Re: [PATCH 6/6 V3] Add age out of guest paused flag

[Marcelo Tosatti <mtosatti@redhat.com>]

On Fri, Nov 04, 2011 at 12:55:12PM -0400, Eric B Munson wrote:
> The KVM_GUEST_PAUSED flag will prevent a guest from compaining about a soft
> lockup but it can mask real soft lockups if the flag isn't cleared when it is
> no longer relevant.  This patch adds a kvm ioctl that the hypervisor will use
> when it resumes a guest to start a timer for aging out the flag.  The time out
> will be specified by the hypervisor in the ioctl call.
> 
> Signed-off-by: Eric B Munson <emunson@mgebm.net>
> Cc: mingo@redhat.com
> Cc: hpa@zytor.com
> Cc: arnd@arndb.de
> Cc: ryanh@linux.vnet.ibm.com
> Cc: aliguori@us.ibm.com
> Cc: mtosatti@redhat.com
> Cc: jeremy.fitzhardinge@citrix.com
> Cc: kvm@vger.kernel.org
> Cc: linux-arch@vger.kernel.org
> Cc: x86@kernel.org
> Cc: linux-kernel@vger.kernel.org
> ---
> Changes from V2:
>  Move function definition to kvm_host.h
> 
>  arch/x86/include/asm/kvm_host.h |    1 +
>  arch/x86/kvm/x86.c              |   32 ++++++++++++++++++++++++++++++++
>  include/linux/kvm.h             |    2 ++
>  include/linux/kvm_host.h        |    2 ++
>  4 files changed, 37 insertions(+), 0 deletions(-)
> 
> diff --git a/arch/x86/include/asm/kvm_host.h b/arch/x86/include/asm/kvm_host.h
> index beb94c6..5cec925 100644
> --- a/arch/x86/include/asm/kvm_host.h
> +++ b/arch/x86/include/asm/kvm_host.h
> @@ -673,6 +673,7 @@ int kvm_pv_mmu_op(struct kvm_vcpu *vcpu, unsigned long bytes,
>  u8 kvm_get_guest_memory_type(struct kvm_vcpu *vcpu, gfn_t gfn);
>  
>  int kvm_set_guest_paused(struct kvm_vcpu *vcpu);
> +void kvm_clear_guest_paused(struct kvm_vcpu *vcpu, unsigned int length);
>  
>  extern bool tdp_enabled;
>  
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index 0127617..ca2187b 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -46,6 +46,7 @@
>  #include <linux/hash.h>
>  #include <linux/pci.h>
>  #include <trace/events/kvm.h>
> +#include <linux/timer.h>
>  
>  #define CREATE_TRACE_POINTS
>  #include "trace.h"
> @@ -3299,6 +3300,15 @@ long kvm_arch_vcpu_ioctl(struct file *filp,
>  		r = kvm_set_guest_paused(vcpu);
>  		break;
>  	}
> +	case KVM_CLEAR_GUEST_PAUSED: {
> +		unsigned int length;
> +		r = -EFAULT;
> +		if (copy_from_user(&length, argp, sizeof length))
> +			goto out;
> +		r = 0;
> +		kvm_clear_guest_paused(vcpu, length);
> +		break;
> +	}
>  	default:
>  		r = -EINVAL;
>  	}
> @@ -6137,6 +6147,28 @@ int kvm_set_guest_paused(struct kvm_vcpu *vcpu)
>  }
>  EXPORT_SYMBOL_GPL(kvm_set_guest_paused);
>  
> +static void kvm_timer_clear_guest_paused(unsigned long vcpu_addr)
> +{
> +	struct kvm_vcpu *vcpu = (struct kvm_vcpu *)vcpu_addr;
> +	struct pvclock_vcpu_time_info *src = &vcpu->arch.hv_clock;
> +	src->flags = src->flags & (~PVCLOCK_GUEST_STOPPED);
> +}
> +
> +/*
> + * Host has resumed the guest, we need to clear the guest paused flag so we
> + * don't mask any real soft lockups.
> + */
> +void kvm_clear_guest_paused(struct kvm_vcpu *vcpu, unsigned int length)
> +{
> +	if (!timer_pending(&vcpu->flag_timer))
> +		setup_timer(&vcpu->flag_timer,
> +			    kvm_timer_clear_guest_paused,
> +			    (unsigned long)vcpu);
> +	mod_timer(&vcpu->flag_timer,
> +		  jiffies + (length * HZ));
> +}
> +EXPORT_SYMBOL_GPL(kvm_clear_guest_paused);
> +

How about setting guest stopped flag when the guest is resumed, and
clearing that flag on the guests watchdog timer interrupt (irrespective
of whether a softlockup has been detected or not) ?

Then this timer is not necessary.

Unrelated: patch 3 is missing on the series, and please post functional
and tested qemu-kvm patches along next time.

Thanks.