Trying Out Nouveau's Accelerated Pascal Support With DRM-Next, Mesa 17.2-dev

[karolherbst]

if for whatever magical reason Nvidia decided to release the PMU signed images, how hard would it be to actually implement support for power management and reclocking?

Depends. If they provide documentation for the PMU Interface, it should be kind of easy. If not, we would need to reverse engineer it.

[karolherbst]

Sorry for stupid question but why this PMU firmware cannot be extracted from propertary Linux or even Windows driver? It is encrypted or something?

Because we first need to figure out where it is. There is no magic sign telling us "PMU FIRMWARE HERE!". It might be compressed as well. Thing is nobody found anything yet.

[Luke]

I don't know how is handled in this specific case, but they are using cryptographic signatures so the general explanation is the same.
If you tamper signed files the signature becomes invalid. I assume that they are signing the whole driver blob or something that prevents people from simply extracting them from the driver (like it was done before).
The card's BIOS is checking for these signatures and refuses to load firmwares not signed correctly.

If NVIDIA releases blob files that are individually signed, then no problem.

This system is common in software, also linux distros use it to ensure that the packages you receive aren't tampered by third parties.

Not all implementations of cryptographically signing software are equally strong. A good cipher can be applied in a sloppy manner. I don't know how good a job Nvidia does with it, but a report that a compromise of Debian's system for signing packages was theoretically possible was enough to induce me to build my kernel and the entire cipher stack cryptsetup uses from source. I don't know how practical that attack would be but I was taking no chances.

Wonder if Nvidia made a similar mistake with their package signing implementation?

[Mangix]

It's really an unfortunate situation. Nvidia's drivers have stupid bugs like not allowing my laptop monitor to turn off or change the brightness whereas nouveau sometimes segfaults and is slow. The latter should be fixable since AFAIK PMU firmware is not needed on my GPU in order to reclock but it's not implemented yet.

[Sonadow]

Just wanting to throw it out here:

I can't build Mesa 17 because it complains that my version of Wayland is too old. (Fedora 23). If I were to build a new version of Wayland, do I also need to rebuild the Xserver as well? Since Xwayland is tied to both xserver and wayland.

[theriddick]

I'm sure NVIDIA will care about Nouveau's once RadeonSI/AMDGPU/whatever starts kicking their ass (and not just in TR hmmkay). Until then I can see why they just don't bother supporting the Nouveau project, that and the fact if they did it probably reveal some rather uncompliant methods for getting the OpenGL performance they do out of many titles.

[birdie]

Sadly there won't be any Pascal reclocking anytime soon, because there is more locked down than it was on Maxwell2, so we need the signed PMU images to even change the voltage or clocks, what we could do without any signed images before.

I guess NVIDIA doesn't want you to fry their GPUs accidentally, so while this is unfortunate, I guess it is warranted.

[karolherbst]

I guess NVIDIA doesn't want you to fry their GPUs accidentally, so while this is unfortunate, I guess it is warranted.

funny, we never fried one and the GPUs are protected against this in hardware.

[squallbayu]

I don't know but, is it still valid to quote Torvalds' word abaut Nvidia?

[Guest]

The community really needs more resources put into nouveau. Every new gnome-shell release is exposing more issues for the NVIDIA proprietary driver. I already have to revert one mutter patch to get it to work properly under X11 with NVIDIA driver.
I am even considering switching to the onboard Intel graphics on my skylake machine.