Documentation/s390/vfio-ccw.rst

   1 ==================================
   2 vfio-ccw: the basic infrastructure
   3 ==================================
   4
   5 Introduction
   6 ------------
   7
   8 Here we describe the vfio support for I/O subchannel devices for
   9 Linux/s390. Motivation for vfio-ccw is to passthrough subchannels to a
  10 virtual machine, while vfio is the means.
  11
  12 Different than other hardware architectures, s390 has defined a unified
  13 I/O access method, which is so called Channel I/O. It has its own access
  14 patterns:
  15
  16 - Channel programs run asynchronously on a separate (co)processor.
  17 - The channel subsystem will access any memory designated by the caller
  18   in the channel program directly, i.e. there is no iommu involved.
  19
  20 Thus when we introduce vfio support for these devices, we realize it
  21 with a mediated device (mdev) implementation. The vfio mdev will be
  22 added to an iommu group, so as to make itself able to be managed by the
  23 vfio framework. And we add read/write callbacks for special vfio I/O
  24 regions to pass the channel programs from the mdev to its parent device
  25 (the real I/O subchannel device) to do further address translation and
  26 to perform I/O instructions.
  27
  28 This document does not intend to explain the s390 I/O architecture in
  29 every detail. More information/reference could be found here:
  30
  31 - A good start to know Channel I/O in general:
  32   https://en.wikipedia.org/wiki/Channel_I/O
  33 - s390 architecture:
  34   s390 Principles of Operation manual (IBM Form. No. SA22-7832)
  35 - The existing QEMU code which implements a simple emulated channel
  36   subsystem could also be a good reference. It makes it easier to follow
  37   the flow.
  38   qemu/hw/s390x/css.c
  39
  40 For vfio mediated device framework:
  41 - Documentation/driver-api/vfio-mediated-device.rst
  42
  43 Motivation of vfio-ccw
  44 ----------------------
  45
  46 Typically, a guest virtualized via QEMU/KVM on s390 only sees
  47 paravirtualized virtio devices via the "Virtio Over Channel I/O
  48 (virtio-ccw)" transport. This makes virtio devices discoverable via
  49 standard operating system algorithms for handling channel devices.
  50
  51 However this is not enough. On s390 for the majority of devices, which
  52 use the standard Channel I/O based mechanism, we also need to provide
  53 the functionality of passing through them to a QEMU virtual machine.
  54 This includes devices that don't have a virtio counterpart (e.g. tape
  55 drives) or that have specific characteristics which guests want to
  56 exploit.
  57
  58 For passing a device to a guest, we want to use the same interface as
  59 everybody else, namely vfio. We implement this vfio support for channel
  60 devices via the vfio mediated device framework and the subchannel device
  61 driver "vfio_ccw".
  62
  63 Access patterns of CCW devices
  64 ------------------------------
  65
  66 s390 architecture has implemented a so called channel subsystem, that
  67 provides a unified view of the devices physically attached to the
  68 systems. Though the s390 hardware platform knows about a huge variety of
  69 different peripheral attachments like disk devices (aka. DASDs), tapes,
  70 communication controllers, etc. They can all be accessed by a well
  71 defined access method and they are presenting I/O completion a unified
  72 way: I/O interruptions.
  73
  74 All I/O requires the use of channel command words (CCWs). A CCW is an
  75 instruction to a specialized I/O channel processor. A channel program is
  76 a sequence of CCWs which are executed by the I/O channel subsystem.  To
  77 issue a channel program to the channel subsystem, it is required to
  78 build an operation request block (ORB), which can be used to point out
  79 the format of the CCW and other control information to the system. The
  80 operating system signals the I/O channel subsystem to begin executing
  81 the channel program with a SSCH (start sub-channel) instruction. The
  82 central processor is then free to proceed with non-I/O instructions
  83 until interrupted. The I/O completion result is received by the
  84 interrupt handler in the form of interrupt response block (IRB).
  85
  86 Back to vfio-ccw, in short:
  87
  88 - ORBs and channel programs are built in guest kernel (with guest
  89   physical addresses).
  90 - ORBs and channel programs are passed to the host kernel.
  91 - Host kernel translates the guest physical addresses to real addresses
  92   and starts the I/O with issuing a privileged Channel I/O instruction
  93   (e.g SSCH).
  94 - channel programs run asynchronously on a separate processor.
  95 - I/O completion will be signaled to the host with I/O interruptions.
  96   And it will be copied as IRB to user space to pass it back to the
  97   guest.
  98
  99 Physical vfio ccw device and its child mdev
 100 -------------------------------------------
 101
 102 As mentioned above, we realize vfio-ccw with a mdev implementation.
 103
 104 Channel I/O does not have IOMMU hardware support, so the physical
 105 vfio-ccw device does not have an IOMMU level translation or isolation.
 106
 107 Subchannel I/O instructions are all privileged instructions. When
 108 handling the I/O instruction interception, vfio-ccw has the software
 109 policing and translation how the channel program is programmed before
 110 it gets sent to hardware.
 111
 112 Within this implementation, we have two drivers for two types of
 113 devices:
 114
 115 - The vfio_ccw driver for the physical subchannel device.
 116   This is an I/O subchannel driver for the real subchannel device.  It
 117   realizes a group of callbacks and registers to the mdev framework as a
 118   parent (physical) device. As a consequence, mdev provides vfio_ccw a
 119   generic interface (sysfs) to create mdev devices. A vfio mdev could be
 120   created by vfio_ccw then and added to the mediated bus. It is the vfio
 121   device that added to an IOMMU group and a vfio group.
 122   vfio_ccw also provides an I/O region to accept channel program
 123   request from user space and store I/O interrupt result for user
 124   space to retrieve. To notify user space an I/O completion, it offers
 125   an interface to setup an eventfd fd for asynchronous signaling.
 126
 127 - The vfio_mdev driver for the mediated vfio ccw device.
 128   This is provided by the mdev framework. It is a vfio device driver for
 129   the mdev that created by vfio_ccw.
 130   It realizes a group of vfio device driver callbacks, adds itself to a
 131   vfio group, and registers itself to the mdev framework as a mdev
 132   driver.
 133   It uses a vfio iommu backend that uses the existing map and unmap
 134   ioctls, but rather than programming them into an IOMMU for a device,
 135   it simply stores the translations for use by later requests. This
 136   means that a device programmed in a VM with guest physical addresses
 137   can have the vfio kernel convert that address to process virtual
 138   address, pin the page and program the hardware with the host physical
 139   address in one step.
 140   For a mdev, the vfio iommu backend will not pin the pages during the
 141   VFIO_IOMMU_MAP_DMA ioctl. Mdev framework will only maintain a database
 142   of the iova<->vaddr mappings in this operation. And they export a
 143   vfio_pin_pages and a vfio_unpin_pages interfaces from the vfio iommu
 144   backend for the physical devices to pin and unpin pages by demand.
 145
 146 Below is a high Level block diagram::
 147
 148  +-------------+
 149  |             |
 150  | +---------+ | mdev_register_driver() +--------------+
 151  | |  Mdev   | +<-----------------------+              |
 152  | |  bus    | |                        | vfio_mdev.ko |
 153  | | driver  | +----------------------->+              |<-> VFIO user
 154  | +---------+ |    probe()/remove()    +--------------+    APIs
 155  |             |
 156  |  MDEV CORE  |
 157  |   MODULE    |
 158  |   mdev.ko   |
 159  | +---------+ | mdev_register_device() +--------------+
 160  | |Physical | +<-----------------------+              |
 161  | | device  | |                        |  vfio_ccw.ko |<-> subchannel
 162  | |interface| +----------------------->+              |     device
 163  | +---------+ |       callback         +--------------+
 164  +-------------+
 165
 166 The process of how these work together.
 167
 168 1. vfio_ccw.ko drives the physical I/O subchannel, and registers the
 169    physical device (with callbacks) to mdev framework.
 170    When vfio_ccw probing the subchannel device, it registers device
 171    pointer and callbacks to the mdev framework. Mdev related file nodes
 172    under the device node in sysfs would be created for the subchannel
 173    device, namely 'mdev_create', 'mdev_destroy' and
 174    'mdev_supported_types'.
 175 2. Create a mediated vfio ccw device.
 176    Use the 'mdev_create' sysfs file, we need to manually create one (and
 177    only one for our case) mediated device.
 178 3. vfio_mdev.ko drives the mediated ccw device.
 179    vfio_mdev is also the vfio device drvier. It will probe the mdev and
 180    add it to an iommu_group and a vfio_group. Then we could pass through
 181    the mdev to a guest.
 182
 183 vfio-ccw I/O region
 184 -------------------
 185
 186 An I/O region is used to accept channel program request from user
 187 space and store I/O interrupt result for user space to retrieve. The
 188 definition of the region is::
 189
 190   struct ccw_io_region {
 191   #define ORB_AREA_SIZE 12
 192           __u8    orb_area[ORB_AREA_SIZE];
 193   #define SCSW_AREA_SIZE 12
 194           __u8    scsw_area[SCSW_AREA_SIZE];
 195   #define IRB_AREA_SIZE 96
 196           __u8    irb_area[IRB_AREA_SIZE];
 197           __u32   ret_code;
 198   } __packed;
 199
 200 While starting an I/O request, orb_area should be filled with the
 201 guest ORB, and scsw_area should be filled with the SCSW of the Virtual
 202 Subchannel.
 203
 204 irb_area stores the I/O result.
 205
 206 ret_code stores a return code for each access of the region.
 207
 208 vfio-ccw operation details
 209 --------------------------
 210
 211 vfio-ccw follows what vfio-pci did on the s390 platform and uses
 212 vfio-iommu-type1 as the vfio iommu backend.
 213
 214 * CCW translation APIs
 215   A group of APIs (start with `cp_`) to do CCW translation. The CCWs
 216   passed in by a user space program are organized with their guest
 217   physical memory addresses. These APIs will copy the CCWs into kernel
 218   space, and assemble a runnable kernel channel program by updating the
 219   guest physical addresses with their corresponding host physical addresses.
 220   Note that we have to use IDALs even for direct-access CCWs, as the
 221   referenced memory can be located anywhere, including above 2G.
 222
 223 * vfio_ccw device driver
 224   This driver utilizes the CCW translation APIs and introduces
 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
 229     VFIO_DEVICE_GET_INFO
 230     VFIO_DEVICE_GET_IRQ_INFO
 231     VFIO_DEVICE_GET_REGION_INFO
 232     VFIO_DEVICE_RESET
 233     VFIO_DEVICE_SET_IRQS
 234
 235   This provides an I/O region, so that the user space program can pass a
 236   channel program to the kernel, to do further CCW translation before
 237   issuing them to a real device.
 238   This also provides the SET_IRQ ioctl to setup an event notifier to
 239   notify the user space program the I/O completion in an asynchronous
 240   way.
 241
 242 The use of vfio-ccw is not limited to QEMU, while QEMU is definitely a
 243 good example to get understand how these patches work. Here is a little
 244 bit more detail how an I/O request triggered by the QEMU guest will be
 245 handled (without error handling).
 246
 247 Explanation:
 248
 249 - Q1-Q7: QEMU side process.
 250 - K1-K5: Kernel side process.
 251
 252 Q1.
 253     Get I/O region info during initialization.
 254
 255 Q2.
 256     Setup event notifier and handler to handle I/O completion.
 257
 258 ... ...
 259
 260 Q3.
 261     Intercept a ssch instruction.
 262 Q4.
 263     Write the guest channel program and ORB to the I/O region.
 264
 265     K1.
 266         Copy from guest to kernel.
 267     K2.
 268         Translate the guest channel program to a host kernel space
 269         channel program, which becomes runnable for a real device.
 270     K3.
 271         With the necessary information contained in the orb passed in
 272         by QEMU, issue the ccwchain to the device.
 273     K4.
 274         Return the ssch CC code.
 275 Q5.
 276     Return the CC code to the guest.
 277
 278 ... ...
 279
 280     K5.
 281         Interrupt handler gets the I/O result and write the result to
 282         the I/O region.
 283     K6.
 284         Signal QEMU to retrieve the result.
 285
 286 Q6.
 287     Get the signal and event handler reads out the result from the I/O
 288     region.
 289 Q7.
 290     Update the irb for the guest.
 291
 292 Limitations
 293 -----------
 294
 295 The current vfio-ccw implementation focuses on supporting basic commands
 296 needed to implement block device functionality (read/write) of DASD/ECKD
 297 device only. Some commands may need special handling in the future, for
 298 example, anything related to path grouping.
 299
 300 DASD is a kind of storage device. While ECKD is a data recording format.
 301 More information for DASD and ECKD could be found here:
 302 https://en.wikipedia.org/wiki/Direct-access_storage_device
 303 https://en.wikipedia.org/wiki/Count_key_data
 304
 305 Together with the corresponding work in QEMU, we can bring the passed
 306 through DASD/ECKD device online in a guest now and use it as a block
 307 device.
 308
 309 While the current code allows the guest to start channel programs via
 310 START SUBCHANNEL, support for HALT SUBCHANNEL or CLEAR SUBCHANNEL is
 311 not yet implemented.
 312
 313 vfio-ccw supports classic (command mode) channel I/O only. Transport
 314 mode (HPF) is not supported.
 315
 316 QDIO subchannels are currently not supported. Classic devices other than
 317 DASD/ECKD might work, but have not been tested.
 318
 319 Reference
 320 ---------
 321 1. ESA/s390 Principles of Operation manual (IBM Form. No. SA22-7832)
 322 2. ESA/390 Common I/O Device Commands manual (IBM Form. No. SA22-7204)
 323 3. https://en.wikipedia.org/wiki/Channel_I/O
 324 4. Documentation/s390/cds.rst
 325 5. Documentation/driver-api/vfio.rst
 326 6. Documentation/driver-api/vfio-mediated-device.rst