Messages in this thread | | | Subject | Re: devfs again, (was RE: USB device allocation) | Date | Mon, 11 Oct 1999 13:50:34 -0300 | From | Horst von Brand <> |
| |
Stephen Frost <sfrost@ns.snowman.net> said: > On Mon, 11 Oct 1999, Bernhard Rosenkraenzer wrote:
[...]
> > Enjoy running rm and mknod whenever you insert a PCMCIA or USB device... > > And of course let's not forget about newbie users who want their > > StarOffice and who don't worry about system internals such as any /dev > > entries? Like it or not, Linux is going to make it to the desktop market, > > and thereby to users who don't have a clue. > > For them, devfs is more important than for us. > > Dynamic /dev is important to them, as is user-notification of device > addition/removal. This is not the same as devfs, devfs is one implementation > of this.
Sorry, no. Newbies want the system to take care of PnP, device-wise. How it does so (a fully populated /dev, a dynamic /dev, Ouija chip, ...) is of no concern to them. Our concern is different: How do we provide said black magic efficiently. IMHO, a dynamic /dev is overkill (you want to have the available devices, nobody says they have to be _just_ the available devices), and underkill (no real persistence, no clean way to add ACLs and whatnot) at the same time. Plus horrible bloat (the kernel will have to contain code for every possible hot-pluggable device type, just in case one of them happens to show up), and no reasonable way of notifying userland about new devices (that being the _real_ need for PnP). Rummaging around under hundreds of devices, or dozens of device clases at least, isn't "reasonable notification" in my book.
[...]
> > What is the extra cost, if you set CONFIG_DEVFS=n ?
> Driver developers having to support devfs, and possibly > major,minor's as well. The problem is, if you only go half-way you loose > one of the (to me) big advantages of devfs, getting rid of major,minor's > in favor of something (arguably) more scalable. As in, if you have to > have your drivers support both it takes away one of devfs's advantages. > If you allow CONFIG_DEVFS=n, then you MUST support major,minor's.
The major,minor representation of devices is completely orthogonal to a dynamic /dev
> > Have you heard of automount? It was originally meant to mount NFS volumes > > when required, but it can (and actually has been) used for other stuff as > > well. > > For example, it can be used to automatically mount CDs when they're > > inserted, or to mount new partitions when they appear in devfs... > > The same can be done from userspace with appropriate kernel > notification...
Userspace will need the appropiate kernel notification anyway, independent of /dev being dynamic or static. Automount works that way today. I've played with automounting my CD (module!) via autofs, works silky. Just unmounting doesn't work right. I'd want to unmount when I try to eject the CD (if no users, that is). That can't work because on PCs the CD drive doesn't notify of that fact. :( -- Horst von Brand vonbrand@sleipnir.valparaiso.cl Casilla 9G, Viña del Mar, Chile +56 32 672616
- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu Please read the FAQ at http://www.tux.org/lkml/
| |