Gentoo Archives: gentoo-dev

From: Walter Dnes <waltdnes@××××××××.org>
To: gentoo-dev@l.g.o
Subject: Re: [gentoo-dev] maintainer-wanted: x11-drivers/nvidia-drivers
Date: Wed, 06 Mar 2013 03:34:13
Message-Id: 20130306033351.GA2687@waltdnes.org
In Reply to: Re: [gentoo-dev] maintainer-wanted: x11-drivers/nvidia-drivers by Greg KH
1 On Tue, Mar 05, 2013 at 04:47:09PM +0800, Greg KH wrote
2 > On Tue, Mar 05, 2013 at 02:01:31AM -0500, Walter Dnes wrote:
3 > > On Mon, Mar 04, 2013 at 03:44:33PM -0100, Carlos Silva wrote
4 > > > On Mon, Mar 4, 2013 at 3:28 PM, Walter Dnes <waltdnes@××××××××.org> wrote:
5 > > >
6 > > > > I'm not a C programmer, let alone a developer, so this may be a stupid
7 > > > > question, but here goes... has anyone ever tried doing a HAL (Hardware
8 > > > > Abstraction Layer) to present a reasonably stable interface to binary
9 > > > > video drivers? Think of it as a shim translating a "pseudo-API" into
10 > > > > "the real API" that the kernel exposes directly. Surely, we can do
11 > > > > better than VESA. Give drivers 2 options...
12 > > > > 1) direct kernel access like now
13 > > > > 2) access via the HAL/shim
14 > > >
15 > > >
16 > > > Just read this file and you'll have the answer:
17 > > > /usr/src/linux/Documentation/stable_api_nonsense.txt
18 > >
19 > > Thanks. That was an eye-opener. If user-space drivers are really
20 > > that slow, we may as well stick with VESA as a fallback.
21 >
22 > Ok, I'll bite, What do you mean by that? Where does the
23 > stable_api_nonsense.txt file talk about userspace drivers?
24 >
25 > greg "I wrote that file" k-h
26
27 My statement was a general response to the entire thread. Sorry, I
28 should've retitled it [REDUX whatever]
29
30 * stable_api_nonsense.txt explained lack of a stable *KERNEL* api
31
32 * Duncan's message talked about slow *USERSPACE* API...
33
34 > Of course it's possible to implement a userspace driver that
35 > wouldn't have the same issues as it'd use the stable userspace API,
36 > but that's generally accepted to be far too high a performance cost
37 > for graphics drivers, regardless of the kernel involved (MS tried
38 > it too for stability reasons and gave up at the performance penalty
39 > they were taking).
40
41 So between your file, and Duncan's message, I saw that...
42 1) a stable kernel API is not possible
43 2) a userspace API is too slow.
44
45 I apologize again for the vagueness in my previous reply.
46
47 --
48 Walter Dnes <waltdnes@××××××××.org>
49 I don't run "desktop environments"; I run useful applications