Date: Fri, 13 Nov 2015 19:47:28 -0200
From: Marcelo Tosatti <mtosatti@redhat.com>
To: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
Cc: pbonzini@redhat.com, kvm@vger.kernel.org, linux-kernel@vger.kernel.org
Subject: Re: [PATCH 05/10] KVM: x86: MMU: Use for_each_rmap_spte macro
 instead of pte_list_walk()
Message-ID: <20151113214727.GA29103@amt.cnet>
References: <20151112204849.ba920599a8426d7196a0df73@lab.ntt.co.jp>
 <20151112205245.6bd773737cfa78422dac5a79@lab.ntt.co.jp>
MIME-Version: 1.0
Content-Type: text/plain; charset=us-ascii
Content-Disposition: inline
In-Reply-To: <20151112205245.6bd773737cfa78422dac5a79@lab.ntt.co.jp>
User-Agent: Mutt/1.5.21 (2010-09-15)
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 3677
Lines: 113

On Thu, Nov 12, 2015 at 08:52:45PM +0900, Takuya Yoshikawa wrote:
> kvm_mmu_mark_parents_unsync() alone uses pte_list_walk(), witch does
> nearly the same as the for_each_rmap_spte macro.  The only difference
> is that is_shadow_present_pte() checks cannot be placed there because
> kvm_mmu_mark_parents_unsync() can be called with a new parent pointer
> whose entry is not set yet.
> 
> By calling mark_unsync() separately for the parent and adding the parent
> pointer to the parent_ptes chain later in kvm_mmu_get_page(), the macro
> works with no problem.
> 
> Signed-off-by: Takuya Yoshikawa <yoshikawa_takuya_b1@lab.ntt.co.jp>
> ---
>  arch/x86/kvm/mmu.c | 36 +++++++++++++-----------------------
>  1 file changed, 13 insertions(+), 23 deletions(-)
> 
> diff --git a/arch/x86/kvm/mmu.c b/arch/x86/kvm/mmu.c
> index e8cfdc4..1691171 100644
> --- a/arch/x86/kvm/mmu.c
> +++ b/arch/x86/kvm/mmu.c
> @@ -1007,26 +1007,6 @@ static void pte_list_remove(u64 *spte, unsigned long *pte_list)
>  	}
>  }
>  
> -typedef void (*pte_list_walk_fn) (u64 *spte);
> -static void pte_list_walk(unsigned long *pte_list, pte_list_walk_fn fn)
> -{
> -	struct pte_list_desc *desc;
> -	int i;
> -
> -	if (!*pte_list)
> -		return;
> -
> -	if (!(*pte_list & 1))
> -		return fn((u64 *)*pte_list);
> -
> -	desc = (struct pte_list_desc *)(*pte_list & ~1ul);
> -	while (desc) {
> -		for (i = 0; i < PTE_LIST_EXT && desc->sptes[i]; ++i)
> -			fn(desc->sptes[i]);
> -		desc = desc->more;
> -	}
> -}
> -
>  static unsigned long *__gfn_to_rmap(gfn_t gfn, int level,
>  				    struct kvm_memory_slot *slot)
>  {
> @@ -1741,7 +1721,12 @@ static struct kvm_mmu_page *kvm_mmu_alloc_page(struct kvm_vcpu *vcpu,
>  static void mark_unsync(u64 *spte);
>  static void kvm_mmu_mark_parents_unsync(struct kvm_mmu_page *sp)
>  {
> -	pte_list_walk(&sp->parent_ptes, mark_unsync);
> +	u64 *sptep;
> +	struct rmap_iterator iter;
> +
> +	for_each_rmap_spte(&sp->parent_ptes, &iter, sptep) {
> +		mark_unsync(sptep);
> +	}
>  }
>  
>  static void mark_unsync(u64 *spte)
> @@ -2111,12 +2096,17 @@ static struct kvm_mmu_page *kvm_mmu_get_page(struct kvm_vcpu *vcpu,

Faulting a spte, and one of the levels of sptes, either


		spte-1
		spte-2
		spte-3

has present bit clear. So we're searching for a guest page to shadow, with
gfn "gfn".

>  		if (sp->unsync && kvm_sync_page_transient(vcpu, sp))
>  			break;

If a shadow for gfn exists, but is unsync, sync guest-page ---to--> kvm
sptes.

> -		mmu_page_add_parent_pte(vcpu, sp, parent_pte);

add "gfn" (actually its "struct kvm_mmu_page *sp" pointer) to
the parent.
>  		if (sp->unsync_children) {
>  			kvm_make_request(KVM_REQ_MMU_SYNC, vcpu);
>  			kvm_mmu_mark_parents_unsync(sp);

kvm_mmu_mark_parents_unsync relied on the links from current level all
the way to top level to mark all levels unsync, so that on guest entry,
KVM_REQ_MMU_SYNC is processed and any level is brought from guest -->
kvm pages. This now fails, because you removed "mmu_page_add_parent_pte"
(the link is not formed all the way to root).

Unless i am missing something, this is not correct.

> -		} else if (sp->unsync)
> +			if (parent_pte)
> +				mark_unsync(parent_pte);
> +		} else if (sp->unsync) {
>  			kvm_mmu_mark_parents_unsync(sp);
> +			if (parent_pte)
> +				mark_unsync(parent_pte);
> +		}
> +		mmu_page_add_parent_pte(vcpu, sp, parent_pte);
>  
>  		__clear_sp_write_flooding_count(sp);
>  		trace_kvm_mmu_get_page(sp, false);

--
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/