On 2024/5/20 16:49, Jan Kara wrote: > On Thu 16-05-24 16:27:25, Zhang Yi wrote: >> On 2024/5/15 8:25, Jan Kara wrote: >>> On Mon 13-05-24 15:21:19, Zhang Yi wrote: >>> Also accessing j_commit_sequence without any >>> lock is theoretically problematic wrt compiler optimization. You should have >>> READ_ONCE() there and the places modifying j_commit_sequence need to use >>> WRITE_ONCE(). >>> >> >> Thanks for pointing this out, but I'm not sure if we have to need READ_ONCE() >> here. IIUC, if we add READ_ONCE(), we could make sure to get the latest >> j_commit_sequence, if not, there is a window (it might becomes larger) that >> we could get the old value and jbd2_transaction_committed() could return false >> even if the given transaction was just committed, but I think the window is >> always there, so it looks like it is not a big problem, is that right? > > Well, all accesses to any memory should use READ_ONCE(), be protected by a > lock, or use types that handle atomicity on assembly level (like atomic_t, > or atomic bit operations and similar). Otherwise the compiler is free to > assume the underlying memory cannot change and generate potentionally > invalid code. In this case, I don't think realistically any compiler will > do it but still it is a good practice and also it saves us from KCSAN > warnings. If you want to know more details about possible problems, see > > tools/memory-model/Documentation/explanation.txt > > chapter "PLAIN ACCESSES AND DATA RACES". > Sure, this document is really helpful, I'll add READ_ONCE() and WRITE_ONCE() here, thanks a lot. Yi.
