Re: [PATCH v2 1/1] mm: Always release pages to the buddy allocator in memblock_free_late().

[Aaron Thompson]

On 2023-01-05 02:48, Ingo Molnar wrote:
> * Aaron Thompson <dev@xxxxxxxxxx> wrote:
>
> > For example, on an Amazon EC2 t3.micro VM (1 GB) booting via EFI:
> >
> > v6.2-rc2:
> >   # grep -E 'Node|spanned|present|managed' /proc/zoneinfo
> >   Node 0, zone      DMA
> >           spanned  4095
> >           present  3999
> >           managed  3840
> >   Node 0, zone    DMA32
> >           spanned  246652
> >           present  245868
> >           managed  178867
> >
> > v6.2-rc2 + patch:
> >   # grep -E 'Node|spanned|present|managed' /proc/zoneinfo
> >   Node 0, zone      DMA
> >           spanned  4095
> >           present  3999
> >           managed  3840
> >   Node 0, zone    DMA32
> >           spanned  246652
> >           present  245868
> >           managed  222816   # +43,949 pages
>
> [ Note the annotation I added to the output - might be useful in the
> changelog too. ]
>
> So this patch adds around +17% of RAM to this 1 GB virtual system? That
> looks rather significant ...
>
> Thanks,
>
> 	Ingo

It is significant, but I wouldn't describe it as being added. I would 
say that the system is currently losing 17% of RAM due to a bug, and 
this patch fixes that bug.

The actual numbers depend on the mappings given by the EFI, so they're 
largely out of our control. As an example, similar VMs that I run with 
the OVMF EFI lose about 3%. I couldn't say for sure which is the 
outlier, but my point is that the specific values are not really the 
focus, this is just an example that shows that the issue can be 
encountered in the wild with real impact. I know I'll be happy to get 
that memory back, whether it is 3% or 17% :)

Thanks,
-- Aaron