Received-SPF: pass (google.com: best guess record for domain of linux-kernel-owner@vger.kernel.org designates 209.132.180.67 as permitted sender) client-ip=209.132.180.67;
Subject: Re: [PATCH RESEND v4 7/9] KVM: VMX: Handle SPP induced vmexit and
 page fault
From:   Paolo Bonzini <pbonzini@redhat.com>
To:     Yang Weijiang <weijiang.yang@intel.com>, kvm@vger.kernel.org,
        linux-kernel@vger.kernel.org, sean.j.christopherson@intel.com
Cc:     mst@redhat.com, rkrcmar@redhat.com, jmattson@google.com,
        yu.c.zhang@intel.com, alazar@bitdefender.com
References: <20190814070403.6588-1-weijiang.yang@intel.com>
 <20190814070403.6588-8-weijiang.yang@intel.com>
 <5f6ba406-17c4-a552-2352-2ff50569aac0@redhat.com>
Message-ID: <fb6cd8b4-eee9-6e58-4047-550811bffd58@redhat.com>
Date:   Mon, 19 Aug 2019 17:04:23 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:60.0) Gecko/20100101
 Thunderbird/60.8.0
MIME-Version: 1.0
In-Reply-To: <5f6ba406-17c4-a552-2352-2ff50569aac0@redhat.com>
Content-Type: text/plain; charset=utf-8
Content-Language: en-US
Content-Transfer-Encoding: 7bit
Sender: linux-kernel-owner@vger.kernel.org
Precedence: bulk

On 19/08/19 16:43, Paolo Bonzini wrote:
>> +			/*
>> +			 * Record write protect fault caused by
>> +			 * Sub-page Protection, let VMI decide
>> +			 * the next step.
>> +			 */
>> +			if (spte & PT_SPP_MASK) {
> Should this be "if (spte & PT_WRITABLE_MASK)" instead?  That is, if the
> page is already writable, the fault must be an SPP fault.

Hmm, no I forgot how SPP works; still, this is *not* correct.  For
example, if SPP marks part of a page as read-write, but KVM wants to
write-protect the whole page for access or dirty tracking, that should
not cause an SPP exit.

So I think that when KVM wants to write-protect the whole page
(wrprot_ad_disabled_spte) it must also clear PT_SPP_MASK; for example it
could save it in bit 53 (PT64_SECOND_AVAIL_BITS_SHIFT + 1).  If the
saved bit is set, fast_page_fault must then set PT_SPP_MASK instead of
PT_WRITABLE_MASK.  On re-entry this will cause an SPP vmexit;
fast_page_fault should never trigger an SPP userspace exit on its own,
all the SPP handling should go through handle_spp.

Paolo