On Wed, Dec 21, 2022 at 6:35 PM Vipin Sharma <vipinsh@xxxxxxxxxx> wrote:
>
> Page table pages of a VM are currently allocated based on the current
> task's NUMA node or its mempolicy. This can cause suboptimal remote
> accesses by the vCPU if it is accessing physical pages local to its NUMA
> node but the page table pages mapping those physcal pages were created
> by some other vCPU which was on different NUMA node or had different
> policy.
>
> Allocate page table pages on the same NUMA node where underlying
> physical page exists. Page table at level 5, 4, and 3 might not end up
> on the same NUMA node as they can span multiple NUMA nodes.
A page table at any level could map memory spanning multiple NUMA
nodes, it just becomes more likely at higher levels.
We're only guaranteed that a page table maps memory all on the same
node if it's a split hugepage.
This change can only guarantee that the page table pages are allocated
on the same node as at least some of the memory they map.
Of course in practice, the above is absolutely correct since we'd
expect to have multi-GB continuous ranges of GFNs allocated on the
same node via huge pages.
And since the root pages are allocated based only on where the thread
allocating them is running, they're not actually guaranteed to be on
the same node as any of the memory they map. (Though they probably
will be.)
>
> Signed-off-by: Vipin Sharma <vipinsh@xxxxxxxxxx>
> ---
> arch/x86/include/asm/kvm_host.h | 2 +-
> arch/x86/kvm/mmu/mmu.c | 63 ++++++++++++++++++++++-----------
> arch/x86/kvm/mmu/paging_tmpl.h | 4 +--
> arch/x86/kvm/mmu/tdp_mmu.c | 11 +++---
> virt/kvm/kvm_main.c | 2 +-
> 5 files changed, 53 insertions(+), 29 deletions(-)
>
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index 293994fabae3..b1f319ad6f89 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -782,7 +782,7 @@ struct kvm_vcpu_arch {
> struct kvm_mmu *walk_mmu;
>
> struct kvm_mmu_memory_cache mmu_pte_list_desc_cache;
> - struct kvm_mmu_memory_cache mmu_shadow_page_cache;
> + struct kvm_mmu_memory_cache mmu_shadow_page_cache[MAX_NUMNODES];
> struct kvm_mmu_memory_cache mmu_shadowed_info_cache;
> struct kvm_mmu_memory_cache mmu_page_header_cache;
>
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index 23a3b82b2384..511c6ef265ee 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -677,24 +677,29 @@ static int mmu_topup_sp_memory_cache(struct kvm_mmu_memory_cache *cache,
>
> static int mmu_topup_memory_caches(struct kvm_vcpu *vcpu, bool maybe_indirect)
> {
> - int r;
> + int r, nid;
>
> /* 1 rmap, 1 parent PTE per level, and the prefetched rmaps. */
> r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache,
> 1 + PT64_ROOT_MAX_LEVEL + PTE_PREFETCH_NUM);
> if (r)
> return r;
> - r = mmu_topup_sp_memory_cache(&vcpu->arch.mmu_shadow_page_cache,
> - &vcpu->arch.mmu_shadow_page_cache_lock,
> - PT64_ROOT_MAX_LEVEL);
> - if (r)
> - return r;
> +
> + for_each_online_node(nid) {
> + r = mmu_topup_sp_memory_cache(&vcpu->arch.mmu_shadow_page_cache[nid],
> + &vcpu->arch.mmu_shadow_page_cache_lock,
> + PT64_ROOT_MAX_LEVEL);
> + if (r)
> + return r;
> + }
> +
> if (maybe_indirect) {
> r = kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_shadowed_info_cache,
> PT64_ROOT_MAX_LEVEL);
> if (r)
> return r;
> }
> +
> return kvm_mmu_topup_memory_cache(&vcpu->arch.mmu_page_header_cache,
> PT64_ROOT_MAX_LEVEL);
> }
> @@ -715,9 +720,14 @@ static void mmu_free_sp_memory_cache(struct kvm_mmu_memory_cache *cache,
>
> static void mmu_free_memory_caches(struct kvm_vcpu *vcpu)
> {
> + int nid;
> +
> kvm_mmu_free_memory_cache(&vcpu->arch.mmu_pte_list_desc_cache);
> - mmu_free_sp_memory_cache(&vcpu->arch.mmu_shadow_page_cache,
> - &vcpu->arch.mmu_shadow_page_cache_lock);
> +
> + for_each_node(nid)
> + mmu_free_sp_memory_cache(&vcpu->arch.mmu_shadow_page_cache[nid],
> + &vcpu->arch.mmu_shadow_page_cache_lock);
> +
Was just trying to think if there could be any issue with memory
leakage if the online nodes changed, though IDK if any hardware does
that.
Still, it might be more robust to use ARRAY_SIZE and cover the whole array.
> kvm_mmu_free_memory_cache(&vcpu->arch.mmu_shadowed_info_cache);
> kvm_mmu_free_memory_cache(&vcpu->arch.mmu_page_header_cache);
> }
> @@ -2256,11 +2266,12 @@ static struct kvm_mmu_page *__kvm_mmu_get_shadow_page(struct kvm *kvm,
>
> static struct kvm_mmu_page *kvm_mmu_get_shadow_page(struct kvm_vcpu *vcpu,
> gfn_t gfn,
> - union kvm_mmu_page_role role)
> + union kvm_mmu_page_role role,
> + int nid)
> {
> struct shadow_page_caches caches = {
> .page_header_cache = &vcpu->arch.mmu_page_header_cache,
> - .shadow_page_cache = &vcpu->arch.mmu_shadow_page_cache,
> + .shadow_page_cache = &vcpu->arch.mmu_shadow_page_cache[nid],
> .shadowed_info_cache = &vcpu->arch.mmu_shadowed_info_cache,
> .shadow_page_cache_lock = &vcpu->arch.mmu_shadow_page_cache_lock
> };
> @@ -2316,15 +2327,19 @@ static union kvm_mmu_page_role kvm_mmu_child_role(u64 *sptep, bool direct,
>
> static struct kvm_mmu_page *kvm_mmu_get_child_sp(struct kvm_vcpu *vcpu,
> u64 *sptep, gfn_t gfn,
> - bool direct, unsigned int access)
> + bool direct, unsigned int access,
> + kvm_pfn_t pfn)
> {
> union kvm_mmu_page_role role;
> + int nid;
>
> if (is_shadow_present_pte(*sptep) && !is_large_pte(*sptep))
> return ERR_PTR(-EEXIST);
>
> role = kvm_mmu_child_role(sptep, direct, access);
> - return kvm_mmu_get_shadow_page(vcpu, gfn, role);
> + nid = kvm_pfn_to_page_table_nid(pfn);
> +
> + return kvm_mmu_get_shadow_page(vcpu, gfn, role, nid);
> }
>
> static void shadow_walk_init_using_root(struct kvm_shadow_walk_iterator *iterator,
> @@ -3208,7 +3223,8 @@ static int direct_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
> if (it.level == fault->goal_level)
> break;
>
> - sp = kvm_mmu_get_child_sp(vcpu, it.sptep, base_gfn, true, ACC_ALL);
> + sp = kvm_mmu_get_child_sp(vcpu, it.sptep, base_gfn, true,
> + ACC_ALL, fault->pfn);
> if (sp == ERR_PTR(-EEXIST))
> continue;
>
> @@ -3636,7 +3652,7 @@ static hpa_t mmu_alloc_root(struct kvm_vcpu *vcpu, gfn_t gfn, int quadrant,
> WARN_ON_ONCE(quadrant && !role.has_4_byte_gpte);
> WARN_ON_ONCE(role.direct && role.has_4_byte_gpte);
>
> - sp = kvm_mmu_get_shadow_page(vcpu, gfn, role);
> + sp = kvm_mmu_get_shadow_page(vcpu, gfn, role, numa_mem_id());
> ++sp->root_count;
>
> return __pa(sp->spt);
> @@ -5952,7 +5968,7 @@ static int __kvm_mmu_create(struct kvm_vcpu *vcpu, struct kvm_mmu *mmu)
>
> int kvm_mmu_create(struct kvm_vcpu *vcpu)
> {
> - int ret;
> + int ret, nid;
>
> INIT_KVM_MMU_MEMORY_CACHE(&vcpu->arch.mmu_pte_list_desc_cache,
> pte_list_desc_cache, NUMA_NO_NODE);
> @@ -5960,8 +5976,9 @@ int kvm_mmu_create(struct kvm_vcpu *vcpu)
> INIT_KVM_MMU_MEMORY_CACHE(&vcpu->arch.mmu_page_header_cache,
> mmu_page_header_cache, NUMA_NO_NODE);
>
> - INIT_KVM_MMU_MEMORY_CACHE(&vcpu->arch.mmu_shadow_page_cache,
> - NULL, NUMA_NO_NODE);
> + for_each_node(nid)
> + INIT_KVM_MMU_MEMORY_CACHE(&vcpu->arch.mmu_shadow_page_cache[nid],
> + NULL, nid);
> spin_lock_init(&vcpu->arch.mmu_shadow_page_cache_lock);
>
> vcpu->arch.mmu = &vcpu->arch.root_mmu;
> @@ -6692,13 +6709,17 @@ void kvm_mmu_invalidate_mmio_sptes(struct kvm *kvm, u64 gen)
> }
>
> static unsigned long mmu_shrink_cache(struct kvm_mmu_memory_cache *cache,
> + int cache_count,
> spinlock_t *cache_lock)
> {
> unsigned long freed = 0;
> + int nid;
>
> spin_lock(cache_lock);
> - if (cache->nobjs)
> - freed = kvm_mmu_empty_memory_cache(cache);
> + for (nid = 0; nid < cache_count; nid++) {
> + if (node_online(nid) && cache[nid].nobjs)
Is there any reason to keep the cache if !node_online(nid)?
Actually, I'd also just drop the cache_count argument and always
iterate over the entire array, only checking nobjs. There's no
guarantee I'm aware of that the set of nodes has a sequential series
of IDs starting at 0 and you'd get a bug if that wasn't the case since
it only iterates to nid < cache_count here but some of the earlier
nids might not have been online.
> + freed += kvm_mmu_empty_memory_cache(&cache[nid]);
> + }
> spin_unlock(cache_lock);
> return freed;
> }
> @@ -6721,13 +6742,15 @@ mmu_shrink_scan(struct shrinker *shrink, struct shrink_control *sc)
> list_move_tail(&kvm->vm_list, &vm_list);
>
> freed += mmu_shrink_cache(&kvm->arch.split_shadow_page_cache,
> + 1,
So lonely.
One.
All by itself,
with only a coma for company.
NIT: This could be merged to the previous or subsequent lines.
> &kvm->arch.split_shadow_page_cache_lock);
>
> if (freed >= sc->nr_to_scan)
> break;
>
> kvm_for_each_vcpu(i, vcpu, kvm) {
> - freed += mmu_shrink_cache(&vcpu->arch.mmu_shadow_page_cache,
> + freed += mmu_shrink_cache(vcpu->arch.mmu_shadow_page_cache,
> + MAX_NUMNODES,
> &vcpu->arch.mmu_shadow_page_cache_lock);
> }
>
> diff --git a/arch/x86/kvm/mmu/paging_tmpl.h b/arch/x86/kvm/mmu/paging_tmpl.h
> index e5662dbd519c..1ceca62ec4cf 100644
> --- a/arch/x86/kvm/mmu/paging_tmpl.h
> +++ b/arch/x86/kvm/mmu/paging_tmpl.h
> @@ -652,7 +652,7 @@ static int FNAME(fetch)(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault,
> table_gfn = gw->table_gfn[it.level - 2];
> access = gw->pt_access[it.level - 2];
> sp = kvm_mmu_get_child_sp(vcpu, it.sptep, table_gfn,
> - false, access);
> + false, access, fault->pfn);
>
> if (sp != ERR_PTR(-EEXIST)) {
> /*
> @@ -708,7 +708,7 @@ static int FNAME(fetch)(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault,
> validate_direct_spte(vcpu, it.sptep, direct_access);
>
> sp = kvm_mmu_get_child_sp(vcpu, it.sptep, base_gfn,
> - true, direct_access);
> + true, direct_access, fault->pfn);
> if (sp == ERR_PTR(-EEXIST))
> continue;
>
> diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
> index 376b8dceb3f9..b5abae2366dd 100644
> --- a/arch/x86/kvm/mmu/tdp_mmu.c
> +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> @@ -259,12 +259,12 @@ static struct kvm_mmu_page *tdp_mmu_next_root(struct kvm *kvm,
> kvm_mmu_page_as_id(_root) != _as_id) { \
> } else
>
> -static struct kvm_mmu_page *tdp_mmu_alloc_sp(struct kvm_vcpu *vcpu)
> +static struct kvm_mmu_page *tdp_mmu_alloc_sp(struct kvm_vcpu *vcpu, int nid)
> {
> struct kvm_mmu_page *sp;
>
> sp = kvm_mmu_memory_cache_alloc(&vcpu->arch.mmu_page_header_cache);
> - sp->spt = kvm_mmu_sp_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache,
> + sp->spt = kvm_mmu_sp_memory_cache_alloc(&vcpu->arch.mmu_shadow_page_cache[nid],
> &vcpu->arch.mmu_shadow_page_cache_lock);
>
> return sp;
> @@ -317,7 +317,7 @@ hpa_t kvm_tdp_mmu_get_vcpu_root_hpa(struct kvm_vcpu *vcpu)
> goto out;
> }
>
> - root = tdp_mmu_alloc_sp(vcpu);
> + root = tdp_mmu_alloc_sp(vcpu, numa_mem_id());
Might be worth calling out somewhere that the root page is just
allocated based on where the thread allocating it runs.
> tdp_mmu_init_sp(root, NULL, 0, role);
>
> refcount_set(&root->tdp_mmu_root_count, 1);
> @@ -1149,7 +1149,7 @@ int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
> struct kvm *kvm = vcpu->kvm;
> struct tdp_iter iter;
> struct kvm_mmu_page *sp;
> - int ret = RET_PF_RETRY;
> + int ret = RET_PF_RETRY, nid;
>
> kvm_mmu_hugepage_adjust(vcpu, fault);
>
> @@ -1178,11 +1178,12 @@ int kvm_tdp_mmu_map(struct kvm_vcpu *vcpu, struct kvm_page_fault *fault)
> !is_large_pte(iter.old_spte))
> continue;
>
> + nid = kvm_pfn_to_page_table_nid(fault->pfn);
> /*
> * The SPTE is either non-present or points to a huge page that
> * needs to be split.
> */
> - sp = tdp_mmu_alloc_sp(vcpu);
> + sp = tdp_mmu_alloc_sp(vcpu, nid);
> tdp_mmu_init_child_sp(sp, &iter);
>
> sp->nx_huge_page_disallowed = fault->huge_page_disallowed;
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index d96c8146e9ba..4f3db7ffeba8 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -415,7 +415,7 @@ static inline void *mmu_memory_cache_alloc_obj(struct kvm_mmu_memory_cache *mc,
> if (mc->kmem_cache)
> return kmem_cache_alloc(mc->kmem_cache, gfp_flags);
> else
> - return (void *)__get_free_page(gfp_flags);
> + return kvm_mmu_get_free_page(mc->node, gfp_flags);
You could do part of this change in the commit that introduced
kvm_mmu_get_free_page too.
> }
>
> int __kvm_mmu_topup_memory_cache(struct kvm_mmu_memory_cache *mc, int capacity, int min)
> --
> 2.39.0.314.g84b9a713c41-goog
>
