[v2,0/5] KVM: mm: count KVM page table pages in pagetable stats

Message

Yosry Ahmed April 4, 2022, 11:41 p.m. UTC

We keep track of several kernel memory stats (total kernel memory, page
tables, stack, vmalloc, etc) on multiple levels (global, per-node,
per-memcg, etc). These stats give insights to users to how much memory
is used by the kernel and for what purposes.

Currently, memory used by kvm for its page tables is not accounted in
the pagetable stats. This patch series accounts the memory pages used by
KVM for page tables in those stats.

The riscv and mips patches are not tested properly due to lack of
resources. Feel free to test or drop them.

Yosry Ahmed (5):
  KVM: mm: add a helper to account page table pages used by KVM.
  KVM: x86: mm: count KVM page table pages in pagetable stats
  KVM: arm64: mm: count KVM page table pages in pagetable stats
  KVM: riscv: mm: count KVM page table pages in pagetable stats
  KVM: mips: mm: count KVM page table pages in pagetable stats

 arch/arm64/kernel/image-vars.h |  3 ++
 arch/arm64/kvm/hyp/pgtable.c   | 50 +++++++++++++++++++++-------------
 arch/mips/kvm/mips.c           |  1 +
 arch/mips/kvm/mmu.c            |  9 +++++-
 arch/riscv/kvm/mmu.c           | 26 +++++++++++++-----
 arch/x86/kvm/mmu/mmu.c         | 16 +++++++++--
 arch/x86/kvm/mmu/tdp_mmu.c     | 16 +++++++++--
 include/linux/kvm_host.h       |  9 ++++++
 8 files changed, 99 insertions(+), 31 deletions(-)

Comments

Sean Christopherson April 5, 2022, 6:44 p.m. UTC | #1

On Mon, Apr 04, 2022, Yosry Ahmed wrote:
> We keep track of several kernel memory stats (total kernel memory, page
> tables, stack, vmalloc, etc) on multiple levels (global, per-node,
> per-memcg, etc). These stats give insights to users to how much memory
> is used by the kernel and for what purposes.
> 
> Currently, memory used by kvm for its page tables is not accounted in
> the pagetable stats. This patch series accounts the memory pages used by
> KVM for page tables in those stats.

It's still not obvious to me that piggybacking NR_PAGETABLE is desirable, probably
because I am quite clueless as to how these stats are used on the backend.  E.g.
why not have a NR_SECONDARY_PAGETABLE entry to track pages used for secondary MMU
page tables?

Yosry Ahmed April 5, 2022, 6:50 p.m. UTC | #2

+Johannes Weiner  +Michal Hocko  +Roman Gushchin

On Tue, Apr 5, 2022 at 11:45 AM Sean Christopherson <[e-mail]> wrote:
>
> On Mon, Apr 04, 2022, Yosry Ahmed wrote:
> > We keep track of several kernel memory stats (total kernel memory, page
> > tables, stack, vmalloc, etc) on multiple levels (global, per-node,
> > per-memcg, etc). These stats give insights to users to how much memory
> > is used by the kernel and for what purposes.
> >
> > Currently, memory used by kvm for its page tables is not accounted in
> > the pagetable stats. This patch series accounts the memory pages used by
> > KVM for page tables in those stats.
>
> It's still not obvious to me that piggybacking NR_PAGETABLE is desirable, probably
> because I am quite clueless as to how these stats are used on the backend.  E.g.
> why not have a NR_SECONDARY_PAGETABLE entry to track pages used for secondary MMU
> page tables?

We can add NR_SECONDARY_PAGETABLE or even NR_KVM_PAGETABLE, but I am
not sure whether this separation is desired on the MM side. Let's see
what MM folks think about this.