-----Original Message----- From: penberg@gmail.com [mailto:penberg@gmail.com] On Behalf Of ext Pekka Enberg Sent: 08 May, 2012 11:03 To: KOSAKI Motohiro Cc: Anton Vorontsov; Minchan Kim; Moiseichuk Leonid (Nokia-MP/Espoo); John
...
That comes from a real-world requirement. See Leonid's email on the topic:
I know, many embedded guys prefer such timer interval. I also have an experience similar logic when I was TV box developer. but I must disagree. Someone hope timer housekeeping complexity into kernel. but I haven't seen any justification.
Leonid?
The "usleep(timeout); read(vmevent_fd)" will eliminate opportunity to use vmevent API for mobile devices. Developers already have to use heartbeat primitives to align/sync timers and update code which is not always simple to do. But the idea is to have user-space wakeup only if we have something change in memory numbers, thus aligned timers will not help much in vmevent case due to memory situation may change a lot in short time. Short depends from software stack but usually it below 1s. To have use-time and wakeups on good level (below 50Hz by e.g. powertop) and allow cpu switch off timers of such short period like 1s are not allowed.
Leonid PS: Sorry, meetings prevent to do interesting things :( I am tracking conversation with quite low understanding how it will be useful for practical needs because user-space developers in 80% cases needs to track simply dirty memory changes i.e. modified pages which cannot be dropped.