On Tue, Aug 08, 2017 at 05:46:16PM +0100, James Morse wrote:
> Instead of supporting SDEI in KVM, and providing a new API to
> control and configure the in-kernel support, allow user-space to
> request particular SMC-CC ranges from guest HVC calls to be handled
> by user-space.
>
> This requires two KVM capabilities, KVM_CAP_ARM_SDEI_1_0 advertises
> that KVM knows how match SDEI SMC-CC calls from a guest. To pass these
> calls to user-space requires this cap to be enabled using
> KVM_CAP_ENABLE_CAP_VM.
>
> Calls are passed with exit_reason = KVM_EXIT_HYPERCALL, the kvm_run
> structure has copies of the first 6 registers and the guest pstate.
>
> Signed-off-by: James Morse <james.morse@xxxxxxx>
> ---
> While I'm in here, remove KVM_CAP_ARM_SET_DEVICE_ADDR's extra entry
> for r=1;break?
>
> Changes from v1:
> * all of it
>
> Documentation/virtual/kvm/api.txt | 12 +++++++++---
> arch/arm64/include/asm/kvm_host.h | 6 ++++++
> arch/arm64/kvm/handle_exit.c | 28 +++++++++++++++++++++++++++-
> include/uapi/linux/kvm.h | 1 +
> virt/kvm/arm/arm.c | 29 +++++++++++++++++++++++++++--
> 5 files changed, 70 insertions(+), 6 deletions(-)
>
> diff --git a/Documentation/virtual/kvm/api.txt b/Documentation/virtual/kvm/api.txt
> index e63a35fafef0..740288d6e894 100644
> --- a/Documentation/virtual/kvm/api.txt
> +++ b/Documentation/virtual/kvm/api.txt
> @@ -1012,7 +1012,7 @@ documentation when it pops into existence).
> 4.37 KVM_ENABLE_CAP
>
> Capability: KVM_CAP_ENABLE_CAP, KVM_CAP_ENABLE_CAP_VM
> -Architectures: x86 (only KVM_CAP_ENABLE_CAP_VM),
> +Architectures: x86, arm, arm64 (only KVM_CAP_ENABLE_CAP_VM),
> mips (only KVM_CAP_ENABLE_CAP), ppc, s390
> Type: vcpu ioctl, vm ioctl (with KVM_CAP_ENABLE_CAP_VM)
> Parameters: struct kvm_enable_cap (in)
> @@ -3540,10 +3540,16 @@ pending operations.
> __u32 pad;
> } hypercall;
>
> -Unused. This was once used for 'hypercall to userspace'. To implement
> -such functionality, use KVM_EXIT_IO (x86) or KVM_EXIT_MMIO (all except s390).
> +This was once used for 'hypercall to userspace'. To implement such
> +functionality, use KVM_EXIT_IO (x86) or KVM_EXIT_MMIO (all except s390).
I suggest you add 'for x86' to these two sentences, otherwise the
following paragraph seems contradicting.
> Note KVM_EXIT_IO is significantly faster than KVM_EXIT_MMIO.
>
> +On arm64 this is used to complete guest hypercalls (HVC) in user space.
> +e.g. Configuring SDEI or communicating with an emulated TEE.
s/e.g./For example,/
> +The required HVC ranges must first be enabled by KVM_CAP_ENABLE_CAP_VM.
> +The 'args' array contains a copy of r0-r5 and 'longmode' contains a copy
> +of the CPSR/PSTATE.
> +
I'm not entirely sure what this means by just looking at this text. Do
you mean that some HVC (and SMC?) calls can still be handled by the
kernel directly (PSCI), and some can be configured to be handled by the
kernel? Where does it specify how these 'ranges' are defined?
What about the immediate field, could userspace not also need to know
this?
> /* KVM_EXIT_TPR_ACCESS */
> struct {
> __u64 rip;
> diff --git a/arch/arm64/include/asm/kvm_host.h b/arch/arm64/include/asm/kvm_host.h
> index 7733492d9a35..77b8436e745e 100644
> --- a/arch/arm64/include/asm/kvm_host.h
> +++ b/arch/arm64/include/asm/kvm_host.h
> @@ -71,8 +71,14 @@ struct kvm_arch {
>
> /* Interrupt controller */
> struct vgic_dist vgic;
> +
> + /* SMC-CC/HVC ranges user space has requested */
> + u32 hvc_passthrough_ranges;
> };
>
> +/* SMC-CC/HVC ranges that can be passed to user space */
> +#define KVM_HVC_RANGE_SDEI 1
Oh, I see you want the kernel to know what constitutes a range for some
functionality and set things up that way.
What are your thoughts on letting userspace configure a range of from/to
values in x0 for some value/range of the immediate field?
> +
> #define KVM_NR_MEM_OBJS 40
>
> /*
> diff --git a/arch/arm64/kvm/handle_exit.c b/arch/arm64/kvm/handle_exit.c
> index 17d8a1677a0b..22eadf2cd82f 100644
> --- a/arch/arm64/kvm/handle_exit.c
> +++ b/arch/arm64/kvm/handle_exit.c
> @@ -21,6 +21,7 @@
>
> #include <linux/kvm.h>
> #include <linux/kvm_host.h>
> +#include <linux/sdei.h>
>
> #include <asm/esr.h>
> #include <asm/kvm_asm.h>
> @@ -34,15 +35,40 @@
>
> typedef int (*exit_handle_fn)(struct kvm_vcpu *, struct kvm_run *);
>
> +#define HVC_PASSTHROUGH(kvm, range) (kvm->arch.hvc_passthrough_ranges & range)
I think this should have parenthesis around (range) ?
Are we not going to support this for SMC passthrough? (for nested virt
for example).
> +
> +/*
> + * The guest made an SMC-CC call that user-space has claimed.
> + */
> +static int populate_hypercall_exit(struct kvm_vcpu *vcpu, struct kvm_run *run)
> +{
> + int i;
> +
> + run->exit_reason = KVM_EXIT_HYPERCALL;
> +
> + for (i = 0 ; i < ARRAY_SIZE(run->hypercall.args); i++)
> + run->hypercall.args[i] = vcpu_get_reg(vcpu, i);
> +
> + run->hypercall.longmode = *vcpu_cpsr(vcpu);
> +
> + return 0;
> +}
> +
> static int handle_hvc(struct kvm_vcpu *vcpu, struct kvm_run *run)
> {
> int ret;
> + struct kvm *kvm = vcpu->kvm;
> + unsigned long x0 = vcpu_get_reg(vcpu, 0);
>
> trace_kvm_hvc_arm64(*vcpu_pc(vcpu), vcpu_get_reg(vcpu, 0),
> kvm_vcpu_hvc_get_imm(vcpu));
> vcpu->stat.hvc_exit_stat++;
>
> - ret = kvm_psci_call(vcpu);
> + if (IS_SDEI_CALL(x0) && HVC_PASSTHROUGH(kvm, KVM_HVC_RANGE_SDEI))
> + ret = populate_hypercall_exit(vcpu, run);
> + else
> + ret = kvm_psci_call(vcpu);
> +
> if (ret < 0) {
> kvm_inject_undefined(vcpu);
> return 1;
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index 6cd63c18708a..8e6bc6ba918d 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -929,6 +929,7 @@ struct kvm_ppc_resize_hpt {
> #define KVM_CAP_PPC_SMT_POSSIBLE 147
> #define KVM_CAP_HYPERV_SYNIC2 148
> #define KVM_CAP_HYPERV_VP_INDEX 149
> +#define KVM_CAP_ARM_SDEI_1_0 150
>
> #ifdef KVM_CAP_IRQ_ROUTING
>
> diff --git a/virt/kvm/arm/arm.c b/virt/kvm/arm/arm.c
> index e9765ee6d769..b8657b68fea7 100644
> --- a/virt/kvm/arm/arm.c
> +++ b/virt/kvm/arm/arm.c
> @@ -206,8 +206,10 @@ int kvm_vm_ioctl_check_extension(struct kvm *kvm, long ext)
> case KVM_CAP_READONLY_MEM:
> case KVM_CAP_MP_STATE:
> case KVM_CAP_IMMEDIATE_EXIT:
> - r = 1;
> - break;
> + case KVM_CAP_ENABLE_CAP_VM:
> +#ifdef CONFIG_ARM64
> + case KVM_CAP_ARM_SDEI_1_0:
> +#endif
> case KVM_CAP_ARM_SET_DEVICE_ADDR:
> r = 1;
> break;
> @@ -1082,6 +1084,21 @@ static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
> }
> }
>
> +static int kvm_vm_ioctl_enable_cap(struct kvm *kvm, struct kvm_enable_cap *req)
> +{
> + int err = -EINVAL;
> +
> + if (req->flags)
> + return err;
> +
> + if (IS_ENABLED(CONFIG_ARM64) && req->cap == KVM_CAP_ARM_SDEI_1_0) {
> + kvm->arch.hvc_passthrough_ranges |= KVM_HVC_RANGE_SDEI;
> + err = 0;
> + }
> +
> + return err;
> +}
> +
> long kvm_arch_vm_ioctl(struct file *filp,
> unsigned int ioctl, unsigned long arg)
> {
> @@ -1118,6 +1135,14 @@ long kvm_arch_vm_ioctl(struct file *filp,
>
> return 0;
> }
> + case KVM_ENABLE_CAP: {
> + struct kvm_enable_cap req;
> +
> + if (copy_from_user(&req, argp, sizeof(req)))
> + return -EFAULT;
> +
> + return kvm_vm_ioctl_enable_cap(kvm, &req);
> + }
> default:
> return -EINVAL;
> }
> --
> 2.13.3
>
Thanks,
-Christoffer
--
To unsubscribe from this list: send the line "unsubscribe devicetree" in
the body of a message to majordomo@xxxxxxxxxxxxxxx
More majordomo info at http://vger.kernel.org/majordomo-info.html
