Maintaining stable stability

Posted Jul 23, 2020 13:07 UTC (Thu) by dowdle (subscriber, #659)
Parent article: Maintaining stable stability

In contemporary Fedora kernels (starting with later 5.6.x kernels into the 5.7.x series), there have been at least two significant regressions I've seen users report in the #fedora channel on the Freenode IRC network. Luckily I've not run into either on the hardware I use. One regresion is with the Intel i915 video driver (https://bugzilla.redhat.com/show_bug.cgi?id=1854167) and another on a particular family of wifi chipsets that I don't recall the make and model of. In both cases they are devices that had long been working in previous kernels and stopped working. I believe in the wifi case there was a kernel mailing list discussion showing they were aware of the patch that introduced the problem but since that patch fixed something else, they didn't want to just revert it... so it was a trade-one-bug-for-another patch . Sorry to be so vague with details but it is early and I'm not on the machine with the IRC logs to look at. Anyhoo... the point is that regressions still seem to be fairly common... and lingering for periods of time. If the regressions were on hardware I owned, I'd be a lot more active with them. The bugzilla link I gave only has the initial reporter's info so while it may appear to be limited in its impact, I know I saw at least three other people with the same breakage... and the work-around in the bug report didn't work for them... but unfortunately they didn't chime in with me-toos in the bug report to get it more attention. I know, I know... anecdotal.

Maintaining stable stability

Posted Jul 23, 2020 13:51 UTC (Thu) by dmoulding (subscriber, #95171) [Link]

I believe the WiFi bug you are referring to may be the one that affected the Intel 3168 family of devices. Or perhaps it was another one...

In any event, I recall I had been running the last few 5.5-rc kernels on my hardware and everything was working fine. Then when 5.5.0 came out, suddenly the WiFi didn't work at all. Within just a few hours of 5.5.0 dropping, I had a patch submitted to the linux-wireless mailing list. But it still took several weeks before the patch found its way into the mainline and subsequently was admitted into stable. So there were a number of 5.5.x releases that had this regression in them.

I felt like it took too long to get that patched for 5.5.x users. While I'm quite sure it doesn't *always* take that long to get patches integrated into stable, if avoiding regressions is highly desirable, then getting patches quickly when they *do* occur should be highly desirable, as well.

Seems like we could improve processes around admitting patches for regressions, especially very obvious ones like an entire family of devices that quit working altogether. But, I'm not really heavily involved in the community, so don't have any specific recommendations of how to do that, only my one anecdote that tells me there seems to be some room for improvement.

(I'm entirely open to the possibility that there already is some way to expedite patches like this, and that in my case because I simply wasn't aware of it, and because I didn't use the right "channels", it took longer than it might have otherwise).

Maintaining stable stability

Posted Jul 23, 2020 20:47 UTC (Thu) by nivedita76 (subscriber, #121790) [Link]

Weren't these both regressions in mainline rather than the stable tree?