[<prev] [next>] [<thread-prev] [day] [month] [year] [list]
Message-ID: <87jzdst6os.wl-maz@kernel.org>
Date: Mon, 28 Oct 2024 18:37:23 +0000
From: Marc Zyngier <maz@...nel.org>
To: Aneesh Kumar K.V <aneesh.kumar@...nel.org>
Cc: linux-kernel@...r.kernel.org,
linux-arm-kernel@...ts.infradead.org,
kvmarm@...ts.linux.dev,
Suzuki K Poulose <Suzuki.Poulose@....com>,
Steven Price <steven.price@....com>,
Will Deacon <will@...nel.org>,
Catalin Marinas <catalin.marinas@....com>,
Mark Rutland <mark.rutland@....com>,
Oliver Upton <oliver.upton@...ux.dev>,
Joey Gouly <joey.gouly@....com>,
Zenghui Yu <yuzenghui@...wei.com>
Subject: Re: [PATCH 4/4] arm64: mte: Use stage-2 NoTagAccess memory attribute if supported
On Mon, 28 Oct 2024 13:28:42 +0000,
Aneesh Kumar K.V <aneesh.kumar@...nel.org> wrote:
>
> Marc Zyngier <maz@...nel.org> writes:
>
> > On Mon, 28 Oct 2024 09:40:14 +0000,
> > "Aneesh Kumar K.V (Arm)" <aneesh.kumar@...nel.org> wrote:
> >>
> >> Currently, the kernel won't start a guest if the MTE feature is enabled
> >> and the guest RAM is backed by memory which doesn't support access tags.
> >> Update this such that the kernel uses the NoTagAccess memory attribute
> >> while mapping pages from VMAs for which MTE is not allowed. The fault
> >> from accessing the access tags with such pages is forwarded to VMM so
> >> that VMM can decide to kill the guest or remap the pages so that
> >> access tag storage is allowed.
> >
> > I only have questions here:
> >
> > - what is the benefit of such approach? why shouldn't that be the
> > kernel's job to fix it?
> >
>
> IMHO leaving that policy decision to VMM makes the kernel changes
> simpler. In most cases, VMM will kill the guest, because these
> restrictions of MTE_ALLOWED are applied at the memslot/vma.
Where is that captured? The whole idea behind FEAT_MTE_PERM was that
it would be the hypervisor's task to lazily allocate MTE-capable
memory as tagged-access would occur.
>
> >
> > - where is the documentation for this new userspace ABI?
> >
>
> I will add the details if we agree that this should be a separate EXIT
> as outlined in this patch.
Woooot??? This isn't a *DETAIL*. This is the very first thing you
should write. Everything *else* is an implementation detail.
>
> >
> > - are you expecting the VMM to create a new memslot for this?
> >
>
> I guess there are examples of configs where some memory regions are
> backed by page cache where we don't directly use those memory regions as
> allocatable memory in the guest. This change allows us to enable MTE in such
> configs.
This doesn't answer my question. What is expected sequence a VMM
should apply to provide tagged memory to the guest?
>
> > - where is the example of a VMM using this?
> >
>
> I do have changes to kvmtool which won't do any fixup on receiving that
> VM exit. I expect other VMM to do the same by default when they get a
> VM exit with an unknown exit_reason. So unless VMM wants to do any
> special handling, we don't need any change in the VMM.
OK, so this has never been tested. I'm sorry, but for something of
this magnitude, with expected userspace interactions, and requiring
VMM handling, I want to see the full end-to-end thing.
>
>
> diff --git a/arm/kvm-cpu.c b/arm/kvm-cpu.c
> index 3b95750ecec7..4760bad07476 100644
> --- a/arm/kvm-cpu.c
> +++ b/arm/kvm-cpu.c
> @@ -239,6 +239,17 @@ static bool handle_memoryfault(struct kvm_cpu *vcpu)
> return true;
> }
>
> +static bool handle_notag_access(struct kvm_cpu *vcpu)
> +{
> + u64 gpa = vcpu->kvm_run->memory_fault.gpa;
> + u64 size = vcpu->kvm_run->memory_fault.size;
> +
> + /* For now VMM just panic */
> + pr_err("Tag Access to a wrong memory region 0x%lx size 0x%lx\n",
> + (unsigned long)gpa, (unsigned long)size);
> + return false;
> +}
> +
> bool kvm_cpu__handle_exit(struct kvm_cpu *vcpu)
> {
> switch (vcpu->kvm_run->exit_reason) {
> @@ -246,6 +257,8 @@ bool kvm_cpu__handle_exit(struct kvm_cpu *vcpu)
> return handle_hypercall(vcpu);
> case KVM_EXIT_MEMORY_FAULT:
> return handle_memoryfault(vcpu);
> + case KVM_EXIT_ARM_NOTAG_ACCESS:
> + return handle_notag_access(vcpu);
> }
>
> return false;
> diff --git a/include/linux/kvm.h b/include/linux/kvm.h
> index 32cff22f0e4d..deef6614f577 100644
> --- a/include/linux/kvm.h
> +++ b/include/linux/kvm.h
> @@ -178,6 +178,7 @@ struct kvm_xen_exit {
> #define KVM_EXIT_NOTIFY 37
> #define KVM_EXIT_LOONGARCH_IOCSR 38
> #define KVM_EXIT_MEMORY_FAULT 39
> +#define KVM_EXIT_ARM_NOTAG_ACCESS 40
>
> /* For KVM_EXIT_INTERNAL_ERROR */
> /* Emulate instruction failed. */
> @@ -429,10 +430,17 @@ struct kvm_run {
> /* KVM_EXIT_MEMORY_FAULT */
> struct {
> #define KVM_MEMORY_EXIT_FLAG_PRIVATE (1ULL << 3)
> +#define KVM_MEMORY_EXIT_FLAG_NOTAGACCESS (1ULL << 4)
> __u64 flags;
> __u64 gpa;
> __u64 size;
> } memory_fault;
> + /* KVM_EXIT_ARM_NOTAG_ACCESS */
> + struct {
> + __u64 flags;
> + __u64 gpa;
> + __u64 size;
> + } notag_access;
> /* Fix the size of the union. */
> char padding[256];
> };
>
> >>
> >> NOTE: We could also use KVM_EXIT_MEMORY_FAULT for this. I chose to
> >> add a new EXIT type because this is arm64 specific exit type.
> >>
> >> Signed-off-by: Aneesh Kumar K.V (Arm) <aneesh.kumar@...nel.org>
> >> ---
> >> arch/arm64/include/asm/kvm_emulate.h | 5 +++++
> >> arch/arm64/include/asm/kvm_pgtable.h | 1 +
> >> arch/arm64/kvm/hyp/pgtable.c | 16 +++++++++++++---
> >> arch/arm64/kvm/mmu.c | 28 ++++++++++++++++++++++------
> >> include/uapi/linux/kvm.h | 7 +++++++
> >> 5 files changed, 48 insertions(+), 9 deletions(-)
> >>
> >> diff --git a/arch/arm64/include/asm/kvm_emulate.h b/arch/arm64/include/asm/kvm_emulate.h
> >> index a601a9305b10..fa0149a0606a 100644
> >> --- a/arch/arm64/include/asm/kvm_emulate.h
> >> +++ b/arch/arm64/include/asm/kvm_emulate.h
> >> @@ -373,6 +373,11 @@ static inline bool kvm_vcpu_trap_is_exec_fault(const struct kvm_vcpu *vcpu)
> >> return kvm_vcpu_trap_is_iabt(vcpu) && !kvm_vcpu_abt_iss1tw(vcpu);
> >> }
> >>
> >> +static inline bool kvm_vcpu_trap_is_tagaccess(const struct kvm_vcpu *vcpu)
> >> +{
> >> + return !!(ESR_ELx_ISS2(kvm_vcpu_get_esr(vcpu)) & ESR_ELx_TagAccess);
> >> +}
> >> +
> >> static __always_inline u8 kvm_vcpu_trap_get_fault(const struct kvm_vcpu *vcpu)
> >> {
> >> return kvm_vcpu_get_esr(vcpu) & ESR_ELx_FSC;
> >> diff --git a/arch/arm64/include/asm/kvm_pgtable.h b/arch/arm64/include/asm/kvm_pgtable.h
> >> index 03f4c3d7839c..5657ac1998ad 100644
> >> --- a/arch/arm64/include/asm/kvm_pgtable.h
> >> +++ b/arch/arm64/include/asm/kvm_pgtable.h
> >> @@ -252,6 +252,7 @@ enum kvm_pgtable_prot {
> >>
> >> KVM_PGTABLE_PROT_DEVICE = BIT(3),
> >> KVM_PGTABLE_PROT_NORMAL_NC = BIT(4),
> >> + KVM_PGTABLE_PROT_NORMAL_NOTAGACCESS = BIT(5),
> >
> > This seems wrong. NOTAGACCESS is a *permission*, not a memory type.
> >
>
> Are you suggesting the name is wrong? The memory attribute value I
> wanted to use is
>
> MemAttr[3:0] = 0b0100 which is Normal, NoTagAccess, writeback cacheable.
>
> I am following the changes similar to KVM_PGTABLE_PROT_NORMAL_NC.
But that's entirely different. This really isn't a memory type. Quite
the opposite, actually. This is a stage-2 permission setting, which
you are conflating with the actual memory type.
Also, I don't see why that should be incompatible with something other
than Normal memory.
>
> >
> >>
> >> KVM_PGTABLE_PROT_SW0 = BIT(55),
> >> KVM_PGTABLE_PROT_SW1 = BIT(56),
> >> diff --git a/arch/arm64/kvm/hyp/pgtable.c b/arch/arm64/kvm/hyp/pgtable.c
> >> index b11bcebac908..bc0d9f08c49a 100644
> >> --- a/arch/arm64/kvm/hyp/pgtable.c
> >> +++ b/arch/arm64/kvm/hyp/pgtable.c
> >> @@ -677,9 +677,11 @@ static int stage2_set_prot_attr(struct kvm_pgtable *pgt, enum kvm_pgtable_prot p
> >> {
> >> kvm_pte_t attr;
> >> u32 sh = KVM_PTE_LEAF_ATTR_LO_S2_SH_IS;
> >> + unsigned long prot_mask = KVM_PGTABLE_PROT_DEVICE |
> >> + KVM_PGTABLE_PROT_NORMAL_NC |
> >> + KVM_PGTABLE_PROT_NORMAL_NOTAGACCESS;
> >>
> >> - switch (prot & (KVM_PGTABLE_PROT_DEVICE |
> >> - KVM_PGTABLE_PROT_NORMAL_NC)) {
> >> + switch (prot & prot_mask) {
> >> case KVM_PGTABLE_PROT_DEVICE | KVM_PGTABLE_PROT_NORMAL_NC:
> >> return -EINVAL;
> >> case KVM_PGTABLE_PROT_DEVICE:
> >> @@ -692,6 +694,12 @@ static int stage2_set_prot_attr(struct kvm_pgtable *pgt, enum kvm_pgtable_prot p
> >> return -EINVAL;
> >> attr = KVM_S2_MEMATTR(pgt, NORMAL_NC);
> >> break;
> >> + case KVM_PGTABLE_PROT_NORMAL_NOTAGACCESS:
> >> + if (system_supports_notagaccess())
> >> + attr = KVM_S2_MEMATTR(pgt, NORMAL_NOTAGACCESS);
> >> + else
> >> + return -EINVAL;
> >> + break;
> >
> > How do you see this working when migrating a VM from one host to
> > another, one that supports FEAT_MTE_PERM and one that doesn't? The
> > current assumptions are that the VMM will replay the *exact same*
> > setup on the target host, and this obviously doesn't work.
> >
>
> I missed looking at kvm migration. I guess I will have to expose this as
> a capability and only allow migration if the target also supports the
> same capability?
I don't think so. This doesn't affect the guest at all, as the guest
doesn't know anything about S2, unless you decide to expose
FEAT_MTE_PERM to NV.
It affects the VMM though, and that's because you are making a
difference in handling between having FEAT_MTE_PERM or not. Given that
this is invisible to userspace, that's not a great design.
Thanks,
M.
--
Without deviation from the norm, progress is not possible.
Powered by blists - more mailing lists