Re: [RFC][PATCH] kvm: add suspend pm-notifier

[Marc Zyngier <maz@kernel.org>]

On Thu, 03 Jun 2021 17:43:15 +0100,
Sergey Senozhatsky <senozhatsky@chromium.org> wrote:
> 
> Add KVM suspend/hibernate PM-notifier which lets architectures
> to implement arch-specific VM suspend code. For instance, on x86
> this sets PVCLOCK_GUEST_STOPPED on all the VCPUs.
> 
> Our case is that user puts the host system into sleep multiple
> times a day (e.g. closes the laptop's lid) so we need a reliable
> way to suspend VMs properly.
> 
> Signed-off-by: Sergey Senozhatsky <senozhatsky@chromium.org>
> ---
>  arch/arm64/kvm/arm.c       |  4 ++++
>  arch/mips/kvm/mips.c       |  4 ++++
>  arch/powerpc/kvm/powerpc.c |  4 ++++
>  arch/s390/kvm/kvm-s390.c   |  4 ++++
>  arch/x86/kvm/x86.c         | 21 ++++++++++++++++++++
>  include/linux/kvm_host.h   |  8 ++++++++
>  virt/kvm/kvm_main.c        | 40 ++++++++++++++++++++++++++++++++++++++
>  7 files changed, 85 insertions(+)
> 
> diff --git a/arch/arm64/kvm/arm.c b/arch/arm64/kvm/arm.c
> index 1126eae27400..547dbe44d039 100644
> --- a/arch/arm64/kvm/arm.c
> +++ b/arch/arm64/kvm/arm.c
> @@ -1311,6 +1311,10 @@ static int kvm_vm_ioctl_set_device_addr(struct kvm *kvm,
>  	}
>  }
>  
> +void kvm_arch_pm_notifier(struct kvm *kvm)
> +{
> +}
> +
>  long kvm_arch_vm_ioctl(struct file *filp,
>  		       unsigned int ioctl, unsigned long arg)
>  {
> diff --git a/arch/mips/kvm/mips.c b/arch/mips/kvm/mips.c
> index 4d4af97dcc88..d4408acd2be6 100644
> --- a/arch/mips/kvm/mips.c
> +++ b/arch/mips/kvm/mips.c
> @@ -980,6 +980,10 @@ void kvm_arch_flush_remote_tlbs_memslot(struct kvm *kvm,
>  	kvm_flush_remote_tlbs(kvm);
>  }
>  
> +void kvm_arch_pm_notifier(struct kvm *kvm)
> +{
> +}
> +
>  long kvm_arch_vm_ioctl(struct file *filp, unsigned int ioctl, unsigned long arg)
>  {
>  	long r;
> diff --git a/arch/powerpc/kvm/powerpc.c b/arch/powerpc/kvm/powerpc.c
> index a2a68a958fa0..96e8a7b6fcf0 100644
> --- a/arch/powerpc/kvm/powerpc.c
> +++ b/arch/powerpc/kvm/powerpc.c
> @@ -2334,6 +2334,10 @@ static int kvmppc_get_cpu_char(struct kvm_ppc_cpu_char *cp)
>  }
>  #endif
>  
> +void kvm_arch_pm_notifier(struct kvm *kvm)
> +{
> +}
> +
>  long kvm_arch_vm_ioctl(struct file *filp,
>                         unsigned int ioctl, unsigned long arg)
>  {
> diff --git a/arch/s390/kvm/kvm-s390.c b/arch/s390/kvm/kvm-s390.c
> index 1296fc10f80c..c5f86fc1e497 100644
> --- a/arch/s390/kvm/kvm-s390.c
> +++ b/arch/s390/kvm/kvm-s390.c
> @@ -2367,6 +2367,10 @@ static int kvm_s390_handle_pv(struct kvm *kvm, struct kvm_pv_cmd *cmd)
>  	return r;
>  }
>  
> +void kvm_arch_pm_notifier(struct kvm *kvm)
> +{
> +}
> +
>  long kvm_arch_vm_ioctl(struct file *filp,
>  		       unsigned int ioctl, unsigned long arg)
>  {
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index bbc4e04e67ad..3f3d6497593f 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -5613,6 +5613,27 @@ static int kvm_vm_ioctl_set_msr_filter(struct kvm *kvm, void __user *argp)
>  	return 0;
>  }
>  
> +void kvm_arch_pm_notifier(struct kvm *kvm)
> +{
> +#ifdef CONFIG_PM
> +	int c;
> +
> +	mutex_lock(&kvm->lock);
> +	for (c = 0; c < kvm->created_vcpus; c++) {
> +		struct kvm_vcpu *vcpu = kvm->vcpus[c];
> +		int r;
> +
> +		if (!vcpu)
> +			continue;

Wouldn't kvm_for_each_vcpu() avoid this kind of checks?

> +		r = kvm_set_guest_paused(vcpu);
> +		if (!r)
> +			continue;
> +		pr_err("Failed to suspend VCPU-%d: %d\n", vcpu->vcpu_id,  r);
> +	}
> +	mutex_unlock(&kvm->lock);
> +#endif
> +}
> +
>  long kvm_arch_vm_ioctl(struct file *filp,
>  		       unsigned int ioctl, unsigned long arg)
>  {
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index 2f34487e21f2..86695320a6b7 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -37,6 +37,8 @@
>  #include <asm/kvm_host.h>
>  #include <linux/kvm_dirty_ring.h>
>  
> +#include <linux/notifier.h>
> +
>  #ifndef KVM_MAX_VCPU_ID
>  #define KVM_MAX_VCPU_ID KVM_MAX_VCPUS
>  #endif
> @@ -579,6 +581,10 @@ struct kvm {
>  	pid_t userspace_pid;
>  	unsigned int max_halt_poll_ns;
>  	u32 dirty_ring_size;
> +
> +#ifdef CONFIG_PM
> +	struct notifier_block pm_notifier;
> +#endif

I'd certainly like to be able to opt out from this on architectures
that do not implement anything useful in the PM callbacks. Please
consider making this an independent config option that individual
archs can buy into.

>  };
>  
>  #define kvm_err(fmt, ...) \
> @@ -992,6 +998,8 @@ int kvm_arch_vcpu_create(struct kvm_vcpu *vcpu);
>  void kvm_arch_vcpu_postcreate(struct kvm_vcpu *vcpu);
>  void kvm_arch_vcpu_destroy(struct kvm_vcpu *vcpu);
>  
> +void kvm_arch_pm_notifier(struct kvm *kvm);
> +
>  #ifdef __KVM_HAVE_ARCH_VCPU_DEBUGFS
>  void kvm_arch_create_vcpu_debugfs(struct kvm_vcpu *vcpu, struct dentry *debugfs_dentry);
>  #endif
> diff --git a/virt/kvm/kvm_main.c b/virt/kvm/kvm_main.c
> index 6b4feb92dc79..86925ab7d162 100644
> --- a/virt/kvm/kvm_main.c
> +++ b/virt/kvm/kvm_main.c
> @@ -51,6 +51,7 @@
>  #include <linux/io.h>
>  #include <linux/lockdep.h>
>  #include <linux/kthread.h>
> +#include <linux/suspend.h>
>  
>  #include <asm/processor.h>
>  #include <asm/ioctl.h>
> @@ -779,6 +780,43 @@ static int kvm_init_mmu_notifier(struct kvm *kvm)
>  
>  #endif /* CONFIG_MMU_NOTIFIER && KVM_ARCH_WANT_MMU_NOTIFIER */
>  
> +#ifdef CONFIG_PM
> +static int kvm_pm_notifier_call(struct notifier_block *bl,
> +				unsigned long state,
> +				void *unused)
> +{
> +	struct kvm *kvm = container_of(bl, struct kvm, pm_notifier);
> +
> +	switch (state) {
> +	case PM_HIBERNATION_PREPARE:
> +	case PM_SUSPEND_PREPARE:
> +		kvm_arch_pm_notifier(kvm);

How about passing the state to the notifier callback? I'd expect it to
be useful to do something on resume too.

> +		break;
> +	}
> +	return NOTIFY_DONE;
> +}
> +
> +static void kvm_init_pm_notifier(struct kvm *kvm)
> +{
> +	kvm->pm_notifier.notifier_call = kvm_pm_notifier_call;
> +	kvm->pm_notifier.priority = INT_MAX;

How is this priority determined?

> +	register_pm_notifier(&kvm->pm_notifier);
> +}
> +
> +static void kvm_destroy_pm_notifier(struct kvm *kvm)
> +{
> +	unregister_pm_notifier(&kvm->pm_notifier);
> +}
> +#else
> +static void kvm_init_pm_notifier(struct kvm *kvm)
> +{
> +}
> +
> +static void kvm_destroy_pm_notifier(struct kvm *kvm)
> +{
> +}
> +#endif /* CONFIG_PM */
> +
>  static struct kvm_memslots *kvm_alloc_memslots(void)
>  {
>  	int i;
> @@ -962,6 +1000,7 @@ static struct kvm *kvm_create_vm(unsigned long type)
>  	mutex_unlock(&kvm_lock);
>  
>  	preempt_notifier_inc();
> +	kvm_init_pm_notifier(kvm);
>  
>  	return kvm;
>  
> @@ -1009,6 +1048,7 @@ static void kvm_destroy_vm(struct kvm *kvm)
>  	int i;
>  	struct mm_struct *mm = kvm->mm;
>  
> +	kvm_destroy_pm_notifier(kvm);
>  	kvm_uevent_notify_change(KVM_EVENT_DESTROY_VM, kvm);
>  	kvm_destroy_vm_debugfs(kvm);
>  	kvm_arch_sync_events(kvm);

Thanks,

	M.

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