[Ksummit-2013-discuss] [ATTEND] Linux VM Infrastructure to support Memory Power Management

[H. Peter Anvin]

On 07/30/2013 11:47 AM, Arjan van de Ven wrote:
> 
> sort of missed it ;-)
> 
> right now we do "we free everything and when all is free, we tell the hw
> to turn it off"
> (and if halfway the freeing we find a page can't free, we just blew away
> a lot of cache for no gain)
> 
> what I am suggesting we do "we keep track of non-clean/mapped pages, and
> we try to unmap them all;
> if all are unmapped and clean, THEN we try to power off. And if power
> off succeeds, we retroactively
> declare the clean unmapped ones as wiped"
> 
> that way, if we DON'T power off... we still have all the content in it.
> 

Somewhat obvious:

1. This will require that power off failure doesn't involve the
   potential loss of contents.

2. We cannot allow a re-add until the wiping is complete.

	-hpa