On Tue, Oct 24, 2006 at 01:28:31PM -0700, Badari Pulavarty wrote: > On Tue, 2006-10-24 at 15:34 -0400, Chris Mason wrote: > > Hello, > > > > Here is a new cut of my O_DIRECT locking rework. It has a much lower > > cpu cost than the last set, and simple benchmarks no longer show a > > regression here in system time. But, the complexity for inserting > > placeholder pages has gone up. > > > > I've also changed the way I test for a place holder page (from > > mm/filemap.c): > > > > static struct address_space placeholder_address_space; > > #define PagePlaceHolder(page) ((page)->mapping == &placeholder_address_space) > > > > This is more stable than the last one but I'm just starting to run race > > and load testing on it. > > > > -chris > > Gave it a spin with simple fsx tests and ran into .. Thanks, I missed an i_mutex change. Let me audit mutexes coming in and out again and resend. I think the new rules should be: reads: don't need i_mutex at all writes: don't need i_mutex for io inside i_size. Oh, and don't mess with i_mutex under DIO_OWN_LOCKING. -chris - To unsubscribe from this list: send the line "unsubscribe linux-fsdevel" in the body of a message to majordomo@xxxxxxxxxxxxxxx More majordomo info at http://vger.kernel.org/majordomo-info.html
