Intel KMS vs. UMS With Ubuntu 10.04

[mendieta]

Any one running Intel's code "the way it's intended" won't have the option for UMS. Only Ubuntu's bastardised version has the feature

Strong wording, ah?. The reason they went for UMS is that some hardware is just not working with KMS, and I think that's a good reason.You want your hardware to work. Then, you may prefer this or that feature, but at a minimum you need the display to ... well, display

[Craig73]

Seems kind of funny that there wasn't more of a focus on the non-3d aspects of the driver, considering that Intel is a desktop no gamer card.

[Apopas]

Seems kind of funny that there wasn't more of a focus on the non-3d aspects of the driver, considering that Intel is a desktop no gamer card.

"...but 2D UXA and X-Video may be a different story and we'll leave that for another article."

[Craig73]

"...but 2D UXA and X-Video may be a different story and we'll leave that for another article."

[sarcasm] Wow... thanks for pointing that out! [/sarcasm]

I did see that - my point is, why right the first article about the least important aspects of the card.

[Apopas]

Because we already had a similar article for the ATIs.

[kurros]

Strong wording, ah?. The reason they went for UMS is that some hardware is just not working with KMS, and I think that's a good reason.You want your hardware to work. Then, you may prefer this or that feature, but at a minimum you need the display to ... well, display

Wouldn't using separate packages and some modalias definitions as they do with the multiple nvidia blobs be preferable to backporting? At least the users would KMS capable hardware would get some of the newer fixes that can't be easily backported.

[Xake]

Wouldn't using separate packages and some modalias definitions as they do with the multiple nvidia blobs be preferable to backporting? At least the users would KMS capable hardware would get some of the newer fixes that can't be easily backported.

Emmm... Nope.

I have a Intel IGP where every thing works extremely fine with KMS until I try to hook it up to a TV over svideo. I have screen output but it gets garbled, and it seems dependent on things like moving windows, 3d and so on (it is not mesa, removed the mesa drivers and issue still there).

With UMS, same driver versions, everything works just fine.

[glasen]

I think they've played with the kernel drm, libdrm, xorg-server as well as the drivers

Ubuntu Lucid uses a vanilla version of the intel driver 2.9.1 (which supports UMS and KMS ootb), and a vanilla version of libdrm (version 2.4.18).

The Xserver is version 1.7.6 plus udev-patches from Xserver 1.8.

The kernels biggest change is the backported the drm-stack from version 2.6.33. The other changes are minimal (e.g. a small patch for ureadahead, a patch for minimizing kernel log output etc.)

In other words :

Ubuntu Lucid is not as heavily patched as some people believe.

[drag]

Every distro patches their stuff. It's expected that they patch it since the vanilla kernel.org kernel is just shit when it comes to stability.

The days were people used the vanilla kernel are long gone. Dead with the 2.4 series.

The DRM stuff is always kinda separate anyways. Historically they always maintained DRM kernel drivers separate from the vanilla kernel that had newer stuff and more features. Out of things to backport, it's probably one of the lesser big deals.


------------------------

The UMS vs KMS thing is stupid, though. Instead of forward porting UMS support the correct thing is just to fix KMS.

Why?

* Because your going to have to fix KMS anyways eventually, so your just going to make more work for yourself in the long run. You might as well get it over with and fix it correctly right now.

* UMS introduces more bugs and things to support on your own, so actually instead of fixing bugs your more then likely create MORE bugs. By forward porting stuff like this your making the software worse. There are good reasons why the Intel folks dropped it.

* Forcing users to choose between modesetting methods to get the hardware working when 98% of the target audience has no f-ng clue what that even means is shitty OS design. This means that even if your forward porting UMS to work around KMS bugs the vast majority of people that will run into those specific KMS bugs will never benefit from the presence of UMS since it's beyond their means to use it.

Users whose TV out is broken are not going to say to themselves 'Oh, maybe I should hack around with /etc/X11/xorg.conf and see if UMS option will make my TV work...'; They are just going to not use TV out.

[Xake]

The UMS vs KMS thing is stupid, though. Instead of forward porting UMS support the correct thing is just to fix KMS.

Why?

* Because your going to have to fix KMS anyways eventually, so your just going to make more work for yourself in the long run. You might as well get it over with and fix it correctly right now.

* UMS introduces more bugs and things to support on your own, so actually instead of fixing bugs your more then likely create MORE bugs. By forward porting stuff like this your making the software worse. There are good reasons why the Intel folks dropped it.

* Forcing users to choose between modesetting methods to get the hardware working when 98% of the target audience has no f-ng clue what that even means is shitty OS design. This means that even if your forward porting UMS to work around KMS bugs the vast majority of people that will run into those specific KMS bugs will never benefit from the presence of UMS since it's beyond their means to use it.

Users whose TV out is broken are not going to say to themselves 'Oh, maybe I should hack around with /etc/X11/xorg.conf and see if UMS option will make my TV work...'; They are just going to not use TV out.

And that is why a driver/distro should use KMS by _default_, but still have a *WORKING ALTERNATIVE* for all those cases when KMS does not work. Remember 9.04 should be supported for a long time, and that without having to make major upgrades to the kernel/drm code just because some vendor decides to drop support for stuff prematurely.