Path: utzoo!attcan!uunet!nbires!ncar!oddjob!mimsy!chris From: chris@mimsy.UUCP (Chris Torek) Newsgroups: comp.unix.questions Subject: Re: 4.1 --> 4.3 BSD Device Driver Conversion Keywords: device driver 4.1 4.3 BSD vax Message-ID: <11461@mimsy.UUCP> Date: 12 May 88 17:43:34 GMT References: <2182@bgsuvax.UUCP> Organization: U of Maryland, Dept. of Computer Science, Coll. Pk., MD 20742 Lines: 124 Here is something I just wrote this morning about it. Converting drivers from 4.1BSD to 4.3BSD is usually fairly easy. Most of the Unibus goo is unchanged. The most noticeable difference is that raw read and write functions went from fooread(dev) dev_t dev; { return (physio(foostrategy, &foobuf[minor(dev)], dev, B_READ, minphys)); } to fooread(dev, uio) dev_t dev; struct uio *uio; { return (physio(foostrategy, &foobuf[minor(dev)], dev, B_READ, minphys, uio)); } If the read and write functions deal directly with the device, rather than going through physio, they must be changed to deal with `uio' rather than global u.u_xxx variables. Similarly, all the error handling in the open and close routines is simplified. Where in 4.1BSD you have fooopen(dev, flag) dev_t dev; int flag; { ... if (something) { u.u_error = EBUSY; return; } ... } you change this to if (something) return (EBUSY); ... return (0); } (The close routine should return 0 for success, but a number of drivers do not, and no one checks as yet.) Finally, ioctl routines tend to get shorter. Where before you would write fooioctl(dev, cmd, addr, flag) dev_t dev; int cmd; caddr_t addr; int flag; { switch (cmd) { case FIOCSETFOO: u.u_error = copyin(addr, (caddr_t)&kernelvar, sizeof(struct foodata)); ... case FIOCGETFOO: u.u_error = copyout((caddr_t)&kernelvar, addr, sizeof(struct foodata)); ... default: u.u_error = ENOTTY; } } this becomes fooioctl(dev, cmd, data, flag) dev_t dev; int cmd; caddr_t addr; int flag; { switch (cmd) { case FIOCSETFOO: kernelvar = *(struct foodata *)data; ... case FIOCGETFOO: *(struct foodata *)data = kernelvar; ... default: return (ENOTTY) } return (0); } The copyin and copyout is all done at a higher level. See for details; mainly, FIOCSETFOO might have been #define FIOCSETFOO _IO(f, 0) so that it would become #define FIOCSETFOO _IOW(f, 0, struct foodata) There is one limitation in the new ioctl mechanism: the size of structures passed via ioctl is limited to <= 127 bytes. If the ioctl needs more data, you still need the copyin/copyout code, but you would say `int error; ... error = copyin(...); if (error)' rather than `u.u_error = copyin(...); if (u.u_error)'. In general, 4.2BSD tried to purge driver references to `u.', so that drivers could call other drivers. The task is not quite finished, but it is getting there. -- In-Real-Life: Chris Torek, Univ of MD Comp Sci Dept (+1 301 454 7163) Domain: chris@mimsy.umd.edu Path: uunet!mimsy!chris