Hey Jason, On Tue, Feb 13, 2024 at 09:47:26AM -0400, Jason Gunthorpe wrote: > On Tue, Feb 13, 2024 at 12:59:52PM +0000, Will Deacon wrote: > > > The typical kernel standard is to fix bugs in patches and only reach > > > for a wholesale revert if the community is struggling with bug > > > fixing. Dmitry already tested removing that hunk, Robin explained the > > > issue, we understand the bug fix is to remove the > > > arm_smmu_init_domain_context() call. Nothing justifies a full scale > > > revert. > > > > I can't say I'm aware of any consensus for how to handle this, to be > > completely honest with you. > > Well, I work in a lot of subsystems and this is a surprise to me and > not something I've seen before. Fix the bug, move forward. Reverts are > a cultural admission of failure. I use threats of a revert as a hammer > to encourage people to pay attention to the bugs. I hardly ever > actually revert things. What does reverting their code say to my > submitters??? Huh. I guess I'm lucky never to have worked in a environment where that is the case. In fact, my experience is quite the opposite: revert first so that things get back to a working state and the developer/submitter has some breathing room to rework the broken code. It's actually fairly blameless if you get it right and when you have a half-functional CI it's pretty much a necessity. Anyway, I digress... So if you see me appearing to be heavy-handed with reverts when dealing with regressions, it's really nothing tactical. Rather, it's purely about keeping the driver in a known functional state. > > > I'll send another patch if you want, but it seems like a waste of all > > > our time. > > > > It's a bug fix, of course it's a waste of time! We're talking minutes > > though, right? > > It becomes hard for maintainers to juggle the tress since you have to > send the revert to -rc and the fix on top of the rc to next. Again, I > will send the patch, just discussing. I've never had any difficulty managing that, so I think we'll be ok. Worst case, I can merge my fixes queue into my next queue. Will
