On Tue, Nov 24, 2015 at 11:08:07AM +0100, Christian König wrote: > On 24.11.2015 10:36, Thierry Reding wrote: > >On Tue, Nov 24, 2015 at 09:59:09AM +0100, Daniel Vetter wrote: > >>On Tue, Nov 24, 2015 at 04:55:19PM +0800, Jammy Zhou wrote: > >>>This series enable the DKMS build of amdgpu driver. > >>For the curious: What's DKMS? > >I believe in this context it's "Dynamic Kernel Module Support": > > > > https://wiki.debian.org/KernelDKMS > > Yeah, correct. > > > > >I'm somewhat surprised that one would have to do something special to > >the kernel build system to "enable" such a build. But perhaps this is > >completely unrelated. > > I'm a bit torn apart on this. On the one hand I'm not sure if we have > drivers upstream explicitly supporting this as well? > > On the other hand it's just the Makefiles which need to be written in a way > which makes them relocatable. E.g. no absolute path like > drivers/gpu/drm/amd/... in them and that's a good idea anyway. > > Maybe just changing the commit message to "don't use absolute paths in the > makefiles" would be sufficient?
Oh I don't mind DKMS support, it makes sense to allow amd to use upstream as the baseline for their enhanced blob driver stack. Just wanted to know what it is. Imo what we shouldn't merge upstream would be compat code to allow amdgpu to be built on old kernels, since that's a lot more invasive than a few Makefile changes. Otoh that problem is largely address with the linux backporting project, which solves the "new driver directoy in old sources problem". So as long as it's just about building free-standing I think it's perfectly fine. -Daniel -- Daniel Vetter Software Engineer, Intel Corporation http://blog.ffwll.ch
