linux/Documentation/scsi/scsi_mid_low_api.txt
<<
>>
Prefs
   1                          Linux Kernel 2.6 series
   2                 SCSI mid_level - lower_level driver interface
   3                 =============================================
   4
   5Introduction
   6============
   7This document outlines the interface between the Linux SCSI mid level and
   8SCSI lower level drivers. Lower level drivers (LLDs) are variously called 
   9host bus adapter (HBA) drivers and host drivers (HD). A "host" in this
  10context is a bridge between a computer IO bus (e.g. PCI or ISA) and a
  11single SCSI initiator port on a SCSI transport. An "initiator" port
  12(SCSI terminology, see SAM-3 at http://www.t10.org) sends SCSI commands
  13to "target" SCSI ports (e.g. disks). There can be many LLDs in a running
  14system, but only one per hardware type. Most LLDs can control one or more
  15SCSI HBAs. Some HBAs contain multiple hosts.
  16
  17In some cases the SCSI transport is an external bus that already has
  18its own subsystem in Linux (e.g. USB and ieee1394). In such cases the
  19SCSI subsystem LLD is a software bridge to the other driver subsystem.
  20Examples are the usb-storage driver (found in the drivers/usb/storage
  21directory) and the ieee1394/sbp2 driver (found in the drivers/ieee1394
  22directory).
  23
  24For example, the aic7xxx LLD controls Adaptec SCSI parallel interface
  25(SPI) controllers based on that company's 7xxx chip series. The aic7xxx
  26LLD can be built into the kernel or loaded as a module. There can only be
  27one aic7xxx LLD running in a Linux system but it may be controlling many 
  28HBAs. These HBAs might be either on PCI daughter-boards or built into 
  29the motherboard (or both). Some aic7xxx based HBAs are dual controllers
  30and thus represent two hosts. Like most modern HBAs, each aic7xxx host
  31has its own PCI device address. [The one-to-one correspondence between
  32a SCSI host and a PCI device is common but not required (e.g. with
  33ISA adapters).]
  34
  35The SCSI mid level isolates an LLD from other layers such as the SCSI
  36upper layer drivers and the block layer.
  37
  38This version of the document roughly matches linux kernel version 2.6.8 .
  39
  40Documentation
  41=============
  42There is a SCSI documentation directory within the kernel source tree, 
  43typically Documentation/scsi . Most documents are in plain
  44(i.e. ASCII) text. This file is named scsi_mid_low_api.txt and can be 
  45found in that directory. A more recent copy of this document may be found
  46at http://web.archive.org/web/20070107183357rn_1/sg.torque.net/scsi/. 
  47Many LLDs are documented there (e.g. aic7xxx.txt). The SCSI mid-level is
  48briefly described in scsi.txt which contains a url to a document 
  49describing the SCSI subsystem in the lk 2.4 series. Two upper level 
  50drivers have documents in that directory: st.txt (SCSI tape driver) and 
  51scsi-generic.txt (for the sg driver).
  52
  53Some documentation (or urls) for LLDs may be found in the C source code
  54or in the same directory as the C source code. For example to find a url
  55about the USB mass storage driver see the 
  56/usr/src/linux/drivers/usb/storage directory.
  57
  58Driver structure
  59================
  60Traditionally an LLD for the SCSI subsystem has been at least two files in
  61the drivers/scsi directory. For example, a driver called "xyz" has a header
  62file "xyz.h" and a source file "xyz.c". [Actually there is no good reason
  63why this couldn't all be in one file; the header file is superfluous.] Some
  64drivers that have been ported to several operating systems have more than
  65two files. For example the aic7xxx driver has separate files for generic 
  66and OS-specific code (e.g. FreeBSD and Linux). Such drivers tend to have
  67their own directory under the drivers/scsi directory.
  68
  69When a new LLD is being added to Linux, the following files (found in the
  70drivers/scsi directory) will need some attention: Makefile and Kconfig .
  71It is probably best to study how existing LLDs are organized.
  72
  73As the 2.5 series development kernels evolve into the 2.6 series
  74production series, changes are being introduced into this interface. An
  75example of this is driver initialization code where there are now 2 models
  76available. The older one, similar to what was found in the lk 2.4 series,
  77is based on hosts that are detected at HBA driver load time. This will be
  78referred to the "passive" initialization model. The newer model allows HBAs
  79to be hot plugged (and unplugged) during the lifetime of the LLD and will
  80be referred to as the "hotplug" initialization model. The newer model is
  81preferred as it can handle both traditional SCSI equipment that is
  82permanently connected as well as modern "SCSI" devices (e.g. USB or
  83IEEE 1394 connected digital cameras) that are hotplugged. Both 
  84initialization models are discussed in the following sections.
  85
  86An LLD interfaces to the SCSI subsystem several ways:
  87  a) directly invoking functions supplied by the mid level
  88  b) passing a set of function pointers to a registration function
  89     supplied by the mid level. The mid level will then invoke these
  90     functions at some point in the future. The LLD will supply
  91     implementations of these functions.
  92  c) direct access to instances of well known data structures maintained
  93     by the mid level
  94
  95Those functions in group a) are listed in a section entitled "Mid level
  96supplied functions" below.
  97
  98Those functions in group b) are listed in a section entitled "Interface
  99functions" below. Their function pointers are placed in the members of
 100"struct scsi_host_template", an instance of which is passed to
 101scsi_host_alloc() ** .  Those interface functions that the LLD does not 
 102wish to supply should have NULL placed in the corresponding member of 
 103struct scsi_host_template.  Defining an instance of struct 
 104scsi_host_template at file scope will cause NULL to be  placed in function
 105 pointer members not explicitly initialized.
 106
 107Those usages in group c) should be handled with care, especially in a
 108"hotplug" environment. LLDs should be aware of the lifetime of instances
 109that are shared with the mid level and other layers.
 110
 111All functions defined within an LLD and all data defined at file scope
 112should be static. For example the slave_alloc() function in an LLD
 113called "xxx" could be defined as 
 114"static int xxx_slave_alloc(struct scsi_device * sdev) { /* code */ }"
 115
 116** the scsi_host_alloc() function is a replacement for the rather vaguely
 117named scsi_register() function in most situations. The scsi_register()
 118and scsi_unregister() functions remain to support legacy LLDs that use
 119the passive initialization model.
 120
 121
 122Hotplug initialization model
 123============================
 124In this model an LLD controls when SCSI hosts are introduced and removed
 125from the SCSI subsystem. Hosts can be introduced as early as driver
 126initialization and removed as late as driver shutdown. Typically a driver
 127will respond to a sysfs probe() callback that indicates an HBA has been
 128detected. After confirming that the new device is one that the LLD wants
 129to control, the LLD will initialize the HBA and then register a new host
 130with the SCSI mid level.
 131
 132During LLD initialization the driver should register itself with the
 133appropriate IO bus on which it expects to find HBA(s) (e.g. the PCI bus).
 134This can probably be done via sysfs. Any driver parameters (especially
 135those that are writable after the driver is loaded) could also be
 136registered with sysfs at this point. The SCSI mid level first becomes
 137aware of an LLD when that LLD registers its first HBA.
 138
 139At some later time, the LLD becomes aware of an HBA and what follows
 140is a typical sequence of calls between the LLD and the mid level.
 141This example shows the mid level scanning the newly introduced HBA for 3 
 142scsi devices of which only the first 2 respond:
 143
 144     HBA PROBE: assume 2 SCSI devices found in scan
 145LLD                   mid level                    LLD
 146===-------------------=========--------------------===------
 147scsi_host_alloc()  -->
 148scsi_add_host()  ---->
 149scsi_scan_host()  -------+
 150                         |
 151                    slave_alloc()
 152                    slave_configure() -->  scsi_adjust_queue_depth()
 153                         |
 154                    slave_alloc()
 155                    slave_configure()
 156                         |
 157                    slave_alloc()   ***
 158                    slave_destroy() ***
 159------------------------------------------------------------
 160
 161If the LLD wants to adjust the default queue settings, it can invoke
 162scsi_adjust_queue_depth() in its slave_configure() routine.
 163
 164*** For scsi devices that the mid level tries to scan but do not
 165    respond, a slave_alloc(), slave_destroy() pair is called.
 166
 167When an HBA is being removed it could be as part of an orderly shutdown
 168associated with the LLD module being unloaded (e.g. with the "rmmod"
 169command) or in response to a "hot unplug" indicated by sysfs()'s
 170remove() callback being invoked. In either case, the sequence is the
 171same:
 172
 173        HBA REMOVE: assume 2 SCSI devices attached
 174LLD                      mid level                 LLD
 175===----------------------=========-----------------===------
 176scsi_remove_host() ---------+
 177                            |
 178                     slave_destroy()
 179                     slave_destroy()
 180scsi_host_put()
 181------------------------------------------------------------
 182                     
 183It may be useful for a LLD to keep track of struct Scsi_Host instances
 184(a pointer is returned by scsi_host_alloc()). Such instances are "owned"
 185by the mid-level.  struct Scsi_Host instances are freed from
 186scsi_host_put() when the reference count hits zero.
 187
 188Hot unplugging an HBA that controls a disk which is processing SCSI
 189commands on a mounted file system is an interesting situation. Reference
 190counting logic is being introduced into the mid level to cope with many
 191of the issues involved. See the section on reference counting below.
 192
 193
 194The hotplug concept may be extended to SCSI devices. Currently, when an
 195HBA is added, the scsi_scan_host() function causes a scan for SCSI devices
 196attached to the HBA's SCSI transport. On newer SCSI transports the HBA
 197may become aware of a new SCSI device _after_ the scan has completed.
 198An LLD can use this sequence to make the mid level aware of a SCSI device:
 199
 200                 SCSI DEVICE hotplug
 201LLD                   mid level                    LLD
 202===-------------------=========--------------------===------
 203scsi_add_device()  ------+
 204                         |
 205                    slave_alloc()
 206                    slave_configure()   [--> scsi_adjust_queue_depth()]
 207------------------------------------------------------------
 208
 209In a similar fashion, an LLD may become aware that a SCSI device has been
 210removed (unplugged) or the connection to it has been interrupted. Some
 211existing SCSI transports (e.g. SPI) may not become aware that a SCSI
 212device has been removed until a subsequent SCSI command fails which will
 213probably cause that device to be set offline by the mid level. An LLD that
 214detects the removal of a SCSI device can instigate its removal from
 215upper layers with this sequence:
 216
 217                  SCSI DEVICE hot unplug
 218LLD                      mid level                 LLD
 219===----------------------=========-----------------===------
 220scsi_remove_device() -------+
 221                            |
 222                     slave_destroy()
 223------------------------------------------------------------
 224
 225It may be useful for an LLD to keep track of struct scsi_device instances
 226(a pointer is passed as the parameter to slave_alloc() and
 227slave_configure() callbacks). Such instances are "owned" by the mid-level.
 228struct scsi_device instances are freed after slave_destroy().
 229
 230
 231Passive initialization model
 232============================
 233These older LLDs include a file called "scsi_module.c" [yes the ".c" is a
 234little surprising] in their source code. For that file to work an
 235instance of struct scsi_host_template with the name "driver_template"
 236needs to be defined. Here is a typical code sequence used in this model:
 237    static struct scsi_host_template driver_template = {
 238        ...
 239    };
 240    #include "scsi_module.c"
 241
 242The scsi_module.c file contains two functions:
 243  - init_this_scsi_driver() which is executed when the LLD is
 244    initialized (i.e. boot time or module load time)
 245  - exit_this_scsi_driver() which is executed when the LLD is shut
 246    down (i.e. module unload time)
 247Note: since these functions are tagged with __init and __exit qualifiers
 248an LLD should not call them explicitly (since the kernel does that).
 249
 250Here is an example of an initialization sequence when two hosts are
 251detected (so detect() returns 2) and the SCSI bus scan on each host
 252finds 1 SCSI device (and a second device does not respond).
 253
 254LLD                      mid level                 LLD
 255===----------------------=========-----------------===------
 256init_this_scsi_driver() ----+
 257                            |
 258                         detect()  -----------------+
 259                            |                       |
 260                            |                scsi_register()
 261                            |                scsi_register()
 262                            |
 263                      slave_alloc()
 264                      slave_configure()  -->  scsi_adjust_queue_depth()
 265                      slave_alloc()   ***
 266                      slave_destroy() ***
 267                            |
 268                      slave_alloc()
 269                      slave_configure()
 270                      slave_alloc()   ***
 271                      slave_destroy() ***
 272------------------------------------------------------------
 273
 274The mid level invokes scsi_adjust_queue_depth() with tagged queuing off and
 275"cmd_per_lun" for that host as the queue length. These settings can be
 276overridden by a slave_configure() supplied by the LLD.
 277
 278*** For scsi devices that the mid level tries to scan but do not
 279    respond, a slave_alloc(), slave_destroy() pair is called.
 280
 281Here is an LLD shutdown sequence:
 282
 283LLD                      mid level                 LLD
 284===----------------------=========-----------------===------
 285exit_this_scsi_driver() ----+
 286                            |
 287                     slave_destroy()
 288                        release()   -->   scsi_unregister()
 289                            |
 290                     slave_destroy()
 291                        release()   -->   scsi_unregister()
 292------------------------------------------------------------
 293
 294An LLD need not define slave_destroy() (i.e. it is optional). 
 295
 296The shortcoming of the "passive initialization model" is that host
 297registration and de-registration are (typically) tied to LLD initialization
 298and shutdown. Once the LLD is initialized then a new host that appears
 299(e.g. via hotplugging) cannot easily be added without a redundant
 300driver shutdown and re-initialization. It may be possible to write an LLD
 301that uses both initialization models.
 302
 303
 304Reference Counting
 305==================
 306The Scsi_Host structure has had reference counting infrastructure added.
 307This effectively spreads the ownership of struct Scsi_Host instances
 308across the various SCSI layers which use them. Previously such instances
 309were exclusively owned by the mid level. LLDs would not usually need to
 310directly manipulate these reference counts but there may be some cases
 311where they do.
 312
 313There are 3 reference counting functions of interest associated with
 314struct Scsi_Host:
 315  - scsi_host_alloc(): returns a pointer to new instance of struct 
 316        Scsi_Host which has its reference count ^^ set to 1
 317  - scsi_host_get(): adds 1 to the reference count of the given instance
 318  - scsi_host_put(): decrements 1 from the reference count of the given
 319        instance. If the reference count reaches 0 then the given instance
 320        is freed
 321
 322The Scsi_device structure has had reference counting infrastructure added.
 323This effectively spreads the ownership of struct Scsi_device instances
 324across the various SCSI layers which use them. Previously such instances
 325were exclusively owned by the mid level. See the access functions declared
 326towards the end of include/scsi/scsi_device.h . If an LLD wants to keep
 327a copy of a pointer to a Scsi_device instance it should use scsi_device_get()
 328to bump its reference count. When it is finished with the pointer it can
 329use scsi_device_put() to decrement its reference count (and potentially
 330delete it).
 331
 332^^ struct Scsi_Host actually has 2 reference counts which are manipulated
 333in parallel by these functions.
 334
 335
 336Conventions
 337===========
 338First, Linus Torvalds's thoughts on C coding style can be found in the
 339Documentation/CodingStyle file. 
 340
 341Next, there is a movement to "outlaw" typedefs introducing synonyms for 
 342struct tags. Both can be still found in the SCSI subsystem, but
 343the typedefs have been moved to a single file, scsi_typedefs.h to
 344make their future removal easier, for example: 
 345"typedef struct scsi_cmnd Scsi_Cmnd;"
 346
 347Also, most C99 enhancements are encouraged to the extent they are supported
 348by the relevant gcc compilers. So C99 style structure and array
 349initializers are encouraged where appropriate. Don't go too far,
 350VLAs are not properly supported yet.  An exception to this is the use of
 351"//" style comments; /*...*/ comments are still preferred in Linux.
 352
 353Well written, tested and documented code, need not be re-formatted to
 354comply with the above conventions. For example, the aic7xxx driver
 355comes to Linux from FreeBSD and Adaptec's own labs. No doubt FreeBSD
 356and Adaptec have their own coding conventions.
 357
 358
 359Mid level supplied functions
 360============================
 361These functions are supplied by the SCSI mid level for use by LLDs.
 362The names (i.e. entry points) of these functions are exported 
 363so an LLD that is a module can access them. The kernel will
 364arrange for the SCSI mid level to be loaded and initialized before any LLD
 365is initialized. The functions below are listed alphabetically and their
 366names all start with "scsi_".
 367
 368Summary:
 369   scsi_activate_tcq - turn on tag command queueing
 370   scsi_add_device - creates new scsi device (lu) instance
 371   scsi_add_host - perform sysfs registration and set up transport class
 372   scsi_adjust_queue_depth - change the queue depth on a SCSI device
 373   scsi_bios_ptable - return copy of block device's partition table
 374   scsi_block_requests - prevent further commands being queued to given host
 375   scsi_deactivate_tcq - turn off tag command queueing
 376   scsi_host_alloc - return a new scsi_host instance whose refcount==1
 377   scsi_host_get - increments Scsi_Host instance's refcount
 378   scsi_host_put - decrements Scsi_Host instance's refcount (free if 0)
 379   scsi_partsize - parse partition table into cylinders, heads + sectors
 380   scsi_register - create and register a scsi host adapter instance.
 381   scsi_remove_device - detach and remove a SCSI device
 382   scsi_remove_host - detach and remove all SCSI devices owned by host
 383   scsi_report_bus_reset - report scsi _bus_ reset observed
 384   scsi_scan_host - scan SCSI bus
 385   scsi_track_queue_full - track successive QUEUE_FULL events 
 386   scsi_unblock_requests - allow further commands to be queued to given host
 387   scsi_unregister - [calls scsi_host_put()]
 388
 389
 390Details:
 391
 392/**
 393 * scsi_activate_tcq - turn on tag command queueing ("ordered" task attribute)
 394 * @sdev:       device to turn on TCQ for
 395 * @depth:      queue depth
 396 *
 397 *      Returns nothing
 398 *
 399 *      Might block: no
 400 *
 401 *      Notes: Eventually, it is hoped depth would be the maximum depth
 402 *      the device could cope with and the real queue depth
 403 *      would be adjustable from 0 to depth.
 404 *
 405 *      Defined (inline) in: include/scsi/scsi_tcq.h
 406 **/
 407void scsi_activate_tcq(struct scsi_device *sdev, int depth)
 408
 409
 410/**
 411 * scsi_add_device - creates new scsi device (lu) instance
 412 * @shost:   pointer to scsi host instance
 413 * @channel: channel number (rarely other than 0)
 414 * @id:      target id number
 415 * @lun:     logical unit number
 416 *
 417 *      Returns pointer to new struct scsi_device instance or 
 418 *      ERR_PTR(-ENODEV) (or some other bent pointer) if something is
 419 *      wrong (e.g. no lu responds at given address)
 420 *
 421 *      Might block: yes
 422 *
 423 *      Notes: This call is usually performed internally during a scsi
 424 *      bus scan when an HBA is added (i.e. scsi_scan_host()). So it
 425 *      should only be called if the HBA becomes aware of a new scsi
 426 *      device (lu) after scsi_scan_host() has completed. If successful
 427 *      this call can lead to slave_alloc() and slave_configure() callbacks
 428 *      into the LLD.
 429 *
 430 *      Defined in: drivers/scsi/scsi_scan.c
 431 **/
 432struct scsi_device * scsi_add_device(struct Scsi_Host *shost, 
 433                                     unsigned int channel,
 434                                     unsigned int id, unsigned int lun)
 435
 436
 437/**
 438 * scsi_add_host - perform sysfs registration and set up transport class
 439 * @shost:   pointer to scsi host instance
 440 * @dev:     pointer to struct device of type scsi class
 441 *
 442 *      Returns 0 on success, negative errno of failure (e.g. -ENOMEM)
 443 *
 444 *      Might block: no
 445 *
 446 *      Notes: Only required in "hotplug initialization model" after a
 447 *      successful call to scsi_host_alloc().  This function does not
 448 *      scan the bus; this can be done by calling scsi_scan_host() or
 449 *      in some other transport-specific way.  The LLD must set up
 450 *      the transport template before calling this function and may only
 451 *      access the transport class data after this function has been called.
 452 *
 453 *      Defined in: drivers/scsi/hosts.c
 454 **/
 455int scsi_add_host(struct Scsi_Host *shost, struct device * dev)
 456
 457
 458/**
 459 * scsi_adjust_queue_depth - allow LLD to change queue depth on a SCSI device
 460 * @sdev:       pointer to SCSI device to change queue depth on
 461 * @tagged:     0 - no tagged queuing
 462 *              MSG_SIMPLE_TAG - simple tagged queuing
 463 *              MSG_ORDERED_TAG - ordered tagged queuing
 464 * @tags        Number of tags allowed if tagged queuing enabled,
 465 *              or number of commands the LLD can queue up
 466 *              in non-tagged mode (as per cmd_per_lun).
 467 *
 468 *      Returns nothing
 469 *
 470 *      Might block: no
 471 *
 472 *      Notes: Can be invoked any time on a SCSI device controlled by this
 473 *      LLD. [Specifically during and after slave_configure() and prior to
 474 *      slave_destroy().] Can safely be invoked from interrupt code. Actual
 475 *      queue depth change may be delayed until the next command is being
 476 *      processed. See also scsi_activate_tcq() and scsi_deactivate_tcq().
 477 *
 478 *      Defined in: drivers/scsi/scsi.c [see source code for more notes]
 479 *
 480 **/
 481void scsi_adjust_queue_depth(struct scsi_device * sdev, int tagged, 
 482                             int tags)
 483
 484
 485/**
 486 * scsi_bios_ptable - return copy of block device's partition table
 487 * @dev:        pointer to block device
 488 *
 489 *      Returns pointer to partition table, or NULL for failure
 490 *
 491 *      Might block: yes
 492 *
 493 *      Notes: Caller owns memory returned (free with kfree() )
 494 *
 495 *      Defined in: drivers/scsi/scsicam.c
 496 **/
 497unsigned char *scsi_bios_ptable(struct block_device *dev)
 498
 499
 500/**
 501 * scsi_block_requests - prevent further commands being queued to given host
 502 *
 503 * @shost: pointer to host to block commands on
 504 *
 505 *      Returns nothing
 506 *
 507 *      Might block: no
 508 *
 509 *      Notes: There is no timer nor any other means by which the requests
 510 *      get unblocked other than the LLD calling scsi_unblock_requests().
 511 *
 512 *      Defined in: drivers/scsi/scsi_lib.c
 513**/
 514void scsi_block_requests(struct Scsi_Host * shost)
 515
 516
 517/**
 518 * scsi_deactivate_tcq - turn off tag command queueing
 519 * @sdev:       device to turn off TCQ for
 520 * @depth:      queue depth (stored in sdev)
 521 *
 522 *      Returns nothing
 523 *
 524 *      Might block: no
 525 *
 526 *      Defined (inline) in: include/scsi/scsi_tcq.h
 527 **/
 528void scsi_deactivate_tcq(struct scsi_device *sdev, int depth)
 529
 530
 531/**
 532 * scsi_host_alloc - create a scsi host adapter instance and perform basic
 533 *                   initialization.
 534 * @sht:        pointer to scsi host template
 535 * @privsize:   extra bytes to allocate in hostdata array (which is the
 536 *              last member of the returned Scsi_Host instance)
 537 *
 538 *      Returns pointer to new Scsi_Host instance or NULL on failure
 539 *
 540 *      Might block: yes
 541 *
 542 *      Notes: When this call returns to the LLD, the SCSI bus scan on
 543 *      this host has _not_ yet been done.
 544 *      The hostdata array (by default zero length) is a per host scratch 
 545 *      area for the LLD's exclusive use.
 546 *      Both associated refcounting objects have their refcount set to 1.
 547 *      Full registration (in sysfs) and a bus scan are performed later when
 548 *      scsi_add_host() and scsi_scan_host() are called.
 549 *
 550 *      Defined in: drivers/scsi/hosts.c .
 551 **/
 552struct Scsi_Host * scsi_host_alloc(struct scsi_host_template * sht,
 553                                   int privsize)
 554
 555
 556/**
 557 * scsi_host_get - increment Scsi_Host instance refcount
 558 * @shost:   pointer to struct Scsi_Host instance
 559 *
 560 *      Returns nothing
 561 *
 562 *      Might block: currently may block but may be changed to not block
 563 *
 564 *      Notes: Actually increments the counts in two sub-objects
 565 *
 566 *      Defined in: drivers/scsi/hosts.c
 567 **/
 568void scsi_host_get(struct Scsi_Host *shost)
 569
 570
 571/**
 572 * scsi_host_put - decrement Scsi_Host instance refcount, free if 0
 573 * @shost:   pointer to struct Scsi_Host instance
 574 *
 575 *      Returns nothing
 576 *
 577 *      Might block: currently may block but may be changed to not block
 578 *
 579 *      Notes: Actually decrements the counts in two sub-objects. If the
 580 *      latter refcount reaches 0, the Scsi_Host instance is freed.
 581 *      The LLD need not worry exactly when the Scsi_Host instance is
 582 *      freed, it just shouldn't access the instance after it has balanced
 583 *      out its refcount usage.
 584 *
 585 *      Defined in: drivers/scsi/hosts.c
 586 **/
 587void scsi_host_put(struct Scsi_Host *shost)
 588
 589
 590/**
 591 * scsi_partsize - parse partition table into cylinders, heads + sectors
 592 * @buf: pointer to partition table
 593 * @capacity: size of (total) disk in 512 byte sectors
 594 * @cyls: outputs number of cylinders calculated via this pointer
 595 * @hds: outputs number of heads calculated via this pointer
 596 * @secs: outputs number of sectors calculated via this pointer
 597 *
 598 *      Returns 0 on success, -1 on failure
 599 *
 600 *      Might block: no
 601 *
 602 *      Notes: Caller owns memory returned (free with kfree() )
 603 *
 604 *      Defined in: drivers/scsi/scsicam.c
 605 **/
 606int scsi_partsize(unsigned char *buf, unsigned long capacity,
 607                  unsigned int *cyls, unsigned int *hds, unsigned int *secs)
 608
 609
 610/**
 611 * scsi_register - create and register a scsi host adapter instance.
 612 * @sht:        pointer to scsi host template
 613 * @privsize:   extra bytes to allocate in hostdata array (which is the
 614 *              last member of the returned Scsi_Host instance)
 615 *
 616 *      Returns pointer to new Scsi_Host instance or NULL on failure
 617 *
 618 *      Might block: yes
 619 *
 620 *      Notes: When this call returns to the LLD, the SCSI bus scan on
 621 *      this host has _not_ yet been done.
 622 *      The hostdata array (by default zero length) is a per host scratch 
 623 *      area for the LLD.
 624 *
 625 *      Defined in: drivers/scsi/hosts.c .
 626 **/
 627struct Scsi_Host * scsi_register(struct scsi_host_template * sht,
 628                                 int privsize)
 629
 630
 631/**
 632 * scsi_remove_device - detach and remove a SCSI device
 633 * @sdev:      a pointer to a scsi device instance
 634 *
 635 *      Returns value: 0 on success, -EINVAL if device not attached
 636 *
 637 *      Might block: yes
 638 *
 639 *      Notes: If an LLD becomes aware that a scsi device (lu) has
 640 *      been removed but its host is still present then it can request
 641 *      the removal of that scsi device. If successful this call will
 642 *      lead to the slave_destroy() callback being invoked. sdev is an 
 643 *      invalid pointer after this call.
 644 *
 645 *      Defined in: drivers/scsi/scsi_sysfs.c .
 646 **/
 647int scsi_remove_device(struct scsi_device *sdev)
 648
 649
 650/**
 651 * scsi_remove_host - detach and remove all SCSI devices owned by host
 652 * @shost:      a pointer to a scsi host instance
 653 *
 654 *      Returns value: 0 on success, 1 on failure (e.g. LLD busy ??)
 655 *
 656 *      Might block: yes
 657 *
 658 *      Notes: Should only be invoked if the "hotplug initialization
 659 *      model" is being used. It should be called _prior_ to  
 660 *      scsi_unregister().
 661 *
 662 *      Defined in: drivers/scsi/hosts.c .
 663 **/
 664int scsi_remove_host(struct Scsi_Host *shost)
 665
 666
 667/**
 668 * scsi_report_bus_reset - report scsi _bus_ reset observed
 669 * @shost: a pointer to a scsi host involved
 670 * @channel: channel (within) host on which scsi bus reset occurred
 671 *
 672 *      Returns nothing
 673 *
 674 *      Might block: no
 675 *
 676 *      Notes: This only needs to be called if the reset is one which
 677 *      originates from an unknown location.  Resets originated by the 
 678 *      mid level itself don't need to call this, but there should be 
 679 *      no harm.  The main purpose of this is to make sure that a
 680 *      CHECK_CONDITION is properly treated.
 681 *
 682 *      Defined in: drivers/scsi/scsi_error.c .
 683 **/
 684void scsi_report_bus_reset(struct Scsi_Host * shost, int channel)
 685
 686
 687/**
 688 * scsi_scan_host - scan SCSI bus
 689 * @shost: a pointer to a scsi host instance
 690 *
 691 *      Might block: yes
 692 *
 693 *      Notes: Should be called after scsi_add_host()
 694 *
 695 *      Defined in: drivers/scsi/scsi_scan.c
 696 **/
 697void scsi_scan_host(struct Scsi_Host *shost)
 698
 699
 700/**
 701 * scsi_track_queue_full - track successive QUEUE_FULL events on given
 702 *                      device to determine if and when there is a need
 703 *                      to adjust the queue depth on the device.
 704 * @sdev:  pointer to SCSI device instance
 705 * @depth: Current number of outstanding SCSI commands on this device,
 706 *         not counting the one returned as QUEUE_FULL.
 707 *
 708 *      Returns 0  - no change needed
 709 *              >0 - adjust queue depth to this new depth
 710 *              -1 - drop back to untagged operation using host->cmd_per_lun
 711 *                   as the untagged command depth
 712 *
 713 *      Might block: no
 714 *
 715 *      Notes: LLDs may call this at any time and we will do "The Right
 716 *              Thing"; interrupt context safe. 
 717 *
 718 *      Defined in: drivers/scsi/scsi.c .
 719 **/
 720int scsi_track_queue_full(struct scsi_device *sdev, int depth)
 721
 722
 723/**
 724 * scsi_unblock_requests - allow further commands to be queued to given host
 725 *
 726 * @shost: pointer to host to unblock commands on
 727 *
 728 *      Returns nothing
 729 *
 730 *      Might block: no
 731 *
 732 *      Defined in: drivers/scsi/scsi_lib.c .
 733**/
 734void scsi_unblock_requests(struct Scsi_Host * shost)
 735
 736
 737/**
 738 * scsi_unregister - unregister and free memory used by host instance
 739 * @shp:        pointer to scsi host instance to unregister.
 740 *
 741 *      Returns nothing
 742 *
 743 *      Might block: no
 744 *
 745 *      Notes: Should not be invoked if the "hotplug initialization
 746 *      model" is being used. Called internally by exit_this_scsi_driver()
 747 *      in the "passive initialization model". Hence a LLD has no need to
 748 *      call this function directly.
 749 *
 750 *      Defined in: drivers/scsi/hosts.c .
 751 **/
 752void scsi_unregister(struct Scsi_Host * shp)
 753
 754
 755
 756
 757Interface Functions
 758===================
 759Interface functions are supplied (defined) by LLDs and their function
 760pointers are placed in an instance of struct scsi_host_template which
 761is passed to scsi_host_alloc() [or scsi_register() / init_this_scsi_driver()].
 762Some are mandatory. Interface functions should be declared static. The
 763accepted convention is that driver "xyz" will declare its slave_configure() 
 764function as:
 765    static int xyz_slave_configure(struct scsi_device * sdev);
 766and so forth for all interface functions listed below.
 767
 768A pointer to this function should be placed in the 'slave_configure' member
 769of a "struct scsi_host_template" instance. A pointer to such an instance
 770should be passed to the mid level's scsi_host_alloc() [or scsi_register() /
 771init_this_scsi_driver()].
 772
 773The interface functions are also described in the include/scsi/scsi_host.h
 774file immediately above their definition point in "struct scsi_host_template".
 775In some cases more detail is given in scsi_host.h than below.
 776
 777The interface functions are listed below in alphabetical order.
 778
 779Summary:
 780   bios_param - fetch head, sector, cylinder info for a disk
 781   detect - detects HBAs this driver wants to control
 782   eh_timed_out - notify the host that a command timer expired
 783   eh_abort_handler - abort given command
 784   eh_bus_reset_handler - issue SCSI bus reset
 785   eh_device_reset_handler - issue SCSI device reset
 786   eh_host_reset_handler - reset host (host bus adapter)
 787   info - supply information about given host
 788   ioctl - driver can respond to ioctls
 789   proc_info - supports /proc/scsi/{driver_name}/{host_no}
 790   queuecommand - queue scsi command, invoke 'done' on completion
 791   release - release all resources associated with given host
 792   slave_alloc - prior to any commands being sent to a new device 
 793   slave_configure - driver fine tuning for given device after attach
 794   slave_destroy - given device is about to be shut down
 795
 796
 797Details:
 798
 799/**
 800 *      bios_param - fetch head, sector, cylinder info for a disk
 801 *      @sdev: pointer to scsi device context (defined in 
 802 *             include/scsi/scsi_device.h)
 803 *      @bdev: pointer to block device context (defined in fs.h)
 804 *      @capacity:  device size (in 512 byte sectors)
 805 *      @params: three element array to place output:
 806 *              params[0] number of heads (max 255)
 807 *              params[1] number of sectors (max 63)
 808 *              params[2] number of cylinders 
 809 *
 810 *      Return value is ignored
 811 *
 812 *      Locks: none
 813 *
 814 *      Calling context: process (sd)
 815 *
 816 *      Notes: an arbitrary geometry (based on READ CAPACITY) is used
 817 *      if this function is not provided. The params array is
 818 *      pre-initialized with made up values just in case this function 
 819 *      doesn't output anything.
 820 *
 821 *      Optionally defined in: LLD
 822 **/
 823    int bios_param(struct scsi_device * sdev, struct block_device *bdev,
 824                   sector_t capacity, int params[3])
 825
 826
 827/**
 828 *      detect - detects HBAs this driver wants to control
 829 *      @shtp: host template for this driver.
 830 *
 831 *      Returns number of hosts this driver wants to control. 0 means no
 832 *      suitable hosts found.
 833 *
 834 *      Locks: none held
 835 *
 836 *      Calling context: process [invoked from init_this_scsi_driver()]
 837 *
 838 *      Notes: First function called from the SCSI mid level on this
 839 *      driver. Upper level drivers (e.g. sd) may not (yet) be present.
 840 *      For each host found, this method should call scsi_register() 
 841 *      [see hosts.c].
 842 *
 843 *      Defined in: LLD (required if "passive initialization mode" is used,
 844 *                       not invoked in "hotplug initialization mode")
 845 **/
 846    int detect(struct scsi_host_template * shtp)
 847
 848
 849/**
 850 *      eh_timed_out - The timer for the command has just fired
 851 *      @scp: identifies command timing out
 852 *
 853 *      Returns:
 854 *
 855 *      EH_HANDLED:             I fixed the error, please complete the command
 856 *      EH_RESET_TIMER:         I need more time, reset the timer and
 857 *                              begin counting again
 858 *      EH_NOT_HANDLED          Begin normal error recovery
 859 *
 860 *
 861 *      Locks: None held
 862 *
 863 *      Calling context: interrupt
 864 *
 865 *      Notes: This is to give the LLD an opportunity to do local recovery.
 866 *      This recovery is limited to determining if the outstanding command
 867 *      will ever complete.  You may not abort and restart the command from
 868 *      this callback.
 869 *
 870 *      Optionally defined in: LLD
 871 **/
 872     int eh_timed_out(struct scsi_cmnd * scp)
 873
 874
 875/**
 876 *      eh_abort_handler - abort command associated with scp
 877 *      @scp: identifies command to be aborted
 878 *
 879 *      Returns SUCCESS if command aborted else FAILED
 880 *
 881 *      Locks: None held
 882 *
 883 *      Calling context: kernel thread
 884 *
 885 *      Notes: Invoked from scsi_eh thread. No other commands will be
 886 *      queued on current host during eh.
 887 *
 888 *      Optionally defined in: LLD
 889 **/
 890     int eh_abort_handler(struct scsi_cmnd * scp)
 891
 892
 893/**
 894 *      eh_bus_reset_handler - issue SCSI bus reset
 895 *      @scp: SCSI bus that contains this device should be reset
 896 *
 897 *      Returns SUCCESS if command aborted else FAILED
 898 *
 899 *      Locks: None held
 900 *
 901 *      Calling context: kernel thread
 902 *
 903 *      Notes: Invoked from scsi_eh thread. No other commands will be
 904 *      queued on current host during eh.
 905 *
 906 *      Optionally defined in: LLD
 907 **/
 908     int eh_bus_reset_handler(struct scsi_cmnd * scp)
 909
 910
 911/**
 912 *      eh_device_reset_handler - issue SCSI device reset
 913 *      @scp: identifies SCSI device to be reset
 914 *
 915 *      Returns SUCCESS if command aborted else FAILED
 916 *
 917 *      Locks: None held
 918 *
 919 *      Calling context: kernel thread
 920 *
 921 *      Notes: Invoked from scsi_eh thread. No other commands will be
 922 *      queued on current host during eh.
 923 *
 924 *      Optionally defined in: LLD
 925 **/
 926     int eh_device_reset_handler(struct scsi_cmnd * scp)
 927
 928
 929/**
 930 *      eh_host_reset_handler - reset host (host bus adapter)
 931 *      @scp: SCSI host that contains this device should be reset
 932 *
 933 *      Returns SUCCESS if command aborted else FAILED
 934 *
 935 *      Locks: None held
 936 *
 937 *      Calling context: kernel thread
 938 *
 939 *      Notes: Invoked from scsi_eh thread. No other commands will be
 940 *      queued on current host during eh. 
 941 *      With the default eh_strategy in place, if none of the _abort_, 
 942 *      _device_reset_, _bus_reset_ or this eh handler function are 
 943 *      defined (or they all return FAILED) then the device in question 
 944 *      will be set offline whenever eh is invoked.
 945 *
 946 *      Optionally defined in: LLD
 947 **/
 948     int eh_host_reset_handler(struct scsi_cmnd * scp)
 949
 950
 951/**
 952 *      info - supply information about given host: driver name plus data
 953 *             to distinguish given host
 954 *      @shp: host to supply information about
 955 *
 956 *      Return ASCII null terminated string. [This driver is assumed to
 957 *      manage the memory pointed to and maintain it, typically for the
 958 *      lifetime of this host.]
 959 *
 960 *      Locks: none
 961 *
 962 *      Calling context: process
 963 *
 964 *      Notes: Often supplies PCI or ISA information such as IO addresses
 965 *      and interrupt numbers. If not supplied struct Scsi_Host::name used
 966 *      instead. It is assumed the returned information fits on one line 
 967 *      (i.e. does not included embedded newlines).
 968 *      The SCSI_IOCTL_PROBE_HOST ioctl yields the string returned by this
 969 *      function (or struct Scsi_Host::name if this function is not
 970 *      available).
 971 *      In a similar manner, init_this_scsi_driver() outputs to the console
 972 *      each host's "info" (or name) for the driver it is registering.
 973 *      Also if proc_info() is not supplied, the output of this function
 974 *      is used instead.
 975 *
 976 *      Optionally defined in: LLD
 977 **/
 978    const char * info(struct Scsi_Host * shp)
 979
 980
 981/**
 982 *      ioctl - driver can respond to ioctls
 983 *      @sdp: device that ioctl was issued for
 984 *      @cmd: ioctl number
 985 *      @arg: pointer to read or write data from. Since it points to
 986 *            user space, should use appropriate kernel functions
 987 *            (e.g. copy_from_user() ). In the Unix style this argument
 988 *            can also be viewed as an unsigned long.
 989 *
 990 *      Returns negative "errno" value when there is a problem. 0 or a
 991 *      positive value indicates success and is returned to the user space.
 992 *
 993 *      Locks: none
 994 *
 995 *      Calling context: process
 996 *
 997 *      Notes: The SCSI subsystem uses a "trickle down" ioctl model.
 998 *      The user issues an ioctl() against an upper level driver
 999 *      (e.g. /dev/sdc) and if the upper level driver doesn't recognize
1000 *      the 'cmd' then it is passed to the SCSI mid level. If the SCSI
1001 *      mid level does not recognize it, then the LLD that controls
1002 *      the device receives the ioctl. According to recent Unix standards
1003 *      unsupported ioctl() 'cmd' numbers should return -ENOTTY.
1004 *
1005 *      Optionally defined in: LLD
1006 **/
1007    int ioctl(struct scsi_device *sdp, int cmd, void *arg)
1008
1009
1010/**
1011 *      proc_info - supports /proc/scsi/{driver_name}/{host_no}
1012 *      @buffer: anchor point to output to (0==writeto1_read0) or fetch from
1013 *               (1==writeto1_read0).
1014 *      @start: where "interesting" data is written to. Ignored when
1015 *              1==writeto1_read0.
1016 *      @offset: offset within buffer 0==writeto1_read0 is actually
1017 *               interested in. Ignored when 1==writeto1_read0 .
1018 *      @length: maximum (or actual) extent of buffer
1019 *      @host_no: host number of interest (struct Scsi_Host::host_no)
1020 *      @writeto1_read0: 1 -> data coming from user space towards driver
1021 *                            (e.g. "echo some_string > /proc/scsi/xyz/2")
1022 *                       0 -> user what data from this driver
1023 *                            (e.g. "cat /proc/scsi/xyz/2")
1024 *
1025 *      Returns length when 1==writeto1_read0. Otherwise number of chars
1026 *      output to buffer past offset.
1027 *
1028 *      Locks: none held
1029 *
1030 *      Calling context: process
1031 *
1032 *      Notes: Driven from scsi_proc.c which interfaces to proc_fs. proc_fs
1033 *      support can now be configured out of the scsi subsystem.
1034 *
1035 *      Optionally defined in: LLD
1036 **/
1037    int proc_info(char * buffer, char ** start, off_t offset, 
1038                  int length, int host_no, int writeto1_read0)
1039
1040
1041/**
1042 *      queuecommand - queue scsi command, invoke scp->scsi_done on completion
1043 *      @shost: pointer to the scsi host object
1044 *      @scp: pointer to scsi command object
1045 *
1046 *      Returns 0 on success.
1047 *
1048 *      If there's a failure, return either:
1049 *
1050 *      SCSI_MLQUEUE_DEVICE_BUSY if the device queue is full, or
1051 *      SCSI_MLQUEUE_HOST_BUSY if the entire host queue is full
1052 *
1053 *      On both of these returns, the mid-layer will requeue the I/O
1054 *
1055 *      - if the return is SCSI_MLQUEUE_DEVICE_BUSY, only that particular
1056 *      device will be paused, and it will be unpaused when a command to
1057 *      the device returns (or after a brief delay if there are no more
1058 *      outstanding commands to it).  Commands to other devices continue
1059 *      to be processed normally.
1060 *
1061 *      - if the return is SCSI_MLQUEUE_HOST_BUSY, all I/O to the host
1062 *      is paused and will be unpaused when any command returns from
1063 *      the host (or after a brief delay if there are no outstanding
1064 *      commands to the host).
1065 *
1066 *      For compatibility with earlier versions of queuecommand, any
1067 *      other return value is treated the same as
1068 *      SCSI_MLQUEUE_HOST_BUSY.
1069 *
1070 *      Other types of errors that are detected immediately may be
1071 *      flagged by setting scp->result to an appropriate value,
1072 *      invoking the scp->scsi_done callback, and then returning 0
1073 *      from this function. If the command is not performed
1074 *      immediately (and the LLD is starting (or will start) the given
1075 *      command) then this function should place 0 in scp->result and
1076 *      return 0.
1077 *
1078 *      Command ownership.  If the driver returns zero, it owns the
1079 *      command and must take responsibility for ensuring the
1080 *      scp->scsi_done callback is executed.  Note: the driver may
1081 *      call scp->scsi_done before returning zero, but after it has
1082 *      called scp->scsi_done, it may not return any value other than
1083 *      zero.  If the driver makes a non-zero return, it must not
1084 *      execute the command's scsi_done callback at any time.
1085 *
1086 *      Locks: up to and including 2.6.36, struct Scsi_Host::host_lock
1087 *             held on entry (with "irqsave") and is expected to be
1088 *             held on return. From 2.6.37 onwards, queuecommand is
1089 *             called without any locks held.
1090 *
1091 *      Calling context: in interrupt (soft irq) or process context
1092 *
1093 *      Notes: This function should be relatively fast. Normally it
1094 *      will not wait for IO to complete. Hence the scp->scsi_done
1095 *      callback is invoked (often directly from an interrupt service
1096 *      routine) some time after this function has returned. In some
1097 *      cases (e.g. pseudo adapter drivers that manufacture the
1098 *      response to a SCSI INQUIRY) the scp->scsi_done callback may be
1099 *      invoked before this function returns.  If the scp->scsi_done
1100 *      callback is not invoked within a certain period the SCSI mid
1101 *      level will commence error processing.  If a status of CHECK
1102 *      CONDITION is placed in "result" when the scp->scsi_done
1103 *      callback is invoked, then the LLD driver should perform
1104 *      autosense and fill in the struct scsi_cmnd::sense_buffer
1105 *      array. The scsi_cmnd::sense_buffer array is zeroed prior to
1106 *      the mid level queuing a command to an LLD.
1107 *
1108 *      Defined in: LLD
1109 **/
1110    int queuecommand(struct Scsi_Host *shost, struct scsi_cmnd * scp)
1111
1112
1113/**
1114 *      release - release all resources associated with given host
1115 *      @shp: host to be released.
1116 *
1117 *      Return value ignored (could soon be a function returning void).
1118 *
1119 *      Locks: none held
1120 *
1121 *      Calling context: process
1122 *
1123 *      Notes: Invoked from scsi_module.c's exit_this_scsi_driver().
1124 *      LLD's implementation of this function should call 
1125 *      scsi_unregister(shp) prior to returning.
1126 *      Only needed for old-style host templates.
1127 *
1128 *      Defined in: LLD (required in "passive initialization model",
1129 *                       should not be defined in hotplug model)
1130 **/
1131    int release(struct Scsi_Host * shp)
1132
1133
1134/**
1135 *      slave_alloc -   prior to any commands being sent to a new device 
1136 *                      (i.e. just prior to scan) this call is made
1137 *      @sdp: pointer to new device (about to be scanned)
1138 *
1139 *      Returns 0 if ok. Any other return is assumed to be an error and
1140 *      the device is ignored.
1141 *
1142 *      Locks: none
1143 *
1144 *      Calling context: process
1145 *
1146 *      Notes: Allows the driver to allocate any resources for a device
1147 *      prior to its initial scan. The corresponding scsi device may not
1148 *      exist but the mid level is just about to scan for it (i.e. send
1149 *      and INQUIRY command plus ...). If a device is found then
1150 *      slave_configure() will be called while if a device is not found
1151 *      slave_destroy() is called.
1152 *      For more details see the include/scsi/scsi_host.h file.
1153 *
1154 *      Optionally defined in: LLD
1155 **/
1156    int slave_alloc(struct scsi_device *sdp)
1157
1158
1159/**
1160 *      slave_configure - driver fine tuning for given device just after it
1161 *                     has been first scanned (i.e. it responded to an
1162 *                     INQUIRY)
1163 *      @sdp: device that has just been attached
1164 *
1165 *      Returns 0 if ok. Any other return is assumed to be an error and
1166 *      the device is taken offline. [offline devices will _not_ have
1167 *      slave_destroy() called on them so clean up resources.]
1168 *
1169 *      Locks: none
1170 *
1171 *      Calling context: process
1172 *
1173 *      Notes: Allows the driver to inspect the response to the initial
1174 *      INQUIRY done by the scanning code and take appropriate action.
1175 *      For more details see the include/scsi/scsi_host.h file.
1176 *
1177 *      Optionally defined in: LLD
1178 **/
1179    int slave_configure(struct scsi_device *sdp)
1180
1181
1182/**
1183 *      slave_destroy - given device is about to be shut down. All
1184 *                      activity has ceased on this device.
1185 *      @sdp: device that is about to be shut down
1186 *
1187 *      Returns nothing
1188 *
1189 *      Locks: none
1190 *
1191 *      Calling context: process
1192 *
1193 *      Notes: Mid level structures for given device are still in place
1194 *      but are about to be torn down. Any per device resources allocated
1195 *      by this driver for given device should be freed now. No further
1196 *      commands will be sent for this sdp instance. [However the device
1197 *      could be re-attached in the future in which case a new instance
1198 *      of struct scsi_device would be supplied by future slave_alloc()
1199 *      and slave_configure() calls.]
1200 *
1201 *      Optionally defined in: LLD
1202 **/
1203    void slave_destroy(struct scsi_device *sdp)
1204
1205
1206
1207Data Structures
1208===============
1209struct scsi_host_template
1210-------------------------
1211There is one "struct scsi_host_template" instance per LLD ***. It is
1212typically initialized as a file scope static in a driver's header file. That
1213way members that are not explicitly initialized will be set to 0 or NULL.
1214Member of interest:
1215    name         - name of driver (may contain spaces, please limit to
1216                   less than 80 characters)
1217    proc_name    - name used in "/proc/scsi/<proc_name>/<host_no>" and
1218                   by sysfs in one of its "drivers" directories. Hence
1219                   "proc_name" should only contain characters acceptable
1220                   to a Unix file name.
1221   (*queuecommand)() - primary callback that the mid level uses to inject
1222                   SCSI commands into an LLD.
1223The structure is defined and commented in include/scsi/scsi_host.h
1224
1225*** In extreme situations a single driver may have several instances
1226    if it controls several different classes of hardware (e.g. an LLD
1227    that handles both ISA and PCI cards and has a separate instance of
1228    struct scsi_host_template for each class).
1229
1230struct Scsi_Host
1231----------------
1232There is one struct Scsi_Host instance per host (HBA) that an LLD
1233controls. The struct Scsi_Host structure has many members in common
1234with "struct scsi_host_template". When a new struct Scsi_Host instance
1235is created (in scsi_host_alloc() in hosts.c) those common members are
1236initialized from the driver's struct scsi_host_template instance. Members
1237of interest:
1238    host_no      - system wide unique number that is used for identifying
1239                   this host. Issued in ascending order from 0.
1240    can_queue    - must be greater than 0; do not send more than can_queue
1241                   commands to the adapter.
1242    this_id      - scsi id of host (scsi initiator) or -1 if not known
1243    sg_tablesize - maximum scatter gather elements allowed by host.
1244                   0 implies scatter gather not supported by host
1245    max_sectors  - maximum number of sectors (usually 512 bytes) allowed
1246                   in a single SCSI command. The default value of 0 leads
1247                   to a setting of SCSI_DEFAULT_MAX_SECTORS (defined in
1248                   scsi_host.h) which is currently set to 1024. So for a
1249                   disk the maximum transfer size is 512 KB when max_sectors
1250                   is not defined. Note that this size may not be sufficient
1251                   for disk firmware uploads.
1252    cmd_per_lun  - maximum number of commands that can be queued on devices
1253                   controlled by the host. Overridden by LLD calls to
1254                   scsi_adjust_queue_depth().
1255    unchecked_isa_dma - 1=>only use bottom 16 MB of ram (ISA DMA addressing
1256                   restriction), 0=>can use full 32 bit (or better) DMA
1257                   address space
1258    use_clustering - 1=>SCSI commands in mid level's queue can be merged,
1259                     0=>disallow SCSI command merging
1260    hostt        - pointer to driver's struct scsi_host_template from which
1261                   this struct Scsi_Host instance was spawned
1262    hostt->proc_name  - name of LLD. This is the driver name that sysfs uses
1263    transportt   - pointer to driver's struct scsi_transport_template instance
1264                   (if any). FC and SPI transports currently supported.
1265    sh_list      - a double linked list of pointers to all struct Scsi_Host
1266                   instances (currently ordered by ascending host_no)
1267    my_devices   - a double linked list of pointers to struct scsi_device 
1268                   instances that belong to this host.
1269    hostdata[0]  - area reserved for LLD at end of struct Scsi_Host. Size
1270                   is set by the second argument (named 'xtr_bytes') to
1271                   scsi_host_alloc() or scsi_register().
1272    vendor_id    - a unique value that identifies the vendor supplying
1273                   the LLD for the Scsi_Host.  Used most often in validating
1274                   vendor-specific message requests.  Value consists of an
1275                   identifier type and a vendor-specific value.
1276                   See scsi_netlink.h for a description of valid formats.
1277
1278The scsi_host structure is defined in include/scsi/scsi_host.h
1279
1280struct scsi_device
1281------------------
1282Generally, there is one instance of this structure for each SCSI logical unit
1283on a host. Scsi devices connected to a host are uniquely identified by a
1284channel number, target id and logical unit number (lun).
1285The structure is defined in include/scsi/scsi_device.h
1286
1287struct scsi_cmnd
1288----------------
1289Instances of this structure convey SCSI commands to the LLD and responses
1290back to the mid level. The SCSI mid level will ensure that no more SCSI
1291commands become queued against the LLD than are indicated by
1292scsi_adjust_queue_depth() (or struct Scsi_Host::cmd_per_lun). There will
1293be at least one instance of struct scsi_cmnd available for each SCSI device.
1294Members of interest:
1295    cmnd         - array containing SCSI command
1296    cmnd_len     - length (in bytes) of SCSI command
1297    sc_data_direction - direction of data transfer in data phase. See
1298                "enum dma_data_direction" in include/linux/dma-mapping.h
1299    request_bufflen - number of data bytes to transfer (0 if no data phase)
1300    use_sg       - ==0 -> no scatter gather list, hence transfer data
1301                          to/from request_buffer
1302                 - >0 ->  scatter gather list (actually an array) in
1303                          request_buffer with use_sg elements
1304    request_buffer - either contains data buffer or scatter gather list
1305                     depending on the setting of use_sg. Scatter gather
1306                     elements are defined by 'struct scatterlist' found
1307                     in include/asm/scatterlist.h .
1308    done         - function pointer that should be invoked by LLD when the
1309                   SCSI command is completed (successfully or otherwise).
1310                   Should only be called by an LLD if the LLD has accepted
1311                   the command (i.e. queuecommand() returned or will return
1312                   0). The LLD may invoke 'done'  prior to queuecommand()
1313                   finishing.
1314    result       - should be set by LLD prior to calling 'done'. A value
1315                   of 0 implies a successfully completed command (and all
1316                   data (if any) has been transferred to or from the SCSI
1317                   target device). 'result' is a 32 bit unsigned integer that
1318                   can be viewed as 4 related bytes. The SCSI status value is
1319                   in the LSB. See include/scsi/scsi.h status_byte(),
1320                   msg_byte(), host_byte() and driver_byte() macros and
1321                   related constants.
1322    sense_buffer - an array (maximum size: SCSI_SENSE_BUFFERSIZE bytes) that
1323                   should be written when the SCSI status (LSB of 'result')
1324                   is set to CHECK_CONDITION (2). When CHECK_CONDITION is
1325                   set, if the top nibble of sense_buffer[0] has the value 7
1326                   then the mid level will assume the sense_buffer array
1327                   contains a valid SCSI sense buffer; otherwise the mid
1328                   level will issue a REQUEST_SENSE SCSI command to
1329                   retrieve the sense buffer. The latter strategy is error
1330                   prone in the presence of command queuing so the LLD should
1331                   always "auto-sense".
1332    device       - pointer to scsi_device object that this command is
1333                   associated with.
1334    resid        - an LLD should set this signed integer to the requested
1335                   transfer length (i.e. 'request_bufflen') less the number
1336                   of bytes that are actually transferred. 'resid' is
1337                   preset to 0 so an LLD can ignore it if it cannot detect
1338                   underruns (overruns should be rare). If possible an LLD
1339                   should set 'resid' prior to invoking 'done'. The most
1340                   interesting case is data transfers from a SCSI target
1341                   device (e.g. READs) that underrun.
1342    underflow    - LLD should place (DID_ERROR << 16) in 'result' if
1343                   actual number of bytes transferred is less than this
1344                   figure. Not many LLDs implement this check and some that
1345                   do just output an error message to the log rather than
1346                   report a DID_ERROR. Better for an LLD to implement
1347                   'resid'.
1348
1349It is recommended that a LLD set 'resid' on data transfers from a SCSI
1350target device (e.g. READs). It is especially important that 'resid' is set
1351when such data transfers have sense keys of MEDIUM ERROR and HARDWARE ERROR
1352(and possibly RECOVERED ERROR). In these cases if a LLD is in doubt how much
1353data has been received then the safest approach is to indicate no bytes have
1354been received. For example: to indicate that no valid data has been received
1355a LLD might use these helpers:
1356    scsi_set_resid(SCpnt, scsi_bufflen(SCpnt));
1357where 'SCpnt' is a pointer to a scsi_cmnd object. To indicate only three 512
1358bytes blocks has been received 'resid' could be set like this:
1359    scsi_set_resid(SCpnt, scsi_bufflen(SCpnt) - (3 * 512));
1360
1361The scsi_cmnd structure is defined in include/scsi/scsi_cmnd.h
1362
1363
1364Locks
1365=====
1366Each struct Scsi_Host instance has a spin_lock called struct 
1367Scsi_Host::default_lock which is initialized in scsi_host_alloc() [found in 
1368hosts.c]. Within the same function the struct Scsi_Host::host_lock pointer
1369is initialized to point at default_lock.  Thereafter lock and unlock
1370operations performed by the mid level use the struct Scsi_Host::host_lock
1371pointer.  Previously drivers could override the host_lock pointer but
1372this is not allowed anymore.
1373
1374
1375Autosense
1376=========
1377Autosense (or auto-sense) is defined in the SAM-2 document as "the
1378automatic return of sense data to the application client coincident
1379with the completion of a SCSI command" when a status of CHECK CONDITION
1380occurs. LLDs should perform autosense. This should be done when the LLD
1381detects a CHECK CONDITION status by either: 
1382    a) instructing the SCSI protocol (e.g. SCSI Parallel Interface (SPI))
1383       to perform an extra data in phase on such responses
1384    b) or, the LLD issuing a REQUEST SENSE command itself
1385
1386Either way, when a status of CHECK CONDITION is detected, the mid level
1387decides whether the LLD has performed autosense by checking struct 
1388scsi_cmnd::sense_buffer[0] . If this byte has an upper nibble of 7 (or 0xf)
1389then autosense is assumed to have taken place. If it has another value (and
1390this byte is initialized to 0 before each command) then the mid level will
1391issue a REQUEST SENSE command.
1392
1393In the presence of queued commands the "nexus" that maintains sense
1394buffer data from the command that failed until a following REQUEST SENSE
1395may get out of synchronization. This is why it is best for the LLD
1396to perform autosense.
1397
1398
1399Changes since lk 2.4 series
1400===========================
1401io_request_lock has been replaced by several finer grained locks. The lock 
1402relevant to LLDs is struct Scsi_Host::host_lock and there is
1403one per SCSI host.
1404
1405The older error handling mechanism has been removed. This means the
1406LLD interface functions abort() and reset() have been removed.
1407The struct scsi_host_template::use_new_eh_code flag has been removed.
1408
1409In the 2.4 series the SCSI subsystem configuration descriptions were 
1410aggregated with the configuration descriptions from all other Linux 
1411subsystems in the Documentation/Configure.help file. In the 2.6 series, 
1412the SCSI subsystem now has its own (much smaller) drivers/scsi/Kconfig
1413file that contains both configuration and help information.
1414
1415struct SHT has been renamed to struct scsi_host_template.
1416
1417Addition of the "hotplug initialization model" and many extra functions
1418to support it.
1419
1420
1421Credits
1422=======
1423The following people have contributed to this document:
1424        Mike Anderson <andmike at us dot ibm dot com>
1425        James Bottomley <James dot Bottomley at hansenpartnership dot com>
1426        Patrick Mansfield <patmans at us dot ibm dot com> 
1427        Christoph Hellwig <hch at infradead dot org>
1428        Doug Ledford <dledford at redhat dot com>
1429        Andries Brouwer <Andries dot Brouwer at cwi dot nl>
1430        Randy Dunlap <rdunlap at xenotime dot net>
1431        Alan Stern <stern at rowland dot harvard dot edu>
1432
1433
1434Douglas Gilbert
1435dgilbert at interlog dot com
143621st September 2004
1437
lxr.linux.no kindly hosted by Redpill Linpro AS, provider of Linux consulting and operations services since 1995.