Subject: Re: kernel & libkvm [was IIci success]
To: None <current-users@NetBSD.ORG>
From: der Mouse <mouse@Collatz.McRCIM.McGill.EDU>
Date: 01/11/1996 18:49:35
>>> Why not? Having a /dev which configures itself sounds like a handy
>>> feature to me.
>> It does have a certain advantage for LKMs, because of the changing
>> major number problem;
> What changing major number problem? That's what the post-load script
> is for.
Hm, I wasn't aware of that. It does address most of the problem.
The fragments it doesn't address are:
- What if the module is more than one device, more than one syscall, or
a device plus a syscall? Is the post-load script executed multiple
times? modload(8) is not clear on this point; its specification for
the argument pattern to the script does not specify what is passed
beyond the first two arguments if the count of devices and syscalls
in the LKM is not one.
- We are then back at the situation of having user-land and the kernel
separately maintaining the same information. SunOS does this with
the mount table, and it's a royal pain; I was really glad to see that
NetBSD allows you to get the mount table from the kernel. (On the
other hand, all of /dev is in this state right now, so I suppose this
is worse only in that it changes comparatively often.)