Re: Thinking aloud.

On 10/20/2012 09:02 AM, Oleksandr Natalenko wrote:
> /me thinks that's great idea, but the main obstacle is (really) extra
> swap-or-so space needed to compare with.
Yeah, extra storage needed is an obstacle, but the new functionality 
could always be made entirely optional. I don't think comparing would be 
so much of an problem. I'm not thinking of loading the old contents and 
comparing, but of getting the SHA1 of the current contents and comparing 
with a SHA1 of the old contents that was either kept in memory or loaded 
prior to starting to write the image. The extra memory needed would just 
be sizeof(SHA1 checksum) * number_of_pages_in_previous_image - at least 
nominally.
> Another obstacle is using something like Gentoo hardened patches, where
> RAM pages are randomized intentionally (AFAIK).
That's as far as userspace is concerned? We're looking at things from 
the kernel perspective, so randomised addresses would have a minimal 
impact. There would have to be some sort of daemon shifting pages around 
in the kernel mapping for hibernation to be impacted.
> P.S. As for me vanilla hibernation really works OK for now (TOI also
> works OK for sure). May be it's time to merge some TOI features into
> mainline?
Well, I'd like to, but when I've tried in the past, I've always found it 
a waste of my time to try.

Regards,

Nigel