On Thu, Jan 21, 2016 at 11:44:14AM +0300, Andrey Ryabinin wrote:
> @@ -1391,6 +1391,101 @@ static inline void bdi_dirty_limits(struct
> backing_dev_info *bdi,
> }
> }
>
> +static void balance_dirty_pages_ub(struct address_space *mapping,
> + unsigned long write_chunk)
> +{
> + long ub_dirty, ub_writeback;
> + long ub_thresh, ub_background_thresh;
> + unsigned long pages_written = 0;
> + unsigned long pause = 1;
> + struct user_beancounter *ub = get_io_ub();
> +
> + struct backing_dev_info *bdi = mapping->backing_dev_info;
> +
> + for (;;) {
> + ub_dirty = ub_stat_get(ub, dirty_pages);
> + ub_writeback = ub_stat_get(ub, writeback_pages);
> +
> + if (!ub_dirty_limits(&ub_background_thresh, &ub_thresh, ub))
> + break;
> +
> + /*
> + * Check thresholds, set dirty_exceeded flags and
> + * start background writeback before throttling.
> + */
We don't use dirty_exceeded flags, please update the comment.
Also, please submit a separate patch removing user_beancounter->ub_flags
and UB_DIRTY_EXCEEDED.
> + if (ub_dirty + ub_writeback <= ub_thresh)
> + break;
> + if (!writeback_in_progress(bdi))
> + bdi_start_background_writeback(bdi);
> +
> + /*
> + * Throttle it only when the background writeback cannot
> + * catch-up. This avoids (excessively) small writeouts
> + * when the bdi limits are ramping up.
> + */
> + if (ub_dirty + ub_writeback <
> + (ub_background_thresh + ub_thresh) / 2)
> + break;
> +
> + if (ub_dirty > ub_thresh) {
> + pages_written = writeback_inodes_wb(&bdi->wb,
> write_chunk,
You need to subtract the number of pages written from write_chunk here,
obviously.
> + WB_REASON_BACKGROUND, ub);
> + ub_dirty = ub_stat_get(ub, dirty_pages);
> + ub_writeback = ub_stat_get(ub, writeback_pages);
> + }
> +
> + /* fixup ub-stat per-cpu drift to avoid false-positive */
> + if (ub_dirty + ub_writeback > ub_thresh &&
> + ub_dirty + ub_writeback - ub_thresh <
> + UB_STAT_BATCH * num_possible_cpus()) {
> + ub_dirty = ub_stat_get_exact(ub, dirty_pages);
> + ub_writeback = ub_stat_get_exact(ub, writeback_pages);
> + }
> +
> + if (ub_dirty + ub_writeback <= ub_thresh)
> + break;
> +
> + if (pages_written >= write_chunk)
> + break; /* We've done our duty */
> +
> + __set_current_state(TASK_KILLABLE);
> + io_schedule_timeout(pause);
> +
> + /*
> + * Increase the delay for each loop, up to our previous
> + * default of taking a 100ms nap.
> + */
> + pause <<= 1;
> + if (pause > HZ / 10)
> + pause = HZ / 10;
> +
> + if (fatal_signal_pending(current))
> + break;
> + }
> +
> + virtinfo_notifier_call(VITYPE_IO, VIRTINFO_IO_BALANCE_DIRTY,
> + (void*)write_chunk);
> +
> + /*
> + * Even if this is filtered writeback for other ub it will write
> + * inodes for this ub, because ub->dirty_exceeded is set.
> + */
Again, stale comment. Update please.
> + if (writeback_in_progress(bdi))
> + return;
> +
> + /*
> + * In laptop mode, we wait until hitting the higher threshold before
> + * starting background writeout, and then write out all the way down
> + * to the lower threshold. So slow writers cause minimal disk activity.
> + *
> + * In normal mode, we start background writeout at the lower
> + * background_thresh, to keep the amount of dirty memory low.
> + */
I don't think we need laptop_mode handling.
> + if ((laptop_mode && pages_written) ||
> + (!laptop_mode && ub_dirty > ub_background_thresh))
> + bdi_start_background_writeback(bdi);
> +}
> +
> /*
> * balance_dirty_pages() must be called by processes which are generating
> dirty
> * data. It looks at the number of dirty pages in the machine and will force
_______________________________________________
Devel mailing list
[email protected]
https://lists.openvz.org/mailman/listinfo/devel
