1 .. include:: <isonum.txt>
7 :Copyright: |copy| 2016, NVIDIA CORPORATION. All rights reserved.
8 :Author: Neo Jia <cjia@nvidia.com>
9 :Author: Kirti Wankhede <kwankhede@nvidia.com>
11 This program is free software; you can redistribute it and/or modify
12 it under the terms of the GNU General Public License version 2 as
13 published by the Free Software Foundation.
16 Virtual Function I/O (VFIO) Mediated devices[1]
17 ===============================================
19 The number of use cases for virtualizing DMA devices that do not have built-in
20 SR_IOV capability is increasing. Previously, to virtualize such devices,
21 developers had to create their own management interfaces and APIs, and then
22 integrate them with user space software. To simplify integration with user space
23 software, we have identified common requirements and a unified management
24 interface for such devices.
26 The VFIO driver framework provides unified APIs for direct device access. It is
27 an IOMMU/device-agnostic framework for exposing direct device access to user
28 space in a secure, IOMMU-protected environment. This framework is used for
29 multiple devices, such as GPUs, network adapters, and compute accelerators. With
30 direct device access, virtual machines or user space applications have direct
31 access to the physical device. This framework is reused for mediated devices.
33 The mediated core driver provides a common interface for mediated device
34 management that can be used by drivers of different devices. This module
35 provides a generic interface to perform these operations:
37 * Create and destroy a mediated device
38 * Add a mediated device to and remove it from a mediated bus driver
39 * Add a mediated device to and remove it from an IOMMU group
41 The mediated core driver also provides an interface to register a bus driver.
42 For example, the mediated VFIO mdev driver is designed for mediated devices and
43 supports VFIO APIs. The mediated bus driver adds a mediated device to and
44 removes it from a VFIO group.
46 The following high-level block diagram shows the main components and interfaces
47 in the VFIO mediated driver framework. The diagram shows NVIDIA, Intel, and IBM
48 devices as examples, as these devices are the first devices to use this module::
52 | +-----------+ | mdev_register_driver() +--------------+
53 | | | +<------------------------+ |
55 | | bus | +------------------------>+ vfio_mdev.ko |<-> VFIO user
56 | | driver | | probe()/remove() | | APIs
57 | | | | +--------------+
63 | +-----------+ | mdev_register_device() +--------------+
64 | | | +<------------------------+ |
65 | | | | | nvidia.ko |<-> physical
66 | | | +------------------------>+ | device
67 | | | | callbacks +--------------+
69 | | device | | mdev_register_device() +--------------+
70 | | interface | |<------------------------+ |
71 | | | | | i915.ko |<-> physical
72 | | | +------------------------>+ | device
73 | | | | callbacks +--------------+
75 | | | | mdev_register_device() +--------------+
76 | | | +<------------------------+ |
77 | | | | | ccw_device.ko|<-> physical
78 | | | +------------------------>+ | device
79 | | | | callbacks +--------------+
84 Registration Interfaces
85 =======================
87 The mediated core driver provides the following types of registration
90 * Registration interface for a mediated bus driver
91 * Physical device driver interface
93 Registration Interface for a Mediated Bus Driver
94 ------------------------------------------------
96 The registration interface for a mediated device driver provides the following
97 structure to represent a mediated device's driver::
100 * struct mdev_driver [2] - Mediated device's driver
101 * @probe: called when new device created
102 * @remove: called when device removed
103 * @driver: device driver structure
106 int (*probe) (struct mdev_device *dev);
107 void (*remove) (struct mdev_device *dev);
108 struct attribute_group **supported_type_groups;
109 struct device_driver driver;
112 A mediated bus driver for mdev should use this structure in the function calls
113 to register and unregister itself with the core driver:
117 extern int mdev_register_driver(struct mdev_driver *drv);
121 extern void mdev_unregister_driver(struct mdev_driver *drv);
123 The mediated bus driver's probe function should create a vfio_device on top of
124 the mdev_device and connect it to an appropriate implementation of
127 When a driver wants to add the GUID creation sysfs to an existing device it has
128 probe'd to then it should call::
130 extern int mdev_register_device(struct device *dev,
131 struct mdev_driver *mdev_driver);
133 This will provide the 'mdev_supported_types/XX/create' files which can then be
134 used to trigger the creation of a mdev_device. The created mdev_device will be
135 attached to the specified driver.
137 When the driver needs to remove itself it calls::
139 extern void mdev_unregister_device(struct device *dev);
141 Which will unbind and destroy all the created mdevs and remove the sysfs files.
143 Mediated Device Management Interface Through sysfs
144 ==================================================
146 The management interface through sysfs enables user space software, such as
147 libvirt, to query and configure mediated devices in a hardware-agnostic fashion.
148 This management interface provides flexibility to the underlying physical
149 device's driver to support features such as:
151 * Mediated device hot plug
152 * Multiple mediated devices in a single virtual machine
153 * Multiple mediated devices from different physical devices
155 Links in the mdev_bus Class Directory
156 -------------------------------------
157 The /sys/class/mdev_bus/ directory contains links to devices that are registered
158 with the mdev core driver.
160 Directories and files under the sysfs for Each Physical Device
161 --------------------------------------------------------------
165 |- [parent physical device]
166 |--- Vendor-specific-attributes [optional]
167 |--- [mdev_supported_types]
171 | | |--- available_instances
178 | | |--- available_instances
185 | |--- available_instances
190 * [mdev_supported_types]
192 The list of currently supported mediated device types and their details.
194 [<type-id>], device_api, and available_instances are mandatory attributes
195 that should be provided by vendor driver.
199 The [<type-id>] name is created by adding the device driver string as a prefix
200 to the string provided by the vendor driver. This format of this name is as
203 sprintf(buf, "%s-%s", dev_driver_string(parent->dev), group->name);
205 (or using mdev_parent_dev(mdev) to arrive at the parent device outside
206 of the core mdev code)
210 This attribute should show which device API is being created, for example,
211 "vfio-pci" for a PCI device.
213 * available_instances
215 This attribute should show the number of devices of type <type-id> that can be
220 This directory contains links to the devices of type <type-id> that have been
225 This attribute should show human readable name. This is optional attribute.
229 This attribute should show brief features/description of the type. This is
232 Directories and Files Under the sysfs for Each mdev Device
233 ----------------------------------------------------------
237 |- [parent phy device]
240 |--- mdev_type {link to its type}
241 |--- vendor-specific-attributes [optional]
243 * remove (write only)
245 Writing '1' to the 'remove' file destroys the mdev device. The vendor driver can
246 fail the remove() callback if that device is active and the vendor driver
247 doesn't support hot unplug.
251 # echo 1 > /sys/bus/mdev/devices/$mdev_UUID/remove
253 Mediated device Hot plug
254 ------------------------
256 Mediated devices can be created and assigned at runtime. The procedure to hot
257 plug a mediated device is the same as the procedure to hot plug a PCI device.
259 Translation APIs for Mediated Devices
260 =====================================
262 The following APIs are provided for translating user pfn to host pfn in a VFIO
265 int vfio_pin_pages(struct vfio_device *device, unsigned long *user_pfn,
266 int npage, int prot, unsigned long *phys_pfn);
268 int vfio_unpin_pages(struct vfio_device *device, unsigned long *user_pfn,
271 These functions call back into the back-end IOMMU module by using the pin_pages
272 and unpin_pages callbacks of the struct vfio_iommu_driver_ops[4]. Currently
273 these callbacks are supported in the TYPE1 IOMMU module. To enable them for
274 other IOMMU backend modules, such as PPC64 sPAPR module, they need to provide
275 these two callback functions.
277 Using the Sample Code
278 =====================
280 mtty.c in samples/vfio-mdev/ directory is a sample driver program to
281 demonstrate how to use the mediated device framework.
283 The sample driver creates an mdev device that simulates a serial port over a PCI
286 1. Build and load the mtty.ko module.
288 This step creates a dummy device, /sys/devices/virtual/mtty/mtty/
290 Files in this device directory in sysfs are similar to the following::
292 # tree /sys/devices/virtual/mtty/mtty/
293 /sys/devices/virtual/mtty/mtty/
294 |-- mdev_supported_types
296 | | |-- available_instances
302 | |-- available_instances
308 | `-- sample_mtty_dev
310 | |-- autosuspend_delay_ms
312 | |-- runtime_active_time
314 | `-- runtime_suspended_time
315 |-- subsystem -> ../../../../class/mtty
318 2. Create a mediated device by using the dummy device that you created in the
321 # echo "83b8f4f2-509f-382f-3c1e-e6bfe0fa1001" > \
322 /sys/devices/virtual/mtty/mtty/mdev_supported_types/mtty-2/create
324 3. Add parameters to qemu-kvm::
327 sysfsdev=/sys/bus/mdev/devices/83b8f4f2-509f-382f-3c1e-e6bfe0fa1001
331 In the Linux guest VM, with no hardware on the host, the device appears
334 # lspci -s 00:05.0 -xxvv
335 00:05.0 Serial controller: Device 4348:3253 (rev 10) (prog-if 02 [16550])
336 Subsystem: Device 4348:3253
338 Control: I/O+ Mem- BusMaster- SpecCycle- MemWINV- VGASnoop- ParErr-
339 Stepping- SERR- FastB2B- DisINTx-
340 Status: Cap- 66MHz- UDF- FastB2B- ParErr- DEVSEL=medium >TAbort-
341 <TAbort- <MAbort- >SERR- <PERR- INTx-
342 Interrupt: pin A routed to IRQ 10
343 Region 0: I/O ports at c150 [size=8]
344 Region 1: I/O ports at c158 [size=8]
345 Kernel driver in use: serial
346 00: 48 43 53 32 01 00 00 02 10 02 00 07 00 00 00 00
347 10: 51 c1 00 00 59 c1 00 00 00 00 00 00 00 00 00 00
348 20: 00 00 00 00 00 00 00 00 00 00 00 00 48 43 53 32
349 30: 00 00 00 00 00 00 00 00 00 00 00 00 0a 01 00 00
351 In the Linux guest VM, dmesg output for the device is as follows:
353 serial 0000:00:05.0: PCI INT A -> Link[LNKA] -> GSI 10 (level, high) -> IRQ 10
354 0000:00:05.0: ttyS1 at I/O 0xc150 (irq = 10) is a 16550A
355 0000:00:05.0: ttyS2 at I/O 0xc158 (irq = 10) is a 16550A
358 5. In the Linux guest VM, check the serial ports::
360 # setserial -g /dev/ttyS*
361 /dev/ttyS0, UART: 16550A, Port: 0x03f8, IRQ: 4
362 /dev/ttyS1, UART: 16550A, Port: 0xc150, IRQ: 10
363 /dev/ttyS2, UART: 16550A, Port: 0xc158, IRQ: 10
365 6. Using minicom or any terminal emulation program, open port /dev/ttyS1 or
366 /dev/ttyS2 with hardware flow control disabled.
368 7. Type data on the minicom terminal or send data to the terminal emulation
369 program and read the data.
371 Data is loop backed from hosts mtty driver.
373 8. Destroy the mediated device that you created::
375 # echo 1 > /sys/bus/mdev/devices/83b8f4f2-509f-382f-3c1e-e6bfe0fa1001/remove
380 1. See Documentation/driver-api/vfio.rst for more information on VFIO.
381 2. struct mdev_driver in include/linux/mdev.h
382 3. struct mdev_parent_ops in include/linux/mdev.h
383 4. struct vfio_iommu_driver_ops in include/linux/vfio.h