Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 23.128.96.18 as permitted sender) client-ip=23.128.96.18;
Message-ID: <43fd49ff-d54d-558f-f5c8-6dcc5dc726b2@redhat.com>
Date:   Fri, 22 Oct 2021 15:33:49 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.1.0
Subject: Re: [PATCH 0/8] KVM: SEV-ES: fixes for string I/O emulation
Content-Language: en-US
To:     Sean Christopherson <seanjc@google.com>
Cc:     linux-kernel@vger.kernel.org, kvm@vger.kernel.org,
        fwilhelm@google.com, oupton@google.com
References: <20211013165616.19846-1-pbonzini@redhat.com>
 <YXH8hmB64gnwxIx6@google.com>
From:   Paolo Bonzini <pbonzini@redhat.com>
In-Reply-To: <YXH8hmB64gnwxIx6@google.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 22/10/21 01:49, Sean Christopherson wrote:
> On Wed, Oct 13, 2021, Paolo Bonzini wrote:
>> Patches 2 to 7 are a bunch of cleanups to emulator_pio_in and
>> emulator_pio_in_out, so that the final SEV code is a little easier
>> to reason on.  Just a little, no big promises.
> IMO, this series goes in the wrong direction and doesn't make the mess any better,
> just different.
> 
> The underlying issue is that kernel_pio() does the completely horrendous thing
> of consuming vcpu->arch.pio.  That leads to the juggling that this series tries
> to clean up, but it's essentially an impossible problem to solve because the
> approach itself is broken.

I agree on this, but I disagree that the series does not make the mess 
any better.  At the very least, the new signatures for 
__emulator_pio_in, complete_emulator_pio_in and emulator_pio_in_out are 
improvements regarding the _role_ of vcpu->arch.pio*:

- complete_emulator_pio_in clearly takes the values from vcpu->arch.pio, 
which _is_ the right thing to do for a complete_userspace_io function. 
This is not clear of emulator_pio_in before the patch

- __emulator_pio_in and emulator_pio_in_out do not take anymore the 
buffer argument, making it clear that they operate on the internal 
pio_data buffer and only complete_emulator_pio_in copies out of it. 
Which yes is horrible, but at least it is clearly visible in the code now.

I managed to clean things up quite satisfactorily with just 6 patches on 
top of these eight, so I'll post the full series as soon as I finish 
testing them.  5.15 can then include these to fix the bug at hand.

Paolo