On Wed, May 25, 2022 at 5:39 PM Sean Christopherson <seanjc@xxxxxxxxxx> wrote:
>
> On Wed, May 25, 2022, Johannes Weiner wrote:
> > On Tue, May 24, 2022 at 03:31:52PM -0700, Yosry Ahmed wrote:
> > > I don't have enough context to say whether we should piggyback KVM MMU
> > > pages to the existing NR_PAGETABLE item, but from a high level it
> > > seems like it would be more helpful if they are a separate stat.
> > > Anyway, I am willing to go with whatever Sean thinks is best.
> >
> > Somebody should work this out and put it into a changelog. It's
> > permanent ABI.
>
> After a lot of waffling, my vote is to add a dedicated NR_SECONDARY_PAGETABLE.
>
> It's somewhat redundant from a KVM perspective, as NR_SECONDARY_PAGETABLE will
> scale with KVM's per-VM pages_{4k,2m,1g} stats unless the guest is doing something
> bizarre, e.g. accessing only 4kb chunks of 2mb pages so that KVM is forced to
> allocate a large number of page tables even though the guest isn't accessing that
> much memory.
>
> But, someone would need to either understand how KVM works to make that connection,
> or know (or be told) to go look at KVM's stats if they're running VMs to better
> decipher the stats.
>
> And even in the little bit of time I played with this, I found having
> nr_page_table_pages side-by-side with nr_secondary_page_table_pages to be very
> informative. E.g. when backing a VM with THP versus HugeTLB,
> nr_secondary_page_table_pages is roughly the same, but nr_page_table_pages is an
> order of a magnitude higher with THP. I'm guessing the THP behavior is due to
> something triggering DoubleMap, but now I want to find out why that's happening.
>
> So while I'm pretty sure a clever user could glean the same info by cross-referencing
> NR_PAGETABLE stats with KVM stats, I think having NR_SECONDARY_PAGETABLE will at the
> very least prove to be helpful for understanding tradeoffs between VM backing types,
> and likely even steer folks towards potential optimizations.
>
> Baseline:
> # grep page_table /proc/vmstat
> nr_page_table_pages 2830
> nr_secondary_page_table_pages 0
>
> THP:
> # grep page_table /proc/vmstat
> nr_page_table_pages 7584
> nr_secondary_page_table_pages 140
>
> HugeTLB:
> # grep page_table /proc/vmstat
> nr_page_table_pages 3153
> nr_secondary_page_table_pages 153
>
Interesting findings! Thanks for taking the time to look into this, Sean!
I will refresh this patchset and summarize the discussion in the
commit message, and also fix some nits on the KVM side. Does this
sound good to everyone?
