Re: [PATCH 3/3] KVM: arm64: selftests: Enable stage-2 in NV preparation functions

From: Wei-Lin Chang

Date: Thu Mar 26 2026 - 17:19:16 EST


On Wed, Mar 25, 2026 at 09:18:20AM +0000, Marc Zyngier wrote:
> On Wed, 25 Mar 2026 00:36:20 +0000,
> Wei-Lin Chang <weilin.chang@xxxxxxx> wrote:
> >
> > Introduce library functions for setting up guest stage-2 page tables,
> > then use that to give L2 an identity mapped stage-2 and enable it.
> >
> > The translation and stage-2 page table built is simple, start level 0,
> > 4 levels, 4KB granules, normal cachable, 48-bit IA, 40-bit OA.
>
> That's a no go. The most common NV-capable out there can realistically
> only do 16kB at S2, and is limited to 36bit IPA. We can't really
> afford to say "too bad" and leave the main development platform behind.

Interesting, I didn't know that! I didn't consider the guest stage-2
granule size < host stage-2 granule size case either.

>
> >
> > The nested page table code is adapted from lib/x86/vmx.c.
>
> I guess this starting point is the main issue.
>
> >
> > Signed-off-by: Wei-Lin Chang <weilin.chang@xxxxxxx>
> > ---
> > .../selftests/kvm/include/arm64/nested.h | 7 ++
> > .../selftests/kvm/include/arm64/processor.h | 9 ++
> > .../testing/selftests/kvm/lib/arm64/nested.c | 97 ++++++++++++++++++-
> > 3 files changed, 111 insertions(+), 2 deletions(-)
> >
> > diff --git a/tools/testing/selftests/kvm/include/arm64/nested.h b/tools/testing/selftests/kvm/include/arm64/nested.h
> > index 739ff2ee0161..0be10a775e48 100644
> > --- a/tools/testing/selftests/kvm/include/arm64/nested.h
> > +++ b/tools/testing/selftests/kvm/include/arm64/nested.h
> > @@ -6,6 +6,13 @@
> > #ifndef SELFTEST_KVM_NESTED_H
> > #define SELFTEST_KVM_NESTED_H
> >
> > +uint64_t get_l1_vtcr(void);
> > +
> > +void nested_map(struct kvm_vm *vm, vm_paddr_t guest_pgd,
> > + uint64_t nested_paddr, uint64_t paddr, uint64_t size);
> > +void nested_map_memslot(struct kvm_vm *vm, vm_paddr_t guest_pgd,
> > + uint32_t memslot);
> > +
> > void prepare_l2_stack(struct kvm_vm *vm, struct kvm_vcpu *vcpu);
> > void prepare_hyp_state(struct kvm_vm *vm, struct kvm_vcpu *vcpu);
> > void prepare_eret_destination(struct kvm_vm *vm, struct kvm_vcpu *vcpu, void *l2_pc);
> > diff --git a/tools/testing/selftests/kvm/include/arm64/processor.h b/tools/testing/selftests/kvm/include/arm64/processor.h
> > index ac97a1c436fc..5de2e932d95a 100644
> > --- a/tools/testing/selftests/kvm/include/arm64/processor.h
> > +++ b/tools/testing/selftests/kvm/include/arm64/processor.h
> > @@ -104,6 +104,15 @@
> > #define TCR_HA (UL(1) << 39)
> > #define TCR_DS (UL(1) << 59)
> >
> > +/* VTCR_EL2 specific flags */
> > +#define VTCR_EL2_T0SZ_BITS(x) ((UL(64) - (x)) << VTCR_EL2_T0SZ_SHIFT)
> > +
> > +#define VTCR_EL2_SL0_LV0_4K (UL(2) << VTCR_EL2_SL0_SHIFT)
> > +#define VTCR_EL2_SL0_LV1_4K (UL(1) << VTCR_EL2_SL0_SHIFT)
> > +#define VTCR_EL2_SL0_LV2_4K (UL(0) << VTCR_EL2_SL0_SHIFT)
> > +
> > +#define VTCR_EL2_PS_40_BITS (UL(2) << VTCR_EL2_PS_SHIFT)
> > +
> > /*
> > * AttrIndx[2:0] encoding (mapping attributes defined in the MAIR* registers).
> > */
> > diff --git a/tools/testing/selftests/kvm/lib/arm64/nested.c b/tools/testing/selftests/kvm/lib/arm64/nested.c
> > index 111d02f44cfe..910f8cd30f96 100644
> > --- a/tools/testing/selftests/kvm/lib/arm64/nested.c
> > +++ b/tools/testing/selftests/kvm/lib/arm64/nested.c
> > @@ -1,8 +1,11 @@
> > // SPDX-License-Identifier: GPL-2.0
> > /*
> > - * ARM64 Nested virtualization helpers
> > + * ARM64 Nested virtualization helpers, nested page table code adapted from
> > + * ../x86/vmx.c.
> > */
> >
> > +#include <linux/sizes.h>
> > +
> > #include "kvm_util.h"
> > #include "nested.h"
> > #include "processor.h"
> > @@ -18,6 +21,87 @@ static void hvc_handler(struct ex_regs *regs)
> > regs->pc = (u64)after_hvc;
> > }
> >
> > +uint64_t get_l1_vtcr(void)
> > +{
> > + return VTCR_EL2_PS_40_BITS | VTCR_EL2_TG0_4K | VTCR_EL2_ORGN0_WBWA |
> > + VTCR_EL2_IRGN0_WBWA | VTCR_EL2_SL0_LV0_4K | VTCR_EL2_T0SZ_BITS(48);
>
> Irk. See above.
>
> > +}
> > +
> > +static void __nested_pg_map(struct kvm_vm *vm, uint64_t guest_pgd,
> > + uint64_t nested_paddr, uint64_t paddr, uint64_t flags)
> > +{
> > + uint8_t attr_idx = flags & (PTE_ATTRINDX_MASK >> PTE_ATTRINDX_SHIFT);
> > + uint64_t pg_attr;
> > + uint64_t *ptep;
> > +
> > + TEST_ASSERT((nested_paddr % vm->page_size) == 0,
> > + "L2 IPA not on page boundary,\n"
> > + " nested_paddr: 0x%lx vm->page_size: 0x%x", nested_paddr, vm->page_size);
> > + TEST_ASSERT((paddr % vm->page_size) == 0,
> > + "Guest physical address not on page boundary,\n"
> > + " paddr: 0x%lx vm->page_size: 0x%x", paddr, vm->page_size);
> > + TEST_ASSERT((paddr >> vm->page_shift) <= vm->max_gfn,
> > + "Physical address beyond maximum supported,\n"
> > + " paddr: 0x%lx vm->max_gfn: 0x%lx vm->page_size: 0x%x",
> > + paddr, vm->max_gfn, vm->page_size);
> > +
> > + ptep = addr_gpa2hva(vm, guest_pgd) + ((nested_paddr >> 39) & 0x1ffu) * 8;
> > + if (!*ptep)
> > + *ptep = (vm_alloc_page_table(vm) & GENMASK(47, 12)) | PGD_TYPE_TABLE | PTE_VALID;
> > + ptep = addr_gpa2hva(vm, *ptep & GENMASK(47, 12)) + ((nested_paddr >> 30) & 0x1ffu) * 8;
> > + if (!*ptep)
> > + *ptep = (vm_alloc_page_table(vm) & GENMASK(47, 12)) | PUD_TYPE_TABLE | PTE_VALID;
> > + ptep = addr_gpa2hva(vm, *ptep & GENMASK(47, 12)) + ((nested_paddr >> 21) & 0x1ffu) * 8;
> > + if (!*ptep)
> > + *ptep = (vm_alloc_page_table(vm) & GENMASK(47, 12)) | PMD_TYPE_TABLE | PTE_VALID;
> > + ptep = addr_gpa2hva(vm, *ptep & GENMASK(47, 12)) + ((nested_paddr >> 12) & 0x1ffu) * 8;
> > +
> > + pg_attr = PTE_AF | PTE_ATTRINDX(attr_idx) | PTE_TYPE_PAGE | PTE_VALID;
> > + pg_attr |= PTE_SHARED;
> > +
> > + *ptep = (paddr & GENMASK(47, 12)) | pg_attr;
>
> Please use named constants, and write a page table generator that is
> independent of page, IA and OA sizes, as advertised to the guest.

Ack, and yes, I should absolutely consider what is advertised to the
guest...

Thanks,
Wei-Lin Chang

>
> Thanks,
>
> M.
>
> --
> Without deviation from the norm, progress is not possible.