Re: [RFC 2/2] x86_64, mm: Reinsert the absent memory

[Ingo Molnar <mingo@kernel.org>]

* H. Peter Anvin <hpa@zytor.com> wrote:

> On 06/23/2013 02:32 AM, Ingo Molnar wrote:
> > 
> > * Ingo Molnar <mingo@kernel.org> wrote:
> > 
> >> Yet another thing to consider would be to implement an initialization 
> >> speedup of 3 orders of magnitude: initialize on the large page (2MB) 
> >> grandularity and on-demand delay the initialization of the 4K granular 
> >> struct pages [but still allocating them] - which I suspect are a good 
> >> chunk of the overhead? That way we could initialize in 2MB steps and speed 
> >> up the 2 hours bootup of 32 TB of RAM to 14 seconds...
> >>
> >> [ The cost would be one more branch in the buddy allocator, to detect
> >>   not-yet-initialized 2 MB chunks as we encounter them. Acceptable I 
> >>   think. ]
> > 
> > One advantage of this scheme would be that we could use it on pretty much 
> > any box, it would provide instant boot time speedups everywhere [a couple 
> > of hundred msecs on a small 4GB box - significant I think] - and would 
> > spread out and parallelize initialization to later stages.
> 
> Even better if we could start at the 1 GB level, which most of these 
> really huge machines will have hardware support for.

That might be a bit too granular: if we hit such an uninitialized block of 
memory we'd have to process 262,144 pages - potentially from an IRQ 
handler that does GFP_ATOMIC... or other latency critical code.

With 2MB we'd have to on-demand initialize 512 pages, which shouldn't show 
up during normal use.

Thanks,

	Ingo