Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:18 as permitted sender) client-ip=2620:137:e000::1:18;
Message-ID: <36978590-c714-d0ba-6e4e-ab0016136381@redhat.com>
Date:   Fri, 22 Apr 2022 12:19:51 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:91.0) Gecko/20100101
 Thunderbird/91.8.0
Content-Language: en-US
To:     Marc Zyngier <maz@kernel.org>
Cc:     Oliver Upton <oupton@google.com>, linux-kernel@vger.kernel.org,
        kvm@vger.kernel.org, will@kernel.org, apatel@ventanamicro.com,
        atishp@rivosinc.com, seanjc@google.com, pgonda@google.com
References: <20220421180443.1465634-1-pbonzini@redhat.com>
 <YmJgIQe+5zGbrxoF@google.com>
 <ef5c6c5b-2ed1-7d4c-e757-ed8bcead5d18@redhat.com>
 <87a6cda13p.wl-maz@kernel.org>
From:   Paolo Bonzini <pbonzini@redhat.com>
Subject: Re: [PATCH 0/4] KVM: fix KVM_EXIT_SYSTEM_EVENT mess
In-Reply-To: <87a6cda13p.wl-maz@kernel.org>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk

On 4/22/22 12:01, Marc Zyngier wrote:
>> These patches are against kvm/next, so that is already what I did. :)
> 
> Can you please post a complete series? It is becoming really hard to
> track what you are doing.

Yes, I'll post a complete series for kvm/master in a second.

>> On the other hand right now the ARM and RISC-V flags are unusable with
>> 32-bit userspace, so we need to fix _something_ in 5.18 as well.
> 
> What 32bit userspace? arm64 doesn't have any that can interact with KVM,
> so I don't see anything to fix on that front.

You're right, ARM64 does not define KVM_COMPAT.  But RISC-V does.

>> For your proposal, all that's missing is a 5.18 patch to add the
>> padding. But since the flags UAPI was completely unused before 5.18
>> and there's no reason to inflict the different naming of fields to
>> userspace.  So I think we want to apply this UAPI change in 5.18 too.
> 
> As it was pointed out already, CrosVM has already started looking at
> the flags. The fact that it was always 0 until now doesn't make it
> less of a UAPI.

I heard that and that's exactly why I dropped the idea of using 
NDATA_VALID in bit 31 of flags, and switched to a capability instead. 
If it is desirable for Android, crosvm can "quirk" that ARM always has 
valid data[0].

Paolo

> I'd like to see a full series that implements the transition before we
> make a decision on this.