Hello there is a patch available for building the nvidia 378.13 driver on Linux 4.10. This most likely will NOT work on 4.11.

Patch is HERE: gist.github.com/tpruzina/c4d9c0ca6bdbb6e78ab2126a7cdf8f1c

Discussion is HERE: devtalk.nvidia.com/default/topic/995636/linux/-patches-378-13-4-10-and-4-11-stageing


I am willing to test this patch so let me know when I can test.

Thanks!

NOTE: There is a new driver 381.09 but have not seen it available for download for Linux yet and it purports 4.10 support out of the box.[/url]
Back to top

the patch is junk, it's trying a patch a directory and file that doesn't exist in 378.13.

the 381.09 patch is junk too, doesn't apply.

the 340 and 304 patches are fine, work, apply, no issues.

tested, I won't spend time debugging junk patches, if you can find some that actually apply I'll add them.



there's no directory or file nvidia-uvm/uvm8_test.c in the driver package, so whoever made that patch was hallucinating.

all the other patches for 378 apply fine. the 381 however fails to find the lines, and just fails overall.
Back to top

Thanks for letting me know...will wait for 381.09 or a successor driver to test. Not sure if 381.09 works out of the box on 4.10.

There is another patch a poster on devtalk mentioned that works for him:

gist.githubusercontent.com/tpruzina/c4d9c0ca6bdbb6e78ab2126a7cdf8f1c/raw/38ccd3224f3205e45982c0c217e23d04f9681b49

In the thread mentioned in my first post there was a command to skip UVM building.

Let me know and sorry to put you on a wild goose chase.
Back to top

kernel/nvidia-uvm/uvm8_test.c

I wish I knew where they got the driver that works with this patch, it's certainly not the standard one, which does not contain that directory, and thus the patch fails, so what on earth are all these people smoking or ingesting that makes them keep posting patches that cannot possibly work on the downloaded driver?

do I simply remove that piece? what happens then?

again,, if it's not clear, you extract the driver package, it has a set of directories, and that directory nvidia-uvm DOES NOT EXIST!!

so what the hell are they talking about in these patches?

usually patches have small glitches, and don't perfectly align, but do work, but this patch cannot work, by design, and cannot apply, so what the hell are they doing?

I can upload it minus the uvm test, but I have no idea what that will do.

again, this is wrong, period, and I don't understand why people are posting patches that cannot work.
Back to top

OK thanks again for the attempt, and I will just wait till the 381 series driver comes out so that I can get 4.10 kernel support. I haven't seen patches for the 375 series either.
Back to top

There's no attempt involved really, all I want is a patch that isn't incoherent, and also a solution to the mystery to why people keep posting patches to drivers that can't possibly work.

I could remove the last part of the patch and it would work fine, but I have exactly zero knowledge about what that would do technically, I rely on people who make patches to actually test them etc, so I'm mystified why they keep posting these patches that can't be applied without failure, do they simply not notice that the patching failed?

technically I could remove the last patch block since the directory/file it's patching doesn't exist in the driver, so in theory it shouldn't matter I guess, but it's very rare to see patches this badly done, even in the script kiddie linux world.
Back to top

Just to try, I removed the uvm8 patching of nonexistent uvm8 files/directory, then tested the patch which applies fine.

I can say nothing if it will work or not in terms of x starting etc, but the patch applies.

so give it a try if you want, I can't verify or confirm anything about the actual functionality, beyond the patch working on the driver package.
Back to top

This time the patch works, and I am able to load the driver successfully. So far no glitches or anything unusual yet when running 4.10.10 kernel.

Not sure why the UVM dir was removed from the nvidia driver package anyway as it's stuff for CUDA. But at the moment driver works fine.

Thanks for taking another stab at this pesky problem and seems nvidia gets behind the 8-ball when it comes to new xorg or kernel releases where ABI gets broken.
Back to top

well I figured given there was no such directory, it couldn't probably matter if that non existent directory was not patched, heh.

thanks for the feedback.

note that the beta driver patch does NOT work, not the same situation, it simply is wrong, offset errors etc.
Back to top

No problem and thanks for letting me test this patch. As for the 381 beta driver that is always changing and I am not sure what build that beta version patch was built on.

As for 4.11 kernel I will be watching for a suitable patch as 4.11 will bring some changes that will break more things.
Back to top