On 2020-11-15 05:52:33 [+0100], Mike Galbraith wrote: > On Sat, 2020-11-14 at 13:24 -0600, Tom Zanussi wrote: > > On Sat, 2020-11-14 at 20:00 +0100, Mike Galbraith wrote: > > > > > __raw_write_seqcount_end() is an integral part of write_sequnlock(), > > > but we do seem to be missing a seqcount_release() in 5.4-rt. > > > > > > > Yep, you're right, it's just the missing seqcount_release() - I'll > > resubmit with just that. > > Or just drop the backport, since it adds annotation, while the original > was fixing existing annotation. > > __raw_write_seqcount_begin() called in 5.4-rt try_write_seqlock() is > not annotated, while write_seqcount_begin() called by the 5.9-rt > version leads to the broken annotation that the original then fixed. That is correct. I was looking at the 5.4-RT series Steven posted and I was under the impression that this patch was correctly missing in previous RT since I even added the stable tag. As Mike said, the previous RT implementation did not use seqlock annotation, they used a spin-lock instead. So the "try_write_seqlock()" did the try-lock annotation. With the reworked seqcount implementation (v5.6-RT time frame) this was solved differently (closer to what upstream does) and the now the annotation was wrong and fixed. Sorry for that. > -Mike Sebastian
