Re: [PATCH v7 05/15] s390/vfio-ap: Callback to get/set vfio device mig state during guest migration

[Anthony Krowiak <akrowiak@linux.ibm.com>]

On 8/12/26 12:04 PM, Jason J. Herne wrote:
>
>
> On 8/7/26 6:18 PM, Anthony Krowiak wrote:
>> Implements two callbacks related to the current state of a vfio-ap 
>> device:
>> * The callback that sets a new migration state of a vfio-ap device 
>> during
>>    live migration of guests with pass-through access to AP devices. This
>>    callback is mandatory for VFIO_DEVICE_FEATURE_MIGRATION support.
>>
>>    The function pointer for this callback is specified via the
>>    'migration_set_state' field of the 'vfio_migration_ops' structure
>>    which is stored with the VFIO device when the 'vfio_device'
>>    structure representing the mediated device is initialized.
>>
>> * The callback that returns the current vfio device migration state 
>> during
>>    live migration of guests with pass-through access to AP devices.
>>
>>    The function pointer for this callback is specified via the
>>    'migration_get_state' field of the 'vfio_migration_ops' structure
>>    which is stored with the VFIO device when the 'vfio_device'
>>    structure representing the mediated device is initialized.
>>
>> Signed-off-by: Anthony Krowiak <akrowiak@linux.ibm.com>
>> ---
>>   drivers/s390/crypto/vfio_ap_migration.c | 130 +++++++++++++++++++++++-
>>   1 file changed, 128 insertions(+), 2 deletions(-)
>>
>> diff --git a/drivers/s390/crypto/vfio_ap_migration.c 
>> b/drivers/s390/crypto/vfio_ap_migration.c
>> index 1613813124b6..af740b177c97 100644
>> --- a/drivers/s390/crypto/vfio_ap_migration.c
>> +++ b/drivers/s390/crypto/vfio_ap_migration.c
>> @@ -128,16 +128,141 @@ static void 
>> vfio_ap_release_resuming_file(struct vfio_ap_migration_data *mig_dat
>>        */
>>   }
>>   +static struct file *
>> +vfio_ap_transition_to_state(struct ap_matrix_mdev *matrix_mdev,
>> +                enum vfio_device_mig_state new_state)
>> +{
>> +    struct vfio_ap_migration_data *mig_data;
>> +    enum vfio_device_mig_state cur_state;
>> +
>> +    lockdep_assert_held(&matrix_dev->mdevs_lock);
>> +    mig_data = matrix_mdev->mig_data;
>> +    cur_state = mig_data->mig_state;
>> +    dev_dbg(matrix_mdev->vdev.dev, "%s: %d -> %d\n", __func__, 
>> cur_state,
>> +        new_state);
>> +
>> +    if (cur_state == VFIO_DEVICE_STATE_STOP &&
>> +        new_state == VFIO_DEVICE_STATE_STOP_COPY) {
>> +        /* TODO */
>> +        return ERR_PTR(-EOPNOTSUPP);
>> +    }
>> +
>> +    if (cur_state == VFIO_DEVICE_STATE_STOP &&
>> +        new_state == VFIO_DEVICE_STATE_RESUMING) {
>> +        /* TODO */
>> +        return ERR_PTR(-EOPNOTSUPP);
>> +    }
>> +
>> +    if ((cur_state == VFIO_DEVICE_STATE_RESUMING &&
>> +         new_state == VFIO_DEVICE_STATE_STOP) ||
>> +        (cur_state == VFIO_DEVICE_STATE_STOP_COPY &&
>> +         new_state == VFIO_DEVICE_STATE_STOP)) {
>> +        /* TODO */
>> +        return ERR_PTR(-EOPNOTSUPP);
>> +    }
>> +
>> +    if ((cur_state == VFIO_DEVICE_STATE_STOP &&
>> +         new_state == VFIO_DEVICE_STATE_RUNNING) ||
>> +        (cur_state == VFIO_DEVICE_STATE_RUNNING &&
>> +         new_state == VFIO_DEVICE_STATE_STOP)) {
>> +        /* TODO */
>> +        return ERR_PTR(-EOPNOTSUPP);
>> +    }
>> +
>> +    /* vfio_mig_get_next_state() does not use arcs other than the 
>> above */
>> +    WARN_ON(true);
>> +
>> +    return ERR_PTR(-EINVAL);
>> +}
>> +
>>   static struct file *vfio_ap_set_state(struct vfio_device *vdev,
>>                         enum vfio_device_mig_state  new_state)
>>   {
>> -    return NULL;
>> +    int ret;
>> +    struct file *filp = NULL;
>> +    struct ap_matrix_mdev *matrix_mdev;
>> +    enum vfio_device_mig_state next_state;
>> +    struct vfio_ap_migration_data *mig_data;
>> +
>> +    matrix_mdev = container_of(vdev, struct ap_matrix_mdev, vdev);
>
> The container_of() pointer swizzle is done under lock in 
> vfio_ap_get_state, but not here. Why? They should probably be handled 
> consistently.

The vdev pointer is stable. It is passed in as a function parameter from the
VFIO core and is guaranteed valid for the callback duration. On the other
hand, in the interest of consistency - as you explained - I'll move it after
taking the mdevs_lock.

>
>> +
>> +    mutex_lock(&matrix_dev->mdevs_lock);
>> +    if (ap_is_se_guest()) {
>> +        dev_err_once(matrix_mdev->vdev.dev,
>> +                 "Migration not allowed from or to a Secure 
>> Execution guest\n");
>> +        mutex_unlock(&matrix_dev->mdevs_lock);
>> +        return ERR_PTR(-EPERM);
>> +    }
>> +
>> +    mig_data = matrix_mdev->mig_data;
>> +
>> +    /*
>> +     * The mig_data pointer is set in the vfio_ap_init_migration_data
>> +     * function which is called when the vfio-ap device fd is opened.
>> +     * Since the implicit pre-open state is RUNNING, a request to set
>> +     * RUNNING is a no-op. Any other state transition is invalid before
>> +     * open_device.
>> +     */
>> +    if (!mig_data) {
>> +        mutex_unlock(&matrix_dev->mdevs_lock);
>> +        if (new_state == VFIO_DEVICE_STATE_RUNNING)
>> +            return NULL;
>> +        return ERR_PTR(-ENODEV);
>> +    }
>> +
>> +    dev_dbg(vdev->dev, "%s -> %d\n", __func__, new_state);
>> +
>> +    while (mig_data->mig_state != VFIO_DEVICE_STATE_ERROR &&
>> +           mig_data->mig_state != new_state) {
>> +        ret = vfio_mig_get_next_state(vdev, mig_data->mig_state,
>> +                          new_state, &next_state);
>> +        if (ret) {
>> +            filp = ERR_PTR(ret);
>> +            break;
>> +        }
>> +
>> +        filp = vfio_ap_transition_to_state(matrix_mdev, next_state);
>> +        if (IS_ERR(filp))
>> +            break;
>> +
>> +        mig_data->mig_state = next_state;
>> +
>> +        if (WARN_ON(filp && new_state != next_state)) {
>> +            fput(filp);
>> +            filp = ERR_PTR(-EINVAL);
>> +            break;
>> +        }
>> +    }
>> +
>> +    mutex_unlock(&matrix_dev->mdevs_lock);
>> +
>> +    return filp;
>>   }
>>     static int vfio_ap_get_state(struct vfio_device *vdev,
>>                    enum vfio_device_mig_state  *current_state)
>>   {
>> -    return -EOPNOTSUPP;
>> +    struct ap_matrix_mdev *matrix_mdev;
>> +    struct vfio_ap_migration_data *mig_data;
>> +
>> +    mutex_lock(&matrix_dev->mdevs_lock);
>> +
>> +    matrix_mdev = container_of(vdev, struct ap_matrix_mdev, vdev);
>> +    mig_data =  matrix_mdev->mig_data;
>
> Why bother with temp variable? You could just access 
> matrix_mdev->mig_data directly.

No other reason than for readability and shortening lines of code that 
access it multiple
times. I see no good reason to remove it.

>
>> +
>> +    /*
>> +     * The mig_data pointer is set in the vfio_ap_init_migration_data
>> +     * function which is called when the vfio-ap device fd is opened.
>> +     * If mig_data is NULL, report RUNNING as the implicit pre-open 
>> state
>> +     * so userspace doesn't need to perform any state transition 
>> before the
>> +     * device becomes active.
>> +     */
>> +    *current_state = (mig_data) ? mig_data->mig_state :
>> +                      VFIO_DEVICE_STATE_RUNNING;
>> +
>> +    mutex_unlock(&matrix_dev->mdevs_lock);
>> +
>> +    return 0;
>>   }
>>     static int vfio_ap_get_data_size(struct vfio_device *vdev,
>> @@ -159,6 +284,7 @@ static const struct vfio_migration_ops 
>> vfio_ap_migration_ops = {
>>    */
>>   void vfio_ap_init_migration_capabilities(struct ap_matrix_mdev 
>> *matrix_mdev)
>>   {
>> +    /* Live guest migration is not supported for SE guests */
>>       if (ap_is_se_guest())
>>           return;
>
> Comment is unrelated to the rest of the patch.

This belongs in patch 03/15. I'll move it.