Message-ID: <b0ed98f6-3586-3151-47e7-9ec86d6c716d@linux.ibm.com>
Date:   Tue, 31 Jan 2023 09:46:18 -0500
MIME-Version: 1.0
User-Agent: Mozilla/5.0 (X11; Linux x86_64; rv:102.0) Gecko/20100101
 Thunderbird/102.4.0
Subject: Re: [PATCH] kvm/vfio: Fix potential deadlock on vfio group_lock
Content-Language: en-US
To:     Jason Gunthorpe <jgg@nvidia.com>
Cc:     Yi Liu <yi.l.liu@intel.com>, alex.williamson@redhat.com,
        pbonzini@redhat.com, mjrosato@linux.ibm.com, kevin.tian@intel.com,
        cohuck@redhat.com, farman@linux.ibm.com, pmorel@linux.ibm.com,
        borntraeger@linux.ibm.com, frankja@linux.ibm.com,
        imbrenda@linux.ibm.com, david@redhat.com, jjherne@linux.ibm.com,
        pasic@linux.ibm.com, zhenyuw@linux.intel.com, zhi.a.wang@intel.com,
        seanjc@google.com, linux-s390@vger.kernel.org, kvm@vger.kernel.org,
        intel-gvt-dev@lists.freedesktop.org,
        intel-gfx@lists.freedesktop.org, linux-kernel@vger.kernel.org
References: <20230120150528.471752-1-yi.l.liu@intel.com>
 <67ec09bf-cb24-34e3-6ec4-1ae87b0738bd@linux.ibm.com>
 <Y9km+xezgD4ovjDX@nvidia.com>
From:   Anthony Krowiak <akrowiak@linux.ibm.com>
In-Reply-To: <Y9km+xezgD4ovjDX@nvidia.com>
Content-Type: text/plain; charset=UTF-8; format=flowed
Content-Transfer-Encoding: 7bit
Precedence: bulk


On 1/31/23 9:34 AM, Jason Gunthorpe wrote:
> On Tue, Jan 31, 2023 at 09:27:54AM -0500, Anthony Krowiak wrote:
>> I encountered a lockdep splat while running some regression tests today (see
>> below). I suspected it might be this patch so I reverted it, rebuilt the
>> kernel and ran the regression tests again; this time, the test ran cleanly.
>> It looks like this patch may not have fixed the problem for which it was
>> intended. Here is the relevant dmesg output:
> Well, it fixes the deadlock it intended to fix and created another one
> :)
>
> It means device drivers cannot obtain the kvm lock from their open
> functions in this new model.
>
> Why does ap need to touch kvm->lock? (via get_update_locks_for_kvm)


We need the kvm->lock because we take the vCPUs out of SIE in order to 
dynamically change values in the APCB.


>
> Maybe you should split that lock and have a dedicated apcb lock?


I don't think that would suffice for taking the vCPUs out of SIE.


>
> Jason