Andrew Morton wrote:

Nick Piggin [off-list ref] wrote:

quoted

Andrew Morton wrote:

quoted

Nick Piggin [off-list ref] wrote:

quoted

Previously the ->protection[] logic was broken. It was difficult to follow
and basically didn't use the asynch reclaim watermarks properly.

eh?

Broken how?

min = (1<<order) + z->protection[alloc_type];

This value is used both as the condition for waking kswapd, and
whether or not to enter synch reclaim.

What should happen is kswapd gets woken at pages_low, and synch
reclaim is started at pages_min.

Are you aware of this:

void wakeup_kswapd(struct zone *zone)
{
	if (zone->free_pages > zone->pages_low)
		return;

?

Err, yes?

quoted

pages_low + protection and pages_min + protection, etc.

Nick, sorry, but I shouldn't have to expend these many braincells
decrypting your work.  Please: much better explanations, more testing
results.  This stuff is fiddly, sensitive and has a habit of blowing up in
our faces weeks later.  We need to be cautious.  The barriers are higher
nowadays.

OK previously, in a nutshell:

	for_each_zone(z) {
		if (z->free_pages < z->protection)
			continue;
		else
			goto got_pg;
	}

	for_each_zone(z)
		wakeup_kswapd(z);

	for_each_zone(z) {
		if (z->free_pages < z->protection)
			continue;
		else
			goto got_pg;
	}

	try_to_free_pages();

	try again;

After my patch:
	for_each_zone(z) {
		if (z->free_pages < z->pages_low + z->protection)
			continue;
		else
			goto got_pg;
	}

	for_each_zone(z)
		wakeup_kswapd(z);

	for_each_zone(z) {
		if (z->free_pages < z->pages_min + z->protection)
			continue;
		else
			goto got_pg;
	}

	try_to_free_pages();

	try again;

Ie, we have the (pages_low - pages_min) buffer after waking kswapd
before entering synch reclaim. Previously there was no buffer. I thought
this was the point of background reclaim. I don't know if I can explain
it any better than that sorry.
--
To unsubscribe, send a message with 'unsubscribe linux-mm' in
the body to majordomo@kvack.org.  For more info on Linux MM,
see: http://www.linux-mm.org/ .
Don't email: <a href=mailto:"aart@kvack.org"> aart@kvack.org </a>