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

[Gleb Natapov <gleb@redhat.com>]

On Fri, Oct 11, 2013 at 06:08:16PM +0800, chai wen wrote:
> On 10/10/2013 11:15 PM, Gleb Natapov wrote:
> > On Thu, Oct 10, 2013 at 09:20:15PM +0800, chai wen wrote:
> >> Hi Gleb
> >>
> >> Thanks for you explanation about async_pf in kvm.
> >> Page pinning is not mandatory in kvm async_pf processing and probably should
> >> be dropped later.this patch drops the FOLL_GET flag when doing GUP in async_pf
> >> and uses a bool (marking result of GUP) to replace the *page in kvm_async_pf
> >> structure to simplify some following async_pf check/clear processing.
> > Thanks, but the patch need to be on top of
> > git://git.kernel.org/pub/scm/virt/kvm/kvm.git queue branch.
> >
> > Some more comment bellow.
> >
> >> Thanks.
> >>
> >> Suggested-by:  gleb@redhat.com
> >> Signed-off-by: chai wen <chaiw.fnst@cn.fujitsu.com>
> >> ---
> >>  arch/x86/kvm/x86.c       |    5 ++---
> >>  include/linux/kvm_host.h |    2 +-
> >>  virt/kvm/async_pf.c      |   16 ++++++----------
> >>  3 files changed, 9 insertions(+), 14 deletions(-)
> >>
> >> diff --git a/arch/x86/kvm/x86.c b/arch/x86/kvm/x86.c
> >> index e5ca72a..0ff7d02 100644
> >> --- a/arch/x86/kvm/x86.c
> >> +++ b/arch/x86/kvm/x86.c
> >> @@ -7282,8 +7282,7 @@ 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))
> >> +	if ((vcpu->arch.mmu.direct_map != work->arch.direct_map))
> >>  		return;
> >>  
> >>  	r = kvm_mmu_reload(vcpu);
> >> @@ -7393,7 +7392,7 @@ 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->got_page)
> >>  		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 0fbbc7a..232a7d0 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 got_page;
> > I would call it wakeup_all.
> >
> >>  	bool done;
> >>  };
> >>  
> >> diff --git a/virt/kvm/async_pf.c b/virt/kvm/async_pf.c
> >> index 8a39dda..0cafdc7 100644
> >> --- a/virt/kvm/async_pf.c
> >> +++ b/virt/kvm/async_pf.c
> >> @@ -56,7 +56,7 @@ void kvm_async_pf_vcpu_init(struct kvm_vcpu *vcpu)
> >>  
> >>  static void async_pf_execute(struct work_struct *work)
> >>  {
> >> -	struct page *page = NULL;
> >> +	long nrpages = 0;
> >>  	struct kvm_async_pf *apf =
> >>  		container_of(work, struct kvm_async_pf, work);
> >>  	struct mm_struct *mm = apf->mm;
> >> @@ -68,13 +68,13 @@ 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);
> >> +	nrpages = 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;
> >> +	apf->got_page = (1==nrpages);
> > Space before and after ==, but this is not needed at all.  It does
> > not matter if GUP failed or not here. We need to wake up guest process
> > that is waiting on a page, so just set wakeup_all to false in kvm_setup_async_pf().
> Hi Gleb
> Thanks for you reply.
> It does not matter if GUP failed or not here. But in the original logic of
> kvm_arch_async_page_ready and kvm_arch_async_page_present, there is
> is_error_page
> checking in them. I am not sure wheher this ‘logic’ should be dropped.
> And in kvm_async_pf_wakeup_all we used a dummy bad page pointer to generate
> boradcast wakeup before. if wakeup_all(bool) is marking the result of
> gup should it be false in this function ?
> 
is_error_page() does not consider NULL as error value.

> thanks
> >
> >>  	apf->done = true;
> >>  	spin_unlock(&vcpu->async_pf.lock);
> >>  
> >> @@ -114,8 +114,6 @@ 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);
> >> @@ -135,14 +133,12 @@ void kvm_check_async_pf_completion(struct kvm_vcpu *vcpu)
> >>  		list_del(&work->link);
> >>  		spin_unlock(&vcpu->async_pf.lock);
> >>  
> >> -		if (work->page)
> >> +		if (work->got_page)
> >>  			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);
> >>  	}
> >>  }
> >> @@ -165,7 +161,7 @@ int kvm_setup_async_pf(struct kvm_vcpu *vcpu, gva_t gva, gfn_t gfn,
> >>  	if (!work)
> >>  		return 0;
> >>  
> >> -	work->page = NULL;
> >> +	work->got_page = false;
> > Sp here do:
> > 	work->wakeup_all = false;
> >
> >>  	work->done = false;
> >>  	work->vcpu = vcpu;
> >>  	work->gva = gva;
> >> @@ -206,7 +202,7 @@ int kvm_async_pf_wakeup_all(struct kvm_vcpu *vcpu)
> >>  	if (!work)
> >>  		return -ENOMEM;
> >>  
> >> -	work->page = KVM_ERR_PTR_BAD_PAGE;
> >> +	work->got_page = false;
> > And here:
> > 	work->wakeup_all = true;
> >>  	INIT_LIST_HEAD(&work->queue); /* for list_del to work */
> >>  
> >>  	spin_lock(&vcpu->async_pf.lock);
> >> -- 
> >> 1.7.1
> > --
> > 			Gleb.
> > --
> > To unsubscribe from this list: send the line "unsubscribe linux-kernel" in
> > the body of a message to majordomo@vger.kernel.org
> > More majordomo info at  http://vger.kernel.org/majordomo-info.html
> > Please read the FAQ at  http://www.tux.org/lkml/
> >
> 
> 
> -- 
> Regards
> 
> chai wen
> 

--
			Gleb.