Re: [RFC PATCH 16/28] arm64: RME: Allow populating initial contents

[Zhi Wang]

On Fri, 10 Mar 2023 15:47:16 +0000
Steven Price <steven.price@xxxxxxx> wrote:

> On 06/03/2023 17:34, Zhi Wang wrote:
> > On Fri, 27 Jan 2023 11:29:20 +0000
> > Steven Price <steven.price@xxxxxxx> wrote:
> >   
> >> The VMM needs to populate the realm with some data before starting (e.g.
> >> a kernel and initrd). This is measured by the RMM and used as part of
> >> the attestation later on.
> >>
> >> Signed-off-by: Steven Price <steven.price@xxxxxxx>
> >> ---
> >>  arch/arm64/kvm/rme.c | 366 +++++++++++++++++++++++++++++++++++++++++++
> >>  1 file changed, 366 insertions(+)
> >>
> >> diff --git a/arch/arm64/kvm/rme.c b/arch/arm64/kvm/rme.c
> >> index 16e0bfea98b1..3405b43e1421 100644
> >> --- a/arch/arm64/kvm/rme.c
> >> +++ b/arch/arm64/kvm/rme.c
> >> @@ -4,6 +4,7 @@
> >>   */
> >>  
> >>  #include <linux/kvm_host.h>
> >> +#include <linux/hugetlb.h>
> >>  
> >>  #include <asm/kvm_emulate.h>
> >>  #include <asm/kvm_mmu.h>
> >> @@ -426,6 +427,359 @@ void kvm_realm_unmap_range(struct kvm *kvm, unsigned long ipa, u64 size)
> >>  	}
> >>  }
> >>  
> >> +static int realm_create_protected_data_page(struct realm *realm,
> >> +					    unsigned long ipa,
> >> +					    struct page *dst_page,
> >> +					    struct page *tmp_page)
> >> +{
> >> +	phys_addr_t dst_phys, tmp_phys;
> >> +	int ret;
> >> +
> >> +	copy_page(page_address(tmp_page), page_address(dst_page));
> >> +
> >> +	dst_phys = page_to_phys(dst_page);
> >> +	tmp_phys = page_to_phys(tmp_page);
> >> +
> >> +	if (rmi_granule_delegate(dst_phys))
> >> +		return -ENXIO;
> >> +
> >> +	ret = rmi_data_create(dst_phys, virt_to_phys(realm->rd), ipa, tmp_phys,
> >> +			      RMI_MEASURE_CONTENT);
> >> +
> >> +	if (RMI_RETURN_STATUS(ret) == RMI_ERROR_RTT) {
> >> +		/* Create missing RTTs and retry */
> >> +		int level = RMI_RETURN_INDEX(ret);
> >> +
> >> +		ret = realm_create_rtt_levels(realm, ipa, level,
> >> +					      RME_RTT_MAX_LEVEL, NULL);
> >> +		if (ret)
> >> +			goto err;
> >> +
> >> +		ret = rmi_data_create(dst_phys, virt_to_phys(realm->rd), ipa,
> >> +				      tmp_phys, RMI_MEASURE_CONTENT);
> >> +	}
> >> +
> >> +	if (ret)
> >> +		goto err;
> >> +
> >> +	return 0;
> >> +
> >> +err:
> >> +	if (WARN_ON(rmi_granule_undelegate(dst_phys))) {
> >> +		/* Page can't be returned to NS world so is lost */
> >> +		get_page(dst_page);
> >> +	}
> >> +	return -ENXIO;
> >> +}
> >> +
> >> +static int fold_rtt(phys_addr_t rd, unsigned long addr, int level,
> >> +		    struct realm *realm)
> >> +{
> >> +	struct rtt_entry rtt;
> >> +	phys_addr_t rtt_addr;
> >> +
> >> +	if (rmi_rtt_read_entry(rd, addr, level, &rtt))
> >> +		return -ENXIO;
> >> +
> >> +	if (rtt.state != RMI_TABLE)
> >> +		return -EINVAL;
> >> +
> >> +	rtt_addr = rmi_rtt_get_phys(&rtt);
> >> +	if (rmi_rtt_fold(rtt_addr, rd, addr, level + 1))
> >> +		return -ENXIO;
> >> +
> >> +	free_delegated_page(realm, rtt_addr);
> >> +
> >> +	return 0;
> >> +}
> >> +
> >> +int realm_map_protected(struct realm *realm,
> >> +			unsigned long hva,
> >> +			unsigned long base_ipa,
> >> +			struct page *dst_page,
> >> +			unsigned long map_size,
> >> +			struct kvm_mmu_memory_cache *memcache)
> >> +{
> >> +	phys_addr_t dst_phys = page_to_phys(dst_page);
> >> +	phys_addr_t rd = virt_to_phys(realm->rd);
> >> +	unsigned long phys = dst_phys;
> >> +	unsigned long ipa = base_ipa;
> >> +	unsigned long size;
> >> +	int map_level;
> >> +	int ret = 0;
> >> +
> >> +	if (WARN_ON(!IS_ALIGNED(ipa, map_size)))
> >> +		return -EINVAL;
> >> +
> >> +	switch (map_size) {
> >> +	case PAGE_SIZE:
> >> +		map_level = 3;
> >> +		break;
> >> +	case RME_L2_BLOCK_SIZE:
> >> +		map_level = 2;
> >> +		break;
> >> +	default:
> >> +		return -EINVAL;
> >> +	}
> >> +
> >> +	if (map_level < RME_RTT_MAX_LEVEL) {
> >> +		/*
> >> +		 * A temporary RTT is needed during the map, precreate it,
> >> +		 * however if there is an error (e.g. missing parent tables)
> >> +		 * this will be handled below.
> >> +		 */
> >> +		realm_create_rtt_levels(realm, ipa, map_level,
> >> +					RME_RTT_MAX_LEVEL, memcache);
> >> +	}
> >> +
> >> +	for (size = 0; size < map_size; size += PAGE_SIZE) {
> >> +		if (rmi_granule_delegate(phys)) {
> >> +			struct rtt_entry rtt;
> >> +
> >> +			/*
> >> +			 * It's possible we raced with another VCPU on the same
> >> +			 * fault. If the entry exists and matches then exit
> >> +			 * early and assume the other VCPU will handle the
> >> +			 * mapping.
> >> +			 */
> >> +			if (rmi_rtt_read_entry(rd, ipa, RME_RTT_MAX_LEVEL, &rtt))
> >> +				goto err;
> >> +
> >> +			// FIXME: For a block mapping this could race at level
> >> +			// 2 or 3...
> >> +			if (WARN_ON((rtt.walk_level != RME_RTT_MAX_LEVEL ||
> >> +				     rtt.state != RMI_ASSIGNED ||
> >> +				     rtt.desc != phys))) {
> >> +				goto err;
> >> +			}
> >> +
> >> +			return 0;
> >> +		}
> >> +
> >> +		ret = rmi_data_create_unknown(phys, rd, ipa);
> >> +
> >> +		if (RMI_RETURN_STATUS(ret) == RMI_ERROR_RTT) {
> >> +			/* Create missing RTTs and retry */
> >> +			int level = RMI_RETURN_INDEX(ret);
> >> +
> >> +			ret = realm_create_rtt_levels(realm, ipa, level,
> >> +						      RME_RTT_MAX_LEVEL,
> >> +						      memcache);
> >> +			WARN_ON(ret);
> >> +			if (ret)
> >> +				goto err_undelegate;
> >> +
> >> +			ret = rmi_data_create_unknown(phys, rd, ipa);
> >> +		}
> >> +		WARN_ON(ret);
> >> +
> >> +		if (ret)
> >> +			goto err_undelegate;
> >> +
> >> +		phys += PAGE_SIZE;
> >> +		ipa += PAGE_SIZE;
> >> +	}
> >> +
> >> +	if (map_size == RME_L2_BLOCK_SIZE)
> >> +		ret = fold_rtt(rd, base_ipa, map_level, realm);
> >> +	if (WARN_ON(ret))
> >> +		goto err;
> >> +
> >> +	return 0;
> >> +
> >> +err_undelegate:
> >> +	if (WARN_ON(rmi_granule_undelegate(phys))) {
> >> +		/* Page can't be returned to NS world so is lost */
> >> +		get_page(phys_to_page(phys));
> >> +	}
> >> +err:
> >> +	while (size > 0) {
> >> +		phys -= PAGE_SIZE;
> >> +		size -= PAGE_SIZE;
> >> +		ipa -= PAGE_SIZE;
> >> +
> >> +		rmi_data_destroy(rd, ipa);
> >> +
> >> +		if (WARN_ON(rmi_granule_undelegate(phys))) {
> >> +			/* Page can't be returned to NS world so is lost */
> >> +			get_page(phys_to_page(phys));
> >> +		}
> >> +	}
> >> +	return -ENXIO;
> >> +}
> >> +  
> > 
> > There seems no caller to the function above. Better move it to the related
> > patch.  
> 
> Indeed this should really be in the next patch - will move as it's very
> confusing having it in this patch (sorry about that).
> 
> >> +static int populate_par_region(struct kvm *kvm,
> >> +			       phys_addr_t ipa_base,
> >> +			       phys_addr_t ipa_end)
> >> +{
> >> +	struct realm *realm = &kvm->arch.realm;
> >> +	struct kvm_memory_slot *memslot;
> >> +	gfn_t base_gfn, end_gfn;
> >> +	int idx;
> >> +	phys_addr_t ipa;
> >> +	int ret = 0;
> >> +	struct page *tmp_page;
> >> +	phys_addr_t rd = virt_to_phys(realm->rd);
> >> +
> >> +	base_gfn = gpa_to_gfn(ipa_base);
> >> +	end_gfn = gpa_to_gfn(ipa_end);
> >> +
> >> +	idx = srcu_read_lock(&kvm->srcu);
> >> +	memslot = gfn_to_memslot(kvm, base_gfn);
> >> +	if (!memslot) {
> >> +		ret = -EFAULT;
> >> +		goto out;
> >> +	}
> >> +
> >> +	/* We require the region to be contained within a single memslot */
> >> +	if (memslot->base_gfn + memslot->npages < end_gfn) {
> >> +		ret = -EINVAL;
> >> +		goto out;
> >> +	}
> >> +
> >> +	tmp_page = alloc_page(GFP_KERNEL);
> >> +	if (!tmp_page) {
> >> +		ret = -ENOMEM;
> >> +		goto out;
> >> +	}
> >> +
> >> +	mmap_read_lock(current->mm);
> >> +
> >> +	ipa = ipa_base;
> >> +
> >> +	while (ipa < ipa_end) {
> >> +		struct vm_area_struct *vma;
> >> +		unsigned long map_size;
> >> +		unsigned int vma_shift;
> >> +		unsigned long offset;
> >> +		unsigned long hva;
> >> +		struct page *page;
> >> +		kvm_pfn_t pfn;
> >> +		int level;
> >> +
> >> +		hva = gfn_to_hva_memslot(memslot, gpa_to_gfn(ipa));
> >> +		vma = vma_lookup(current->mm, hva);
> >> +		if (!vma) {
> >> +			ret = -EFAULT;
> >> +			break;
> >> +		}
> >> +
> >> +		if (is_vm_hugetlb_page(vma))
> >> +			vma_shift = huge_page_shift(hstate_vma(vma));
> >> +		else
> >> +			vma_shift = PAGE_SHIFT;
> >> +
> >> +		map_size = 1 << vma_shift;
> >> +
> >> +		/*
> >> +		 * FIXME: This causes over mapping, but there's no good
> >> +		 * solution here with the ABI as it stands
> >> +		 */
> >> +		ipa = ALIGN_DOWN(ipa, map_size);
> >> +
> >> +		switch (map_size) {
> >> +		case RME_L2_BLOCK_SIZE:
> >> +			level = 2;
> >> +			break;
> >> +		case PAGE_SIZE:
> >> +			level = 3;
> >> +			break;
> >> +		default:
> >> +			WARN_ONCE(1, "Unsupport vma_shift %d", vma_shift);
> >> +			ret = -EFAULT;
> >> +			break;
> >> +		}
> >> +
> >> +		pfn = gfn_to_pfn_memslot(memslot, gpa_to_gfn(ipa));
> >> +
> >> +		if (is_error_pfn(pfn)) {
> >> +			ret = -EFAULT;
> >> +			break;
> >> +		}
> >> +
> >> +		ret = rmi_rtt_init_ripas(rd, ipa, level);
> >> +		if (RMI_RETURN_STATUS(ret) == RMI_ERROR_RTT) {
> >> +			ret = realm_create_rtt_levels(realm, ipa,
> >> +						      RMI_RETURN_INDEX(ret),
> >> +						      level, NULL);
> >> +			if (ret)
> >> +				break;
> >> +			ret = rmi_rtt_init_ripas(rd, ipa, level);
> >> +			if (ret) {
> >> +				ret = -ENXIO;
> >> +				break;
> >> +			}
> >> +		}
> >> +
> >> +		if (level < RME_RTT_MAX_LEVEL) {
> >> +			/*
> >> +			 * A temporary RTT is needed during the map, precreate
> >> +			 * it, however if there is an error (e.g. missing
> >> +			 * parent tables) this will be handled in the
> >> +			 * realm_create_protected_data_page() call.
> >> +			 */
> >> +			realm_create_rtt_levels(realm, ipa, level,
> >> +						RME_RTT_MAX_LEVEL, NULL);
> >> +		}
> >> +
> >> +		page = pfn_to_page(pfn);
> >> +
> >> +		for (offset = 0; offset < map_size && !ret;
> >> +		     offset += PAGE_SIZE, page++) {
> >> +			phys_addr_t page_ipa = ipa + offset;
> >> +
> >> +			ret = realm_create_protected_data_page(realm, page_ipa,
> >> +							       page, tmp_page);
> >> +		}
> >> +		if (ret)
> >> +			goto err_release_pfn;
> >> +
> >> +		if (level == 2) {
> >> +			ret = fold_rtt(rd, ipa, level, realm);
> >> +			if (ret)
> >> +				goto err_release_pfn;
> >> +		}
> >> +
> >> +		ipa += map_size;  
> >   
> >> +		kvm_set_pfn_accessed(pfn);
> >> +		kvm_set_pfn_dirty(pfn);  
> > 
> > kvm_release_pfn_dirty() has already called kvm_set_pfn_{accessed, dirty}().  
> 
> Will remove those calls.
> 
> >> +		kvm_release_pfn_dirty(pfn);
> >> +err_release_pfn:
> >> +		if (ret) {
> >> +			kvm_release_pfn_clean(pfn);
> >> +			break;
> >> +		}
> >> +	}
> >> +
> >> +	mmap_read_unlock(current->mm);
> >> +	__free_page(tmp_page);
> >> +
> >> +out:
> >> +	srcu_read_unlock(&kvm->srcu, idx);
> >> +	return ret;
> >> +}
> >> +
> >> +static int kvm_populate_realm(struct kvm *kvm,
> >> +			      struct kvm_cap_arm_rme_populate_realm_args *args)
> >> +{
> >> +	phys_addr_t ipa_base, ipa_end;
> >> +  
> > 
> > Check kvm_is_realm(kvm) here or in the kvm_realm_enable_cap().  
> 
> I'm going to update kvm_vm_ioctl_enable_cap() to check kvm_is_realm() so
> we won't get here.
> 
> >> +	if (kvm_realm_state(kvm) != REALM_STATE_NEW)
> >> +		return -EBUSY;  
> > 
> > Maybe -EINVAL? The realm hasn't been created (RMI_REALM_CREATE is not called
> > yet). The userspace shouldn't reach this path.  
> 
> Well user space can attempt to populate in the ACTIVE state - which is
> where the idea of 'busy' comes from. Admittedly it's a little confusing
> when RMI_REALM_CREATE hasn't been called.
> 
> I'm not particularly bothered about the return code, but it's useful to
> have a different code to -EINVAL as it's not an invalid argument, but
> calling at the wrong time. I can't immediately see a better error code
> though.
> 
The reason why I feel -EBUSY is little bit off is EBUSY usually indicates
something is already initialized and currently running, then another
calling path wanna to operate it. 

I took a look on the ioctls in arch/arm64/kvm/arm.c. It seems people have
different opinions for calling execution path at a wrong time:

For example:

long kvm_arch_vcpu_ioctl()
...
        case KVM_GET_REG_LIST: {
                struct kvm_reg_list __user *user_list = argp;
                struct kvm_reg_list reg_list;
                unsigned n;

                r = -ENOEXEC;
                if (unlikely(!kvm_vcpu_initialized(vcpu)))
                        break;

                r = -EPERM;
                if (!kvm_arm_vcpu_is_finalized(vcpu))
                        break;

If we have to choose one, I prefer -ENOEXEC as -EPERM is stranger. But
personally my vote goes to -EINVAL.

> Steve
> 
> >> +
> >> +	if (!IS_ALIGNED(args->populate_ipa_base, PAGE_SIZE) ||
> >> +	    !IS_ALIGNED(args->populate_ipa_size, PAGE_SIZE))
> >> +		return -EINVAL;
> >> +
> >> +	ipa_base = args->populate_ipa_base;
> >> +	ipa_end = ipa_base + args->populate_ipa_size;
> >> +
> >> +	if (ipa_end < ipa_base)
> >> +		return -EINVAL;
> >> +
> >> +	return populate_par_region(kvm, ipa_base, ipa_end);
> >> +}
> >> +
> >>  static int set_ipa_state(struct kvm_vcpu *vcpu,
> >>  			 unsigned long ipa,
> >>  			 unsigned long end,
> >> @@ -748,6 +1102,18 @@ int kvm_realm_enable_cap(struct kvm *kvm, struct kvm_enable_cap *cap)
> >>  		r = kvm_init_ipa_range_realm(kvm, &args);
> >>  		break;
> >>  	}
> >> +	case KVM_CAP_ARM_RME_POPULATE_REALM: {
> >> +		struct kvm_cap_arm_rme_populate_realm_args args;
> >> +		void __user *argp = u64_to_user_ptr(cap->args[1]);
> >> +
> >> +		if (copy_from_user(&args, argp, sizeof(args))) {
> >> +			r = -EFAULT;
> >> +			break;
> >> +		}
> >> +
> >> +		r = kvm_populate_realm(kvm, &args);
> >> +		break;
> >> +	}
> >>  	default:
> >>  		r = -EINVAL;
> >>  		break;  
> >   
>