer.c) be nonzero?
If so, then we probably need a bitfield to indicate "free".
Otherwise we can use 0 ("no device") as value.

Andries

Re: mm and Oops

[Andreas Dilger]

Andries writes:
> After sending util-linux out, I booted a kernel that had kdev_t
> a pointer type, to see whether that still works.
> And all (minus md/lvm/nfs that didnt compile)...

Yes, LVM totally abuses kdev_t (assumes = dev_t in user space).
Changing kdev_t should force this to be cleaned up.

> The second one is the use of a special constant B_FREE
> as device value to indicate that the buffer is free.
> I'll look at this tomorrow but perhaps someone knows:
> must the constant B_FREE (used only in fs/buffer.c) be nonzero?
> If so, then we probably need a bitfield to indicate "free".
> Otherwise we can use 0 ("no device") as value.

Isn't a device = 0 used for NFS root?

Cheers, Andreas
-- 
Andreas Dilger  \ "If a man ate a pound of pasta and a pound of antipasto,
                 \  would they cancel out, leaving him still hungry?"
http://www-mddsp.enel.ucalgary.ca/People/adilger/               -- Dogbert