On Wed, 15 Oct 2003 20:38:44 +0000, Raymond Jennings wrote: > >Oh well, I hope it was at least worth brainstorming.
Brainstorming is (almost) never a bad idea. > >XFree86 *might* wish to consider a "modulette" to cover things that userland >CAN'T do, like AGP, DMA, IRQ, and so on. AGP stuff can be done in usermode. Most the DRI drivers DO include a kernel module for handling DMA and interrupts. The idea of a "generic" DMA/IRQ handler is somewhat attractive, but the various graphics chips are so very different that doing anything generically is quite difficult. >Or maybe the modulette could grant I/O privileges on behalf of an X >server that opens it (thus the X server doesn't require root privileges)? You get the same spoofing issue here. If an unprivileged XFree86 server can gain access to the kernel module, then any arbitrary unprivileged application can do so as well. You really need some way to identify the XFree86 server as "trusted". In Linux today, the only mechanism for doing that is suid root. >Does the notion of a kernel module have ANY merit at all? Or was the idea >complete garbage? As we have said, many of the drivers DO have kernel modules for implementing OpenGL acceleration. However, there is a tradeoff. You're getting additional functionality, in exchange for an operating system dependency and the inherent stability risks in moving stuff to the kernel. There is clearly a threshhold beyond which the tradeoff makes good sense. My key point is that the threshhold needs to be set rather high. It's not that the kernel idea is unconditionally bad. It's just that, for the typical 2D driver, the gain isn't worth the pain. -- - Tim Roberts, [EMAIL PROTECTED] Providenza & Boekelheide, Inc. _______________________________________________ Devel mailing list [EMAIL PROTECTED] http://XFree86.Org/mailman/listinfo/devel
