On 22 September 2017 at 15:02, James Hogarth <james.hogarth@gmail.com> wrote:


On 22 September 2017 at 14:50, Gary Gatling <gsgatlin@ncsu.edu> wrote:


On Fri, Sep 22, 2017 at 9:41 AM, James Hogarth <james.hogarth@gmail.com> wrote:


Pretty sure the last testing I did with the details form Hans's blog[0] the behaviour was that if the nvidia driver failed then the nouveau driver was a fallback (rather than the older instructions that totally blacklisted it leaving no GPU at all).

Of course this is only helpful when the GPU is supported by nouveau ;)

Note I haven't tested the negativo driver yet on the stabalization copr as I prefer the bumblebee experience since that keeps the discrete GPU powered down saving battery and temperatures when not in use ... and the negativo driver flips everything over to running on the discrete GPU and ignores the Intel integrated one.

Given that it's fundamentally the same driver though I expect the experience to be the same - willing to test and report back :)


Hello,

I am looking at the patch referenced earlier in the thread. There was a problem with a missing file in nvidia's archive (nv_compiler.h) I am unsure if it was really needed.

I will try to test it with bumblebee-nvidia when I have time this weekend on 4.12 and 4.13 beta.

Thank you.
 

_


Do keep in mind that the patching which is done to get it working (note that patch still failed compiling for me on 4.13.2 ... no patch was needed on 4.13.1) also has to change the license type so that it works, as a non-GPL symbol in use was marked GPL apparently ... which of course means it cannot be distributed by Negativo as a "fix" ...

Checking the comments on the Negativo nvidia-driver site[0] it appears that there's already reports of it not working on F27 with that kernel:

> ANARCHISTCAT
> September 20, 2017 at 12:52 am
> With Fedora 27 nvidia driver 384.69-1.fc27 and kernel 4.13.2-300.fc27.x86_64 i got with akmod an error and its unable to compile the driver 2017/09/20 00:40:28 akmodsbuild: MODPOST 4 modules
> 2017/09/20 00:40:28 akmodsbuild: FATAL: modpost: GPL-incompatible module nvidia.ko uses GPL-only symbol ‘lockdep_init_map’
> 2017/09/20 00:40:28 akmodsbuild: make[2]: *** [scripts/Makefile.modpost:91: __modpost] Error 1
> 2017/09/20 00:40:28 akmodsbuild: make[1]: *** [Makefile:1520: modules] Error 2
> 2017/09/20 00:40:28 akmodsbuild: make[1]: Leaving directory ‘/usr/src/kernels/4.13.2-300.fc27.x86_64’





And just to confirm my own experiences ....

With the negativo driver the build fails:

  CC [M]  /var/lib/dkms/nvidia/384.69/build/nvidia-drm/nv-pci-table.o
ld -r -o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-interface.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-frontend.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-instance.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-acpi.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-chrdev.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-cray.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-dma.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-gvi.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-i2c.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-mempool.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-mmap.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-p2p.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-pat.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-procfs.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-usermap.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-vm.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-vtophys.o /var/lib/dkms/nvidia/384.69/build/nvidia/os-interface.o /var/lib/dkms/nvidia/384.69/build/nvidia/os-mlock.o /var/lib/dkms/nvidia/384.69/build/nvidia/os-pci.o /var/lib/dkms/nvidia/384.69/build/nvidia/os-registry.o /var/lib/dkms/nvidia/384.69/build/nvidia/os-usermap.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-modeset-interface.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-pci-table.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-kthread-q.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-kthread-q-selftest.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv-memdbg.o /var/lib/dkms/nvidia/384.69/build/nvidia/nv_uvm_interface.o /var/lib/dkms/nvidia/384.69/build/nvidia/nvlink_linux.o
ld -r -o /var/lib/dkms/nvidia/384.69/build/nvidia-modeset/nv-modeset-interface.o /var/lib/dkms/nvidia/384.69/build/nvidia-modeset/nvidia-modeset-linux.o
  LD [M]  /var/lib/dkms/nvidia/384.69/build/nvidia.o
  LD [M]  /var/lib/dkms/nvidia/384.69/build/nvidia-uvm.o
  LD [M]  /var/lib/dkms/nvidia/384.69/build/nvidia-modeset.o
  LD [M]  /var/lib/dkms/nvidia/384.69/build/nvidia-drm.o
  Building modules, stage 2.
  MODPOST 4 modules
FATAL: modpost: GPL-incompatible module nvidia.ko uses GPL-only symbol 'lockdep_init_map'
make[2]: *** [scripts/Makefile.modpost:91: __modpost] Error 1
make[1]: *** [Makefile:1520: modules] Error 2
make[1]: Leaving directory '/usr/src/kernels/4.13.2-200.fc26.x86_64'
make: *** [Makefile:81: modules] Error 2


However it looks like it did compile what failed before and it's just the symbol license issue... so I'm not sure what he might have patched already...

It did fail gracefully to loading the nouveau driver according to PRIME=1 instead of the nvidia driver (of course my primary driver with no nvidia is intel)... so I think the experience won't drop users at a blank screen if they are using the "supported" install method for the nvidia drivers.

They just won't have the performance etc until the driver gets an update or that symbol from the kernel has its exported license reverted ...