2Linux for S/390 and zSeries
   5Common Device Support (CDS)
   6Device Driver I/O Support Routines
   9        - Ingo Adlung
  10        - Cornelia Huck
  12Copyright, IBM Corp. 1999-2002
  17This document describes the common device support routines for Linux/390.
  18Different than other hardware architectures, ESA/390 has defined a unified
  19I/O access method. This gives relief to the device drivers as they don't
  20have to deal with different bus types, polling versus interrupt
  21processing, shared versus non-shared interrupt processing, DMA versus port
  22I/O (PIO), and other hardware features more. However, this implies that
  23either every single device driver needs to implement the hardware I/O
  24attachment functionality itself, or the operating system provides for a
  25unified method to access the hardware, providing all the functionality that
  26every single device driver would have to provide itself.
  28The document does not intend to explain the ESA/390 hardware architecture in
  29every detail.This information can be obtained from the ESA/390 Principles of
  30Operation manual (IBM Form. No. SA22-7201).
  32In order to build common device support for ESA/390 I/O interfaces, a
  33functional layer was introduced that provides generic I/O access methods to
  34the hardware.
  36The common device support layer comprises the I/O support routines defined
  37below. Some of them implement common Linux device driver interfaces, while
  38some of them are ESA/390 platform specific.
  41  In order to write a driver for S/390, you also need to look into the interface
  42  described in Documentation/s390/driver-model.rst.
  44Note for porting drivers from 2.4:
  46The major changes are:
  48* The functions use a ccw_device instead of an irq (subchannel).
  49* All drivers must define a ccw_driver (see driver-model.txt) and the associated
  50  functions.
  51* request_irq() and free_irq() are no longer done by the driver.
  52* The oper_handler is (kindof) replaced by the probe() and set_online() functions
  53  of the ccw_driver.
  54* The not_oper_handler is (kindof) replaced by the remove() and set_offline()
  55  functions of the ccw_driver.
  56* The channel device layer is gone.
  57* The interrupt handlers must be adapted to use a ccw_device as argument.
  58  Moreover, they don't return a devstat, but an irb.
  59* Before initiating an io, the options must be set via ccw_device_set_options().
  60* Instead of calling read_dev_chars()/read_conf_data(), the driver issues
  61  the channel program and handles the interrupt itself.
  64   get commands from extended sense data.
  66ccw_device_start(), ccw_device_start_timeout(), ccw_device_start_key(), ccw_device_start_key_timeout()
  67   initiate an I/O request.
  70   resume channel program execution.
  73   terminate the current I/O request processed on the device.
  76   generic interrupt routine. This function is called by the interrupt entry
  77   routine whenever an I/O interrupt is presented to the system. The do_IRQ()
  78   routine determines the interrupt status and calls the device specific
  79   interrupt handler according to the rules (flags) defined during I/O request
  80   initiation with do_IO().
  82The next chapters describe the functions other than do_IRQ() in more details.
  83The do_IRQ() interface is not described, as it is called from the Linux/390
  84first level interrupt handler only and does not comprise a device driver
  85callable interface. Instead, the functional description of do_IO() also
  86describes the input to the device specific interrupt handler.
  89        All explanations apply also to the 64 bit architecture s390x.
  92Common Device Support (CDS) for Linux/390 Device Drivers
  95General Information
  98The following chapters describe the I/O related interface routines the
  99Linux/390 common device support (CDS) provides to allow for device specific
 100driver implementations on the IBM ESA/390 hardware platform. Those interfaces
 101intend to provide the functionality required by every device driver
 102implementation to allow to drive a specific hardware device on the ESA/390
 103platform. Some of the interface routines are specific to Linux/390 and some
 104of them can be found on other Linux platforms implementations too.
 105Miscellaneous function prototypes, data declarations, and macro definitions
 106can be found in the architecture specific C header file
 109Overview of CDS interface concepts
 112Different to other hardware platforms, the ESA/390 architecture doesn't define
 113interrupt lines managed by a specific interrupt controller and bus systems
 114that may or may not allow for shared interrupts, DMA processing, etc.. Instead,
 115the ESA/390 architecture has implemented a so called channel subsystem, that
 116provides a unified view of the devices physically attached to the systems.
 117Though the ESA/390 hardware platform knows about a huge variety of different
 118peripheral attachments like disk devices (aka. DASDs), tapes, communication
 119controllers, etc. they can all be accessed by a well defined access method and
 120they are presenting I/O completion a unified way : I/O interruptions. Every
 121single device is uniquely identified to the system by a so called subchannel,
 122where the ESA/390 architecture allows for 64k devices be attached.
 124Linux, however, was first built on the Intel PC architecture, with its two
 125cascaded 8259 programmable interrupt controllers (PICs), that allow for a
 126maximum of 15 different interrupt lines. All devices attached to such a system
 127share those 15 interrupt levels. Devices attached to the ISA bus system must
 128not share interrupt levels (aka. IRQs), as the ISA bus bases on edge triggered
 129interrupts. MCA, EISA, PCI and other bus systems base on level triggered
 130interrupts, and therewith allow for shared IRQs. However, if multiple devices
 131present their hardware status by the same (shared) IRQ, the operating system
 132has to call every single device driver registered on this IRQ in order to
 133determine the device driver owning the device that raised the interrupt.
 135Up to kernel 2.4, Linux/390 used to provide interfaces via the IRQ (subchannel).
 136For internal use of the common I/O layer, these are still there. However,
 137device drivers should use the new calling interface via the ccw_device only.
 139During its startup the Linux/390 system checks for peripheral devices. Each
 140of those devices is uniquely defined by a so called subchannel by the ESA/390
 141channel subsystem. While the subchannel numbers are system generated, each
 142subchannel also takes a user defined attribute, the so called device number.
 143Both subchannel number and device number cannot exceed 65535. During sysfs
 144initialisation, the information about control unit type and device types that
 145imply specific I/O commands (channel command words - CCWs) in order to operate
 146the device are gathered. Device drivers can retrieve this set of hardware
 147information during their initialization step to recognize the devices they
 148support using the information saved in the struct ccw_device given to them.
 149This methods implies that Linux/390 doesn't require to probe for free (not
 150armed) interrupt request lines (IRQs) to drive its devices with. Where
 151applicable, the device drivers can use issue the READ DEVICE CHARACTERISTICS
 152ccw to retrieve device characteristics in its online routine.
 154In order to allow for easy I/O initiation the CDS layer provides a
 155ccw_device_start() interface that takes a device specific channel program (one
 156or more CCWs) as input sets up the required architecture specific control blocks
 157and initiates an I/O request on behalf of the device driver. The
 158ccw_device_start() routine allows to specify whether it expects the CDS layer
 159to notify the device driver for every interrupt it observes, or with final status
 160only. See ccw_device_start() for more details. A device driver must never issue
 161ESA/390 I/O commands itself, but must use the Linux/390 CDS interfaces instead.
 163For long running I/O request to be canceled, the CDS layer provides the
 164ccw_device_halt() function. Some devices require to initially issue a HALT
 165SUBCHANNEL (HSCH) command without having pending I/O requests. This function is
 166also covered by ccw_device_halt().
 169get_ciw() - get command information word
 171This call enables a device driver to get information about supported commands
 172from the extended SenseID data.
 176  struct ciw *
 177  ccw_device_get_ciw(struct ccw_device *cdev, __u32 cmd);
 179====  ========================================================
 180cdev  The ccw_device for which the command is to be retrieved.
 181cmd   The command type to be retrieved.
 182====  ========================================================
 184ccw_device_get_ciw() returns:
 186=====  ================================================================
 187 NULL  No extended data available, invalid device or command not found.
 188!NULL  The command requested.
 189=====  ================================================================
 193  ccw_device_start() - Initiate I/O Request
 195The ccw_device_start() routines is the I/O request front-end processor. All
 196device driver I/O requests must be issued using this routine. A device driver
 197must not issue ESA/390 I/O commands itself. Instead the ccw_device_start()
 198routine provides all interfaces required to drive arbitrary devices.
 200This description also covers the status information passed to the device
 201driver's interrupt handler as this is related to the rules (flags) defined
 202with the associated I/O request when calling ccw_device_start().
 206  int ccw_device_start(struct ccw_device *cdev,
 207                       struct ccw1 *cpa,
 208                       unsigned long intparm,
 209                       __u8 lpm,
 210                       unsigned long flags);
 211  int ccw_device_start_timeout(struct ccw_device *cdev,
 212                               struct ccw1 *cpa,
 213                               unsigned long intparm,
 214                               __u8 lpm,
 215                               unsigned long flags,
 216                               int expires);
 217  int ccw_device_start_key(struct ccw_device *cdev,
 218                           struct ccw1 *cpa,
 219                           unsigned long intparm,
 220                           __u8 lpm,
 221                           __u8 key,
 222                           unsigned long flags);
 223  int ccw_device_start_key_timeout(struct ccw_device *cdev,
 224                                   struct ccw1 *cpa,
 225                                   unsigned long intparm,
 226                                   __u8 lpm,
 227                                   __u8 key,
 228                                   unsigned long flags,
 229                                   int expires);
 231============= =============================================================
 232cdev          ccw_device the I/O is destined for
 233cpa           logical start address of channel program
 234user_intparm  user specific interrupt information; will be presented
 235              back to the device driver's interrupt handler. Allows a
 236              device driver to associate the interrupt with a
 237              particular I/O request.
 238lpm           defines the channel path to be used for a specific I/O
 239              request. A value of 0 will make cio use the opm.
 240key           the storage key to use for the I/O (useful for operating on a
 241              storage with a storage key != default key)
 242flag          defines the action to be performed for I/O processing
 243expires       timeout value in jiffies. The common I/O layer will terminate
 244              the running program after this and call the interrupt handler
 245              with ERR_PTR(-ETIMEDOUT) as irb.
 246============= =============================================================
 248Possible flag values are:
 250========================= =============================================
 251DOIO_ALLOW_SUSPEND        channel program may become suspended
 252DOIO_DENY_PREFETCH        don't allow for CCW prefetch; usually
 253                          this implies the channel program might
 254                          become modified
 255DOIO_SUPPRESS_INTER       don't call the handler on intermediate status
 256========================= =============================================
 258The cpa parameter points to the first format 1 CCW of a channel program::
 260  struct ccw1 {
 261        __u8  cmd_code;/* command code */
 262        __u8  flags;   /* flags, like IDA addressing, etc. */
 263        __u16 count;   /* byte count */
 264        __u32 cda;     /* data address */
 265  } __attribute__ ((packed,aligned(8)));
 267with the following CCW flags values defined:
 269=================== =========================
 270CCW_FLAG_DC         data chaining
 271CCW_FLAG_CC         command chaining
 272CCW_FLAG_SLI        suppress incorrect length
 273CCW_FLAG_SKIP       skip
 274CCW_FLAG_PCI        PCI
 275CCW_FLAG_IDA        indirect addressing
 276CCW_FLAG_SUSPEND    suspend
 277=================== =========================
 280Via ccw_device_set_options(), the device driver may specify the following
 281options for the device:
 283========================= ======================================
 284DOIO_EARLY_NOTIFICATION   allow for early interrupt notification
 285DOIO_REPORT_ALL           report all interrupt conditions
 286========================= ======================================
 289The ccw_device_start() function returns:
 291======== ======================================================================
 292      0  successful completion or request successfully initiated
 293 -EBUSY  The device is currently processing a previous I/O request, or there is
 294         a status pending at the device.
 295-ENODEV  cdev is invalid, the device is not operational or the ccw_device is
 296         not online.
 297======== ======================================================================
 299When the I/O request completes, the CDS first level interrupt handler will
 300accumulate the status in a struct irb and then call the device interrupt handler.
 301The intparm field will contain the value the device driver has associated with a
 302particular I/O request. If a pending device status was recognized,
 303intparm will be set to 0 (zero). This may happen during I/O initiation or delayed
 304by an alert status notification. In any case this status is not related to the
 305current (last) I/O request. In case of a delayed status notification no special
 306interrupt will be presented to indicate I/O completion as the I/O request was
 307never started, even though ccw_device_start() returned with successful completion.
 309The irb may contain an error value, and the device driver should check for this
 312========== =================================================================
 313-ETIMEDOUT the common I/O layer terminated the request after the specified
 314           timeout value
 315-EIO       the common I/O layer terminated the request due to an error state
 316========== =================================================================
 318If the concurrent sense flag in the extended status word (esw) in the irb is
 319set, the field erw.scnt in the esw describes the number of device specific
 320sense bytes available in the extended control word irb->scsw.ecw[]. No device
 321sensing by the device driver itself is required.
 323The device interrupt handler can use the following definitions to investigate
 324the primary unit check source coded in sense byte 0 :
 326======================= ====
 327SNS0_CMD_REJECT         0x80
 329SNS0_BUS_OUT_CHECK      0x20
 331SNS0_DATA_CHECK         0x08
 332SNS0_OVERRUN            0x04
 333SNS0_INCOMPL_DOMAIN     0x01
 334======================= ====
 336Depending on the device status, multiple of those values may be set together.
 337Please refer to the device specific documentation for details.
 339The irb->scsw.cstat field provides the (accumulated) subchannel status :
 341========================= ============================
 342SCHN_STAT_PCI             program controlled interrupt
 343SCHN_STAT_INCORR_LEN      incorrect length
 344SCHN_STAT_PROG_CHECK      program check
 345SCHN_STAT_PROT_CHECK      protection check
 346SCHN_STAT_CHN_DATA_CHK    channel data check
 347SCHN_STAT_CHN_CTRL_CHK    channel control check
 348SCHN_STAT_INTF_CTRL_CHK   interface control check
 349SCHN_STAT_CHAIN_CHECK     chaining check
 350========================= ============================
 352The irb->scsw.dstat field provides the (accumulated) device status :
 354===================== =================
 355DEV_STAT_ATTENTION    attention
 356DEV_STAT_STAT_MOD     status modifier
 357DEV_STAT_CU_END       control unit end
 358DEV_STAT_BUSY         busy
 359DEV_STAT_CHN_END      channel end
 360DEV_STAT_DEV_END      device end
 361DEV_STAT_UNIT_CHECK   unit check
 362DEV_STAT_UNIT_EXCEP   unit exception
 363===================== =================
 365Please see the ESA/390 Principles of Operation manual for details on the
 366individual flag meanings.
 368Usage Notes:
 370ccw_device_start() must be called disabled and with the ccw device lock held.
 372The device driver is allowed to issue the next ccw_device_start() call from
 373within its interrupt handler already. It is not required to schedule a
 374bottom-half, unless a non deterministically long running error recovery procedure
 375or similar needs to be scheduled. During I/O processing the Linux/390 generic
 376I/O device driver support has already obtained the IRQ lock, i.e. the handler
 377must not try to obtain it again when calling ccw_device_start() or we end in a
 378deadlock situation!
 380If a device driver relies on an I/O request to be completed prior to start the
 381next it can reduce I/O processing overhead by chaining a NoOp I/O command
 382CCW_CMD_NOOP to the end of the submitted CCW chain. This will force Channel-End
 383and Device-End status to be presented together, with a single interrupt.
 384However, this should be used with care as it implies the channel will remain
 385busy, not being able to process I/O requests for other devices on the same
 386channel. Therefore e.g. read commands should never use this technique, as the
 387result will be presented by a single interrupt anyway.
 389In order to minimize I/O overhead, a device driver should use the
 390DOIO_REPORT_ALL  only if the device can report intermediate interrupt
 391information prior to device-end the device driver urgently relies on. In this
 392case all I/O interruptions are presented to the device driver until final
 393status is recognized.
 395If a device is able to recover from asynchronously presented I/O errors, it can
 396perform overlapping I/O using the DOIO_EARLY_NOTIFICATION flag. While some
 397devices always report channel-end and device-end together, with a single
 398interrupt, others present primary status (channel-end) when the channel is
 399ready for the next I/O request and secondary status (device-end) when the data
 400transmission has been completed at the device.
 402Above flag allows to exploit this feature, e.g. for communication devices that
 403can handle lost data on the network to allow for enhanced I/O processing.
 405Unless the channel subsystem at any time presents a secondary status interrupt,
 406exploiting this feature will cause only primary status interrupts to be
 407presented to the device driver while overlapping I/O is performed. When a
 408secondary status without error (alert status) is presented, this indicates
 409successful completion for all overlapping ccw_device_start() requests that have
 410been issued since the last secondary (final) status.
 412Channel programs that intend to set the suspend flag on a channel command word
 413(CCW)  must start the I/O operation with the DOIO_ALLOW_SUSPEND option or the
 414suspend flag will cause a channel program check. At the time the channel program
 415becomes suspended an intermediate interrupt will be generated by the channel
 418ccw_device_resume() - Resume Channel Program Execution
 420If a device driver chooses to suspend the current channel program execution by
 421setting the CCW suspend flag on a particular CCW, the channel program execution
 422is suspended. In order to resume channel program execution the CIO layer
 423provides the ccw_device_resume() routine.
 427  int ccw_device_resume(struct ccw_device *cdev);
 429====  ================================================
 430cdev  ccw_device the resume operation is requested for
 431====  ================================================
 433The ccw_device_resume() function returns:
 435=========   ==============================================
 436        0   suspended channel program is resumed
 437   -EBUSY   status pending
 438  -ENODEV   cdev invalid or not-operational subchannel
 439  -EINVAL   resume function not applicable
 440-ENOTCONN   there is no I/O request pending for completion
 441=========   ==============================================
 443Usage Notes:
 445Please have a look at the ccw_device_start() usage notes for more details on
 446suspended channel programs.
 448ccw_device_halt() - Halt I/O Request Processing
 450Sometimes a device driver might need a possibility to stop the processing of
 451a long-running channel program or the device might require to initially issue
 452a halt subchannel (HSCH) I/O command. For those purposes the ccw_device_halt()
 453command is provided.
 455ccw_device_halt() must be called disabled and with the ccw device lock held.
 459  int ccw_device_halt(struct ccw_device *cdev,
 460                      unsigned long intparm);
 462=======  =====================================================
 463cdev     ccw_device the halt operation is requested for
 464intparm  interruption parameter; value is only used if no I/O
 465         is outstanding, otherwise the intparm associated with
 466         the I/O request is returned
 467=======  =====================================================
 469The ccw_device_halt() function returns:
 471=======  ==============================================================
 472      0  request successfully initiated
 473-EBUSY   the device is currently busy, or status pending.
 474-ENODEV  cdev invalid.
 475-EINVAL  The device is not operational or the ccw device is not online.
 476=======  ==============================================================
 478Usage Notes:
 480A device driver may write a never-ending channel program by writing a channel
 481program that at its end loops back to its beginning by means of a transfer in
 482channel (TIC)   command (CCW_CMD_TIC). Usually this is performed by network
 483device drivers by setting the PCI CCW flag (CCW_FLAG_PCI). Once this CCW is
 484executed a program controlled interrupt (PCI) is generated. The device driver
 485can then perform an appropriate action. Prior to interrupt of an outstanding
 486read to a network device (with or without PCI flag) a ccw_device_halt()
 487is required to end the pending operation.
 491  ccw_device_clear() - Terminage I/O Request Processing
 493In order to terminate all I/O processing at the subchannel, the clear subchannel
 494(CSCH) command is used. It can be issued via ccw_device_clear().
 496ccw_device_clear() must be called disabled and with the ccw device lock held.
 500  int ccw_device_clear(struct ccw_device *cdev, unsigned long intparm);
 502======= ===============================================
 503cdev    ccw_device the clear operation is requested for
 504intparm interruption parameter (see ccw_device_halt())
 505======= ===============================================
 507The ccw_device_clear() function returns:
 509=======  ==============================================================
 510      0  request successfully initiated
 511-ENODEV  cdev invalid
 512-EINVAL  The device is not operational or the ccw device is not online.
 513=======  ==============================================================
 515Miscellaneous Support Routines
 518This chapter describes various routines to be used in a Linux/390 device
 519driver programming environment.
 523Get the address of the device specific lock. This is then used in
 524spin_lock() / spin_unlock() calls.
 528  __u8 ccw_device_get_path_mask(struct ccw_device *cdev);
 530Get the mask of the path currently available for cdev.