Re: [PATCH 7/9] kvm: s390: Add CPU dump functionality

[Claudio Imbrenda <imbrenda@linux.ibm.com>]

On Thu, 28 Apr 2022 13:01:00 +0000
Janosch Frank <frankja@linux.ibm.com> wrote:

> The previous patch introduced the per-VM dump functions now let's
> focus on dumping the VCPU state via the newly introduced
> KVM_S390_PV_CPU_COMMAND ioctl which mirrors the VM UV ioctl and can be
> extended with new commands later.
> 
> Signed-off-by: Janosch Frank <frankja@linux.ibm.com>
> ---
>  arch/s390/kvm/kvm-s390.c | 73 ++++++++++++++++++++++++++++++++++++++++
>  arch/s390/kvm/kvm-s390.h |  1 +
>  arch/s390/kvm/pv.c       | 16 +++++++++
>  include/uapi/linux/kvm.h |  5 +++
>  4 files changed, 95 insertions(+)
> 
> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
> index 8984e8db33b4..d15ce38bef14 100644
> --- a/arch/s390/kvm/kvm-s390.c
> +++ b/arch/s390/kvm/kvm-s390.c
> @@ -5149,6 +5149,52 @@ long kvm_arch_vcpu_async_ioctl(struct file *filp,
>  	return -ENOIOCTLCMD;
>  }
>  
> +static int kvm_s390_handle_pv_vcpu_dump(struct kvm_vcpu *vcpu,
> +					struct kvm_pv_cmd *cmd)
> +{
> +	struct kvm_s390_pv_dmp dmp;
> +	void *data;
> +	int ret;
> +
> +	/* Dump initialization is a prerequisite */
> +	if (!vcpu->kvm->arch.pv.dumping)
> +		return -EINVAL;
> +
> +	if (copy_from_user(&dmp, (__u8 __user *)cmd->data, sizeof(dmp)))
> +		return -EFAULT;
> +
> +	/* We only handle this subcmd right now */
> +	if (dmp.subcmd != KVM_PV_DUMP_CPU)
> +		return -EINVAL;
> +
> +	/* CPU dump length is the same as create cpu storage donation. */
> +	if (dmp.buff_len != uv_info.guest_cpu_stor_len)
> +		return -EINVAL;
> +
> +	data = vzalloc(uv_info.guest_cpu_stor_len);
> +	if (!data)
> +		return -ENOMEM;
> +
> +	ret = kvm_s390_pv_dump_cpu(vcpu, data, &cmd->rc, &cmd->rrc);
> +
> +	VCPU_EVENT(vcpu, 3, "PROTVIRT DUMP CPU %d rc %x rrc %x",
> +		   vcpu->vcpu_id, cmd->rc, cmd->rrc);
> +
> +	if (ret) {
> +		vfree(data);
> +		return -EINVAL;
> +	}
> +
> +	/* On success copy over the dump data */
> +	if (copy_to_user((__u8 __user *)dmp.buff_addr, data, uv_info.guest_cpu_stor_len)) {
> +		vfree(data);
> +		return -EFAULT;
> +	}
> +
> +	vfree(data);
> +	return 0;
> +}
> +
>  long kvm_arch_vcpu_ioctl(struct file *filp,
>  			 unsigned int ioctl, unsigned long arg)
>  {
> @@ -5313,6 +5359,33 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
>  					   irq_state.len);
>  		break;
>  	}
> +	case KVM_S390_PV_CPU_COMMAND: {
> +		struct kvm_pv_cmd cmd;
> +
> +		r = -EINVAL;
> +		if (!is_prot_virt_host())
> +			break;
> +
> +		r = -EFAULT;
> +		if (copy_from_user(&cmd, argp, sizeof(cmd)))
> +			break;
> +
> +		r = -EINVAL;
> +		if (cmd.flags)
> +			break;
> +
> +		/* We only handle this cmd right now */
> +		if (cmd.cmd != KVM_PV_DUMP)
> +			break;
> +
> +		r = kvm_s390_handle_pv_vcpu_dump(vcpu, &cmd);
> +
> +		/* Always copy over UV rc / rrc data */
> +		if (copy_to_user((__u8 __user *)argp, &cmd.rc,
> +				 sizeof(cmd.rc) + sizeof(cmd.rrc)))
> +			r = -EFAULT;
> +		break;
> +	}
>  	default:
>  		r = -ENOTTY;
>  	}
> diff --git a/arch/s390/kvm/kvm-s390.h b/arch/s390/kvm/kvm-s390.h
> index 2868dd0bba25..a39815184350 100644
> --- a/arch/s390/kvm/kvm-s390.h
> +++ b/arch/s390/kvm/kvm-s390.h
> @@ -250,6 +250,7 @@ int kvm_s390_pv_set_sec_parms(struct kvm *kvm, void *hdr, u64 length, u16 *rc,
>  int kvm_s390_pv_unpack(struct kvm *kvm, unsigned long addr, unsigned long size,
>  		       unsigned long tweak, u16 *rc, u16 *rrc);
>  int kvm_s390_pv_set_cpu_state(struct kvm_vcpu *vcpu, u8 state);
> +int kvm_s390_pv_dump_cpu(struct kvm_vcpu *vcpu, void *buff, u16 *rc, u16 *rrc);
>  int kvm_s390_pv_dump_stor_state(struct kvm *kvm, void __user *buff_user,
>  				u64 *gaddr, u64 buff_user_len, u16 *rc, u16 *rrc);
>  
> diff --git a/arch/s390/kvm/pv.c b/arch/s390/kvm/pv.c
> index d1635ed50078..9ab8192b9b23 100644
> --- a/arch/s390/kvm/pv.c
> +++ b/arch/s390/kvm/pv.c
> @@ -299,6 +299,22 @@ int kvm_s390_pv_set_cpu_state(struct kvm_vcpu *vcpu, u8 state)
>  	return 0;
>  }
>  
> +int kvm_s390_pv_dump_cpu(struct kvm_vcpu *vcpu, void *buff, u16 *rc, u16 *rrc)
> +{
> +	struct uv_cb_dump_cpu uvcb = {
> +		.header.cmd = UVC_CMD_DUMP_CPU,
> +		.header.len = sizeof(uvcb),
> +		.cpu_handle = vcpu->arch.pv.handle,
> +		.dump_area_origin = (u64)buff,
> +	};
> +	int cc;
> +
> +	cc = uv_call_sched(0, (u64)&uvcb);

it's a small amount of data, but you use the _sched variant?

and, why aren't you using the _sched variant in the previous patch (for
DUMP_COMPLETE)?

to be clear: I think the right thing is to always use the _sched
variant unless there is a good reason not to (so please fix the previous
patch)

> +	*rc = uvcb.header.rc;
> +	*rrc = uvcb.header.rrc;
> +	return cc;
> +}
> +
>  /* Size of the cache for the storage state dump data. 1MB for now */
>  #define DUMP_BUFF_LEN HPAGE_SIZE
>  
> diff --git a/include/uapi/linux/kvm.h b/include/uapi/linux/kvm.h
> index b34850907291..108bc7b7a71b 100644
> --- a/include/uapi/linux/kvm.h
> +++ b/include/uapi/linux/kvm.h
> @@ -1144,6 +1144,7 @@ struct kvm_ppc_resize_hpt {
>  #define KVM_CAP_S390_MEM_OP_EXTENSION 211
>  #define KVM_CAP_PMU_CAPABILITY 212
>  #define KVM_CAP_DISABLE_QUIRKS2 213
> +#define KVM_CAP_S390_PROTECTED_DUMP 214
>  
>  #ifdef KVM_CAP_IRQ_ROUTING
>  
> @@ -1649,6 +1650,7 @@ enum pv_cmd_dmp_id {
>  	KVM_PV_DUMP_INIT,
>  	KVM_PV_DUMP_CONFIG_STOR_STATE,
>  	KVM_PV_DUMP_COMPLETE,
> +	KVM_PV_DUMP_CPU,
>  };
>  
>  struct kvm_s390_pv_dmp {
> @@ -2110,4 +2112,7 @@ struct kvm_stats_desc {
>  /* Available with KVM_CAP_XSAVE2 */
>  #define KVM_GET_XSAVE2		  _IOR(KVMIO,  0xcf, struct kvm_xsave)
>  
> +/* Available with KVM_CAP_S390_PROTECTED_DUMP */
> +#define KVM_S390_PV_CPU_COMMAND	_IOWR(KVMIO, 0xd0, struct kvm_pv_cmd)
> +
>  #endif /* __LINUX_KVM_H */