Re: [PATCH] KVM: arm/arm64: Skip updating page table entry if no change

From: Marc Zyngier
Date: Fri Aug 10 2018 - 10:41:04 EST


Hi Punit,

On Fri, 10 Aug 2018 12:13:00 +0100,
Punit Agrawal <punit.agrawal@xxxxxxx> wrote:
>
> Contention on updating a page table entry by a large number of vcpus
> can lead to duplicate work when handling stage 2 page faults. As the
> page table update follows the break-before-make requirement of the
> architecture, it can lead to repeated refaults due to clearing the
> entry and flushing the tlbs.
>
> This problem is more likely when -
>
> * there are large number of vcpus
> * the mapping is large block mapping
>
> such as when using PMD hugepages (512MB) with 64k pages.
>
> Fix this by skipping the page table update if there is no change in
> the entry being updated.
>
> Signed-off-by: Punit Agrawal <punit.agrawal@xxxxxxx>
> Cc: Marc Zyngier <marc.zyngier@xxxxxxx>
> Cc: Christoffer Dall <christoffer.dall@xxxxxxx>
> Cc: Suzuki Poulose <suzuki.poulose@xxxxxxx>

This definitely deserves a Cc to stable, and a Fixes: tag.

> --
> Hi,
>
> This problem was reported by a user when testing PUD hugepages. During
> VM restore when all threads are running cpu intensive workload, the
> refauting was causing the VM to not make any forward progress.
>
> This patch fixes the problem for PMD and PTE page fault handling.
>
> Thanks,
> Punit
>
> Change-Id: I04c9aa8b9fbada47deb1a171c9959f400a0d2a21
> ---
> virt/kvm/arm/mmu.c | 16 ++++++++++++++++
> 1 file changed, 16 insertions(+)
>
> diff --git a/virt/kvm/arm/mmu.c b/virt/kvm/arm/mmu.c
> index 1d90d79706bd..a66a5441ca2f 100644
> --- a/virt/kvm/arm/mmu.c
> +++ b/virt/kvm/arm/mmu.c
> @@ -1027,6 +1027,18 @@ static int stage2_set_pmd_huge(struct kvm *kvm, struct kvm_mmu_memory_cache
> VM_BUG_ON(pmd_present(*pmd) && pmd_pfn(*pmd) != pmd_pfn(*new_pmd));
>
> old_pmd = *pmd;
> + /*
> + * Multiple vcpus faulting on the same PMD entry, can lead to
> + * them sequentially updating the PMD with the same
> + * value. Following the break-before-make (pmd_clear()
> + * followed by tlb_flush()) process can hinder forward
> + * progress due to refaults generated on missing translations.
> + *
> + * Skip updating the page table if the entry is unchanged.
> + */
> + if (pmd_val(old_pmd) == pmd_val(*new_pmd))
> + return 0;

Shouldn't you take this opportunity to also refactor it with the above
VM_BUG_ON and the below pmd_present? At the moment, we end-up testing
pmd_present twice, and your patch is awfully similar to the VM_BUG_ON
one.

> +
> if (pmd_present(old_pmd)) {
> pmd_clear(pmd);
> kvm_tlb_flush_vmid_ipa(kvm, addr);
> @@ -1101,6 +1113,10 @@ static int stage2_set_pte(struct kvm *kvm, struct kvm_mmu_memory_cache *cache,
>
> /* Create 2nd stage page table mapping - Level 3 */
> old_pte = *pte;
> + /* Skip page table update if there is no change */
> + if (pte_val(old_pte) == pte_val(*new_pte))
> + return 0;
> +
> if (pte_present(old_pte)) {
> kvm_set_pte(pte, __pte(0));
> kvm_tlb_flush_vmid_ipa(kvm, addr);

Thanks,

M.

--
Jazz is not dead, it just smell funny.