Eliminate the "pass" variable in the page daemon control loop.

[markj]

Summary

Only scan the inactive queue when the PID controller outputs a positive
value. Move the lowmem eventhandler invocation to a separate subroutine.
Fix up some stale comments.

Now we pause() for 0.1s after an inactive queue scan failed to meet its
target. In a separate change I'd like to pause() only if the active
queue scan met its target *and* the inactive queue scan failed to meet
its target.

As a side effect, this fixes a serious bug: after r329882, we never
reset pass to 0, so we are invoking the lowmem handlers every 10
seconds even when no memory shortage is taking place.

Test Plan

So far I've only verified that we boot.

[markj]

Ping?

[alc]

In D15491#327604, @markj wrote:

Ping?

After a cursory reading, I'll say that I like the changes. I'll try to read it carefully in a day or so.

[markj]

In D15491#327721, @alc wrote:

In D15491#327604, @markj wrote:

Ping?

After a cursory reading, I'll say that I like the changes. I'll try to read it carefully in a day or so.

Thanks! No rush.

I just realized that this change has the same goal as D13644. The discussion there is still relevant; in particular, with the PID controller change it now doesn't make sense for v_free_target to be set as high as it is: the controller will produce a positive output as soon as v_free_count < v_free_target, and the page daemon runs the controller once every 100ms. In other words, we start freeing pages more or less immediately after v_free_count drops below v_free_target. I plan to address that, but in a separate change.

[jeff]

In D15491#327763, @markj wrote:

I just realized that this change has the same goal as D13644. The discussion there is still relevant; in particular, with the PID controller change it now doesn't make sense for v_free_target to be set as high as it is: the controller will produce a positive output as soon as v_free_count < v_free_target, and the page daemon runs the controller once every 100ms. In other words, we start freeing pages more or less immediately after v_free_count drops below v_free_target. I plan to address that, but in a separate change.

The problem with lowering the v_free_target is you end up with stalls under heavy memory consumption. It may be an ok tradeoff for more memory if you can change it appreciably. We could also change the rise time for the PID controller to help.

As we discussed, we really need to be more conservative about when we call reclaim. I think we have some good options to pursue but I don't want to hold this fix up. I had the pass fix in my branch for months and forgot to commit.

[alc]

head/sys/vm/vm_pageout.c

1351

The comment still mentions "pass".

[markj]

head/sys/vm/vm_pageout.c

1351

How about, "Returns true if enough pages were freed by the inactive queue scan to address the shortage."?

[alc]

head/sys/vm/vm_pageout.c

1351

I would suggest moving the phrase "the inactive queue" to the first sentence, and then the second sentence can be very short and simple.