[0/3] avoid unnecessary memslot rmap walks

Message

Anthony Yznaga June 2, 2020, 8:07 p.m. UTC

While investigating optimizing qemu start time for large memory guests
I found that kvm_mmu_slot_apply_flags() is walking rmaps to update
existing sptes when creating or moving a slot but that there won't be
any existing sptes to update and any sptes inserted once the new memslot
is visible won't need updating.  I can't find any reason for this not to
be the case, but I've taken a more cautious approach to fixing this by
dividing things into three patches.

Anthony Yznaga (3):
  KVM: x86: remove unnecessary rmap walk of read-only memslots
  KVM: x86: avoid unnecessary rmap walks when creating/moving slots
  KVM: x86: minor code refactor and comments fixup around dirty logging

 arch/x86/kvm/x86.c | 106 +++++++++++++++++++++++++----------------------------
 1 file changed, 49 insertions(+), 57 deletions(-)

Comments

Paolo Bonzini June 4, 2020, 6:42 p.m. UTC | #1

On 02/06/20 22:07, Anthony Yznaga wrote:
> While investigating optimizing qemu start time for large memory guests
> I found that kvm_mmu_slot_apply_flags() is walking rmaps to update
> existing sptes when creating or moving a slot but that there won't be
> any existing sptes to update and any sptes inserted once the new memslot
> is visible won't need updating.  I can't find any reason for this not to
> be the case, but I've taken a more cautious approach to fixing this by
> dividing things into three patches.
> 
> Anthony Yznaga (3):
>   KVM: x86: remove unnecessary rmap walk of read-only memslots
>   KVM: x86: avoid unnecessary rmap walks when creating/moving slots
>   KVM: x86: minor code refactor and comments fixup around dirty logging
> 
>  arch/x86/kvm/x86.c | 106 +++++++++++++++++++++++++----------------------------
>  1 file changed, 49 insertions(+), 57 deletions(-)
> 

Queued, thanks.

Paolo