On Fri, 6 Oct 2000, Andrew Doran wrote:

> Matthew Jacob <mjacob@feral.com> wrote:
> 
> > Why do you need extra tags. Why can't you use
> > the ioctl entry point?
> 
> Because these methods should not be accessible to user-space programs.

The ioctl entry point is used by the midlayer- not necessarily by user
ioctls (module to module ioctl). Frankly, this is much better than extending
the structure randomly.

This is currently used, for example, by the midlayer to tell an adapter that
it can enable tagged/sync/wide for a device.

> 
> > The explanation as to purpose isn't quite clear to me. Is prescan
> > an 'activation'?
> 
> It's so that I can (a) tell my adapter to rescan the bus and (b) my driver
> to rebuild its target/LUN -> device ID map, based upon the results of (a).

This is of general enough interest that even a user ioctl would be worthwhile.
It could be useful for the Fibre Channel stuff when you want it to refresh
it's port database (getting around obscure F/W bugs where it didn't see a
fabric change).

Why is this any different than any of the stumblings toward dynamic attachment
that has been occurring so far?

I dunno- Jason keeps claiming he's got a bunch of stuff all sitting in his
scsipi branch that will address some of these kinds of issues- I just haven't
had the time to go see.

I *do* know, however, that what you want is just a plain adapter ioctl
command- really no need to change the structure.

-matt