Re: kill i_dev

> We can kill i_dev.  
> BTW, watch out for socket.c use of ->i_dev.

Yes, I looked at that but concluded that someone (you?)
had added the assignment just to preserve the guarantee
previously given by get_empty_inode() at the moment it
was replaced by new_inode(). But it is superfluous, I think.

> However, rdev and [cb]dev will have to remain separate.

We can fight later on that one. My point of view is
that just like i_dev is a field in i_sb, also i_rdev
can be retrieved from i_[cb]dev.

Something else is kdev_t. I liked it when it was a pointer.
Now it is just garbage and the kernel is full of conversions
to and from. Is there any reason not to throw out all of kdev_t?
That is, is there a reason to have a kdev_t different from dev_t?

Andries