On Thu, Jun 29, 2023 at 08:48:42PM +0200, Greg KH wrote: > On Thu, Jun 29, 2023 at 06:38:38PM +0100, Mark Brown wrote: > > As discussed before your tolerance for risk in stable is *far* higher > > than mine, if there's any value in doing this at all it's probably > > within what would get taken but that doesn't mean that it's something > > that it's sensible to highlight as an important fix like tagging for > > stable does. It's extremely unclear that it fits the severity criteria > > that are supposed to be being applied to stable, though obviously the > > documentation doesn't fit the actual practice these days. > It's not a matter of "tolerance for risk", it's a "if this change is > good enough for future releases, why isn't it good enough for older > releases as well?" > As you know, we don't break user interfaces, so either this is a break > or it isn't, stable trees have nothing to do with it as a normal user > would "hit" this when updating to run Linus's tree, just as easily as > they would "hit" it updating their stable kernel version. You know as well as I do that we have a bunch of interfaces where things end up getting dynamically numbered as they appear, and provided to userspace together with identifying information that allows userspace to figure out what's what in a stable fashion even though the numbers might change. Like I said earlier in the thread this is one of them, better hardware support also has some risk of disturbing things (and some of the numbering is going to be hotplug dependent, though this patch isn't likely to run into that particular bit of things). ABI stability is a continuum, from for example things relying on race conditions or other timing things that were lucky they ever worked to changes in interfaces that break clear and documented guarantees. Reliance on stability is similar, and how much of an issue it is when something does change and someone notices is going to vary depending on what changed and why. While the risk here seems low if the reasoning is just to make things neater then it's even harder to justify for a stable kernel than it is for mainline. Note also that the patch is still under discussion for mainline...
Attachment:
signature.asc
Description: PGP signature
