A pair of suspend2 patches
Posted Jul 30, 2005 8:36 UTC (Sat) by NCunningham
Parent article: A pair of suspend2 patches
"...simply flushing all pages to disk prior to beginning the suspend process would do the trick..."
Hi. This is precisely the point I've trying to make: Flushing all pages to disk prior to beginning the suspend process wouldn't do the trick. It is entirely possible that other processes will submit I/O while or after the flush runs. In addition, if other processes are submitting I/O while the sync is running, the sync may not return until after the other processes have submitted an awfully large amount of I/O (depending on patterns). I've been told that work is in progress to address this, but it still leaves the first problem.
The only way to address the race is to remove the race: stop processes submitting new I/O, sync the I/O that is there and then get on with suspending.
to post comments)