ATI Kernel Power Management Moves A Bit More

[cklein]

Hello everybody,

I am running:
- user-space: Ubuntu Karmic + Xorg/drm/mesa from Lucid
- kernel-space: drm-radeon-testing + patches 0001 to 0030
on an "ATI Technologies Inc Mobility Radeon HD 2400".

It seems to me that the vblank is "slightly" missed. When the power state changes the screen jumps a little bit. With drm-radeon-testing (w/o patches) a horizontal line appears randomly on the screen. Also, dmesg says:

[   52.622540] [drm] Requested: e: 30000 m: 40000 p: 16
[   52.653440] [drm] Setting: e: 30000
[   52.664992] [drm] not in vbl for pm change 00010002 00000000 at entry
[   52.665452] [drm] Setting: m: 40000
[   54.760315] [drm] Requested: e: 45000 m: 40000 p: 16
[   58.462530] [drm] Requested: e: 30000 m: 40000 p: 16
[   59.062600] [drm] Requested: e: 11000 m: 30000 p: 16
[   59.262789] [drm] Setting: e: 11000
[   59.269060] [drm] not in vbl for pm change 00020002 00000000 at exit
[   59.269062] [drm] Setting: m: 30000
[   60.460075] [drm] Requested: e: 30000 m: 40000 p: 16
[   60.460305] [drm] Setting: e: 30000
[   60.466072] [drm] not in vbl for pm change 00020002 00000000 at entry
[   60.466547] [drm] not in vbl for pm change 00020002 00000000 at exit
[   60.466549] [drm] Setting: m: 40000
[   61.162599] [drm] Requested: e: 11000 m: 30000 p: 16
[   61.362791] [drm] Setting: e: 11000

The power consumptions seems 200mA bigger than fglrx.

Anyway, thanks for all the work, the open-source stack with KMS works like a charm.

Cheers,
Cristi.

[pingufunkybeat]

This is great news.

So, this stuff is not/will not be in the 34 kernel. But some powersaving is already in there.

What will the .34 support then? Will it give dynamic PM, just not to the fullest extent (which will then come with .35)?

[Louise]

It's hard to say. We're trying to make sure that the open source drivers are able to do all the same power-saving things as fglrx (as we discover them ) but it's going to be a bit like 3D -- it may not be worth adding all of the complexity if you can get 80% of the benefit with 20% of the work.

On the other hand I wouldn't be surprised if some scenarios specific to consumer users were handled *better* than fglrx so that the overall results were considered superior. Power management is one of the areas where the environment differs widely across the OSes, so a Linux-specific open source solution could also have an edge by not having to be based on common, OS-independent code.

I think it's safe to say it will be "pretty good" anyways

Awesome stuff!

How many percentage would you say have been done now compared to the TODO list?

[bridgman]

I guess the main open issue for power reduction is voltage control, but agd5f and mjg59 seem to be making progress on that right now.

[So7t]

ATI attitude to Linux users is great peace of S***!!! Sorry , but im so tired of so much disrespect!!
Please forgive me with my "offtopic" reply, but since 2003 i bought 7 products from ATI and none of them have a perfect integration!!
ATI MUST SPEND a little more money ON LINUX if they want MY MONEY AGAIN!!!
Greetings to all no paid developers that contribute with their heroic work on ATI chipsets!!! God Bless you!!

[monraaf]

Note that other than the internal thermal stuff on newer r6xx+, the information for power management has been available for a while now. the atombios data tables for r5xx and below has been available for years, and the data tables for r6xx+ have been available for several months.

Well, the parser has not been updated to handle powerplay tables with format revision 5, as used in some (most/all?) Evergreen cards. Also there's no info about them in atombios.h. Forcing atomdis to treat these tables as revision 4, it appears they are very similar, but something must have changed, otherwise they wouldn't have bumped the format revision.

So users with these cards are missing out on all the new pm stuff, unless they start hacking the code themselves.

[Louise]

I guess the main open issue for power reduction is voltage control, but agd5f and mjg59 seem to be making progress on that right now.

Great stuff!

Although I was just getting used to have my bedroom GPU heated =)

Looking at the patches, it seams that R100-R600 (and R700?) now have all these PM features. Does that mean that R800 will get these PM features for free, when R800 supports KMS?

Or is PM hand crafted for each chip?

[bridgman]

There are definitely differences from generation to generation but I *think* the main breakpoints were in the middle of the 6xx generation and then more changes between 6xx and 7xx. We'll know more in a month or so, but AFAIK Evergreen should be able to leverage most of the 7xx code.

Remember, there is no R800, just Evergreen

[yotambien]

What does the ForceLowPowerMode option do? After enabling it I still see with rovclock that the core and memory clocks are at their maximum values...

[RealNC]

I know that I sound like a broken record now, but how would I go about applying the kernel 2.6.34 DRM drivers on top of kernel 2.6.33? Is that possible?