On 13/04/2023 9:13 pm, Julien Grall wrote: > > On 13/04/2023 20:22, Andrew Cooper wrote: >> Despite rcu_read_unlock() being fully inlineable, the optimiser >> cannot fold >> these exit paths, because of the various compiler barriers providing RCU >> safety. Help the compiler out. > > Please mention which compiler(s) (including version) you used. > >> >> This compiles to marginally better code in all cases. > So the code itself is fine with me. But this raises a few questions. > If this is marginal, then why are you doing it? What's your end goal?
I happened to be working in the area while fixing a bug. But am not justifying "because I judged it to be worth doing" further; it is entirely self evident from the fact I sent the patch. Whether you meant it to be or not, the request comes across as insulting, and is not something which should be made of any submitter. But as this kind of thing has come up before, any further debate on the matter can be made to the code of conduct board. A better phrasing might have been "I'm sorry, I don't understand. Why is this an improvement?" But I'm only guessing as to what this issue is. But moving to the technicals aspects, in an attempt to help this along. Do you understand what folding the exit paths means? It's a term which is used frequently enough on list that it ought to be taken for granted, and is what in my opinion makes the patch entirely self-evident. > Lastly what do you mean by "all cases"? Is it all arch? All compilers? Yes. > > Anyway, if this pattern is important (TBD why), then I think we should > update the CODING_STYLE with some guidance. Otherwise, we may > introduce similar patterns (we already have some). Perhaps, but I don't have the time, energy, or willing to dive into the viper pit which is trying to make any changes to that document at all. Especially when there's a laundry list of more important topics that ought to take priority... ~Andrew
