Messages in this thread |  | | From | (Robert Millner) | Subject | Re: modules problems | Date | 28 Sep 1996 16:47:12 -0400 |
| |
"AC" == Alan Cox <alan@cymru.net> writes: AC> Its an inevitability of an efficient interface. In the SMP case its very AC> very hard not to have different structures. To start with your hardware AC> is in places quite different. We could always use the SMP build but that AC> would punish 99.5% or so of the folks who just happen to have only one AC> CPU.
After hacking a driver for the Jorway 73a SCSI-CAMAC controler under Solaris 2.4, I'd agree totally. There is a tremendous amount of overhead in assuming that the driver need be MT safe. Solaris makes things have to happen this way for a sound reason but it makes the driver 10X heavier and more complex than we need it for an SS10 (that they won't let me run Linux on 8-(). I can see benefits as well for enforcing such a scheme even in monoprocessor situations, but one can code to those benefits without having that much complexity.
On the other hand, if a company wants to support Linux, then who am I to complain. They should keep in mind what hardware base their code will run on. Maby push an MT safe (heavy but works everywhere) and a non MT safe (smaller and lighter) driver.
Rob
|  |