Message-ID: <76600ed5-60cc-c530-56db-43f7026d8c8e@linux.ibm.com>
Date:   Wed, 30 Sep 2020 08:59:36 -0400
From:   Tony Krowiak <akrowiak@...ux.ibm.com>
To:     Halil Pasic <pasic@...ux.ibm.com>
Cc:     linux-s390@...r.kernel.org, linux-kernel@...r.kernel.org,
        kvm@...r.kernel.org, freude@...ux.ibm.com, borntraeger@...ibm.com,
        cohuck@...hat.com, mjrosato@...ux.ibm.com,
        alex.williamson@...hat.com, kwankhede@...dia.com,
        fiuczy@...ux.ibm.com, frankja@...ux.ibm.com, david@...hat.com,
        imbrenda@...ux.ibm.com, hca@...ux.ibm.com, gor@...ux.ibm.com
Subject: Re: [PATCH v10 09/16] s390/vfio-ap: allow assignment of unavailable
 AP queues to mdev device



On 9/26/20 7:49 PM, Halil Pasic wrote:
> On Fri, 21 Aug 2020 15:56:09 -0400
> Tony Krowiak <akrowiak@...ux.ibm.com> wrote:
>
>> The current implementation does not allow assignment of an AP adapter or
>> domain to an mdev device if the APQNs resulting from the assignment
>> do not reference AP queue devices that are bound to the vfio_ap device
>> driver. This patch allows assignment of AP resources to the matrix mdev as
>> long as the APQNs resulting from the assignment:
>>     1. Are not reserved by the AP BUS for use by the zcrypt device drivers.
>>     2. Are not assigned to another matrix mdev.
>>
>> The rationale behind this is twofold:
>>     1. The AP architecture does not preclude assignment of APQNs to an AP
>>        configuration that are not available to the system.
>>     2. APQNs that do not reference a queue device bound to the vfio_ap
>>        device driver will not be assigned to the guest's CRYCB, so the
>>        guest will not get access to queues not bound to the vfio_ap driver.
>>
>> Signed-off-by: Tony Krowiak <akrowiak@...ux.ibm.com>
>> ---
>>   drivers/s390/crypto/vfio_ap_ops.c | 212 +++++-------------------------
>>   1 file changed, 35 insertions(+), 177 deletions(-)
>>
>> diff --git a/drivers/s390/crypto/vfio_ap_ops.c b/drivers/s390/crypto/vfio_ap_ops.c
>> index eaf4e9eab6cb..24fd47e43b80 100644
>> --- a/drivers/s390/crypto/vfio_ap_ops.c
>> +++ b/drivers/s390/crypto/vfio_ap_ops.c
>> @@ -1,4 +1,3 @@
>> -// SPDX-License-Identifier: GPL-2.0+
> Probably not intentional, or?

Definitely not intentional. I'll restore it.

>
>>   /*
>>    * Adjunct processor matrix VFIO device driver callbacks.
>>    *
>> @@ -420,122 +419,6 @@ static struct attribute_group *vfio_ap_mdev_type_groups[] = {
>>   	NULL,
>>   };
>>   
>> -struct vfio_ap_queue_reserved {
>> -	unsigned long *apid;
>> -	unsigned long *apqi;
>> -	bool reserved;
>> -};
>> -
>> -/**
>> - * vfio_ap_has_queue
>> - *
>> - * @dev: an AP queue device
>> - * @data: a struct vfio_ap_queue_reserved reference
>> - *
>> - * Flags whether the AP queue device (@dev) has a queue ID containing the APQN,
>> - * apid or apqi specified in @data:
>> - *
>> - * - If @data contains both an apid and apqi value, then @data will be flagged
>> - *   as reserved if the APID and APQI fields for the AP queue device matches
>> - *
>> - * - If @data contains only an apid value, @data will be flagged as
>> - *   reserved if the APID field in the AP queue device matches
>> - *
>> - * - If @data contains only an apqi value, @data will be flagged as
>> - *   reserved if the APQI field in the AP queue device matches
>> - *
>> - * Returns 0 to indicate the input to function succeeded. Returns -EINVAL if
>> - * @data does not contain either an apid or apqi.
>> - */
>> -static int vfio_ap_has_queue(struct device *dev, void *data)
>> -{
>> -	struct vfio_ap_queue_reserved *qres = data;
>> -	struct ap_queue *ap_queue = to_ap_queue(dev);
>> -	ap_qid_t qid;
>> -	unsigned long id;
>> -
>> -	if (qres->apid && qres->apqi) {
>> -		qid = AP_MKQID(*qres->apid, *qres->apqi);
>> -		if (qid == ap_queue->qid)
>> -			qres->reserved = true;
>> -	} else if (qres->apid && !qres->apqi) {
>> -		id = AP_QID_CARD(ap_queue->qid);
>> -		if (id == *qres->apid)
>> -			qres->reserved = true;
>> -	} else if (!qres->apid && qres->apqi) {
>> -		id = AP_QID_QUEUE(ap_queue->qid);
>> -		if (id == *qres->apqi)
>> -			qres->reserved = true;
>> -	} else {
>> -		return -EINVAL;
>> -	}
>> -
>> -	return 0;
>> -}
>> -
>> -/**
>> - * vfio_ap_verify_queue_reserved
>> - *
>> - * @matrix_dev: a mediated matrix device
>> - * @apid: an AP adapter ID
>> - * @apqi: an AP queue index
>> - *
>> - * Verifies that the AP queue with @apid/@...i is reserved by the VFIO AP device
>> - * driver according to the following rules:
>> - *
>> - * - If both @apid and @apqi are not NULL, then there must be an AP queue
>> - *   device bound to the vfio_ap driver with the APQN identified by @apid and
>> - *   @apqi
>> - *
>> - * - If only @apid is not NULL, then there must be an AP queue device bound
>> - *   to the vfio_ap driver with an APQN containing @apid
>> - *
>> - * - If only @apqi is not NULL, then there must be an AP queue device bound
>> - *   to the vfio_ap driver with an APQN containing @apqi
>> - *
>> - * Returns 0 if the AP queue is reserved; otherwise, returns -EADDRNOTAVAIL.
>> - */
>> -static int vfio_ap_verify_queue_reserved(unsigned long *apid,
>> -					 unsigned long *apqi)
>> -{
>> -	int ret;
>> -	struct vfio_ap_queue_reserved qres;
>> -
>> -	qres.apid = apid;
>> -	qres.apqi = apqi;
>> -	qres.reserved = false;
>> -
>> -	ret = driver_for_each_device(&matrix_dev->vfio_ap_drv->driver, NULL,
>> -				     &qres, vfio_ap_has_queue);
>> -	if (ret)
>> -		return ret;
>> -
>> -	if (qres.reserved)
>> -		return 0;
>> -
>> -	return -EADDRNOTAVAIL;
>> -}
>> -
>> -static int
>> -vfio_ap_mdev_verify_queues_reserved_for_apid(struct ap_matrix_mdev *matrix_mdev,
>> -					     unsigned long apid)
>> -{
>> -	int ret;
>> -	unsigned long apqi;
>> -	unsigned long nbits = matrix_mdev->matrix.aqm_max + 1;
>> -
>> -	if (find_first_bit_inv(matrix_mdev->matrix.aqm, nbits) >= nbits)
>> -		return vfio_ap_verify_queue_reserved(&apid, NULL);
>> -
>> -	for_each_set_bit_inv(apqi, matrix_mdev->matrix.aqm, nbits) {
>> -		ret = vfio_ap_verify_queue_reserved(&apid, &apqi);
>> -		if (ret)
>> -			return ret;
>> -	}
>> -
>> -	return 0;
>> -}
>> -
>>   #define MDEV_SHARING_ERR "Userspace may not re-assign queue %02lx.%04lx " \
>>   			 "already assigned to %s"
>>   
>> @@ -572,6 +455,11 @@ static int vfio_ap_mdev_verify_no_sharing(struct ap_matrix_mdev *matrix_mdev,
>>   	DECLARE_BITMAP(aqm, AP_DOMAINS);
>>   
>>   	list_for_each_entry(lstdev, &matrix_dev->mdev_list, node) {
>> +		/*
>> +		 * If either of the input masks belongs to the mdev to which an
>> +		 * AP resource is being assigned, then we don't need to verify
>> +		 * that mdev's masks.
>> +		 */
>>   		if (matrix_mdev == lstdev)
>>   			continue;
>>   
> Seems unrelated.

What seems unrelated? The matrix_mdev passed in is the mdev to which 
assignment is
being made. This function is verifying that no APQN assigned to the 
matrix_mdev is
assigned to any other mdev. Since we are looping through all mdevs here, 
we are
skipping the verification if the current mdev being examined is the same 
as the matrix_mdev
to which the assignment is being made. Maybe I'm not understanding your 
point here.

>
>> @@ -597,6 +485,20 @@ static int vfio_ap_mdev_verify_no_sharing(struct ap_matrix_mdev *matrix_mdev,
>>   	return 0;
>>   }
>>   
>> +static int vfio_ap_mdev_validate_masks(struct ap_matrix_mdev *matrix_mdev,
>> +				       unsigned long *mdev_apm,
>> +				       unsigned long *mdev_aqm)
>> +{
>> +	DECLARE_BITMAP(apm, AP_DEVICES);
>> +	DECLARE_BITMAP(aqm, AP_DOMAINS);
>> +
>> +	if (bitmap_and(apm, mdev_apm, ap_perms.apm, AP_DEVICES) &&
>> +	    bitmap_and(aqm, mdev_aqm, ap_perms.aqm, AP_DOMAINS))
> Isn't ap_perms supposed to be protected by ap_perms_mutex? In theory
> you could end up with a torn write (catch the a[pq]mask_commit() with
> its pants down, in a sense that only a part of the memcpy was done (and
> became observable on the other CPU doing this validate).

Good catch. I should probably use the 
ap_apqn_in_matrix_owned_by_def_drv(apm, aqm)
function in ap_bus.c.

>
>> +		return -EADDRNOTAVAIL;
>> +
>> +	return vfio_ap_mdev_verify_no_sharing(matrix_mdev, mdev_apm, mdev_aqm);
>> +}
>> +
>>   /**
>>    * vfio_ap_mdev_filter_matrix
>>    *
>> @@ -882,33 +784,21 @@ static ssize_t assign_adapter_store(struct device *dev,
>>   	if (apid > matrix_mdev->matrix.apm_max)
>>   		return -ENODEV;
>>   
>> -	/*
>> -	 * Set the bit in the AP mask (APM) corresponding to the AP adapter
>> -	 * number (APID). The bits in the mask, from most significant to least
>> -	 * significant bit, correspond to APIDs 0-255.
>> -	 */
>> -	mutex_lock(&matrix_dev->lock);
>> -
>> -	ret = vfio_ap_mdev_verify_queues_reserved_for_apid(matrix_mdev, apid);
>> -	if (ret)
>> -		goto done;
>> -
>>   	memset(apm, 0, sizeof(apm));
>>   	set_bit_inv(apid, apm);
>>   
>> -	ret = vfio_ap_mdev_verify_no_sharing(matrix_mdev, apm,
>> -					     matrix_mdev->matrix.aqm);
>> -	if (ret)
>> -		goto done;
>> -
>> +	mutex_lock(&matrix_dev->lock);
>> +	ret = vfio_ap_mdev_validate_masks(matrix_mdev, apm,
>> +					  matrix_mdev->matrix.aqm);
>> +	if (ret) {
>> +		mutex_unlock(&matrix_dev->lock);
>> +		return ret;
>> +	}
> At this point the ap_perms may have already changed, or?

Both this function and the in_use callback take
the matrix_dev->lock; therefore, the ap_perms will not be changed until
getting an answer from the in_use callback which will be blocked until
this assignment function releases the lock. Does that sound about
right?

>
>>   	set_bit_inv(apid, matrix_mdev->matrix.apm);
>>   	vfio_ap_mdev_link_queues(matrix_mdev, LINK_APID, apid);
>> -	ret = count;
>> -
>> -done:
>>   	mutex_unlock(&matrix_dev->lock);
>>   
>> -	return ret;
>> +	return count;
>>   }
>>   static DEVICE_ATTR_WO(assign_adapter);
>>   
>> @@ -958,26 +848,6 @@ static ssize_t unassign_adapter_store(struct device *dev,
>>   }
>>   static DEVICE_ATTR_WO(unassign_adapter);
>>   
>> -static int
>> -vfio_ap_mdev_verify_queues_reserved_for_apqi(struct ap_matrix_mdev *matrix_mdev,
>> -					     unsigned long apqi)
>> -{
>> -	int ret;
>> -	unsigned long apid;
>> -	unsigned long nbits = matrix_mdev->matrix.apm_max + 1;
>> -
>> -	if (find_first_bit_inv(matrix_mdev->matrix.apm, nbits) >= nbits)
>> -		return vfio_ap_verify_queue_reserved(NULL, &apqi);
>> -
>> -	for_each_set_bit_inv(apid, matrix_mdev->matrix.apm, nbits) {
>> -		ret = vfio_ap_verify_queue_reserved(&apid, &apqi);
>> -		if (ret)
>> -			return ret;
>> -	}
>> -
>> -	return 0;
>> -}
>> -
>>   /**
>>    * assign_domain_store
>>    *
>> @@ -1031,28 +901,21 @@ static ssize_t assign_domain_store(struct device *dev,
>>   	if (apqi > max_apqi)
>>   		return -ENODEV;
>>   
>> -	mutex_lock(&matrix_dev->lock);
>> -
>> -	ret = vfio_ap_mdev_verify_queues_reserved_for_apqi(matrix_mdev, apqi);
>> -	if (ret)
>> -		goto done;
>> -
>>   	memset(aqm, 0, sizeof(aqm));
>>   	set_bit_inv(apqi, aqm);
>>   
>> -	ret = vfio_ap_mdev_verify_no_sharing(matrix_mdev,
>> -					     matrix_mdev->matrix.apm, aqm);
>> -	if (ret)
>> -		goto done;
>> -
>> +	mutex_lock(&matrix_dev->lock);
>> +	ret = vfio_ap_mdev_validate_masks(matrix_mdev, matrix_mdev->matrix.apm,
>> +					  aqm);
>> +	if (ret) {
>> +		mutex_unlock(&matrix_dev->lock);
>> +		return ret;
>> +	}
>>   	set_bit_inv(apqi, matrix_mdev->matrix.aqm);
>>   	vfio_ap_mdev_link_queues(matrix_mdev, LINK_APQI, apqi);
>> -	ret = count;
>> -
>> -done:
>>   	mutex_unlock(&matrix_dev->lock);
>>   
>> -	return ret;
>> +	return count;
>>   }
>>   static DEVICE_ATTR_WO(assign_domain);
>>   
>> @@ -1139,11 +1002,6 @@ static ssize_t assign_control_domain_store(struct device *dev,
>>   	if (id > matrix_mdev->matrix.adm_max)
>>   		return -ENODEV;
>>   
>> -	/* Set the bit in the ADM (bitmask) corresponding to the AP control
>> -	 * domain number (id). The bits in the mask, from most significant to
>> -	 * least significant, correspond to IDs 0 up to the one less than the
>> -	 * number of control domains that can be assigned.
>> -	 */
>>   	mutex_lock(&matrix_dev->lock);
>>   	set_bit_inv(id, matrix_mdev->matrix.adm);
>>   	mutex_unlock(&matrix_dev->lock);

Powered by blists - more mailing lists