diff mbox series

[v10,23/50] KVM: SEV: Make AVIC backing, VMSA and VMCB memory allocation SNP safe

Message ID 20231016132819.1002933-24-michael.roth@amd.com (mailing list archive)
State New
Headers show
Series Add AMD Secure Nested Paging (SEV-SNP) Hypervisor Support | expand

Commit Message

Michael Roth Oct. 16, 2023, 1:27 p.m. UTC
From: Brijesh Singh <brijesh.singh@amd.com>

Implement a workaround for an SNP erratum where the CPU will incorrectly
signal an RMP violation #PF if a hugepage (2mb or 1gb) collides with the
RMP entry of a VMCB, VMSA or AVIC backing page.

When SEV-SNP is globally enabled, the CPU marks the VMCB, VMSA, and AVIC
backing pages as "in-use" via a reserved bit in the corresponding RMP
entry after a successful VMRUN. This is done for _all_ VMs, not just
SNP-Active VMs.

If the hypervisor accesses an in-use page through a writable
translation, the CPU will throw an RMP violation #PF. On early SNP
hardware, if an in-use page is 2mb aligned and software accesses any
part of the associated 2mb region with a hupage, the CPU will
incorrectly treat the entire 2mb region as in-use and signal a spurious
RMP violation #PF.

The recommended is to not use the hugepage for the VMCB, VMSA or
AVIC backing page for similar reasons. Add a generic allocator that will
ensure that the page returns is not hugepage (2mb or 1gb) and is safe to
be used when SEV-SNP is enabled. Also implement similar handling for the
VMCB/VMSA pages of nested guests.

Co-developed-by: Marc Orr <marcorr@google.com>
Signed-off-by: Marc Orr <marcorr@google.com>
Reported-by: Alper Gun <alpergun@google.com> # for nested VMSA case
Co-developed-by: Ashish Kalra <ashish.kalra@amd.com>
Signed-off-by: Ashish Kalra <ashish.kalra@amd.com>
Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
[mdr: squash in nested guest handling from Ashish]
Signed-off-by: Michael Roth <michael.roth@amd.com>
---
 arch/x86/include/asm/kvm-x86-ops.h |  1 +
 arch/x86/include/asm/kvm_host.h    |  1 +
 arch/x86/kvm/lapic.c               |  5 ++++-
 arch/x86/kvm/svm/nested.c          |  2 +-
 arch/x86/kvm/svm/sev.c             | 33 ++++++++++++++++++++++++++++++
 arch/x86/kvm/svm/svm.c             | 17 ++++++++++++---
 arch/x86/kvm/svm/svm.h             |  1 +
 7 files changed, 55 insertions(+), 5 deletions(-)

Comments

Vlastimil Babka Dec. 11, 2023, 1:24 p.m. UTC | #1
On 10/16/23 15:27, Michael Roth wrote:
> From: Brijesh Singh <brijesh.singh@amd.com>
> 
> Implement a workaround for an SNP erratum where the CPU will incorrectly
> signal an RMP violation #PF if a hugepage (2mb or 1gb) collides with the
> RMP entry of a VMCB, VMSA or AVIC backing page.
> 
> When SEV-SNP is globally enabled, the CPU marks the VMCB, VMSA, and AVIC
> backing pages as "in-use" via a reserved bit in the corresponding RMP
> entry after a successful VMRUN. This is done for _all_ VMs, not just
> SNP-Active VMs.
> 
> If the hypervisor accesses an in-use page through a writable
> translation, the CPU will throw an RMP violation #PF. On early SNP
> hardware, if an in-use page is 2mb aligned and software accesses any
> part of the associated 2mb region with a hupage, the CPU will
> incorrectly treat the entire 2mb region as in-use and signal a spurious
> RMP violation #PF.
> 
> The recommended is to not use the hugepage for the VMCB, VMSA or
> AVIC backing page for similar reasons. Add a generic allocator that will
> ensure that the page returns is not hugepage (2mb or 1gb) and is safe to

This is a bit confusing wording as we are not avoiding "using a
hugepage" but AFAIU, avoiding using a (4k) page that has a hugepage
aligned physical address, right?

> be used when SEV-SNP is enabled. Also implement similar handling for the
> VMCB/VMSA pages of nested guests.
> 
> Co-developed-by: Marc Orr <marcorr@google.com>
> Signed-off-by: Marc Orr <marcorr@google.com>
> Reported-by: Alper Gun <alpergun@google.com> # for nested VMSA case
> Co-developed-by: Ashish Kalra <ashish.kalra@amd.com>
> Signed-off-by: Ashish Kalra <ashish.kalra@amd.com>
> Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
> [mdr: squash in nested guest handling from Ashish]
> Signed-off-by: Michael Roth <michael.roth@amd.com>
> ---

<snip>

> +
> +struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu)
> +{
> +	unsigned long pfn;
> +	struct page *p;
> +
> +	if (!cpu_feature_enabled(X86_FEATURE_SEV_SNP))
> +		return alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +
> +	/*
> +	 * Allocate an SNP safe page to workaround the SNP erratum where
> +	 * the CPU will incorrectly signal an RMP violation  #PF if a
> +	 * hugepage (2mb or 1gb) collides with the RMP entry of VMCB, VMSA
> +	 * or AVIC backing page. The recommeded workaround is to not use the
> +	 * hugepage.

Same here "not use the hugepage"

> +	 *
> +	 * Allocate one extra page, use a page which is not 2mb aligned
> +	 * and free the other.

This makes more sense.

> +	 */
> +	p = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO, 1);
> +	if (!p)
> +		return NULL;
> +
> +	split_page(p, 1);

Yeah I think that's a sensible use of split_page(), as we don't have
support for forcefully non-aligned allocations or specific "page
coloring" in the page allocator.
So even with my wording concerns:

Acked-by: Vlastimil Babka <vbabka@suse.cz>

> +
> +	pfn = page_to_pfn(p);
> +	if (IS_ALIGNED(pfn, PTRS_PER_PMD))
> +		__free_page(p++);
> +	else
> +		__free_page(p + 1);
> +
> +	return p;
> +}
> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> index 1e7fb1ea45f7..8e4ef0cd968a 100644
> --- a/arch/x86/kvm/svm/svm.c
> +++ b/arch/x86/kvm/svm/svm.c
> @@ -706,7 +706,7 @@ static int svm_cpu_init(int cpu)
>  	int ret = -ENOMEM;
>  
>  	memset(sd, 0, sizeof(struct svm_cpu_data));
> -	sd->save_area = alloc_page(GFP_KERNEL | __GFP_ZERO);
> +	sd->save_area = snp_safe_alloc_page(NULL);
>  	if (!sd->save_area)
>  		return ret;
>  
> @@ -1425,7 +1425,7 @@ static int svm_vcpu_create(struct kvm_vcpu *vcpu)
>  	svm = to_svm(vcpu);
>  
>  	err = -ENOMEM;
> -	vmcb01_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +	vmcb01_page = snp_safe_alloc_page(vcpu);
>  	if (!vmcb01_page)
>  		goto out;
>  
> @@ -1434,7 +1434,7 @@ static int svm_vcpu_create(struct kvm_vcpu *vcpu)
>  		 * SEV-ES guests require a separate VMSA page used to contain
>  		 * the encrypted register state of the guest.
>  		 */
> -		vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +		vmsa_page = snp_safe_alloc_page(vcpu);
>  		if (!vmsa_page)
>  			goto error_free_vmcb_page;
>  
> @@ -4876,6 +4876,16 @@ static int svm_vm_init(struct kvm *kvm)
>  	return 0;
>  }
>  
> +static void *svm_alloc_apic_backing_page(struct kvm_vcpu *vcpu)
> +{
> +	struct page *page = snp_safe_alloc_page(vcpu);
> +
> +	if (!page)
> +		return NULL;
> +
> +	return page_address(page);
> +}
> +
>  static struct kvm_x86_ops svm_x86_ops __initdata = {
>  	.name = KBUILD_MODNAME,
>  
> @@ -5007,6 +5017,7 @@ static struct kvm_x86_ops svm_x86_ops __initdata = {
>  
>  	.vcpu_deliver_sipi_vector = svm_vcpu_deliver_sipi_vector,
>  	.vcpu_get_apicv_inhibit_reasons = avic_vcpu_get_apicv_inhibit_reasons,
> +	.alloc_apic_backing_page = svm_alloc_apic_backing_page,
>  };
>  
>  /*
> diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
> index c13070d00910..b7b8bf73cbb9 100644
> --- a/arch/x86/kvm/svm/svm.h
> +++ b/arch/x86/kvm/svm/svm.h
> @@ -694,6 +694,7 @@ void sev_es_vcpu_reset(struct vcpu_svm *svm);
>  void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
>  void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa);
>  void sev_es_unmap_ghcb(struct vcpu_svm *svm);
> +struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu);
>  
>  /* vmenter.S */
>
Kalra, Ashish Dec. 12, 2023, midnight UTC | #2
Hello Vlastimil,

On 12/11/2023 7:24 AM, Vlastimil Babka wrote:
> On 10/16/23 15:27, Michael Roth wrote:
>> From: Brijesh Singh <brijesh.singh@amd.com>
>>
>> Implement a workaround for an SNP erratum where the CPU will incorrectly
>> signal an RMP violation #PF if a hugepage (2mb or 1gb) collides with the
>> RMP entry of a VMCB, VMSA or AVIC backing page.
>>
>> When SEV-SNP is globally enabled, the CPU marks the VMCB, VMSA, and AVIC
>> backing pages as "in-use" via a reserved bit in the corresponding RMP
>> entry after a successful VMRUN. This is done for _all_ VMs, not just
>> SNP-Active VMs.
>>
>> If the hypervisor accesses an in-use page through a writable
>> translation, the CPU will throw an RMP violation #PF. On early SNP
>> hardware, if an in-use page is 2mb aligned and software accesses any
>> part of the associated 2mb region with a hupage, the CPU will
>> incorrectly treat the entire 2mb region as in-use and signal a spurious
>> RMP violation #PF.
>>
>> The recommended is to not use the hugepage for the VMCB, VMSA or
>> AVIC backing page for similar reasons. Add a generic allocator that will
>> ensure that the page returns is not hugepage (2mb or 1gb) and is safe to
> 
> This is a bit confusing wording as we are not avoiding "using a
> hugepage" but AFAIU, avoiding using a (4k) page that has a hugepage
> aligned physical address, right?

Yes.

> 
>> be used when SEV-SNP is enabled. Also implement similar handling for the
>> VMCB/VMSA pages of nested guests.
>>
>> Co-developed-by: Marc Orr <marcorr@google.com>
>> Signed-off-by: Marc Orr <marcorr@google.com>
>> Reported-by: Alper Gun <alpergun@google.com> # for nested VMSA case
>> Co-developed-by: Ashish Kalra <ashish.kalra@amd.com>
>> Signed-off-by: Ashish Kalra <ashish.kalra@amd.com>
>> Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
>> [mdr: squash in nested guest handling from Ashish]
>> Signed-off-by: Michael Roth <michael.roth@amd.com>
>> ---
> 
> <snip>
> 
>> +
>> +struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu)
>> +{
>> +	unsigned long pfn;
>> +	struct page *p;
>> +
>> +	if (!cpu_feature_enabled(X86_FEATURE_SEV_SNP))
>> +		return alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
>> +
>> +	/*
>> +	 * Allocate an SNP safe page to workaround the SNP erratum where
>> +	 * the CPU will incorrectly signal an RMP violation  #PF if a
>> +	 * hugepage (2mb or 1gb) collides with the RMP entry of VMCB, VMSA
>> +	 * or AVIC backing page. The recommeded workaround is to not use the
>> +	 * hugepage.
> 
> Same here "not use the hugepage"
> 
>> +	 *
>> +	 * Allocate one extra page, use a page which is not 2mb aligned
>> +	 * and free the other.
> 
> This makes more sense.
> 
>> +	 */
>> +	p = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO, 1);
>> +	if (!p)
>> +		return NULL;
>> +
>> +	split_page(p, 1);
> > Yeah I think that's a sensible use of split_page(), as we don't have
> support for forcefully non-aligned allocations or specific "page
> coloring" in the page allocator.

Yes, using split_page() allows us to free the additionally allocated 
page individually.

Thanks,
Ashish

> So even with my wording concerns:
> 
> Acked-by: Vlastimil Babka <vbabka@suse.cz>
Paolo Bonzini Dec. 13, 2023, 1:31 p.m. UTC | #3
On 10/16/23 15:27, Michael Roth wrote:
> From: Brijesh Singh <brijesh.singh@amd.com>
> 
> Implement a workaround for an SNP erratum where the CPU will incorrectly
> signal an RMP violation #PF if a hugepage (2mb or 1gb) collides with the
> RMP entry of a VMCB, VMSA or AVIC backing page.
> 
> When SEV-SNP is globally enabled, the CPU marks the VMCB, VMSA, and AVIC
> backing pages as "in-use" via a reserved bit in the corresponding RMP
> entry after a successful VMRUN. This is done for _all_ VMs, not just
> SNP-Active VMs.
> 
> If the hypervisor accesses an in-use page through a writable
> translation, the CPU will throw an RMP violation #PF. On early SNP
> hardware, if an in-use page is 2mb aligned and software accesses any
> part of the associated 2mb region with a hupage, the CPU will
> incorrectly treat the entire 2mb region as in-use and signal a spurious
> RMP violation #PF.

I don't understand if this can happen even if SEV-SNP is not in use, 
just because it is supported on the host?  If so, should this be Cc'd to 
stable?  (I can tweak the wording and submit it).

Paolo
Paolo Bonzini Dec. 13, 2023, 6:45 p.m. UTC | #4
On 10/16/23 15:27, Michael Roth wrote:
> From: Brijesh Singh <brijesh.singh@amd.com>
> 
> Implement a workaround for an SNP erratum where the CPU will incorrectly
> signal an RMP violation #PF if a hugepage (2mb or 1gb) collides with the
> RMP entry of a VMCB, VMSA or AVIC backing page.
> 
> When SEV-SNP is globally enabled, the CPU marks the VMCB, VMSA, and AVIC
> backing pages as "in-use" via a reserved bit in the corresponding RMP
> entry after a successful VMRUN. This is done for _all_ VMs, not just
> SNP-Active VMs.
> 
> If the hypervisor accesses an in-use page through a writable
> translation, the CPU will throw an RMP violation #PF. On early SNP
> hardware, if an in-use page is 2mb aligned and software accesses any
> part of the associated 2mb region with a hupage, the CPU will
> incorrectly treat the entire 2mb region as in-use and signal a spurious
> RMP violation #PF.
> 
> The recommended is to not use the hugepage for the VMCB, VMSA or
> AVIC backing page for similar reasons. Add a generic allocator that will
> ensure that the page returns is not hugepage (2mb or 1gb) and is safe to
> be used when SEV-SNP is enabled. Also implement similar handling for the
> VMCB/VMSA pages of nested guests.
> 
> Co-developed-by: Marc Orr <marcorr@google.com>
> Signed-off-by: Marc Orr <marcorr@google.com>
> Reported-by: Alper Gun <alpergun@google.com> # for nested VMSA case
> Co-developed-by: Ashish Kalra <ashish.kalra@amd.com>
> Signed-off-by: Ashish Kalra <ashish.kalra@amd.com>
> Signed-off-by: Brijesh Singh <brijesh.singh@amd.com>
> [mdr: squash in nested guest handling from Ashish]
> Signed-off-by: Michael Roth <michael.roth@amd.com>

Based on the discussion with Borislav, please move this earlier in the 
series, before patch 6.

Paolo

> ---
>   arch/x86/include/asm/kvm-x86-ops.h |  1 +
>   arch/x86/include/asm/kvm_host.h    |  1 +
>   arch/x86/kvm/lapic.c               |  5 ++++-
>   arch/x86/kvm/svm/nested.c          |  2 +-
>   arch/x86/kvm/svm/sev.c             | 33 ++++++++++++++++++++++++++++++
>   arch/x86/kvm/svm/svm.c             | 17 ++++++++++++---
>   arch/x86/kvm/svm/svm.h             |  1 +
>   7 files changed, 55 insertions(+), 5 deletions(-)
> 
> diff --git a/arch/x86/include/asm/kvm-x86-ops.h b/arch/x86/include/asm/kvm-x86-ops.h
> index f1505a5fa781..4ef2eca14287 100644
> --- a/arch/x86/include/asm/kvm-x86-ops.h
> +++ b/arch/x86/include/asm/kvm-x86-ops.h
> @@ -136,6 +136,7 @@ KVM_X86_OP(vcpu_deliver_sipi_vector)
>   KVM_X86_OP_OPTIONAL_RET0(vcpu_get_apicv_inhibit_reasons);
>   KVM_X86_OP_OPTIONAL_RET0(gmem_prepare)
>   KVM_X86_OP_OPTIONAL(gmem_invalidate)
> +KVM_X86_OP_OPTIONAL(alloc_apic_backing_page)
>   
>   #undef KVM_X86_OP
>   #undef KVM_X86_OP_OPTIONAL
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index fa401cb1a552..a3983271ea28 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -1763,6 +1763,7 @@ struct kvm_x86_ops {
>   
>   	int (*gmem_prepare)(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn, int max_order);
>   	void (*gmem_invalidate)(kvm_pfn_t start, kvm_pfn_t end);
> +	void *(*alloc_apic_backing_page)(struct kvm_vcpu *vcpu);
>   };
>   
>   struct kvm_x86_nested_ops {
> diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
> index dcd60b39e794..631a554c0f48 100644
> --- a/arch/x86/kvm/lapic.c
> +++ b/arch/x86/kvm/lapic.c
> @@ -2810,7 +2810,10 @@ int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
>   
>   	vcpu->arch.apic = apic;
>   
> -	apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
> +	if (kvm_x86_ops.alloc_apic_backing_page)
> +		apic->regs = static_call(kvm_x86_alloc_apic_backing_page)(vcpu);
> +	else
> +		apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
>   	if (!apic->regs) {
>   		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
>   		       vcpu->vcpu_id);
> diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
> index dd496c9e5f91..1f9a3f9eb985 100644
> --- a/arch/x86/kvm/svm/nested.c
> +++ b/arch/x86/kvm/svm/nested.c
> @@ -1194,7 +1194,7 @@ int svm_allocate_nested(struct vcpu_svm *svm)
>   	if (svm->nested.initialized)
>   		return 0;
>   
> -	vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +	vmcb02_page = snp_safe_alloc_page(&svm->vcpu);
>   	if (!vmcb02_page)
>   		return -ENOMEM;
>   	svm->nested.vmcb02.ptr = page_address(vmcb02_page);
> diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
> index 088b32657f46..1cfb9232fc74 100644
> --- a/arch/x86/kvm/svm/sev.c
> +++ b/arch/x86/kvm/svm/sev.c
> @@ -3211,3 +3211,36 @@ void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
>   		break;
>   	}
>   }
> +
> +struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu)
> +{
> +	unsigned long pfn;
> +	struct page *p;
> +
> +	if (!cpu_feature_enabled(X86_FEATURE_SEV_SNP))
> +		return alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +
> +	/*
> +	 * Allocate an SNP safe page to workaround the SNP erratum where
> +	 * the CPU will incorrectly signal an RMP violation  #PF if a
> +	 * hugepage (2mb or 1gb) collides with the RMP entry of VMCB, VMSA
> +	 * or AVIC backing page. The recommeded workaround is to not use the
> +	 * hugepage.
> +	 *
> +	 * Allocate one extra page, use a page which is not 2mb aligned
> +	 * and free the other.
> +	 */
> +	p = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO, 1);
> +	if (!p)
> +		return NULL;
> +
> +	split_page(p, 1);
> +
> +	pfn = page_to_pfn(p);
> +	if (IS_ALIGNED(pfn, PTRS_PER_PMD))
> +		__free_page(p++);
> +	else
> +		__free_page(p + 1);
> +
> +	return p;
> +}
> diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
> index 1e7fb1ea45f7..8e4ef0cd968a 100644
> --- a/arch/x86/kvm/svm/svm.c
> +++ b/arch/x86/kvm/svm/svm.c
> @@ -706,7 +706,7 @@ static int svm_cpu_init(int cpu)
>   	int ret = -ENOMEM;
>   
>   	memset(sd, 0, sizeof(struct svm_cpu_data));
> -	sd->save_area = alloc_page(GFP_KERNEL | __GFP_ZERO);
> +	sd->save_area = snp_safe_alloc_page(NULL);
>   	if (!sd->save_area)
>   		return ret;
>   
> @@ -1425,7 +1425,7 @@ static int svm_vcpu_create(struct kvm_vcpu *vcpu)
>   	svm = to_svm(vcpu);
>   
>   	err = -ENOMEM;
> -	vmcb01_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +	vmcb01_page = snp_safe_alloc_page(vcpu);
>   	if (!vmcb01_page)
>   		goto out;
>   
> @@ -1434,7 +1434,7 @@ static int svm_vcpu_create(struct kvm_vcpu *vcpu)
>   		 * SEV-ES guests require a separate VMSA page used to contain
>   		 * the encrypted register state of the guest.
>   		 */
> -		vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +		vmsa_page = snp_safe_alloc_page(vcpu);
>   		if (!vmsa_page)
>   			goto error_free_vmcb_page;
>   
> @@ -4876,6 +4876,16 @@ static int svm_vm_init(struct kvm *kvm)
>   	return 0;
>   }
>   
> +static void *svm_alloc_apic_backing_page(struct kvm_vcpu *vcpu)
> +{
> +	struct page *page = snp_safe_alloc_page(vcpu);
> +
> +	if (!page)
> +		return NULL;
> +
> +	return page_address(page);
> +}
> +
>   static struct kvm_x86_ops svm_x86_ops __initdata = {
>   	.name = KBUILD_MODNAME,
>   
> @@ -5007,6 +5017,7 @@ static struct kvm_x86_ops svm_x86_ops __initdata = {
>   
>   	.vcpu_deliver_sipi_vector = svm_vcpu_deliver_sipi_vector,
>   	.vcpu_get_apicv_inhibit_reasons = avic_vcpu_get_apicv_inhibit_reasons,
> +	.alloc_apic_backing_page = svm_alloc_apic_backing_page,
>   };
>   
>   /*
> diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
> index c13070d00910..b7b8bf73cbb9 100644
> --- a/arch/x86/kvm/svm/svm.h
> +++ b/arch/x86/kvm/svm/svm.h
> @@ -694,6 +694,7 @@ void sev_es_vcpu_reset(struct vcpu_svm *svm);
>   void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
>   void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa);
>   void sev_es_unmap_ghcb(struct vcpu_svm *svm);
> +struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu);
>   
>   /* vmenter.S */
>
Borislav Petkov Dec. 18, 2023, 2:57 p.m. UTC | #5
Just typos:

On Mon, Oct 16, 2023 at 08:27:52AM -0500, Michael Roth wrote:
> From: Brijesh Singh <brijesh.singh@amd.com>
> 
> Implement a workaround for an SNP erratum where the CPU will incorrectly
> signal an RMP violation #PF if a hugepage (2mb or 1gb) collides with the
> RMP entry of a VMCB, VMSA or AVIC backing page.
> 
> When SEV-SNP is globally enabled, the CPU marks the VMCB, VMSA, and AVIC
> backing pages as "in-use" via a reserved bit in the corresponding RMP
> entry after a successful VMRUN. This is done for _all_ VMs, not just
> SNP-Active VMs.
> 
> If the hypervisor accesses an in-use page through a writable
> translation, the CPU will throw an RMP violation #PF. On early SNP
> hardware, if an in-use page is 2mb aligned and software accesses any
> part of the associated 2mb region with a hupage, the CPU will

"hugepage"

> incorrectly treat the entire 2mb region as in-use and signal a spurious
> RMP violation #PF.
> 
> The recommended is to not use the hugepage for the VMCB, VMSA or

s/recommended/recommendation/
s/the hugepage/a hugepage/

> AVIC backing page for similar reasons. Add a generic allocator that will
> ensure that the page returns is not hugepage (2mb or 1gb) and is safe to

"... the page returned is not a hugepage..."

...

> +struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu)
> +{
> +	unsigned long pfn;
> +	struct page *p;
> +
> +	if (!cpu_feature_enabled(X86_FEATURE_SEV_SNP))
> +		return alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
> +
> +	/*
> +	 * Allocate an SNP safe page to workaround the SNP erratum where
> +	 * the CPU will incorrectly signal an RMP violation  #PF if a
> +	 * hugepage (2mb or 1gb) collides with the RMP entry of VMCB, VMSA
> +	 * or AVIC backing page. The recommeded workaround is to not use the

"recommended"

> +	 * hugepage.
> +	 *
> +	 * Allocate one extra page, use a page which is not 2mb aligned
> +	 * and free the other.
> +	 */
> +	p = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO, 1);
> +	if (!p)
> +		return NULL;
> +
> +	split_page(p, 1);
> +
> +	pfn = page_to_pfn(p);
> +	if (IS_ALIGNED(pfn, PTRS_PER_PMD))
> +		__free_page(p++);
> +	else
> +		__free_page(p + 1);
> +
> +	return p;
> +}
diff mbox series

Patch

diff --git a/arch/x86/include/asm/kvm-x86-ops.h b/arch/x86/include/asm/kvm-x86-ops.h
index f1505a5fa781..4ef2eca14287 100644
--- a/arch/x86/include/asm/kvm-x86-ops.h
+++ b/arch/x86/include/asm/kvm-x86-ops.h
@@ -136,6 +136,7 @@  KVM_X86_OP(vcpu_deliver_sipi_vector)
 KVM_X86_OP_OPTIONAL_RET0(vcpu_get_apicv_inhibit_reasons);
 KVM_X86_OP_OPTIONAL_RET0(gmem_prepare)
 KVM_X86_OP_OPTIONAL(gmem_invalidate)
+KVM_X86_OP_OPTIONAL(alloc_apic_backing_page)
 
 #undef KVM_X86_OP
 #undef KVM_X86_OP_OPTIONAL
diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
index fa401cb1a552..a3983271ea28 100644
--- a/arch/x86/include/asm/kvm_host.h
+++ b/arch/x86/include/asm/kvm_host.h
@@ -1763,6 +1763,7 @@  struct kvm_x86_ops {
 
 	int (*gmem_prepare)(struct kvm *kvm, kvm_pfn_t pfn, gfn_t gfn, int max_order);
 	void (*gmem_invalidate)(kvm_pfn_t start, kvm_pfn_t end);
+	void *(*alloc_apic_backing_page)(struct kvm_vcpu *vcpu);
 };
 
 struct kvm_x86_nested_ops {
diff --git a/arch/x86/kvm/lapic.c b/arch/x86/kvm/lapic.c
index dcd60b39e794..631a554c0f48 100644
--- a/arch/x86/kvm/lapic.c
+++ b/arch/x86/kvm/lapic.c
@@ -2810,7 +2810,10 @@  int kvm_create_lapic(struct kvm_vcpu *vcpu, int timer_advance_ns)
 
 	vcpu->arch.apic = apic;
 
-	apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
+	if (kvm_x86_ops.alloc_apic_backing_page)
+		apic->regs = static_call(kvm_x86_alloc_apic_backing_page)(vcpu);
+	else
+		apic->regs = (void *)get_zeroed_page(GFP_KERNEL_ACCOUNT);
 	if (!apic->regs) {
 		printk(KERN_ERR "malloc apic regs error for vcpu %x\n",
 		       vcpu->vcpu_id);
diff --git a/arch/x86/kvm/svm/nested.c b/arch/x86/kvm/svm/nested.c
index dd496c9e5f91..1f9a3f9eb985 100644
--- a/arch/x86/kvm/svm/nested.c
+++ b/arch/x86/kvm/svm/nested.c
@@ -1194,7 +1194,7 @@  int svm_allocate_nested(struct vcpu_svm *svm)
 	if (svm->nested.initialized)
 		return 0;
 
-	vmcb02_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
+	vmcb02_page = snp_safe_alloc_page(&svm->vcpu);
 	if (!vmcb02_page)
 		return -ENOMEM;
 	svm->nested.vmcb02.ptr = page_address(vmcb02_page);
diff --git a/arch/x86/kvm/svm/sev.c b/arch/x86/kvm/svm/sev.c
index 088b32657f46..1cfb9232fc74 100644
--- a/arch/x86/kvm/svm/sev.c
+++ b/arch/x86/kvm/svm/sev.c
@@ -3211,3 +3211,36 @@  void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector)
 		break;
 	}
 }
+
+struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu)
+{
+	unsigned long pfn;
+	struct page *p;
+
+	if (!cpu_feature_enabled(X86_FEATURE_SEV_SNP))
+		return alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
+
+	/*
+	 * Allocate an SNP safe page to workaround the SNP erratum where
+	 * the CPU will incorrectly signal an RMP violation  #PF if a
+	 * hugepage (2mb or 1gb) collides with the RMP entry of VMCB, VMSA
+	 * or AVIC backing page. The recommeded workaround is to not use the
+	 * hugepage.
+	 *
+	 * Allocate one extra page, use a page which is not 2mb aligned
+	 * and free the other.
+	 */
+	p = alloc_pages(GFP_KERNEL_ACCOUNT | __GFP_ZERO, 1);
+	if (!p)
+		return NULL;
+
+	split_page(p, 1);
+
+	pfn = page_to_pfn(p);
+	if (IS_ALIGNED(pfn, PTRS_PER_PMD))
+		__free_page(p++);
+	else
+		__free_page(p + 1);
+
+	return p;
+}
diff --git a/arch/x86/kvm/svm/svm.c b/arch/x86/kvm/svm/svm.c
index 1e7fb1ea45f7..8e4ef0cd968a 100644
--- a/arch/x86/kvm/svm/svm.c
+++ b/arch/x86/kvm/svm/svm.c
@@ -706,7 +706,7 @@  static int svm_cpu_init(int cpu)
 	int ret = -ENOMEM;
 
 	memset(sd, 0, sizeof(struct svm_cpu_data));
-	sd->save_area = alloc_page(GFP_KERNEL | __GFP_ZERO);
+	sd->save_area = snp_safe_alloc_page(NULL);
 	if (!sd->save_area)
 		return ret;
 
@@ -1425,7 +1425,7 @@  static int svm_vcpu_create(struct kvm_vcpu *vcpu)
 	svm = to_svm(vcpu);
 
 	err = -ENOMEM;
-	vmcb01_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
+	vmcb01_page = snp_safe_alloc_page(vcpu);
 	if (!vmcb01_page)
 		goto out;
 
@@ -1434,7 +1434,7 @@  static int svm_vcpu_create(struct kvm_vcpu *vcpu)
 		 * SEV-ES guests require a separate VMSA page used to contain
 		 * the encrypted register state of the guest.
 		 */
-		vmsa_page = alloc_page(GFP_KERNEL_ACCOUNT | __GFP_ZERO);
+		vmsa_page = snp_safe_alloc_page(vcpu);
 		if (!vmsa_page)
 			goto error_free_vmcb_page;
 
@@ -4876,6 +4876,16 @@  static int svm_vm_init(struct kvm *kvm)
 	return 0;
 }
 
+static void *svm_alloc_apic_backing_page(struct kvm_vcpu *vcpu)
+{
+	struct page *page = snp_safe_alloc_page(vcpu);
+
+	if (!page)
+		return NULL;
+
+	return page_address(page);
+}
+
 static struct kvm_x86_ops svm_x86_ops __initdata = {
 	.name = KBUILD_MODNAME,
 
@@ -5007,6 +5017,7 @@  static struct kvm_x86_ops svm_x86_ops __initdata = {
 
 	.vcpu_deliver_sipi_vector = svm_vcpu_deliver_sipi_vector,
 	.vcpu_get_apicv_inhibit_reasons = avic_vcpu_get_apicv_inhibit_reasons,
+	.alloc_apic_backing_page = svm_alloc_apic_backing_page,
 };
 
 /*
diff --git a/arch/x86/kvm/svm/svm.h b/arch/x86/kvm/svm/svm.h
index c13070d00910..b7b8bf73cbb9 100644
--- a/arch/x86/kvm/svm/svm.h
+++ b/arch/x86/kvm/svm/svm.h
@@ -694,6 +694,7 @@  void sev_es_vcpu_reset(struct vcpu_svm *svm);
 void sev_vcpu_deliver_sipi_vector(struct kvm_vcpu *vcpu, u8 vector);
 void sev_es_prepare_switch_to_guest(struct sev_es_save_area *hostsa);
 void sev_es_unmap_ghcb(struct vcpu_svm *svm);
+struct page *snp_safe_alloc_page(struct kvm_vcpu *vcpu);
 
 /* vmenter.S */