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;
Subject: Re: [PATCH v4 2/2] s390/vfio-ap: control access to PQAP(AQIC)
 interception handler
To:     Jason Gunthorpe <jgg@nvidia.com>
Cc:     jjherne@linux.ibm.com, linux-s390@vger.kernel.org,
        linux-kernel@vger.kernel.org, borntraeger@de.ibm.com,
        cohuck@redhat.com, pasic@linux.vnet.ibm.com,
        alex.williamson@redhat.com, kwankhede@nvidia.com,
        frankja@linux.ibm.com, david@redhat.com, imbrenda@linux.ibm.com,
        hca@linux.ibm.com
References: <20210521193648.940864-1-akrowiak@linux.ibm.com>
 <20210521193648.940864-3-akrowiak@linux.ibm.com>
 <5d15fdf2-aee8-4e6c-c3e1-f07c76ce5974@linux.ibm.com>
 <e2bed0a6-f5e2-0a69-22b9-1b304cbe1362@linux.ibm.com>
 <20210525131912.GW1002214@nvidia.com>
From:   Tony Krowiak <akrowiak@linux.ibm.com>
Message-ID: <6542986f-b20e-3f41-b96c-70f0ce42af2d@linux.ibm.com>
Date:   Tue, 25 May 2021 11:08:22 -0400
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:78.0) Gecko/20100101
 Thunderbird/78.8.1
MIME-Version: 1.0
In-Reply-To: <20210525131912.GW1002214@nvidia.com>
Content-Type: text/plain; charset=utf-8; format=flowed
Content-Transfer-Encoding: 8bit
Content-Language: en-US
Precedence: bulk


On 5/25/21 9:19 AM, Jason Gunthorpe wrote:
> On Tue, May 25, 2021 at 09:16:30AM -0400, Tony Krowiak wrote:
>>
>> On 5/24/21 10:37 AM, Jason J. Herne wrote:
>>> On 5/21/21 3:36 PM, Tony Krowiak wrote:
>>>> The function pointer to the handler that processes interception of the
>>>> PQAP instruction is contained in the mdev. If the mdev is removed and
>>>> its storage de-allocated during the processing of the PQAP instruction,
>>>> the function pointer could get wiped out before the function is called
>>>> because there is currently nothing that controls access to it.
>>>>
>>>> This patch introduces two new functions:
>>>> * The kvm_arch_crypto_register_hook() function registers a function
>>>> pointer
>>>>     for processing intercepted crypto instructions.
>>>> * The kvm_arch_crypto_register_hook() function un-registers a function
>>>>     pointer that was previously registered.
>>> Typo: You meant kvm_arch_crypto_UNregister_hook() in the second bullet.
>>>
>>>
>>> Just one overall observation on this one. The whole hook system seems
>>> kind of over-engineered if this is our only use for it. It looks like a
>>> kvm_s390_crypto_hook is meant to link a specific module with a function
>>> pointer. Do we really need this concept?
>>>
>>> I think a simpler design could be to just place a mutex and a function
>>> pointer in the kvm_s390_crypto struct. Then you can grab the mutex in
>>> vfio_ap_ops.c when registering/unregistering. You would also grab the
>>> mutex in priv.c when calling the function pointer. What I am suggesting
>>> is essentially the exact same scheme you have implemented here, but
>>> simpler and with less infrastructure.
>> That would be great, however; when I implemented something similar, it
>> resulted in a
>> lockdep splat between the lock used to protect the hook and the
>> matrix_dev->lock used to
>> protect updates to matrix_mdev (including the freeing thereof). After
>> pulling what little hair
>> I have left out, this seemed like a reasonable solution, over-engineered
>> though it may be.
>> If somebody has a simpler solution, I'm all ears.
> Why can't you put the locks in the right order? It looked trivial, I'm confused.

Because the handle_pqap() function in priv.c does not have access to the
matrix_dev lock.

>
> Jason