On 14/12/2020 19:29, Jens Axboe wrote: > On 12/14/20 12:09 PM, Bijan Mottahedeh wrote: >> Just a ping. Anything I can do to facilitate the review, please let me >> know. > > I'll get to this soon - sorry that this means that it'll miss 5.11, but > I wanted to make sure that we get this absolutely right. It is > definitely an interesting and useful feature, but worth spending the > necessary time on to ensure we don't have any mistakes we'll regret > later. I'll take a look as I familiarised myself with it before. Also, io-wq punting is probably needed to be fixed first. > For your question, yes I think we could add sqe->update_flags (something > like that) and union it with the other flags, and add a flag that means > we're updating buffers instead of files. A bit iffy with the naming of > the opcode itself, but probably still a useful way to go. #define OPCODE_UPDATE_RESOURCES OPCODE_UPDATE_FILES With define + documenting that they're same IMHO should be fine. > > I'd also love to see a bunch of test cases for this that exercise all > parts of it. > -- Pavel Begunkov
