Re: TTY changes to 2.1.65

Richard Masoner (richardm@cd.com)
Wed, 26 Nov 1997 09:35:39 -0600 (CST)


> The implementation should be fairly simple... all syscalls to devices
> with the "generic" major would get multiplexed (based upon their
> minor) to the "real" major handler,

Nice idea; unforunately the number of bits in the minor number is
already a limiting factor in Linux.

It's not unusual for some of my customers to hang a thousand tty's on a
single host: Linux's 8-bit minors (? if I remember correctly) is
already a serious limitation in these circumstances; to further cripple
Linux by using the minor to determine the device driver would be a
mistake, IMHO. In fact, the way to get around the tiny minor number
limitation is by assigning multiple major numbers to the same driver.

Are there any plans to go the route that SGI has gone with their
hardware graphs (which I *really* like) or other schemes which
basically allow you to hang an unlimited amount of hardware to a single
machine? For those who are unfamiliar with hardware graphs, Silicon
Graphics has basically pitched the idea of major and minor numbers and
cdev- and bdev- switch tables and replaced it all with directed
graphs. It is *really* cool attaching 100 16-port async devices to a
single host and watching things fly!

I do like the changes which have been implemented thus far and which
Linus has folded in. Writing tty drivers for Linux is far easier than
writing them for any other Un*x-like O.S. I've done. Of course, the
fact that I don't have to reverse-engineer unassembled RISC code to
figure out what's going on helps a lot too :-)

Richard Masoner

--
Richard F. Masoner                         Central Data Corporation
richardm@cd.com                            Champaign, IL 61821  USA
+1 217 359 8010                            http://www.cd.com/