On Fri, Jan 22, 2021 at 05:20:32PM +0100, Christoph Hellwig wrote: > This takes the approach from Dave, but adds a new flag instead of abusing > the nowait one, and keeps a simpler calling convention for iomap_dio_rw. Hm. I realized while putting together for-next branches that I really would have preferred the three iomap patches at the start so that I could push those parts through the iomap tree. The changes required to resequence the series is minor and the iomap changes (AFAICT) are inert if the calling fs doesn't set IOMAP_DIO_OVERWRITE_ONLY, so I think it's low risk to push the iomap changes into iomap-for-next as a 5.12 thing. The rest of the xfs patches in this series would form the basis of a second week pull request (or not) since I think I ought to evaluate the effects on performance for a little longer. That said, if you /don't/ want me to pull any of this for 5.12, holler and I'll pull back. :) --D > Changes since v3: > - further comment improvements > - micro-optimize an alignment check > > Changes since v2: > - add another sanity check in __iomap_dio_rw > - rename IOMAP_DIO_UNALIGNED to IOMAP_DIO_OVERWRITE_ONLY > - use more consistent parameter naming > - improve a few comments > - cleanup an if statement > > Changes since v1: > - rename the new flags > - add an EOF check for subblock I/O > - minor cleanups