| blob | 90b3dfead81bd8390d8fd806129904e50cdc05e8 |
1 vfio-ccw: the basic infrastructure
2 ==================================
4 Introduction
5 ------------
7 Here we describe the vfio support for I/O subchannel devices for
8 Linux/s390. Motivation for vfio-ccw is to passthrough subchannels to a
9 virtual machine, while vfio is the means.
11 Different than other hardware architectures, s390 has defined a unified
12 I/O access method, which is so called Channel I/O. It has its own access
13 patterns:
14 - Channel programs run asynchronously on a separate (co)processor.
15 - The channel subsystem will access any memory designated by the caller
16 in the channel program directly, i.e. there is no iommu involved.
17 Thus when we introduce vfio support for these devices, we realize it
18 with a mediated device (mdev) implementation. The vfio mdev will be
19 added to an iommu group, so as to make itself able to be managed by the
20 vfio framework. And we add read/write callbacks for special vfio I/O
21 regions to pass the channel programs from the mdev to its parent device
22 (the real I/O subchannel device) to do further address translation and
23 to perform I/O instructions.
25 This document does not intend to explain the s390 I/O architecture in
26 every detail. More information/reference could be found here:
27 - A good start to know Channel I/O in general:
28 https://en.wikipedia.org/wiki/Channel_I/O
29 - s390 architecture:
30 s390 Principles of Operation manual (IBM Form. No. SA22-7832)
31 - The existing Qemu code which implements a simple emulated channel
32 subsystem could also be a good reference. It makes it easier to follow
33 the flow.
34 qemu/hw/s390x/css.c
36 For vfio mediated device framework:
37 - Documentation/vfio-mediated-device.txt
39 Motivation of vfio-ccw
40 ----------------------
42 Currently, a guest virtualized via qemu/kvm on s390 only sees
43 paravirtualized virtio devices via the "Virtio Over Channel I/O
44 (virtio-ccw)" transport. This makes virtio devices discoverable via
45 standard operating system algorithms for handling channel devices.
47 However this is not enough. On s390 for the majority of devices, which
48 use the standard Channel I/O based mechanism, we also need to provide
49 the functionality of passing through them to a Qemu virtual machine.
50 This includes devices that don't have a virtio counterpart (e.g. tape
51 drives) or that have specific characteristics which guests want to
52 exploit.
54 For passing a device to a guest, we want to use the same interface as
55 everybody else, namely vfio. Thus, we would like to introduce vfio
56 support for channel devices. And we would like to name this new vfio
57 device "vfio-ccw".
59 Access patterns of CCW devices
60 ------------------------------
62 s390 architecture has implemented a so called channel subsystem, that
63 provides a unified view of the devices physically attached to the
64 systems. Though the s390 hardware platform knows about a huge variety of
65 different peripheral attachments like disk devices (aka. DASDs), tapes,
66 communication controllers, etc. They can all be accessed by a well
67 defined access method and they are presenting I/O completion a unified
68 way: I/O interruptions.
70 All I/O requires the use of channel command words (CCWs). A CCW is an
71 instruction to a specialized I/O channel processor. A channel program is
72 a sequence of CCWs which are executed by the I/O channel subsystem. To
73 issue a channel program to the channel subsystem, it is required to
74 build an operation request block (ORB), which can be used to point out
75 the format of the CCW and other control information to the system. The
76 operating system signals the I/O channel subsystem to begin executing
77 the channel program with a SSCH (start sub-channel) instruction. The
78 central processor is then free to proceed with non-I/O instructions
79 until interrupted. The I/O completion result is received by the
80 interrupt handler in the form of interrupt response block (IRB).
82 Back to vfio-ccw, in short:
83 - ORBs and channel programs are built in guest kernel (with guest
84 physical addresses).
85 - ORBs and channel programs are passed to the host kernel.
86 - Host kernel translates the guest physical addresses to real addresses
87 and starts the I/O with issuing a privileged Channel I/O instruction
88 (e.g SSCH).
89 - channel programs run asynchronously on a separate processor.
90 - I/O completion will be signaled to the host with I/O interruptions.
91 And it will be copied as IRB to user space to pass it back to the
92 guest.
94 Physical vfio ccw device and its child mdev
95 -------------------------------------------
97 As mentioned above, we realize vfio-ccw with a mdev implementation.
99 Channel I/O does not have IOMMU hardware support, so the physical
100 vfio-ccw device does not have an IOMMU level translation or isolation.
102 Sub-channel I/O instructions are all privileged instructions, When
103 handling the I/O instruction interception, vfio-ccw has the software
104 policing and translation how the channel program is programmed before
105 it gets sent to hardware.
107 Within this implementation, we have two drivers for two types of
108 devices:
109 - The vfio_ccw driver for the physical subchannel device.
110 This is an I/O subchannel driver for the real subchannel device. It
111 realizes a group of callbacks and registers to the mdev framework as a
112 parent (physical) device. As a consequence, mdev provides vfio_ccw a
113 generic interface (sysfs) to create mdev devices. A vfio mdev could be
114 created by vfio_ccw then and added to the mediated bus. It is the vfio
115 device that added to an IOMMU group and a vfio group.
116 vfio_ccw also provides an I/O region to accept channel program
117 request from user space and store I/O interrupt result for user
118 space to retrieve. To notify user space an I/O completion, it offers
119 an interface to setup an eventfd fd for asynchronous signaling.
121 - The vfio_mdev driver for the mediated vfio ccw device.
122 This is provided by the mdev framework. It is a vfio device driver for
123 the mdev that created by vfio_ccw.
124 It realize a group of vfio device driver callbacks, adds itself to a
125 vfio group, and registers itself to the mdev framework as a mdev
126 driver.
127 It uses a vfio iommu backend that uses the existing map and unmap
128 ioctls, but rather than programming them into an IOMMU for a device,
129 it simply stores the translations for use by later requests. This
130 means that a device programmed in a VM with guest physical addresses
131 can have the vfio kernel convert that address to process virtual
132 address, pin the page and program the hardware with the host physical
133 address in one step.
134 For a mdev, the vfio iommu backend will not pin the pages during the
135 VFIO_IOMMU_MAP_DMA ioctl. Mdev framework will only maintain a database
136 of the iova<->vaddr mappings in this operation. And they export a
137 vfio_pin_pages and a vfio_unpin_pages interfaces from the vfio iommu
138 backend for the physical devices to pin and unpin pages by demand.
140 Below is a high Level block diagram.
142 +-------------+
143 | |
144 | +---------+ | mdev_register_driver() +--------------+
145 | | Mdev | +<-----------------------+ |
146 | | bus | | | vfio_mdev.ko |
147 | | driver | +----------------------->+ |<-> VFIO user
148 | +---------+ | probe()/remove() +--------------+ APIs
149 | |
150 | MDEV CORE |
151 | MODULE |
152 | mdev.ko |
153 | +---------+ | mdev_register_device() +--------------+
154 | |Physical | +<-----------------------+ |
155 | | device | | | vfio_ccw.ko |<-> subchannel
156 | |interface| +----------------------->+ | device
157 | +---------+ | callback +--------------+
158 +-------------+
160 The process of how these work together.
161 1. vfio_ccw.ko drives the physical I/O subchannel, and registers the
162 physical device (with callbacks) to mdev framework.
163 When vfio_ccw probing the subchannel device, it registers device
164 pointer and callbacks to the mdev framework. Mdev related file nodes
165 under the device node in sysfs would be created for the subchannel
166 device, namely 'mdev_create', 'mdev_destroy' and
167 'mdev_supported_types'.
168 2. Create a mediated vfio ccw device.
169 Use the 'mdev_create' sysfs file, we need to manually create one (and
170 only one for our case) mediated device.
171 3. vfio_mdev.ko drives the mediated ccw device.
172 vfio_mdev is also the vfio device drvier. It will probe the mdev and
173 add it to an iommu_group and a vfio_group. Then we could pass through
174 the mdev to a guest.
176 vfio-ccw I/O region
177 -------------------
179 An I/O region is used to accept channel program request from user
180 space and store I/O interrupt result for user space to retrieve. The
181 defination of the region is:
183 struct ccw_io_region {
184 #define ORB_AREA_SIZE 12
185 __u8 orb_area[ORB_AREA_SIZE];
186 #define SCSW_AREA_SIZE 12
187 __u8 scsw_area[SCSW_AREA_SIZE];
188 #define IRB_AREA_SIZE 96
189 __u8 irb_area[IRB_AREA_SIZE];
190 __u32 ret_code;
191 } __packed;
193 While starting an I/O request, orb_area should be filled with the
194 guest ORB, and scsw_area should be filled with the SCSW of the Virtual
195 Subchannel.
197 irb_area stores the I/O result.
199 ret_code stores a return code for each access of the region.
201 vfio-ccw patches overview
202 -------------------------
204 For now, our patches are rebased on the latest mdev implementation.
205 vfio-ccw follows what vfio-pci did on the s390 paltform and uses
206 vfio-iommu-type1 as the vfio iommu backend. It's a good start to launch
207 the code review for vfio-ccw. Note that the implementation is far from
208 complete yet; but we'd like to get feedback for the general
209 architecture.
211 * CCW translation APIs
212 - Description:
213 These introduce a group of APIs (start with 'cp_') to do CCW
214 translation. The CCWs passed in by a user space program are
215 organized with their guest physical memory addresses. These APIs
216 will copy the CCWs into the kernel space, and assemble a runnable
217 kernel channel program by updating the guest physical addresses with
218 their corresponding host physical addresses.
219 - Patches:
220 vfio: ccw: introduce channel program interfaces
222 * vfio_ccw device driver
223 - Description:
224 The following patches utilizes the CCW translation APIs and introduce
225 vfio_ccw, which is the driver for the I/O subchannel devices you want
226 to pass through.
227 vfio_ccw implements the following vfio ioctls:
228 VFIO_DEVICE_GET_INFO
229 VFIO_DEVICE_GET_IRQ_INFO
230 VFIO_DEVICE_GET_REGION_INFO
231 VFIO_DEVICE_RESET
232 VFIO_DEVICE_SET_IRQS
233 This provides an I/O region, so that the user space program can pass a
234 channel program to the kernel, to do further CCW translation before
235 issuing them to a real device.
236 This also provides the SET_IRQ ioctl to setup an event notifier to
237 notify the user space program the I/O completion in an asynchronous
238 way.
239 - Patches:
240 vfio: ccw: basic implementation for vfio_ccw driver
241 vfio: ccw: introduce ccw_io_region
242 vfio: ccw: realize VFIO_DEVICE_GET_REGION_INFO ioctl
243 vfio: ccw: realize VFIO_DEVICE_RESET ioctl
244 vfio: ccw: realize VFIO_DEVICE_G(S)ET_IRQ_INFO ioctls
246 The user of vfio-ccw is not limited to Qemu, while Qemu is definitely a
247 good example to get understand how these patches work. Here is a little
248 bit more detail how an I/O request triggered by the Qemu guest will be
249 handled (without error handling).
251 Explanation:
252 Q1-Q7: Qemu side process.
253 K1-K5: Kernel side process.
255 Q1. Get I/O region info during initialization.
256 Q2. Setup event notifier and handler to handle I/O completion.
258 ... ...
260 Q3. Intercept a ssch instruction.
261 Q4. Write the guest channel program and ORB to the I/O region.
262 K1. Copy from guest to kernel.
263 K2. Translate the guest channel program to a host kernel space
264 channel program, which becomes runnable for a real device.
265 K3. With the necessary information contained in the orb passed in
266 by Qemu, issue the ccwchain to the device.
267 K4. Return the ssch CC code.
268 Q5. Return the CC code to the guest.
270 ... ...
272 K5. Interrupt handler gets the I/O result and write the result to
273 the I/O region.
274 K6. Signal Qemu to retrieve the result.
275 Q6. Get the signal and event handler reads out the result from the I/O
276 region.
277 Q7. Update the irb for the guest.
279 Limitations
280 -----------
282 The current vfio-ccw implementation focuses on supporting basic commands
283 needed to implement block device functionality (read/write) of DASD/ECKD
284 device only. Some commands may need special handling in the future, for
285 example, anything related to path grouping.
287 DASD is a kind of storage device. While ECKD is a data recording format.
288 More information for DASD and ECKD could be found here:
289 https://en.wikipedia.org/wiki/Direct-access_storage_device
290 https://en.wikipedia.org/wiki/Count_key_data
292 Together with the corresponding work in Qemu, we can bring the passed
293 through DASD/ECKD device online in a guest now and use it as a block
294 device.
296 Reference
297 ---------
298 1. ESA/s390 Principles of Operation manual (IBM Form. No. SA22-7832)
299 2. ESA/390 Common I/O Device Commands manual (IBM Form. No. SA22-7204)
300 3. https://en.wikipedia.org/wiki/Channel_I/O
301 4. Documentation/s390/cds.txt
302 5. Documentation/vfio.txt
303 6. Documentation/vfio-mediated-device.txt