Re: [PATCH RESEND] Drop-FOLL_GET-in-GUP-when-doing-async_pf-in-kvm

[Gleb Natapov <gleb@redhat.com>]

On Mon, Oct 14, 2013 at 10:22:33PM +0800, chai wen wrote:
> 
> Hi Gleb
> 
> I am sorry for my mistake before.
> This new patch is based on 'git://git.kernel.org/pub/scm/virt/kvm/kvm.git'.
> Page pinning is not mandatory in kvm async_pf processing and probably
> should be dropped later. And we don't mind whether the GUP is failed or
> not. What we need to do is to wake up guests process that is waitting on
> a page. So drop the FOLL_GET flag in GUP, and do some simplifying in
> async_pf check/clear processing.
> To trigger the happening of async_pf, it needs some memory stress in system
> for doing swap pressure. I simply run some big block dd processes in system
> before. And it works as I see some async_pf happening and the vm works well.
> But I do not know the exact occasion of the async_pf happening.
> Thanks.
> 
Applied, thanks.

> Suggested-by: Gleb Natapov <gleb@redhat.com>
> Signed-off-by: Gu zheng <guz.fnst@cn.fujitsu.com>
> Signed-off-by: chai wen <chaiw.fnst@cn.fujitsu.com>
> ---
>  arch/x86/kvm/x86.c         |    4 ++--
>  include/linux/kvm_host.h   |    2 +-
>  include/trace/events/kvm.h |   10 ++++------
>  virt/kvm/async_pf.c        |   17 +++++------------
>  4 files changed, 12 insertions(+), 21 deletions(-)
> 
> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> index c951c71..edf2a07 100644
> --- a/arch/x86/kvm/x86.c
> +++ b/arch/x86/kvm/x86.c
> @@ -7298,7 +7298,7 @@ void kvm_arch_async_page_ready(struct kvm_vcpu *vcpu, struct kvm_async_pf *work)
>  	int r;
>  
>  	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map) ||
> -	      is_error_page(work->page))
> +	      work->wakeup_all)
>  		return;
>  
>  	r = kvm_mmu_reload(vcpu);
> @@ -7408,7 +7408,7 @@ void kvm_arch_async_page_present(struct kvm_vcpu *vcpu,
>  	struct x86_exception fault;
>  
>  	trace_kvm_async_pf_ready(work->arch.token, work->gva);
> -	if (is_error_page(work->page))
> +	if (work->wakeup_all)
>  		work->arch.token = ~0; /* broadcast wakeup */
>  	else
>  		kvm_del_async_pf_gfn(vcpu, work->arch.gfn);
> diff --git a/include/linux/kvm_host.h b/include/linux/kvm_host.h
> index 7c961e1..5841e14 100644
> --- a/include/linux/kvm_host.h
> +++ b/include/linux/kvm_host.h
> @@ -189,7 +189,7 @@ struct kvm_async_pf {
>  	gva_t gva;
>  	unsigned long addr;
>  	struct kvm_arch_async_pf arch;
> -	struct page *page;
> +	bool   wakeup_all;
>  };
>  
>  void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu);
> diff --git a/include/trace/events/kvm.h b/include/trace/events/kvm.h
> index 7005d11..131a0bd 100644
> --- a/include/trace/events/kvm.h
> +++ b/include/trace/events/kvm.h
> @@ -296,23 +296,21 @@ DEFINE_EVENT(kvm_async_pf_nopresent_ready, kvm_async_pf_ready,
>  
>  TRACE_EVENT(
>  	kvm_async_pf_completed,
> -	TP_PROTO(unsigned long address, struct page *page, u64 gva),
> -	TP_ARGS(address, page, gva),
> +	TP_PROTO(unsigned long address, u64 gva),
> +	TP_ARGS(address, gva),
>  
>  	TP_STRUCT__entry(
>  		__field(unsigned long, address)
> -		__field(pfn_t, pfn)
>  		__field(u64, gva)
>  		),
>  
>  	TP_fast_assign(
>  		__entry->address = address;
> -		__entry->pfn = page ? page_to_pfn(page) : 0;
>  		__entry->gva = gva;
>  		),
>  
> -	TP_printk("gva %#llx address %#lx pfn %#llx",  __entry->gva,
> -		  __entry->address, __entry->pfn)
> +	TP_printk("gva %#llx address %#lx",  __entry->gva,
> +		  __entry->address)
>  );
>  
>  #endif
> diff --git a/virt/kvm/async_pf.c b/virt/kvm/async_pf.c
> index b197950..8631d9c 100644
> --- a/virt/kvm/async_pf.c
> +++ b/virt/kvm/async_pf.c
> @@ -56,7 +56,6 @@ void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
>  
>  static void async_pf_execute(struct work_struct *work)
>  {
> -	struct page *page = NULL;
>  	struct kvm_async_pf *apf =
>  		container_of(work, struct kvm_async_pf, work);
>  	struct mm_struct *mm = apf->mm;
> @@ -68,13 +67,12 @@ static void async_pf_execute(struct work_struct *work)
>  
>  	use_mm(mm);
>  	down_read(&mm->mmap_sem);
> -	get_user_pages(current, mm, addr, 1, 1, 0, &page, NULL);
> +	get_user_pages(current, mm, addr, 1, 1, 0, NULL, NULL);
>  	up_read(&mm->mmap_sem);
>  	unuse_mm(mm);
>  
>  	spin_lock(&vcpu->async_pf.lock);
>  	list_add_tail(&apf->link, &vcpu->async_pf.done);
> -	apf->page = page;
>  	spin_unlock(&vcpu->async_pf.lock);
>  
>  	/*
> @@ -82,7 +80,7 @@ static void async_pf_execute(struct work_struct *work)
>  	 * this point
>  	 */
>  
> -	trace_kvm_async_pf_completed(addr, page, gva);
> +	trace_kvm_async_pf_completed(addr, gva);
>  
>  	if (waitqueue_active(&vcpu->wq))
>  		wake_up_interruptible(&vcpu->wq);
> @@ -112,8 +110,6 @@ void kvm_clear_async_pf_completion_queue(struct kvm_vcpu *vcpu)
>  			list_entry(vcpu->async_pf.done.next,
>  				   typeof(*work), link);
>  		list_del(&work->link);
> -		if (!is_error_page(work->page))
> -			kvm_release_page_clean(work->page);
>  		kmem_cache_free(async_pf_cache, work);
>  	}
>  	spin_unlock(&vcpu->async_pf.lock);
> @@ -133,14 +129,11 @@ void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
>  		list_del(&work->link);
>  		spin_unlock(&vcpu->async_pf.lock);
>  
> -		if (work->page)
> -			kvm_arch_async_page_ready(vcpu, work);
> +		kvm_arch_async_page_ready(vcpu, work);
>  		kvm_arch_async_page_present(vcpu, work);
>  
>  		list_del(&work->queue);
>  		vcpu->async_pf.queued--;
> -		if (!is_error_page(work->page))
> -			kvm_release_page_clean(work->page);
>  		kmem_cache_free(async_pf_cache, work);
>  	}
>  }
> @@ -163,7 +156,7 @@ int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
>  	if (!work)
>  		return 0;
>  
> -	work->page = NULL;
> +	work->wakeup_all = false;
>  	work->vcpu = vcpu;
>  	work->gva = gva;
>  	work->addr = gfn_to_hva(vcpu->kvm, gfn);
> @@ -203,7 +196,7 @@ int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
>  	if (!work)
>  		return -ENOMEM;
>  
> -	work->page = KVM_ERR_PTR_BAD_PAGE;
> +	work->wakeup_all = true;
>  	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
>  
>  	spin_lock(&vcpu->async_pf.lock);
> -- 
> 1.7.1

--
			Gleb.