On Wed, 24 Nov 2021 at 22:44, Jiri Kosina <jikos@xxxxxxxxxx> wrote: > On Wed, 24 Nov 2021, Marco Elver wrote: > > > Oh dear, I don't think we were ready -- this patch would cause all > > kinds of issues: > > https://lkml.kernel.org/r/CANpmjNOHN7SWu-pKGr9EBb3=in2AWiGmqNb6sYwhebGtRk+1uQ@xxxxxxxxxxxxxx > > Yeah, I wasn't expecting it to be queued right away after you've raised > your concerns indeed :) > > I still hate the fact that it's pretending that real BUG() is happening, > while it's in fact not, in passion (yes, it did bite me :) ), with > passion. Well, unfortunately the term "BUG" is overloaded in the kernel, which is probably part of the problem. I think we just have to live with that. The other thing is, you have a lot more to complain about: git grep 'pr_.*BUG:' ;-) Most of them don't panic either (and aren't BUG()s). > How many known tools are there that are relying on this format? I don't know. Off the top of my head, various bots (syzbot, Intel test robot?), also internal CI pipelines; some personal scripts (although they look for "BUG:"); probably many more. The reality is that this format is "the interface" now with which tools extract errors from dmesg. Changing it is akin to changing an API. Would you be happier if we documented everything in one file? Though given git grep 'pr.*BUG:', this is probably a huge task. But maybe we just have to list the ones that panic by default or can be made to panic with a flag?
