On Wed, Jan 15, 2020 at 10:05:00PM -0800, Dan Williams wrote: > On Wed, Jan 15, 2020 at 9:39 PM Darrick J. Wong <darrick.wong@xxxxxxxxxx> wrote: > [..] > > > attempts to minimize software cache effects for both I/O and > > > memory mappings of this file. It requires a file system which > > > has been configured to support DAX. > > > > > > DAX generally assumes all accesses are via cpu load / store > > > instructions which can minimize overhead for small accesses, but > > > may adversely affect cpu utilization for large transfers. > > > > > > File I/O is done directly to/from user-space buffers and memory > > > mapped I/O may be performed with direct memory mappings that > > > bypass kernel page cache. > > > > > > While the DAX property tends to result in data being transferred > > > synchronously, it does not give the same guarantees of > > > synchronous I/O where data and the necessary metadata are > > > transferred together. > > > > (I'm frankly not sure that synchronous I/O actually guarantees that the > > metadata has hit stable storage...) > > Oh? That text was motivated by the open(2) man page description of O_SYNC. Ugh. "synchronous I/O" means two different things, depending on context. In the AIO context, it means "process context waits for operation completion direct", but in the O_SYNC context, it means "we guarantee data integrity for each I/O submitted". Indeed, O_SYNC AIO is a thing. i.e. we can do an "async sync write" to guarantee data integrity without directly waiting for it. Now try describing that only using the words "synchronous write" to describe both behaviours. :) IOWs, if you are talking about data integrity, you need to explicitly say "O_SYNC semantics", not "synchronous write", because "synchronous write" is totally ambiguous without the O_SYNC context of the open(2) man page... Cheers, Dave. -- Dave Chinner david@xxxxxxxxxxxxx
