Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::3:2 as permitted sender) client-ip=2620:137:e000::3:2;
Message-ID: <5d9aadbd-975b-4c4d-ba18-ac6e0fb07ba1@linux.intel.com>
Date:   Tue, 21 Nov 2023 17:57:28 +0800
MIME-Version: 1.0
User-Agent: Mozilla Thunderbird
Subject: Re: [PATCH v6 11/16] KVM: x86/tdp_mmu: Split the large page when zap
 leaf
To:     isaku.yamahata@intel.com
Cc:     kvm@vger.kernel.org, linux-kernel@vger.kernel.org,
        isaku.yamahata@gmail.com, Paolo Bonzini <pbonzini@redhat.com>,
        erdemaktas@google.com, Sean Christopherson <seanjc@google.com>,
        Sagi Shahar <sagis@google.com>,
        David Matlack <dmatlack@google.com>,
        Kai Huang <kai.huang@intel.com>,
        Zhi Wang <zhi.wang.linux@gmail.com>, chen.bo@intel.com,
        hang.yuan@intel.com, tina.zhang@intel.com,
        Xiaoyao Li <xiaoyao.li@intel.com>
References: <cover.1699368363.git.isaku.yamahata@intel.com>
 <8b43a9203c34b5330c4ea5901da5dac3458ac98d.1699368363.git.isaku.yamahata@intel.com>
From:   Binbin Wu <binbin.wu@linux.intel.com>
In-Reply-To: <8b43a9203c34b5330c4ea5901da5dac3458ac98d.1699368363.git.isaku.yamahata@intel.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 11/7/2023 11:00 PM, isaku.yamahata@intel.com wrote:
> From: Xiaoyao Li <xiaoyao.li@intel.com>
>
> When TDX enabled, a large page cannot be zapped if it contains mixed
> pages. In this case, it has to split the large page.
>
> Signed-off-by: Xiaoyao Li <xiaoyao.li@intel.com>
> ---
>   arch/x86/kvm/Kconfig            |  1 +
>   arch/x86/kvm/mmu/mmu.c          |  6 +--
>   arch/x86/kvm/mmu/mmu_internal.h |  9 +++++
>   arch/x86/kvm/mmu/tdp_mmu.c      | 68 +++++++++++++++++++++++++++++++--
>   4 files changed, 78 insertions(+), 6 deletions(-)
>
> diff --git a/arch/x86/kvm/Kconfig b/arch/x86/kvm/Kconfig
> index b0f103641547..557479737962 100644
> --- a/arch/x86/kvm/Kconfig
> +++ b/arch/x86/kvm/Kconfig
> @@ -93,6 +93,7 @@ config KVM_INTEL
>   	tristate "KVM for Intel (and compatible) processors support"
>   	depends on KVM && IA32_FEAT_CTL
>   	select KVM_SW_PROTECTED_VM if INTEL_TDX_HOST
> +	select KVM_GENERIC_MEMORY_ATTRIBUTES if INTEL_TDX_HOST
>   	select KVM_PRIVATE_MEM if INTEL_TDX_HOST
>   	help
>   	  Provides support for KVM on processors equipped with Intel's VT
> diff --git a/arch/x86/kvm/mmu/mmu.c b/arch/x86/kvm/mmu/mmu.c
> index 265177cedf37..0bf043812644 100644
> --- a/arch/x86/kvm/mmu/mmu.c
> +++ b/arch/x86/kvm/mmu/mmu.c
> @@ -7463,8 +7463,8 @@ bool kvm_arch_pre_set_memory_attributes(struct kvm *kvm,
>   	return kvm_unmap_gfn_range(kvm, range);
>   }
>   
> -static bool hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn,
> -				int level)
> +bool kvm_hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn,
> +			     int level)
>   {
>   	return lpage_info_slot(gfn, slot, level)->disallow_lpage & KVM_LPAGE_MIXED_FLAG;
>   }
> @@ -7491,7 +7491,7 @@ static bool hugepage_has_attrs(struct kvm *kvm, struct kvm_memory_slot *slot,
>   		return kvm_range_has_memory_attributes(kvm, start, end, attrs);
>   
>   	for (gfn = start; gfn < end; gfn += KVM_PAGES_PER_HPAGE(level - 1)) {
> -		if (hugepage_test_mixed(slot, gfn, level - 1) ||
> +		if (kvm_hugepage_test_mixed(slot, gfn, level - 1) ||
>   		    attrs != kvm_get_memory_attributes(kvm, gfn))
>   			return false;
>   	}
> diff --git a/arch/x86/kvm/mmu/mmu_internal.h b/arch/x86/kvm/mmu/mmu_internal.h
> index 1da98be74ad2..653e96769956 100644
> --- a/arch/x86/kvm/mmu/mmu_internal.h
> +++ b/arch/x86/kvm/mmu/mmu_internal.h
> @@ -460,4 +460,13 @@ void *mmu_memory_cache_alloc(struct kvm_mmu_memory_cache *mc);
>   void track_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
>   void untrack_possible_nx_huge_page(struct kvm *kvm, struct kvm_mmu_page *sp);
>   
> +#ifdef CONFIG_KVM_GENERIC_MEMORY_ATTRIBUTES
> +bool kvm_hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn, int level);
> +#else
> +static inline bool kvm_hugepage_test_mixed(struct kvm_memory_slot *slot, gfn_t gfn, int level)
> +{
> +	return false;
> +}
> +#endif
> +
>   #endif /* __KVM_X86_MMU_INTERNAL_H */
> diff --git a/arch/x86/kvm/mmu/tdp_mmu.c b/arch/x86/kvm/mmu/tdp_mmu.c
> index 7873e9ee82ad..a209a67decae 100644
> --- a/arch/x86/kvm/mmu/tdp_mmu.c
> +++ b/arch/x86/kvm/mmu/tdp_mmu.c
> @@ -964,6 +964,14 @@ bool kvm_tdp_mmu_zap_sp(struct kvm *kvm, struct kvm_mmu_page *sp)
>   	return true;
>   }
>   
> +
> +static struct kvm_mmu_page *tdp_mmu_alloc_sp_for_split(struct kvm *kvm,
> +						       struct tdp_iter *iter,
> +						       bool shared);
> +
> +static int tdp_mmu_split_huge_page(struct kvm *kvm, struct tdp_iter *iter,
> +				   struct kvm_mmu_page *sp, bool shared);
> +
>   /*
>    * If can_yield is true, will release the MMU lock and reschedule if the
>    * scheduler needs the CPU or there is contention on the MMU lock. If this
> @@ -975,13 +983,15 @@ static bool tdp_mmu_zap_leafs(struct kvm *kvm, struct kvm_mmu_page *root,
>   			      gfn_t start, gfn_t end, bool can_yield, bool flush,
>   			      bool zap_private)
>   {
> +	bool is_private = is_private_sp(root);
> +	struct kvm_mmu_page *split_sp = NULL;
>   	struct tdp_iter iter;
>   
>   	end = min(end, tdp_mmu_max_gfn_exclusive());
>   
>   	lockdep_assert_held_write(&kvm->mmu_lock);
>   
> -	WARN_ON_ONCE(zap_private && !is_private_sp(root));
> +	WARN_ON_ONCE(zap_private && !is_private);
>   	if (!zap_private && is_private_sp(root))
Can use is_private instead of is_private_sp(root) here as well.

>   		return false;
>   
> @@ -1006,12 +1016,66 @@ static bool tdp_mmu_zap_leafs(struct kvm *kvm, struct kvm_mmu_page *root,
>   		    !is_last_spte(iter.old_spte, iter.level))
>   			continue;
>   
> +		if (is_private && kvm_gfn_shared_mask(kvm) &&
> +		    is_large_pte(iter.old_spte)) {
> +			gfn_t gfn = iter.gfn & ~kvm_gfn_shared_mask(kvm);
> +			gfn_t mask = KVM_PAGES_PER_HPAGE(iter.level) - 1;
> +			struct kvm_memory_slot *slot;
> +			struct kvm_mmu_page *sp;
> +
> +			slot = gfn_to_memslot(kvm, gfn);
> +			if (kvm_hugepage_test_mixed(slot, gfn, iter.level) ||
> +			    (gfn & mask) < start ||
> +			    end < (gfn & mask) + KVM_PAGES_PER_HPAGE(iter.level)) {
> +				WARN_ON_ONCE(!can_yield);
> +				if (split_sp) {
> +					sp = split_sp;
> +					split_sp = NULL;
> +					sp->role = tdp_iter_child_role(&iter);
> +				} else {
> +					WARN_ON(iter.yielded);
> +					if (flush && can_yield) {
> +						kvm_flush_remote_tlbs(kvm);
> +						flush = false;
> +					}
Is it necessary to do the flush here?

> +					sp = tdp_mmu_alloc_sp_for_split(kvm, &iter, false);
> +					if (iter.yielded) {
> +						split_sp = sp;
> +						continue;
> +					}
> +				}
> +				KVM_BUG_ON(!sp, kvm);
> +
> +				tdp_mmu_init_sp(sp, iter.sptep, iter.gfn);
> +				if (tdp_mmu_split_huge_page(kvm, &iter, sp, false)) {
> +					kvm_flush_remote_tlbs(kvm);
> +					flush = false;
Why it needs to flush TLB immediately if tdp_mmu_split_huge_page() fails?

Also, when KVM MMU write lock is held, it seems tdp_mmu_split_huge_page()
will not fail. But let's assume this condition can be triggered, since 
sp is local
variable, it will lost its value after continue, and split_sp is also NULL,
it will try to allocate a new sp, memory leakage here?

> +					/* force retry on this gfn. */
> +					iter.yielded = true;
> +				} else
> +					flush = true;
> +				continue;
> +			}
> +		}
> +
>   		tdp_mmu_iter_set_spte(kvm, &iter, SHADOW_NONPRESENT_VALUE);
>   		flush = true;
>   	}
>   
>   	rcu_read_unlock();
>   
> +	if (split_sp) {
> +		WARN_ON(!can_yield);
> +		if (flush) {
> +			kvm_flush_remote_tlbs(kvm);
> +			flush = false;
> +		}
Same here, why we need to do the flush here?
Can we delay it till the caller do the flush?

> +
> +		write_unlock(&kvm->mmu_lock);
> +		tdp_mmu_free_sp(split_sp);
> +		write_lock(&kvm->mmu_lock);
> +	}
> +
>   	/*
>   	 * Because this flow zaps _only_ leaf SPTEs, the caller doesn't need
>   	 * to provide RCU protection as no 'struct kvm_mmu_page' will be freed.
> @@ -1606,8 +1670,6 @@ static struct kvm_mmu_page *tdp_mmu_alloc_sp_for_split(struct kvm *kvm,
>   
>   	KVM_BUG_ON(kvm_mmu_page_role_is_private(role) !=
>   		   is_private_sptep(iter->sptep), kvm);
> -	/* TODO: Large page isn't supported for private SPTE yet. */
> -	KVM_BUG_ON(kvm_mmu_page_role_is_private(role), kvm);
>   
>   	/*
>   	 * Since we are allocating while under the MMU lock we have to be