Subject: Re: [PART1 RFC v4 09/11] svm: Do not expose x2APIC when enable AVIC
To: =?UTF-8?B?UmFkaW0gS3LEjW3DocWZ?= <rkrcmar@redhat.com>,
        Suravee Suthikulpanit <Suravee.Suthikulpanit@amd.com>
References: <1460017232-17429-1-git-send-email-Suravee.Suthikulpanit@amd.com>
 <1460017232-17429-10-git-send-email-Suravee.Suthikulpanit@amd.com>
 <20160411205423.GA29328@potion.brq.redhat.com>
Cc: joro@8bytes.org, bp@alien8.de, gleb@kernel.org, alex.williamson@redhat.com,
        kvm@vger.kernel.org, linux-kernel@vger.kernel.org, wei@redhat.com,
        sherry.hurwitz@amd.com
From: Paolo Bonzini <pbonzini@redhat.com>
Message-ID: <570D720C.8020603@redhat.com>
Date: Wed, 13 Apr 2016 00:09:16 +0200
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:38.0) Gecko/20100101
 Thunderbird/38.7.1
MIME-Version: 1.0
In-Reply-To: <20160411205423.GA29328@potion.brq.redhat.com>
Content-Type: text/plain; charset=UTF-8
Content-Transfer-Encoding: 8bit
Sender: linux-kernel-owner@vger.kernel.org
Content-Length: 976
Lines: 31


On 11/04/2016 22:54, Radim Krčmář wrote:
>> >  
>> >  static void svm_set_supported_cpuid(u32 func, struct kvm_cpuid_entry2 *entry)
>> >  {
>> >  	switch (func) {
>> > +	case 0x00000001:
> ("case 1:" or "case 0x1:" would be easier to read.)
> 
>> > +		if (avic)
>> > +			entry->ecx &= ~bit(X86_FEATURE_X2APIC);
>> > +		break;
> 
> ---
> A rant for the unlikely case I get back to fix the broader situation:
> Only one of these two additions is needed.  If we do the second one,
> then userspace should not set X2APIC, therefore the first one is
> useless.
> 
> Omitting the second one allows userspace to clear apicv_active and set
> X86_FEATURE_X2APIC, but it needs a non-intuitive order of ioctls, so I
> think we should have the second one.
> 
> The problem is that KVM doesn't seems to check whether userspace sets
> cpuid that is a subset of supported ones, so omitting the first one
> needlessly expands the space for potential failures.

Yes, we need both.

Paolo