DRM Work Piling Up For The Linux 3.4 Kernel

[phoronix]

Phoronix: DRM Work Piling Up For The Linux 3.4 Kernel

While it looks like there's still another week before the Linux 3.3 kernel will be released and thus marking the merge window for the Linux 3.4 kernel opening, here's some of the DRM graphics changes you can expect to see merged...

http://www.phoronix.com/vr.php?view=MTA2OTM

[mcgreg]

What about PCI-E 2.0 Support? Or is it already supported in linux 3.3?
I believe I read it caused a major speedup in some cases.

mcgreg

[Azpegath]

What about PCI-E 2.0 Support? Or is it already supported in linux 3.3?
I believe I read it caused a major speedup in some cases.

mcgreg

It's already in 3.2 I believe. I've at least enabled it in my 3.2 kernel, and haven't seen any error messages regarding it. Michael says (in this article)

Taking full advantage of PCI Express 2.0 for Radeon graphics cards involves setting a few registers related to the PCI-E link width and speed control. The code has been there for the supported generations of Radeon hardware for many months now, but it is not on by default. Like the Intel RC6 and Active State Power Management problems in the Linux kernel, PCI-E Gen2 is not on by default at this time due to a few hardware-specific problems. For some problematic systems, having the Radeon DRM with PCI Express 2.0 support enabled will cause a corrupted screen upon mode-setting. The problem comes down to compatibility issues with some motherboards. For over a year now the PCI-E Gen2 support has been concealed behind a kernel module parameter.

Those wishing to take advantage of PCI Express 2.0 with supported motherboards (nearly all Intel and AMD motherboards from 2008 and newer) and graphics cards (Radeon HD 3000 series and newer) must boot the kernel with the "radeon.pcie_gen2=1" on the kernel command line (or probe the radeon DRM with "pcie_gen2=1"), otherwise the graphics card will still be operating at PCI Express 1.0 speeds. Even with the forthcoming Linux 3.3 kernel, PCI-E 2.0 support is still off by default with the Radeon driver.

Regarding the last line, I'd say he's implicitly stating that it's (at least) in the 3.2 kernel already.

[mcgreg]

Thank you Azpegath!

Ahh now that I see this, I believe I've read this one already some time ago.
But somehow I've kind of understood the remaining "issues" would be solved for the next releases (linux 3.3 or linux 3.4) and set it as default if the hardware is capable.

Well, gonna test it now and see if there any advantage.

mcgreg

[mcgreg]

I've tested a few application but only from one I have some numbers

With nexuiz I've done a timedemo and the results are

without PCI-E 2.0: 23 / 37 / 83
with PCI-E 2.0: 33 / 49 / 119

so in average it is ~ 33% faster.
And since I haven't seen any disadvantages yet, I'll keep it that way for now.

Looking forward to see "full" 2d-colortiling with kernel 3.4.

mcgreg

[Azpegath]

I've tested a few application but only from one I have some numbers

With nexuiz I've done a timedemo and the results are

without PCI-E 2.0: 23 / 37 / 83
with PCI-E 2.0: 33 / 49 / 119

so in average it is ~ 33% faster.
And since I haven't seen any disadvantages yet, I'll keep it that way for now.

Looking forward to see "full" 2d-colortiling with kernel 3.4.

mcgreg

Great to see improvements! And I'm with you on the 2d-colortiling thing.

[LLStarks]

I was hoping that more PRIME bits would make this merge.