2017-11-27 17:04 GMT+08:00 Michal Hocko <mhocko@xxxxxxxx>: > On Mon 27-11-17 16:54:39, Yafang Shao wrote: >> 2017-11-27 16:52 GMT+08:00 Michal Hocko <mhocko@xxxxxxxx>: >> > On Mon 27-11-17 16:49:34, Yafang Shao wrote: >> >> 2017-11-27 16:37 GMT+08:00 Michal Hocko <mhocko@xxxxxxxx>: >> >> > On Mon 27-11-17 16:32:42, Yafang Shao wrote: >> >> >> 2017-11-27 16:29 GMT+08:00 Yafang Shao <laoar.shao@xxxxxxxxx>: >> > [...] >> >> >> > It will help us to find the error if we don't change these values like this. >> >> >> > >> >> >> >> >> >> And actually it help us find another issue that when availble_memroy >> >> >> is too small, the thresh and bg_thresh will be 0, that's absolutely >> >> >> wrong. >> >> > >> >> > Why is it wrong? >> >> > -- >> >> >> >> For example, the writeback threads will be wakeup on every write. >> >> I don't think it is meaningful to wakeup the writeback thread when the >> >> dirty pages is very low. >> > >> > Well, this is a corner situation when we are out of memory basically. >> > Doing a wake up on the flusher is the least of your problem. So _why_ >> > exactly is this is problem? >> > -- >> >> Are you _sure_ this is the least of the problem on this corner situation ? > > I am not _sure_ and that is why I'm _asking_ _you_ and you seem to come > up with reasons which don't make me really convinced. > > If we wake up flusher which have nothing to do they will simply back > off. That is what have to do anyway because dirty data can be truncated > at any time, right? > Under this condition it is better not to wakeup the bdi writeback. That's why I return the min value 100 to avoid doing this. Maybe 100 is not enough. It should be geater IMHO. >> Why wakeup the bdi writeback ? why not just kill the program and flush >> the date into the Disk when we do oom ? > > I really do not see how this is related to the discussion here. OOM > killer doesn't flush anything. It kills a task and that is it. Moreover > we do not flush any data from direct context at all. We simply rely on > flushers to do the work. > -- That's my fault. Just killing the program is enough. Thanks Yafang -- To unsubscribe, send a message with 'unsubscribe linux-mm' in the body to majordomo@xxxxxxxxx. For more info on Linux MM, see: http://www.linux-mm.org/ . Don't email: <a href=mailto:"dont@xxxxxxxxx";> email@xxxxxxxxx </a>
