kev009
commented
6 years ago Yeah if there is a desire it wouldn't be hard to have staging and legacy drivers in ports. I think that's a pretty good solution for all concerned going forward.
Some philosophical perspective.. when I had a BSD team at $job-1, my primary concern for us when focused on community stewardship was the past decade of hardware as the sweet spot (which aligned nicely with the business needs), and I was generally sympathetic to time spent on the past 20 years of HW. To put that into perspective, 1998, that is Pentium IIs and maybe some of the NICs that were excluded like fxp. This is why we brought lem(4) into iflib e1000 for instance.
WRT other BSDs, NetBSD is structured a lot better to maintain a lot of toolchains, archs, and devices. Linux' breadth is simply dizzying. There are different values, community size and contributors, and it reflects on the agility and interests of the respective communities. There are definitely trade offs.
I am very opposed to using LKPI for Ethernet drivers. It's not necessary because the surface area is small whereas drm rivals the entire rest of any kernel.
bsdimp
brooksdavis
cschuber![https://assets-cdn.github.com/images/email/message_cards/header.png","avatar_image_url":"https://assets-cdn.github.com/images/email/message_cards/avatar.png","action":{"name":"Open](https://assets-cdn.github.com/images/email/message_cards/header.png","avatar_image_url":"https://assets-cdn.github.com/images/email/message_cards/avatar.png","action":{"name":"Open){kind=link}
Please don't take this as anything but an observation.
First of all, I fully understand the position taken by this paper. There are only so many resources available to The Project. This kind of reminds me of direction I was given when I was working as an MVS systems programmer writing kernel code for the IBM mainframe. My director at the time had an 80/20 rule whereby he did not want to pay [my salary] for features that only 20% of the install base might want to use. I fully understand this and as a FreeBSD developer, with very limited time myself: I have more projects I want to finish in my pipeline than I have time to sit down focus my attention on in the near future. I'm sure other people in The Project are most likely in the same position. This position of the paper is totally understandable.
OTOH, I see the drivers that are being proposed removed here are supported by the other BSDs and many Linux distros, including Red Hat (& Fedora). I don't think we want to place ourselves in the position or have the reputation that we support only* the newest of hardware. Maybe when the drivers are removed they be moved to ports and supported on a best efforts basis, like drm will be. This is certainly an option.
Another option might be to a Linux KPI shim, like DRM is using, to support, using Linux drivers, NICs we may not want to support in base ourselves or support new hardware until we have better (better performing) native support. (The lag between when Linux supports new hardware and FreeBSD does is sometimes noticeable and people have asked me this in person on many occasions.)
Finally, I'm not saying "let's not do this." I fully understand the issues. I am asking where to from here?
Thoughts?