[Qemu-devel] [PATCH v2] KVM: pci-assign: do not map smm memory slot pages

[Herongguang (Stephen)]

From f6f0ee6831488bef7af841cb86f3d85a04848fe5 Mon Sep 17 00:00:00 2001
From: herongguang <address@hidden>
Date: Mon, 27 Mar 2017 15:08:59 +0800
Subject: [PATCH] KVM: pci-assign: do not map smm memory slot pages
 in vt-d page table

or VM memory are not put thus leaked in kvm_iommu_unmap_memslots() when
destroy VM.

This is consistent with current vfio implementation.
---
 virt/kvm/kvm_main.c | 2 +-
 1 file changed, 1 insertion(+), 1 deletion(-)

diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
index 482612b..9018d06 100644
--- a/virt/kvm/kvm_main.c
+++ b/virt/kvm/kvm_main.c
@@ -1052,7 +1052,7 @@ int __kvm_set_memory_region(struct kvm *kvm,
         * changes) is disallowed above, so any other attribute changes getting
         * here can be skipped.
         */
-       if ((change == KVM_MR_CREATE) || (change == KVM_MR_MOVE)) {
+       if ((as_id == 0) && ((change == KVM_MR_CREATE) || (change == 
KVM_MR_MOVE))) {
                r = kvm_iommu_map_pages(kvm, &new);
                return r;
        }
--
1.7.12.4

On 2017/3/25 19:14, herongguang wrote:
> or pages are not unmaped and freed
>
> Signed-off-by: herongguang <address@hidden>
> ---
>   arch/x86/kvm/iommu.c | 6 ++++--
>   1 file changed, 4 insertions(+), 2 deletions(-)
>
> Well, do we should change pci-assign to not map SMM slots instead? Like vfio.
>
> diff --git a/arch/x86/kvm/iommu.c b/arch/x86/kvm/iommu.c
> index b181426..5b931bb 100644
> --- a/arch/x86/kvm/iommu.c
> +++ b/arch/x86/kvm/iommu.c
> @@ -320,15 +320,17 @@ void kvm_iommu_unmap_pages(struct kvm *kvm, struct 
> kvm_memory_slot *slot)
>   static int kvm_iommu_unmap_memslots(struct kvm *kvm)
>   {
>         int idx;
> -       struct kvm_memslots *slots;
> +       struct kvm_memslots *slots, *smm_slots;
>         struct kvm_memory_slot *memslot;
>
>         idx = srcu_read_lock(&kvm->srcu);
>         slots = kvm_memslots(kvm);
> -
>         kvm_for_each_memslot(memslot, slots)
>                 kvm_iommu_unmap_pages(kvm, memslot);
>
> +       smm_slots = __kvm_memslots(kvm, 1);
> +       kvm_for_each_memslot(memslot, smm_slots)
> +               kvm_iommu_unmap_pages(kvm, memslot);
>         srcu_read_unlock(&kvm->srcu, idx);
>
>         if (kvm->arch.iommu_noncoherent)