mmap the acceleration registers? Are you aware, that on Intel machines the
granularity of mmap is 4KB? And are you also aware, that most cards are in
fact broken, and have "dangerous" registers near (= distance less than 4KB)
the other registers?
How do you get safe acceleration on that kind of card without having the
acceleration support in the kernel (for those broken cards only, of course)?
Maybe by having a system daemon that has mlockall():ed its pages, is
running with a real-time priority and has used iopl()? That sounds quite
kernelish to me..
If you happen to need cli/sti, you *have to* mlock your pages, or the thing
will going crash & burn someday. This still leaves all the fancy things like
DMA support out..
I'm not advocating GGI specifically, but the concept generally. GGI might
have flaws, but the basic design should be quite clean. The basic idea is to
have all things needed in the kernel, nothing else. For good cards, that
means nothing new on the kernel side. For bad cards, you'll have to load a
module or run the apps as root.
-- Who is General Failure and why is he reading my hard disk? -- Felix von Leitner- To unsubscribe from this list: send the line "unsubscribe linux-kernel" in the body of a message to majordomo@vger.rutgers.edu