Hi Oliver,
On Mon, Mar 14, 2022 at 12:41 PM Oliver Upton <oupton@xxxxxxxxxx> wrote:
>
> On Thu, Feb 24, 2022 at 05:25:51PM +0000, Raghavendra Rao Ananta wrote:
> > KVM regularly introduces new hypercall services to the guests without
> > any consent from the userspace. This means, the guests can observe
> > hypercall services in and out as they migrate across various host
> > kernel versions. This could be a major problem if the guest
> > discovered a hypercall, started using it, and after getting migrated
> > to an older kernel realizes that it's no longer available. Depending
> > on how the guest handles the change, there's a potential chance that
> > the guest would just panic.
> >
> > As a result, there's a need for the userspace to elect the services
> > that it wishes the guest to discover. It can elect these services
> > based on the kernels spread across its (migration) fleet. To remedy
> > this, extend the existing firmware psuedo-registers, such as
> > KVM_REG_ARM_PSCI_VERSION, for all the hypercall services available.
> >
> > These firmware registers are categorized based on the service call
> > owners, and unlike the existing firmware psuedo-registers, they hold
> > the features supported in the form of a bitmap.
> >
> > During the VM initialization, the registers holds an upper-limit of
> > the features supported by the corresponding registers. It's expected
> > that the VMMs discover the features provided by each register via
> > GET_ONE_REG, and writeback the desired values using SET_ONE_REG.
> > KVM allows this modification only until the VM has started.
> >
> > Older userspace code can simply ignore the capability and the
> > hypercall services will be exposed unconditionally to the guests, thus
> > ensuring backward compatibility.
> >
> > In this patch, the framework adds the register only for ARM's standard
> > secure services (owner value 4). Currently, this includes support only
> > for ARM True Random Number Generator (TRNG) service, with bit-0 of the
> > register representing mandatory features of v1.0. The register is also
> > added to the kvm_arm_vm_scope_fw_regs[] list as it maintains its state
> > per-VM. Other services are momentarily added in the upcoming patches.
> >
> > Signed-off-by: Raghavendra Rao Ananta <rananta@xxxxxxxxxx>
> > ---
> > arch/arm64/include/asm/kvm_host.h | 12 +++++
> > arch/arm64/include/uapi/asm/kvm.h | 8 ++++
> > arch/arm64/kvm/arm.c | 8 ++++
> > arch/arm64/kvm/guest.c | 1 +
> > arch/arm64/kvm/hypercalls.c | 78 +++++++++++++++++++++++++++++++
> > include/kvm/arm_hypercalls.h | 4 ++
> > 6 files changed, 111 insertions(+)
> >
> > diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> > index e823571e50cc..1909ced3208f 100644
> > --- a/arch/arm64/include/asm/kvm_host.h
> > +++ b/arch/arm64/include/asm/kvm_host.h
> > @@ -101,6 +101,15 @@ struct kvm_s2_mmu {
> > struct kvm_arch_memory_slot {
> > };
> >
> > +/**
> > + * struct kvm_hvc_desc: KVM ARM64 hypercall descriptor
> > + *
> > + * @hvc_std_bmap: Bitmap of standard secure service calls
> > + */
> > +struct kvm_hvc_desc {
>
> nit: maybe call this structure kvm_hypercall_features? When nested comes
> along guests will need to use the SVC conduit as HVC traps are always
> taken to EL2. Same will need to be true for virtual EL2.
>
Sure, I can rename it to be more generic.
> > + u64 hvc_std_bmap;
> > +};
> > +
> > struct kvm_arch {
> > struct kvm_s2_mmu mmu;
> >
> > @@ -142,6 +151,9 @@ struct kvm_arch {
> >
> > /* Capture first run of the VM */
> > bool has_run_once;
> > +
> > + /* Hypercall firmware register' descriptor */
> > + struct kvm_hvc_desc hvc_desc;
> > };
> >
> > struct kvm_vcpu_fault_info {
> > diff --git a/arch/arm64/include/uapi/asm/kvm.h b/arch/arm64/include/uapi/asm/kvm.h
> > index c35447cc0e0c..2decc30d6b84 100644
> > --- a/arch/arm64/include/uapi/asm/kvm.h
> > +++ b/arch/arm64/include/uapi/asm/kvm.h
> > @@ -287,6 +287,14 @@ struct kvm_arm_copy_mte_tags {
> > #define KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_NOT_REQUIRED 3
> > #define KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2_ENABLED (1U << 4)
> >
> > +/* Bitmap firmware registers, extension to the existing psuedo-register space */
> > +#define KVM_REG_ARM_FW_BMAP KVM_REG_ARM_FW_REG(0xff00)
>
> What is the motivation for moving the bitmap register indices so far
> away from the rest of the firmware regs?
>
The original motivation to create a sub-space came from Reiji's
comment on v3 [1] so that user-space can distinguish between bitmapped
and regular fw registers.
As with the spacing, I thought a 50/50 split would do a good job of
avoiding collisions. Do you have any recommendations here?
> > +#define KVM_REG_ARM_FW_BMAP_REG(r) (KVM_REG_ARM_FW_BMAP | (r))
>
> If you are still going to use the index offset, just pass 'r' through to
> the other macro:
>
> #define KVM_REG_ARM_FW_BMAP_REG(r) KVM_REG_ARM_FW_REG(0xff00 + r)
>
I'm sorry, what's the advantage of doing this?
> > +#define KVM_REG_ARM_STD_BMAP KVM_REG_ARM_FW_BMAP_REG(0)
> > +#define KVM_REG_ARM_STD_BIT_TRNG_V1_0 BIT(0)
> > +#define KVM_REG_ARM_STD_BMAP_BIT_MAX 0 /* Last valid bit */
>
> Implementation details such as this probably shouldn't live in UAPI
> headers. We'll likely need to bump the value in the future.
>
Wouldn't the macros act as helpers to userspace as well?
I agree with your point about the limited spacing, and we may need to
expand/shrink as needed, but isn't that a general register-space
problem?
> > +
> > /* SVE registers */
> > #define KVM_REG_ARM64_SVE (0x15 << KVM_REG_ARM_COPROC_SHIFT)
> >
> > diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> > index f61cd8d57eae..e9f9edb1cf55 100644
> > --- a/arch/arm64/kvm/arm.c
> > +++ b/arch/arm64/kvm/arm.c
> > @@ -156,6 +156,7 @@ int kvm_arch_init_vm(struct kvm *kvm, unsigned long type)
> > kvm->arch.max_vcpus = kvm_arm_default_max_vcpus();
> >
> > set_default_spectre(kvm);
> > + kvm_arm_init_hypercalls(kvm);
> >
> > return ret;
> > out_free_stage2_pgd:
> > @@ -635,7 +636,14 @@ int kvm_arch_vcpu_run_pid_change(struct kvm_vcpu *vcpu)
> > if (kvm_vm_is_protected(kvm))
> > kvm_call_hyp_nvhe(__pkvm_vcpu_init_traps, vcpu);
> >
> > + /*
> > + * Grab kvm->lock such that the reader of has_run_once can finish
> > + * the necessary operation atomically, such as deciding whether to
> > + * block the writes to the firmware registers if the VM has run once.
> > + */
> > + mutex_lock(&kvm->lock);
> > kvm->arch.has_run_once = true;
> > + mutex_unlock(&kvm->lock);
>
> Shouldn't this have just grabbed the kvm lock in patch 04/13?
>
It could have, and I agree it's a little igly. But thinking that
there's are no parallel threads/consumers when it was initially
implemented gave me no reason to grab the locks.
> > return ret;
> > }
> > diff --git a/arch/arm64/kvm/guest.c b/arch/arm64/kvm/guest.c
> > index eb061e64a7a5..d66e6c742bbe 100644
> > --- a/arch/arm64/kvm/guest.c
> > +++ b/arch/arm64/kvm/guest.c
> > @@ -65,6 +65,7 @@ const struct kvm_stats_header kvm_vcpu_stats_header = {
> > static const u64 kvm_arm_vm_scope_fw_regs[] = {
> > KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1,
> > KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2,
> > + KVM_REG_ARM_STD_BMAP,
> > };
> >
> > /**
> > diff --git a/arch/arm64/kvm/hypercalls.c b/arch/arm64/kvm/hypercalls.c
> > index 8624e6964940..48c126c3da72 100644
> > --- a/arch/arm64/kvm/hypercalls.c
> > +++ b/arch/arm64/kvm/hypercalls.c
> > @@ -58,6 +58,29 @@ static void kvm_ptp_get_time(struct kvm_vcpu *vcpu, u64 *val)
> > val[3] = lower_32_bits(cycles);
> > }
> >
> > +static bool kvm_arm_fw_reg_feat_enabled(u64 reg_bmap, u64 feat_bit)
> > +{
> > + return reg_bmap & feat_bit;
> > +}
> > +
> > +static bool kvm_hvc_call_supported(struct kvm_vcpu *vcpu, u32 func_id)
> > +{
> > + struct kvm_hvc_desc *hvc_desc = &vcpu->kvm->arch.hvc_desc;
> > +
> > + switch (func_id) {
> > + case ARM_SMCCC_TRNG_VERSION:
> > + case ARM_SMCCC_TRNG_FEATURES:
> > + case ARM_SMCCC_TRNG_GET_UUID:
> > + case ARM_SMCCC_TRNG_RND32:
> > + case ARM_SMCCC_TRNG_RND64:
> > + return kvm_arm_fw_reg_feat_enabled(hvc_desc->hvc_std_bmap,
> > + KVM_REG_ARM_STD_BIT_TRNG_V1_0);
> > + default:
> > + /* By default, allow the services that aren't listed here */
> > + return true;
>
> I think your default case should really return false. It keeps people
> honest when they add new patches to set up a new hypercall bit (no bit?
> no call!)
>
> That of course requires that you only return false once all of the
> preexisting hypercalls are enumerated, otherwise such a patch would
> cause a regression in isolation.
>
I agree, but not all hypercalls may have a bit. For example, guarding
ARM_SMCCC_VERSION_FUNC_ID doesn't make sense and may not be gated by
the bit. Hence, the idea was to do an initial check if the func_id is
in fact gated. If yes, check the bit, else let the actual
implementation take care of it.
> > +}
> > +
> > int kvm_hvc_call_handler(struct kvm_vcpu *vcpu)
> > {
> > u32 func_id = smccc_get_function(vcpu);
> > @@ -65,6 +88,9 @@ int kvm_hvc_call_handler(struct kvm_vcpu *vcpu)
> > u32 feature;
> > gpa_t gpa;
> >
> > + if (!kvm_hvc_call_supported(vcpu, func_id))
> > + goto out;
> > +
> > switch (func_id) {
> > case ARM_SMCCC_VERSION_FUNC_ID:
> > val[0] = ARM_SMCCC_VERSION_1_1;
> > @@ -143,6 +169,7 @@ int kvm_hvc_call_handler(struct kvm_vcpu *vcpu)
> > return kvm_psci_call(vcpu);
> > }
> >
> > +out:
> > smccc_set_retval(vcpu, val[0], val[1], val[2], val[3]);
> > return 1;
> > }
> > @@ -151,8 +178,16 @@ static const u64 kvm_arm_fw_reg_ids[] = {
> > KVM_REG_ARM_PSCI_VERSION,
> > KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_1,
> > KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2,
> > + KVM_REG_ARM_STD_BMAP,
> > };
> >
> > +void kvm_arm_init_hypercalls(struct kvm *kvm)
> > +{
> > + struct kvm_hvc_desc *hvc_desc = &kvm->arch.hvc_desc;
> > +
> > + hvc_desc->hvc_std_bmap = ARM_SMCCC_STD_FEATURES;
> > +}
> > +
> > int kvm_arm_get_fw_num_regs(struct kvm_vcpu *vcpu)
> > {
> > return ARRAY_SIZE(kvm_arm_fw_reg_ids);
> > @@ -220,6 +255,7 @@ static int get_kernel_wa_level(u64 regid)
> >
> > int kvm_arm_get_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > {
> > + struct kvm_hvc_desc *hvc_desc = &vcpu->kvm->arch.hvc_desc;
> > void __user *uaddr = (void __user *)(long)reg->addr;
> > u64 val, reg_id = reg->id;
> >
> > @@ -233,6 +269,9 @@ int kvm_arm_get_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > case KVM_REG_ARM_SMCCC_ARCH_WORKAROUND_2:
> > val = get_kernel_wa_level(reg_id) & KVM_REG_FEATURE_LEVEL_MASK;
> > break;
> > + case KVM_REG_ARM_STD_BMAP:
> > + val = READ_ONCE(hvc_desc->hvc_std_bmap);
> > + break;
> > default:
> > return -ENOENT;
> > }
> > @@ -243,6 +282,43 @@ int kvm_arm_get_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > return 0;
> > }
> >
> > +static int kvm_arm_set_fw_reg_bmap(struct kvm_vcpu *vcpu, u64 reg_id, u64 val)
> > +{
> > + int ret = 0;
> > + struct kvm *kvm = vcpu->kvm;
> > + struct kvm_hvc_desc *hvc_desc = &kvm->arch.hvc_desc;
> > + u64 *fw_reg_bmap, fw_reg_features;
>
> nit: use reverse fir tree ordering for locals (longest line first,
> shortest last).
>
Got it.
> > + switch (reg_id) {
> > + case KVM_REG_ARM_STD_BMAP:
> > + fw_reg_bmap = &hvc_desc->hvc_std_bmap;
> > + fw_reg_features = ARM_SMCCC_STD_FEATURES;
> > + break;
> > + default:
> > + return -ENOENT;
> > + }
> > +
> > + /* Check for unsupported bit */
> > + if (val & ~fw_reg_features)
> > + return -EINVAL;
> > +
> > + mutex_lock(&kvm->lock);
> > +
> > + /*
> > + * If the VM (any vCPU) has already started running, return success
> > + * if there's no change in the value. Else, return -EBUSY.
>
> How about returning -EINVAL instead? We already do this for
> KVM_ARM_VCPU_INIT if userspace uses a different target than the one
> previously set.
>
Returning EBUSY could help userspace distinguish the cause of failure
better as compared to EINVAL, since technically, the arguments
supplied by the user are in fact valid. It's just that the timing is
wrong.
> > + */
> > + if (kvm_arm_vm_has_run_once(&kvm->arch)) {
> > + ret = *fw_reg_bmap != val ? -EBUSY : 0;
> > + goto out;
> > + }
> > +
> > + WRITE_ONCE(*fw_reg_bmap, val);
> > +out:
> > + mutex_unlock(&kvm->lock);
> > + return ret;
> > +}
> > +
> > int kvm_arm_set_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > {
> > void __user *uaddr = (void __user *)(long)reg->addr;
> > @@ -321,6 +397,8 @@ int kvm_arm_set_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg)
> > return -EINVAL;
> >
> > return 0;
> > + case KVM_REG_ARM_STD_BMAP:
> > + return kvm_arm_set_fw_reg_bmap(vcpu, reg_id, val);
> > default:
> > return -ENOENT;
> > }
> > diff --git a/include/kvm/arm_hypercalls.h b/include/kvm/arm_hypercalls.h
> > index 5d38628a8d04..64d30b452809 100644
> > --- a/include/kvm/arm_hypercalls.h
> > +++ b/include/kvm/arm_hypercalls.h
> > @@ -6,6 +6,9 @@
> >
> > #include <asm/kvm_emulate.h>
> >
> > +#define ARM_SMCCC_STD_FEATURES \
> > + GENMASK_ULL(KVM_REG_ARM_STD_BMAP_BIT_MAX, 0)
> > +
>
> This probably needs KVM_ somewhere in its name for the sake of scoping.
>
RIght, I can append that.
> > int kvm_hvc_call_handler(struct kvm_vcpu *vcpu);
> >
> > static inline u32 smccc_get_function(struct kvm_vcpu *vcpu)
> > @@ -42,6 +45,7 @@ static inline void smccc_set_retval(struct kvm_vcpu *vcpu,
> >
> > struct kvm_one_reg;
> >
> > +void kvm_arm_init_hypercalls(struct kvm *kvm);
> > int kvm_arm_get_fw_num_regs(struct kvm_vcpu *vcpu);
> > int kvm_arm_copy_fw_reg_indices(struct kvm_vcpu *vcpu, u64 __user *uindices);
> > int kvm_arm_get_fw_reg(struct kvm_vcpu *vcpu, const struct kvm_one_reg *reg);
> > --
> > 2.35.1.473.g83b2b277ed-goog
> >
Regards,
Raghavendra
[1]: https://lore.kernel.org/linux-arm-kernel/CAJHc60y6b-scY8zcPuLnjGtr6HzSBnmhi2mCnmkNm4nTxgMTUQ@xxxxxxxxxxxxxx/
