RFC: One X Server To Rule Them All

RFC: One X Server To Rule Them All

[Burton, Ross]

Hi,

It's bugged me for a while that oe-core is shipping both "big" X.org
(in two variations) and "small" kdrive.  A quick review shows:

- oe-core is using X.org for the qemu images on all platforms
- meta-yocto is using big X.org for Atom, "lite" Xorg for beagleboard,
and kdrive for mpc8315e/routerstationpro.
- meta-ti is using xserver-xorg (and not the lite variation for
beagle).  It also ships a binary kdrive-based Xsgx, but that doesn't
require kdrive in oe-core.

I looked at a few other BSPs but the ones I found appeared to leave X
server choice to the user.  I also see that meta-oe is shipping
another X.org server.

Some consolidation is required, obviously.

Xorg lite is just X.org with --disable-dri --disable-glx
--disable-dga.  I need to investigate how much a difference that
actually makes to the server binary and if it can be controlled by
machine features or more granular packaging.

Here's my plan:
1) Investigate differences between xorg/xorg-lite/kdrive, producing a
feature matrix and binary size comparison.  Then assuming it's fine:
2) Remove xorg-lite, switch beagleboard in meta-yocto to X.org as per
meta-ti's example
3) Remove kdrive, switch meta-yocto's remaining machines  to X.org
4) Review differences between oe-core and meta-oe's X.org, merge,
unification at last, street party.

I'm sure the combined members of oe-core and oe-devel have something
to say about this.  Comments welcome!

I'd obviously need to test X.org on the MPC8315e and RouterStation
Pro, so if anyone has one of these set up then offering to test would
be much appreciated.

Cheers,
Ross

Re: RFC: One X Server To Rule Them All

[Martin Jansa]

[-- Attachment #1: Type: text/plain, Size: 2129 bytes --]

On Fri, Aug 17, 2012 at 01:12:45PM +0100, Burton, Ross wrote:
> Hi,
> 
> It's bugged me for a while that oe-core is shipping both "big" X.org
> (in two variations) and "small" kdrive.  A quick review shows:
> 
> - oe-core is using X.org for the qemu images on all platforms
> - meta-yocto is using big X.org for Atom, "lite" Xorg for beagleboard,
> and kdrive for mpc8315e/routerstationpro.
> - meta-ti is using xserver-xorg (and not the lite variation for
> beagle).  It also ships a binary kdrive-based Xsgx, but that doesn't
> require kdrive in oe-core.
> 
> I looked at a few other BSPs but the ones I found appeared to leave X
> server choice to the user.  I also see that meta-oe is shipping
> another X.org server.

Is it? where? I've merged it to oe-core about year ago..

Cheers,

> Some consolidation is required, obviously.
> 
> Xorg lite is just X.org with --disable-dri --disable-glx
> --disable-dga.  I need to investigate how much a difference that
> actually makes to the server binary and if it can be controlled by
> machine features or more granular packaging.
> 
> Here's my plan:
> 1) Investigate differences between xorg/xorg-lite/kdrive, producing a
> feature matrix and binary size comparison.  Then assuming it's fine:
> 2) Remove xorg-lite, switch beagleboard in meta-yocto to X.org as per
> meta-ti's example
> 3) Remove kdrive, switch meta-yocto's remaining machines  to X.org
> 4) Review differences between oe-core and meta-oe's X.org, merge,
> unification at last, street party.
> 
> I'm sure the combined members of oe-core and oe-devel have something
> to say about this.  Comments welcome!
> 
> I'd obviously need to test X.org on the MPC8315e and RouterStation
> Pro, so if anyone has one of these set up then offering to test would
> be much appreciated.
> 
> Cheers,
> Ross
> 
> _______________________________________________
> Openembedded-devel mailing list
> Openembedded-devel@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-devel

-- 
Martin 'JaMa' Jansa     jabber: Martin.Jansa@gmail.com

[-- Attachment #2: Digital signature --]
[-- Type: application/pgp-signature, Size: 205 bytes --]

Re: RFC: One X Server To Rule Them All

[Burton, Ross]

On 17 August 2012 13:16, Martin Jansa <martin.jansa@gmail.com> wrote:
> Is it? where? I've merged it to oe-core about year ago..

You're right, I saw xserver-common and somehow ignored the "common" bit.

Ross

Re: RFC: One X Server To Rule Them All

[Burton, Ross]

On 17 August 2012 13:12, Burton, Ross <ross.burton@intel.com> wrote:
> 1) Investigate differences between xorg/xorg-lite/kdrive, producing a
> feature matrix and binary size comparison

Some rough numbers: a core-image-sato rootfs for RouterStation Pro
using kdrive is 25959477 bytes.  Change kdrive to Xorg (Xfbdev and a
few important modules) and it grows to 26342094 bytes, approximately
380kb growth.

It appears that we're also shipping libexa in the core server package,
when I think that can be split into a separate package as fbdev isn't
using it.  That will give us another 90kb back.

I'm happy with this size increase as a trade-off for using an X server
that is actually maintained.

Ross

Re: [OE-core] RFC: One X Server To Rule Them All

[Saul Wold]

On 08/22/2012 08:15 AM, Burton, Ross wrote:
> On 17 August 2012 13:12, Burton, Ross <ross.burton@intel.com> wrote:
>> 1) Investigate differences between xorg/xorg-lite/kdrive, producing a
>> feature matrix and binary size comparison
>
> Some rough numbers: a core-image-sato rootfs for RouterStation Pro
> using kdrive is 25959477 bytes.  Change kdrive to Xorg (Xfbdev and a
> few important modules) and it grows to 26342094 bytes, approximately
> 380kb growth.
>
> It appears that we're also shipping libexa in the core server package,
> when I think that can be split into a separate package as fbdev isn't
> using it.  That will give us another 90kb back.
>
> I'm happy with this size increase as a trade-off for using an X server
> that is actually maintained.
>
+1

> Ross
>
> _______________________________________________
> Openembedded-core mailing list
> Openembedded-core@lists.openembedded.org
> http://lists.linuxtogo.org/cgi-bin/mailman/listinfo/openembedded-core
>
>

Re: RFC: One X Server To Rule Them All

[Burton, Ross]

On 22 August 2012 16:15, Burton, Ross <ross.burton@intel.com> wrote:
> Some rough numbers: a core-image-sato rootfs for RouterStation Pro
> using kdrive is 25959477 bytes.  Change kdrive to Xorg (Xfbdev and a
> few important modules) and it grows to 26342094 bytes, approximately
> 380kb growth.

Forgot to say - as I'm still working on this, this was the size
difference in the rootfs.tar.bz2 so that's compressed difference.
Expect a breakdown of files and size differences shortly.

Ross

Re: RFC: One X Server To Rule Them All

[Burton, Ross]

More numbers!

The interesting delta in the root file system is as follows:

-1617464 ./usr/bin/Xfbdev

Bye kdrive.

-9892 ./usr/lib/libts-1.0.so.0.0.0
+8344 ./usr/lib/libts-1.0.so.0.0.0

Now *this* is weird.

+34948 ./usr/lib/libpciaccess.so.0.11.1
+17088 ./usr/lib/xorg/modules/libfbdevhw.so
+19544 ./usr/lib/xorg/modules/drivers/fbdev_drv.so
+30492 ./usr/lib/xorg/modules/libshadowfb.so
+135580 ./usr/lib/xorg/modules/libfb.so
+24492 ./usr/lib/xorg/modules/libvbe.so
+20804 ./usr/lib/xorg/modules/input/kbd_drv.so
+42980 ./usr/lib/xorg/modules/input/mouse_drv.so
+47936 ./usr/lib/xorg/modules/input/evdev_drv.so
+98908 ./usr/lib/xorg/modules/extensions/libextmod.so
+19532 ./usr/lib/xorg/modules/extensions/libdbe.so
+27144 ./usr/lib/xorg/modules/libvgahw.so
+26984 ./usr/lib/xorg/modules/libshadow.so
+1099 ./usr/share/X11/xorg.conf.d/10-evdev.conf
+17315 ./usr/share/X11/rgb.txt
+13616 ./usr/bin/cvt
+7476 ./usr/bin/showrgb
+1983908 ./usr/bin/Xorg
+526 ./etc/X11/xorg.conf

That's a grand total of 2578716 bytes, or an overall gain of 938kb.

I expect the dbe extension can disappear, and you'll notice that to
cover all bases the keyboard, mouse, *and* evdev input modules are
present - we don't need all three.  The rgb.txt database was pulled in
but that's entirely optional and wasn't pulled in by kdrive.

That concludes any analysis of killing kdrive.  Patch series coming
shortly to action it, followed by the same analysis on
xserver-xorg-lite.

Ross

Re: RFC: One X Server To Rule Them All

[Burton, Ross]

On 22 August 2012 17:18, Burton, Ross <ross.burton@intel.com> wrote:
> -9892 ./usr/lib/libts-1.0.so.0.0.0
> +8344 ./usr/lib/libts-1.0.so.0.0.0
>
> Now *this* is weird.

Any ideas as to why this happened are greatly welcomed...

> +19532 ./usr/lib/xorg/modules/extensions/libdbe.so

Confirmed that this is 1990s HP legacy.  Gone.

Ross

Re: RFC: One X Server To Rule Them All

[Paul Eggleton]

On Wednesday 22 August 2012 17:39:45 Burton, Ross wrote:
> On 22 August 2012 17:18, Burton, Ross <ross.burton@intel.com> wrote:
> > -9892 ./usr/lib/libts-1.0.so.0.0.0
> > +8344 ./usr/lib/libts-1.0.so.0.0.0
> > 
> > Now *this* is weird.

Nothing off the top of my head, but is there any difference in the tslib 
configure/compile logs?

Cheers,
Paul

-- 

Paul Eggleton
Intel Open Source Technology Centre

Re: RFC: One X Server To Rule Them All

[Burton, Ross]

On 22 August 2012 17:45, Paul Eggleton <paul.eggleton@linux.intel.com> wrote:
> On Wednesday 22 August 2012 17:39:45 Burton, Ross wrote:
>> On 22 August 2012 17:18, Burton, Ross <ross.burton@intel.com> wrote:
>> > -9892 ./usr/lib/libts-1.0.so.0.0.0
>> > +8344 ./usr/lib/libts-1.0.so.0.0.0
>> >
>> > Now *this* is weird.
>
> Nothing off the top of my head, but is there any difference in the tslib
> configure/compile logs?

There's only one configure/compile log in temp...  It's below X in the
stack so shouldn't have rebuilt.

Ross