1a6546f89SThomas Gleixner.. SPDX-License-Identifier: GPL-2.0-only
2baa293e9SMauro Carvalho Chehab.. include:: <isonum.txt>
3baa293e9SMauro Carvalho Chehab
4baa293e9SMauro Carvalho Chehab=====================
5baa293e9SMauro Carvalho ChehabVFIO Mediated devices
6baa293e9SMauro Carvalho Chehab=====================
7baa293e9SMauro Carvalho Chehab
8baa293e9SMauro Carvalho Chehab:Copyright: |copy| 2016, NVIDIA CORPORATION. All rights reserved.
9baa293e9SMauro Carvalho Chehab:Author: Neo Jia <cjia@nvidia.com>
10baa293e9SMauro Carvalho Chehab:Author: Kirti Wankhede <kwankhede@nvidia.com>
11baa293e9SMauro Carvalho Chehab
12baa293e9SMauro Carvalho Chehab
13baa293e9SMauro Carvalho Chehab
14baa293e9SMauro Carvalho ChehabVirtual Function I/O (VFIO) Mediated devices[1]
15baa293e9SMauro Carvalho Chehab===============================================
16baa293e9SMauro Carvalho Chehab
17baa293e9SMauro Carvalho ChehabThe number of use cases for virtualizing DMA devices that do not have built-in
18baa293e9SMauro Carvalho ChehabSR_IOV capability is increasing. Previously, to virtualize such devices,
19baa293e9SMauro Carvalho Chehabdevelopers had to create their own management interfaces and APIs, and then
20baa293e9SMauro Carvalho Chehabintegrate them with user space software. To simplify integration with user space
21baa293e9SMauro Carvalho Chehabsoftware, we have identified common requirements and a unified management
22baa293e9SMauro Carvalho Chehabinterface for such devices.
23baa293e9SMauro Carvalho Chehab
24baa293e9SMauro Carvalho ChehabThe VFIO driver framework provides unified APIs for direct device access. It is
25baa293e9SMauro Carvalho Chehaban IOMMU/device-agnostic framework for exposing direct device access to user
26baa293e9SMauro Carvalho Chehabspace in a secure, IOMMU-protected environment. This framework is used for
27baa293e9SMauro Carvalho Chehabmultiple devices, such as GPUs, network adapters, and compute accelerators. With
28baa293e9SMauro Carvalho Chehabdirect device access, virtual machines or user space applications have direct
29baa293e9SMauro Carvalho Chehabaccess to the physical device. This framework is reused for mediated devices.
30baa293e9SMauro Carvalho Chehab
31baa293e9SMauro Carvalho ChehabThe mediated core driver provides a common interface for mediated device
32baa293e9SMauro Carvalho Chehabmanagement that can be used by drivers of different devices. This module
33baa293e9SMauro Carvalho Chehabprovides a generic interface to perform these operations:
34baa293e9SMauro Carvalho Chehab
35baa293e9SMauro Carvalho Chehab* Create and destroy a mediated device
36baa293e9SMauro Carvalho Chehab* Add a mediated device to and remove it from a mediated bus driver
37baa293e9SMauro Carvalho Chehab* Add a mediated device to and remove it from an IOMMU group
38baa293e9SMauro Carvalho Chehab
39baa293e9SMauro Carvalho ChehabThe mediated core driver also provides an interface to register a bus driver.
40baa293e9SMauro Carvalho ChehabFor example, the mediated VFIO mdev driver is designed for mediated devices and
41baa293e9SMauro Carvalho Chehabsupports VFIO APIs. The mediated bus driver adds a mediated device to and
42baa293e9SMauro Carvalho Chehabremoves it from a VFIO group.
43baa293e9SMauro Carvalho Chehab
44baa293e9SMauro Carvalho ChehabThe following high-level block diagram shows the main components and interfaces
45baa293e9SMauro Carvalho Chehabin the VFIO mediated driver framework. The diagram shows NVIDIA, Intel, and IBM
46baa293e9SMauro Carvalho Chehabdevices as examples, as these devices are the first devices to use this module::
47baa293e9SMauro Carvalho Chehab
48baa293e9SMauro Carvalho Chehab     +---------------+
49baa293e9SMauro Carvalho Chehab     |               |
50baa293e9SMauro Carvalho Chehab     | +-----------+ |  mdev_register_driver() +--------------+
51baa293e9SMauro Carvalho Chehab     | |           | +<------------------------+              |
52baa293e9SMauro Carvalho Chehab     | |  mdev     | |                         |              |
53baa293e9SMauro Carvalho Chehab     | |  bus      | +------------------------>+ vfio_mdev.ko |<-> VFIO user
54baa293e9SMauro Carvalho Chehab     | |  driver   | |     probe()/remove()    |              |    APIs
55baa293e9SMauro Carvalho Chehab     | |           | |                         +--------------+
56baa293e9SMauro Carvalho Chehab     | +-----------+ |
57baa293e9SMauro Carvalho Chehab     |               |
58baa293e9SMauro Carvalho Chehab     |  MDEV CORE    |
59baa293e9SMauro Carvalho Chehab     |   MODULE      |
60baa293e9SMauro Carvalho Chehab     |   mdev.ko     |
6189345d51SChristoph Hellwig     | +-----------+ |  mdev_register_parent() +--------------+
62baa293e9SMauro Carvalho Chehab     | |           | +<------------------------+              |
63*32328681SChristoph Hellwig     | |           | |                         | ccw_device.ko|<-> physical
64baa293e9SMauro Carvalho Chehab     | |           | +------------------------>+              |    device
65baa293e9SMauro Carvalho Chehab     | |           | |        callbacks        +--------------+
66baa293e9SMauro Carvalho Chehab     | | Physical  | |
6789345d51SChristoph Hellwig     | |  device   | |  mdev_register_parent() +--------------+
68baa293e9SMauro Carvalho Chehab     | | interface | |<------------------------+              |
69baa293e9SMauro Carvalho Chehab     | |           | |                         |  i915.ko     |<-> physical
70baa293e9SMauro Carvalho Chehab     | |           | +------------------------>+              |    device
71baa293e9SMauro Carvalho Chehab     | |           | |        callbacks        +--------------+
72baa293e9SMauro Carvalho Chehab     | +-----------+ |
73baa293e9SMauro Carvalho Chehab     +---------------+
74baa293e9SMauro Carvalho Chehab
75baa293e9SMauro Carvalho Chehab
76baa293e9SMauro Carvalho ChehabRegistration Interfaces
77baa293e9SMauro Carvalho Chehab=======================
78baa293e9SMauro Carvalho Chehab
79baa293e9SMauro Carvalho ChehabThe mediated core driver provides the following types of registration
80baa293e9SMauro Carvalho Chehabinterfaces:
81baa293e9SMauro Carvalho Chehab
82baa293e9SMauro Carvalho Chehab* Registration interface for a mediated bus driver
83baa293e9SMauro Carvalho Chehab* Physical device driver interface
84baa293e9SMauro Carvalho Chehab
85baa293e9SMauro Carvalho ChehabRegistration Interface for a Mediated Bus Driver
86baa293e9SMauro Carvalho Chehab------------------------------------------------
87baa293e9SMauro Carvalho Chehab
8888a21f26SJason GunthorpeThe registration interface for a mediated device driver provides the following
89baa293e9SMauro Carvalho Chehabstructure to represent a mediated device's driver::
90baa293e9SMauro Carvalho Chehab
91baa293e9SMauro Carvalho Chehab     /*
92baa293e9SMauro Carvalho Chehab      * struct mdev_driver [2] - Mediated device's driver
93baa293e9SMauro Carvalho Chehab      * @probe: called when new device created
94baa293e9SMauro Carvalho Chehab      * @remove: called when device removed
95baa293e9SMauro Carvalho Chehab      * @driver: device driver structure
96baa293e9SMauro Carvalho Chehab      */
97baa293e9SMauro Carvalho Chehab     struct mdev_driver {
982a3d15f2SJason Gunthorpe	     int  (*probe)  (struct mdev_device *dev);
992a3d15f2SJason Gunthorpe	     void (*remove) (struct mdev_device *dev);
100f2fbc72eSChristoph Hellwig	     unsigned int (*get_available)(struct mdev_type *mtype);
101685a1537SChristoph Hellwig	     ssize_t (*show_description)(struct mdev_type *mtype, char *buf);
102baa293e9SMauro Carvalho Chehab	     struct device_driver    driver;
103baa293e9SMauro Carvalho Chehab     };
104baa293e9SMauro Carvalho Chehab
105baa293e9SMauro Carvalho ChehabA mediated bus driver for mdev should use this structure in the function calls
106baa293e9SMauro Carvalho Chehabto register and unregister itself with the core driver:
107baa293e9SMauro Carvalho Chehab
108baa293e9SMauro Carvalho Chehab* Register::
109baa293e9SMauro Carvalho Chehab
110d1877e63SAlex Williamson    int mdev_register_driver(struct mdev_driver *drv);
111baa293e9SMauro Carvalho Chehab
112baa293e9SMauro Carvalho Chehab* Unregister::
113baa293e9SMauro Carvalho Chehab
114d1877e63SAlex Williamson    void mdev_unregister_driver(struct mdev_driver *drv);
115baa293e9SMauro Carvalho Chehab
1166b42f491SJason GunthorpeThe mediated bus driver's probe function should create a vfio_device on top of
1176b42f491SJason Gunthorpethe mdev_device and connect it to an appropriate implementation of
1186b42f491SJason Gunthorpevfio_device_ops.
119baa293e9SMauro Carvalho Chehab
12088a21f26SJason GunthorpeWhen a driver wants to add the GUID creation sysfs to an existing device it has
12188a21f26SJason Gunthorpeprobe'd to then it should call::
122baa293e9SMauro Carvalho Chehab
12389345d51SChristoph Hellwig    int mdev_register_parent(struct mdev_parent *parent, struct device *dev,
1246b42f491SJason Gunthorpe			struct mdev_driver *mdev_driver);
125baa293e9SMauro Carvalho Chehab
12688a21f26SJason GunthorpeThis will provide the 'mdev_supported_types/XX/create' files which can then be
12788a21f26SJason Gunthorpeused to trigger the creation of a mdev_device. The created mdev_device will be
12888a21f26SJason Gunthorpeattached to the specified driver.
12988a21f26SJason Gunthorpe
13088a21f26SJason GunthorpeWhen the driver needs to remove itself it calls::
131baa293e9SMauro Carvalho Chehab
13289345d51SChristoph Hellwig    void mdev_unregister_parent(struct mdev_parent *parent);
133baa293e9SMauro Carvalho Chehab
13488a21f26SJason GunthorpeWhich will unbind and destroy all the created mdevs and remove the sysfs files.
135baa293e9SMauro Carvalho Chehab
136baa293e9SMauro Carvalho ChehabMediated Device Management Interface Through sysfs
137baa293e9SMauro Carvalho Chehab==================================================
138baa293e9SMauro Carvalho Chehab
139baa293e9SMauro Carvalho ChehabThe management interface through sysfs enables user space software, such as
140baa293e9SMauro Carvalho Chehablibvirt, to query and configure mediated devices in a hardware-agnostic fashion.
141baa293e9SMauro Carvalho ChehabThis management interface provides flexibility to the underlying physical
142baa293e9SMauro Carvalho Chehabdevice's driver to support features such as:
143baa293e9SMauro Carvalho Chehab
144baa293e9SMauro Carvalho Chehab* Mediated device hot plug
145baa293e9SMauro Carvalho Chehab* Multiple mediated devices in a single virtual machine
146baa293e9SMauro Carvalho Chehab* Multiple mediated devices from different physical devices
147baa293e9SMauro Carvalho Chehab
148baa293e9SMauro Carvalho ChehabLinks in the mdev_bus Class Directory
149baa293e9SMauro Carvalho Chehab-------------------------------------
150baa293e9SMauro Carvalho ChehabThe /sys/class/mdev_bus/ directory contains links to devices that are registered
151baa293e9SMauro Carvalho Chehabwith the mdev core driver.
152baa293e9SMauro Carvalho Chehab
153baa293e9SMauro Carvalho ChehabDirectories and files under the sysfs for Each Physical Device
154baa293e9SMauro Carvalho Chehab--------------------------------------------------------------
155baa293e9SMauro Carvalho Chehab
156baa293e9SMauro Carvalho Chehab::
157baa293e9SMauro Carvalho Chehab
158baa293e9SMauro Carvalho Chehab  |- [parent physical device]
159baa293e9SMauro Carvalho Chehab  |--- Vendor-specific-attributes [optional]
160baa293e9SMauro Carvalho Chehab  |--- [mdev_supported_types]
161baa293e9SMauro Carvalho Chehab  |     |--- [<type-id>]
162baa293e9SMauro Carvalho Chehab  |     |   |--- create
163baa293e9SMauro Carvalho Chehab  |     |   |--- name
164baa293e9SMauro Carvalho Chehab  |     |   |--- available_instances
165baa293e9SMauro Carvalho Chehab  |     |   |--- device_api
166baa293e9SMauro Carvalho Chehab  |     |   |--- description
167baa293e9SMauro Carvalho Chehab  |     |   |--- [devices]
168baa293e9SMauro Carvalho Chehab  |     |--- [<type-id>]
169baa293e9SMauro Carvalho Chehab  |     |   |--- create
170baa293e9SMauro Carvalho Chehab  |     |   |--- name
171baa293e9SMauro Carvalho Chehab  |     |   |--- available_instances
172baa293e9SMauro Carvalho Chehab  |     |   |--- device_api
173baa293e9SMauro Carvalho Chehab  |     |   |--- description
174baa293e9SMauro Carvalho Chehab  |     |   |--- [devices]
175baa293e9SMauro Carvalho Chehab  |     |--- [<type-id>]
176baa293e9SMauro Carvalho Chehab  |          |--- create
177baa293e9SMauro Carvalho Chehab  |          |--- name
178baa293e9SMauro Carvalho Chehab  |          |--- available_instances
179baa293e9SMauro Carvalho Chehab  |          |--- device_api
180baa293e9SMauro Carvalho Chehab  |          |--- description
181baa293e9SMauro Carvalho Chehab  |          |--- [devices]
182baa293e9SMauro Carvalho Chehab
183baa293e9SMauro Carvalho Chehab* [mdev_supported_types]
184baa293e9SMauro Carvalho Chehab
185baa293e9SMauro Carvalho Chehab  The list of currently supported mediated device types and their details.
186baa293e9SMauro Carvalho Chehab
187baa293e9SMauro Carvalho Chehab  [<type-id>], device_api, and available_instances are mandatory attributes
188baa293e9SMauro Carvalho Chehab  that should be provided by vendor driver.
189baa293e9SMauro Carvalho Chehab
190baa293e9SMauro Carvalho Chehab* [<type-id>]
191baa293e9SMauro Carvalho Chehab
192baa293e9SMauro Carvalho Chehab  The [<type-id>] name is created by adding the device driver string as a prefix
193baa293e9SMauro Carvalho Chehab  to the string provided by the vendor driver. This format of this name is as
194baa293e9SMauro Carvalho Chehab  follows::
195baa293e9SMauro Carvalho Chehab
196baa293e9SMauro Carvalho Chehab	sprintf(buf, "%s-%s", dev_driver_string(parent->dev), group->name);
197baa293e9SMauro Carvalho Chehab
198baa293e9SMauro Carvalho Chehab* device_api
199baa293e9SMauro Carvalho Chehab
200290aac5dSJason Gunthorpe  This attribute shows which device API is being created, for example,
201baa293e9SMauro Carvalho Chehab  "vfio-pci" for a PCI device.
202baa293e9SMauro Carvalho Chehab
203baa293e9SMauro Carvalho Chehab* available_instances
204baa293e9SMauro Carvalho Chehab
205f2fbc72eSChristoph Hellwig  This attribute shows the number of devices of type <type-id> that can be
206baa293e9SMauro Carvalho Chehab  created.
207baa293e9SMauro Carvalho Chehab
208baa293e9SMauro Carvalho Chehab* [device]
209baa293e9SMauro Carvalho Chehab
210baa293e9SMauro Carvalho Chehab  This directory contains links to the devices of type <type-id> that have been
211baa293e9SMauro Carvalho Chehab  created.
212baa293e9SMauro Carvalho Chehab
213baa293e9SMauro Carvalho Chehab* name
214baa293e9SMauro Carvalho Chehab
2150bc79069SChristoph Hellwig  This attribute shows a human readable name.
216baa293e9SMauro Carvalho Chehab
217baa293e9SMauro Carvalho Chehab* description
218baa293e9SMauro Carvalho Chehab
219685a1537SChristoph Hellwig  This attribute can show brief features/description of the type. This is an
220baa293e9SMauro Carvalho Chehab  optional attribute.
221baa293e9SMauro Carvalho Chehab
222baa293e9SMauro Carvalho ChehabDirectories and Files Under the sysfs for Each mdev Device
223baa293e9SMauro Carvalho Chehab----------------------------------------------------------
224baa293e9SMauro Carvalho Chehab
225baa293e9SMauro Carvalho Chehab::
226baa293e9SMauro Carvalho Chehab
227baa293e9SMauro Carvalho Chehab  |- [parent phy device]
228baa293e9SMauro Carvalho Chehab  |--- [$MDEV_UUID]
229baa293e9SMauro Carvalho Chehab         |--- remove
230baa293e9SMauro Carvalho Chehab         |--- mdev_type {link to its type}
231baa293e9SMauro Carvalho Chehab         |--- vendor-specific-attributes [optional]
232baa293e9SMauro Carvalho Chehab
233baa293e9SMauro Carvalho Chehab* remove (write only)
234baa293e9SMauro Carvalho Chehab
235baa293e9SMauro Carvalho ChehabWriting '1' to the 'remove' file destroys the mdev device. The vendor driver can
236baa293e9SMauro Carvalho Chehabfail the remove() callback if that device is active and the vendor driver
237baa293e9SMauro Carvalho Chehabdoesn't support hot unplug.
238baa293e9SMauro Carvalho Chehab
239baa293e9SMauro Carvalho ChehabExample::
240baa293e9SMauro Carvalho Chehab
241baa293e9SMauro Carvalho Chehab	# echo 1 > /sys/bus/mdev/devices/$mdev_UUID/remove
242baa293e9SMauro Carvalho Chehab
243baa293e9SMauro Carvalho ChehabMediated device Hot plug
244baa293e9SMauro Carvalho Chehab------------------------
245baa293e9SMauro Carvalho Chehab
246baa293e9SMauro Carvalho ChehabMediated devices can be created and assigned at runtime. The procedure to hot
247baa293e9SMauro Carvalho Chehabplug a mediated device is the same as the procedure to hot plug a PCI device.
248baa293e9SMauro Carvalho Chehab
249baa293e9SMauro Carvalho ChehabTranslation APIs for Mediated Devices
250baa293e9SMauro Carvalho Chehab=====================================
251baa293e9SMauro Carvalho Chehab
252baa293e9SMauro Carvalho ChehabThe following APIs are provided for translating user pfn to host pfn in a VFIO
253baa293e9SMauro Carvalho Chehabdriver::
254baa293e9SMauro Carvalho Chehab
25544abdd16SNicolin Chen	int vfio_pin_pages(struct vfio_device *device, dma_addr_t iova,
25634a255e6SNicolin Chen				  int npage, int prot, struct page **pages);
257baa293e9SMauro Carvalho Chehab
25844abdd16SNicolin Chen	void vfio_unpin_pages(struct vfio_device *device, dma_addr_t iova,
259baa293e9SMauro Carvalho Chehab				    int npage);
260baa293e9SMauro Carvalho Chehab
261baa293e9SMauro Carvalho ChehabThese functions call back into the back-end IOMMU module by using the pin_pages
262baa293e9SMauro Carvalho Chehaband unpin_pages callbacks of the struct vfio_iommu_driver_ops[4]. Currently
263baa293e9SMauro Carvalho Chehabthese callbacks are supported in the TYPE1 IOMMU module. To enable them for
264baa293e9SMauro Carvalho Chehabother IOMMU backend modules, such as PPC64 sPAPR module, they need to provide
265baa293e9SMauro Carvalho Chehabthese two callback functions.
266baa293e9SMauro Carvalho Chehab
267baa293e9SMauro Carvalho ChehabReferences
268baa293e9SMauro Carvalho Chehab==========
269baa293e9SMauro Carvalho Chehab
270baa293e9SMauro Carvalho Chehab1. See Documentation/driver-api/vfio.rst for more information on VFIO.
271baa293e9SMauro Carvalho Chehab2. struct mdev_driver in include/linux/mdev.h
272baa293e9SMauro Carvalho Chehab3. struct mdev_parent_ops in include/linux/mdev.h
273baa293e9SMauro Carvalho Chehab4. struct vfio_iommu_driver_ops in include/linux/vfio.h
274