[<prev] [next>] [<thread-prev] [thread-next>] [day] [month] [year] [list]
Message-ID: <20210525151117.GB1002214@nvidia.com>
Date:   Tue, 25 May 2021 12:11:17 -0300
From:   Jason Gunthorpe <jgg@...dia.com>
To:     Tony Krowiak <akrowiak@...ux.ibm.com>
Cc:     jjherne@...ux.ibm.com, linux-s390@...r.kernel.org,
        linux-kernel@...r.kernel.org, borntraeger@...ibm.com,
        cohuck@...hat.com, pasic@...ux.vnet.ibm.com,
        alex.williamson@...hat.com, kwankhede@...dia.com,
        frankja@...ux.ibm.com, david@...hat.com, imbrenda@...ux.ibm.com,
        hca@...ux.ibm.com
Subject: Re: [PATCH v4 2/2] s390/vfio-ap: control access to PQAP(AQIC)
 interception handler
On Tue, May 25, 2021 at 11:08:22AM -0400, Tony Krowiak wrote:
> > 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.
Based on the sketch  made the handle_pqap() should only handle the
arch.crypto.rwsem.
When it calls the hook it gets the matrix dev
This sets the lock order as always: rwsem then matrix_dev
Of the other two places:
@@ -352,8 +352,7 @@ static int vfio_ap_mdev_create(struct mdev_device *mdev)
+       down_write(&&vcpu->kvm->arch.crypto.rwsem);
        mutex_lock(&matrix_dev->lock);
Obviously correct
@@ -1202,7 +1203,9 @@ static void vfio_ap_mdev_unset_kvm(struct ap_matrix_mdev *matrix_mdev)
                mutex_lock(&matrix_dev->lock);
                vfio_ap_mdev_reset_queues(matrix_mdev->mdev);
+               down_write(&matrix_mdev->kvm->arch.crypto.rwsem);
                matrix_mdev->kvm->arch.crypto.pqap_hook = NULL;
+               up_write(&matrix_mdev->kvm->arch.crypto.rwsem);
This is inverted
Just move the down_write up two lines
What is missing?
Jason
Powered by blists - more mailing lists