1 .. SPDX-License-Identifier: GPL-2.0-only
2 .. include:: <isonum.txt>
8 :Copyright: |copy| 2016, NVIDIA CORPORATION. All rights reserved.
9 :Author: Neo Jia <cjia@nvidia.com>
10 :Author: Kirti Wankhede <kwankhede@nvidia.com>
14 Virtual Function I/O (VFIO) Mediated devices[1]
15 ===============================================
17 The number of use cases for virtualizing DMA devices that do not have built-in
18 SR_IOV capability is increasing. Previously, to virtualize such devices,
19 developers had to create their own management interfaces and APIs, and then
20 integrate them with user space software. To simplify integration with user space
21 software, we have identified common requirements and a unified management
22 interface for such devices.
24 The VFIO driver framework provides unified APIs for direct device access. It is
25 an IOMMU/device-agnostic framework for exposing direct device access to user
26 space in a secure, IOMMU-protected environment. This framework is used for
27 multiple devices, such as GPUs, network adapters, and compute accelerators. With
28 direct device access, virtual machines or user space applications have direct
29 access to the physical device. This framework is reused for mediated devices.
31 The mediated core driver provides a common interface for mediated device
32 management that can be used by drivers of different devices. This module
33 provides a generic interface to perform these operations:
35 * Create and destroy a mediated device
36 * Add a mediated device to and remove it from a mediated bus driver
37 * Add a mediated device to and remove it from an IOMMU group
39 The mediated core driver also provides an interface to register a bus driver.
40 For example, the mediated VFIO mdev driver is designed for mediated devices and
41 supports VFIO APIs. The mediated bus driver adds a mediated device to and
42 removes it from a VFIO group.
44 The following high-level block diagram shows the main components and interfaces
45 in the VFIO mediated driver framework. The diagram shows NVIDIA, Intel, and IBM
46 devices as examples, as these devices are the first devices to use this module::
50 | +-----------+ | mdev_register_driver() +--------------+
51 | | | +<------------------------+ |
53 | | bus | +------------------------>+ vfio_mdev.ko |<-> VFIO user
54 | | driver | | probe()/remove() | | APIs
55 | | | | +--------------+
61 | +-----------+ | mdev_register_parent() +--------------+
62 | | | +<------------------------+ |
63 | | | | | ccw_device.ko|<-> physical
64 | | | +------------------------>+ | device
65 | | | | callbacks +--------------+
67 | | device | | mdev_register_parent() +--------------+
68 | | interface | |<------------------------+ |
69 | | | | | i915.ko |<-> physical
70 | | | +------------------------>+ | device
71 | | | | callbacks +--------------+
76 Registration Interfaces
77 =======================
79 The mediated core driver provides the following types of registration
82 * Registration interface for a mediated bus driver
83 * Physical device driver interface
85 Registration Interface for a Mediated Bus Driver
86 ------------------------------------------------
88 The registration interface for a mediated device driver provides the following
89 structure to represent a mediated device's driver::
92 * struct mdev_driver [2] - Mediated device's driver
93 * @probe: called when new device created
94 * @remove: called when device removed
95 * @driver: device driver structure
98 int (*probe) (struct mdev_device *dev);
99 void (*remove) (struct mdev_device *dev);
100 unsigned int (*get_available)(struct mdev_type *mtype);
101 ssize_t (*show_description)(struct mdev_type *mtype, char *buf);
102 struct device_driver driver;
105 A mediated bus driver for mdev should use this structure in the function calls
106 to register and unregister itself with the core driver:
110 int mdev_register_driver(struct mdev_driver *drv);
114 void mdev_unregister_driver(struct mdev_driver *drv);
116 The mediated bus driver's probe function should create a vfio_device on top of
117 the mdev_device and connect it to an appropriate implementation of
120 When a driver wants to add the GUID creation sysfs to an existing device it has
121 probe'd to then it should call::
123 int mdev_register_parent(struct mdev_parent *parent, struct device *dev,
124 struct mdev_driver *mdev_driver);
126 This will provide the 'mdev_supported_types/XX/create' files which can then be
127 used to trigger the creation of a mdev_device. The created mdev_device will be
128 attached to the specified driver.
130 When the driver needs to remove itself it calls::
132 void mdev_unregister_parent(struct mdev_parent *parent);
134 Which will unbind and destroy all the created mdevs and remove the sysfs files.
136 Mediated Device Management Interface Through sysfs
137 ==================================================
139 The management interface through sysfs enables user space software, such as
140 libvirt, to query and configure mediated devices in a hardware-agnostic fashion.
141 This management interface provides flexibility to the underlying physical
142 device's driver to support features such as:
144 * Mediated device hot plug
145 * Multiple mediated devices in a single virtual machine
146 * Multiple mediated devices from different physical devices
148 Links in the mdev_bus Class Directory
149 -------------------------------------
150 The /sys/class/mdev_bus/ directory contains links to devices that are registered
151 with the mdev core driver.
153 Directories and files under the sysfs for Each Physical Device
154 --------------------------------------------------------------
158 |- [parent physical device]
159 |--- Vendor-specific-attributes [optional]
160 |--- [mdev_supported_types]
164 | | |--- available_instances
171 | | |--- available_instances
178 | |--- available_instances
183 * [mdev_supported_types]
185 The list of currently supported mediated device types and their details.
187 [<type-id>], device_api, and available_instances are mandatory attributes
188 that should be provided by vendor driver.
192 The [<type-id>] name is created by adding the device driver string as a prefix
193 to the string provided by the vendor driver. This format of this name is as
196 sprintf(buf, "%s-%s", dev_driver_string(parent->dev), group->name);
200 This attribute shows which device API is being created, for example,
201 "vfio-pci" for a PCI device.
203 * available_instances
205 This attribute shows the number of devices of type <type-id> that can be
210 This directory contains links to the devices of type <type-id> that have been
215 This attribute shows a human readable name.
219 This attribute can show brief features/description of the type. This is an
222 Directories and Files Under the sysfs for Each mdev Device
223 ----------------------------------------------------------
227 |- [parent phy device]
230 |--- mdev_type {link to its type}
231 |--- vendor-specific-attributes [optional]
233 * remove (write only)
235 Writing '1' to the 'remove' file destroys the mdev device. The vendor driver can
236 fail the remove() callback if that device is active and the vendor driver
237 doesn't support hot unplug.
241 # echo 1 > /sys/bus/mdev/devices/$mdev_UUID/remove
243 Mediated device Hot plug
244 ------------------------
246 Mediated devices can be created and assigned at runtime. The procedure to hot
247 plug a mediated device is the same as the procedure to hot plug a PCI device.
249 Translation APIs for Mediated Devices
250 =====================================
252 The following APIs are provided for translating user pfn to host pfn in a VFIO
255 int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
256 int npage, int prot, struct page **pages);
258 void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,
261 These functions call back into the back-end IOMMU module by using the pin_pages
262 and unpin_pages callbacks of the struct vfio_iommu_driver_ops[4]. Currently
263 these callbacks are supported in the TYPE1 IOMMU module. To enable them for
264 other IOMMU backend modules, such as PPC64 sPAPR module, they need to provide
265 these two callback functions.
270 1. See Documentation/driver-api/vfio.rst for more information on VFIO.
271 2. struct mdev_driver in include/linux/mdev.h
272 3. struct mdev_parent_ops in include/linux/mdev.h
273 4. struct vfio_iommu_driver_ops in include/linux/vfio.h