Received-SPF: pass (google.com: domain of linux-kernel-owner@vger.kernel.org designates 2620:137:e000::1:20 as permitted sender) client-ip=2620:137:e000::1:20;
Message-ID: <f6032e45-5d14-ae05-5383-f6ea1e8a4070@redhat.com>
Date:   Tue, 25 Oct 2022 12:01:29 +0200
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.2.1
Subject: Re: [PATCH 3/4] KVM: introduce memory transaction semaphore
Content-Language: en-US
To:     Emanuele Giuseppe Esposito <eesposit@redhat.com>,
        kvm@vger.kernel.org
Cc:     Jonathan Corbet <corbet@lwn.net>,
        Maxim Levitsky <mlevitsk@redhat.com>,
        Sean Christopherson <seanjc@google.com>,
        Thomas Gleixner <tglx@linutronix.de>,
        Ingo Molnar <mingo@redhat.com>, Borislav Petkov <bp@alien8.de>,
        Dave Hansen <dave.hansen@linux.intel.com>,
        David Hildenbrand <david@redhat.com>, x86@kernel.org,
        "H. Peter Anvin" <hpa@zytor.com>, linux-doc@vger.kernel.org,
        linux-kernel@vger.kernel.org
References: <20221022154819.1823133-1-eesposit@redhat.com>
 <20221022154819.1823133-4-eesposit@redhat.com>
 <62500f94-b95b-1e16-4aa2-f67905fab01a@redhat.com>
 <fba8e829-0d28-8f4d-a8ce-84d533009eb9@redhat.com>
From:   Paolo Bonzini <pbonzini@redhat.com>
In-Reply-To: <fba8e829-0d28-8f4d-a8ce-84d533009eb9@redhat.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 8bit
Precedence: bulk

On 10/24/22 14:57, Emanuele Giuseppe Esposito wrote:
> 
> 
> Am 23/10/2022 um 19:50 schrieb Paolo Bonzini:
>> On 10/22/22 17:48, Emanuele Giuseppe Esposito wrote:
>>> +static DECLARE_RWSEM(memory_transaction);
>>
>> This cannot be global, it must be per-struct kvm.  Otherwise one VM can
>> keep the rwsem indefinitely while a second VM hangs in
>> KVM_KICK_ALL_RUNNING_VCPUS.
>>
>> It can also be changed to an SRCU (with the down_write+up_write sequence
>> changed to synchronize_srcu_expedited) which has similar characteristics
>> to your use of the rwsem.
>>
> 
> Makes sense, but why synchronize_srcu_expedited and not synchronize_srcu?

Because (thanks to the kick) you expect the grace period to end almost 
immediately, and synchronize_srcu() will slow down sensibly the changes 
to the memory map.

Paolo