Bugzilla

(In reply to Benjamin FRANCOIS from comment #26)
> Hello Martin, thank you for your response,
> 
> 1. Wouldn't this have an impact on performance, especially WebGL? Also I am not sure there is a way we'd be able to do that without reaching into our 1000+ users $HOME directories and Firefox profiles, which is not that simple to do at scale. 

Yes, widget.dmabuf.enabled disables WebGL dmabuf backend and leads to ~20% performance penalty.
But I don't know if dmabuf is used on our setup with NVIDIA anyway - it's supposed to work with new drivers and was disabled for old ones.

I *think* what you see now it's just moved dmabuf check so you hit it early before it's disabled due to GBM failure.
You may try to compare WebGL performance with widget.dmabuf.enabled=false at 141.0 and 140.0 without it (I guess it will be the same).

> 2. This does indeed seem to work and allows Firefox, however I am wary of side effects of straight up removing files provided by the Nvidia driver. Could this break something else?

I frankly have no idea. There must be a reason the file is there but I don't use NVIDIA drivers and also I'm not in touch with NVIDIA gfx team.

> 3. I'll discuss this with our senior engineers, there probably was a reason why we dropped the Firefox RHEL RPM in the past in favor of upstream downloaded through autopkg. If that reason is no more, I have no qualms using the ESR release from Red Hat.

Red Hat is going to ship ESR 140.0 as an update and also talks to NVIDIA folks so they should be able to solve this issue for you. I don't think Mozilla as upstream should add extra patches for old and known broken NVIDIA drivers.

(In reply to Benjamin FRANCOIS from comment #26)
> Hello Martin, thank you for your response,
> 
> 1. Wouldn't this have an impact on performance, especially WebGL? Also I am not sure there is a way we'd be able to do that without reaching into our 1000+ users $HOME directories and Firefox profiles, which is not that simple to do at scale. 

Yes, widget.dmabuf.enabled disables WebGL dmabuf backend and leads to ~20% performance penalty.
But I don't know if dmabuf is used on our setup with NVIDIA anyway - it's supposed to work with new drivers and was disabled for old ones.

I *think* what you see now it's just moved dmabuf check so you hit it early before it's disabled due to GBM failure.
You may try to compare WebGL performance with widget.dmabuf.enabled=false at 141.0 and 140.0 without it (I guess it will be the same).

> 2. This does indeed seem to work and allows Firefox, however I am wary of side effects of straight up removing files provided by the Nvidia driver. Could this break something else?

Frankly I have no idea. There must be a reason the file is there but I don't use NVIDIA drivers and also I'm not in touch with NVIDIA gfx team.

> 3. I'll discuss this with our senior engineers, there probably was a reason why we dropped the Firefox RHEL RPM in the past in favor of upstream downloaded through autopkg. If that reason is no more, I have no qualms using the ESR release from Red Hat.

Red Hat is going to ship ESR 140.0 as an update and also talks to NVIDIA folks so they should be able to solve this issue for you. I don't think Mozilla as upstream should add extra patches for old and known broken NVIDIA drivers.