1.. SPDX-License-Identifier: GPL-2.0 2 3============================================= 4SCSI mid_level - lower_level driver interface 5============================================= 6 7Introduction 8============ 9This document outlines the interface between the Linux SCSI mid level and 10SCSI lower level drivers. Lower level drivers (LLDs) are variously called 11host bus adapter (HBA) drivers and host drivers (HD). A "host" in this 12context is a bridge between a computer IO bus (e.g. PCI or ISA) and a 13single SCSI initiator port on a SCSI transport. An "initiator" port 14(SCSI terminology, see SAM-3 at http://www.t10.org) sends SCSI commands 15to "target" SCSI ports (e.g. disks). There can be many LLDs in a running 16system, but only one per hardware type. Most LLDs can control one or more 17SCSI HBAs. Some HBAs contain multiple hosts. 18 19In some cases the SCSI transport is an external bus that already has 20its own subsystem in Linux (e.g. USB and ieee1394). In such cases the 21SCSI subsystem LLD is a software bridge to the other driver subsystem. 22Examples are the usb-storage driver (found in the drivers/usb/storage 23directory) and the ieee1394/sbp2 driver (found in the drivers/ieee1394 24directory). 25 26For example, the aic7xxx LLD controls Adaptec SCSI parallel interface 27(SPI) controllers based on that company's 7xxx chip series. The aic7xxx 28LLD can be built into the kernel or loaded as a module. There can only be 29one aic7xxx LLD running in a Linux system but it may be controlling many 30HBAs. These HBAs might be either on PCI daughter-boards or built into 31the motherboard (or both). Some aic7xxx based HBAs are dual controllers 32and thus represent two hosts. Like most modern HBAs, each aic7xxx host 33has its own PCI device address. [The one-to-one correspondence between 34a SCSI host and a PCI device is common but not required (e.g. with 35ISA adapters).] 36 37The SCSI mid level isolates an LLD from other layers such as the SCSI 38upper layer drivers and the block layer. 39 40This version of the document roughly matches linux kernel version 2.6.8 . 41 42Documentation 43============= 44There is a SCSI documentation directory within the kernel source tree, 45typically Documentation/scsi . Most documents are in reStructuredText 46format. This file is named scsi_mid_low_api.rst and can be 47found in that directory. A more recent copy of this document may be found 48at https://docs.kernel.org/scsi/scsi_mid_low_api.html. Many LLDs are 49documented in Documentation/scsi (e.g. aic7xxx.rst). The SCSI mid-level is 50briefly described in scsi.rst which contains a URL to a document describing 51the SCSI subsystem in the Linux Kernel 2.4 series. Two upper level 52drivers have documents in that directory: st.rst (SCSI tape driver) and 53scsi-generic.rst (for the sg driver). 54 55Some documentation (or URLs) for LLDs may be found in the C source code 56or in the same directory as the C source code. For example to find a URL 57about the USB mass storage driver see the 58/usr/src/linux/drivers/usb/storage directory. 59 60Driver structure 61================ 62Traditionally an LLD for the SCSI subsystem has been at least two files in 63the drivers/scsi directory. For example, a driver called "xyz" has a header 64file "xyz.h" and a source file "xyz.c". [Actually there is no good reason 65why this couldn't all be in one file; the header file is superfluous.] Some 66drivers that have been ported to several operating systems have more than 67two files. For example the aic7xxx driver has separate files for generic 68and OS-specific code (e.g. FreeBSD and Linux). Such drivers tend to have 69their own directory under the drivers/scsi directory. 70 71When a new LLD is being added to Linux, the following files (found in the 72drivers/scsi directory) will need some attention: Makefile and Kconfig . 73It is probably best to study how existing LLDs are organized. 74 75As the 2.5 series development kernels evolve into the 2.6 series 76production series, changes are being introduced into this interface. An 77example of this is driver initialization code where there are now 2 models 78available. The older one, similar to what was found in the lk 2.4 series, 79is based on hosts that are detected at HBA driver load time. This will be 80referred to the "passive" initialization model. The newer model allows HBAs 81to be hot plugged (and unplugged) during the lifetime of the LLD and will 82be referred to as the "hotplug" initialization model. The newer model is 83preferred as it can handle both traditional SCSI equipment that is 84permanently connected as well as modern "SCSI" devices (e.g. USB or 85IEEE 1394 connected digital cameras) that are hotplugged. Both 86initialization models are discussed in the following sections. 87 88An LLD interfaces to the SCSI subsystem several ways: 89 90 a) directly invoking functions supplied by the mid level 91 b) passing a set of function pointers to a registration function 92 supplied by the mid level. The mid level will then invoke these 93 functions at some point in the future. The LLD will supply 94 implementations of these functions. 95 c) direct access to instances of well known data structures maintained 96 by the mid level 97 98Those functions in group a) are listed in a section entitled "Mid level 99supplied functions" below. 100 101Those functions in group b) are listed in a section entitled "Interface 102functions" below. Their function pointers are placed in the members of 103"struct scsi_host_template", an instance of which is passed to 104scsi_host_alloc(). Those interface functions that the LLD does not 105wish to supply should have NULL placed in the corresponding member of 106struct scsi_host_template. Defining an instance of struct 107scsi_host_template at file scope will cause NULL to be placed in function 108pointer members not explicitly initialized. 109 110Those usages in group c) should be handled with care, especially in a 111"hotplug" environment. LLDs should be aware of the lifetime of instances 112that are shared with the mid level and other layers. 113 114All functions defined within an LLD and all data defined at file scope 115should be static. For example the sdev_init() function in an LLD 116called "xxx" could be defined as 117``static int xxx_sdev_init(struct scsi_device * sdev) { /* code */ }`` 118 119 120Hotplug initialization model 121============================ 122In this model an LLD controls when SCSI hosts are introduced and removed 123from the SCSI subsystem. Hosts can be introduced as early as driver 124initialization and removed as late as driver shutdown. Typically a driver 125will respond to a sysfs probe() callback that indicates an HBA has been 126detected. After confirming that the new device is one that the LLD wants 127to control, the LLD will initialize the HBA and then register a new host 128with the SCSI mid level. 129 130During LLD initialization the driver should register itself with the 131appropriate IO bus on which it expects to find HBA(s) (e.g. the PCI bus). 132This can probably be done via sysfs. Any driver parameters (especially 133those that are writable after the driver is loaded) could also be 134registered with sysfs at this point. The SCSI mid level first becomes 135aware of an LLD when that LLD registers its first HBA. 136 137At some later time, the LLD becomes aware of an HBA and what follows 138is a typical sequence of calls between the LLD and the mid level. 139This example shows the mid level scanning the newly introduced HBA for 3 140scsi devices of which only the first 2 respond:: 141 142 HBA PROBE: assume 2 SCSI devices found in scan 143 LLD mid level LLD 144 ===-------------------=========--------------------===------ 145 scsi_host_alloc() --> 146 scsi_add_host() ----> 147 scsi_scan_host() -------+ 148 | 149 sdev_init() 150 sdev_configure() --> scsi_change_queue_depth() 151 | 152 sdev_init() 153 sdev_configure() 154 | 155 sdev_init() *** 156 sdev_destroy() *** 157 158 159 *** For scsi devices that the mid level tries to scan but do not 160 respond, a sdev_init(), sdev_destroy() pair is called. 161 162If the LLD wants to adjust the default queue settings, it can invoke 163scsi_change_queue_depth() in its sdev_configure() routine. 164 165When an HBA is being removed it could be as part of an orderly shutdown 166associated with the LLD module being unloaded (e.g. with the "rmmod" 167command) or in response to a "hot unplug" indicated by sysfs()'s 168remove() callback being invoked. In either case, the sequence is the 169same:: 170 171 HBA REMOVE: assume 2 SCSI devices attached 172 LLD mid level LLD 173 ===----------------------=========-----------------===------ 174 scsi_remove_host() ---------+ 175 | 176 sdev_destroy() 177 sdev_destroy() 178 scsi_host_put() 179 180It may be useful for a LLD to keep track of struct Scsi_Host instances 181(a pointer is returned by scsi_host_alloc()). Such instances are "owned" 182by the mid-level. struct Scsi_Host instances are freed from 183scsi_host_put() when the reference count hits zero. 184 185Hot unplugging an HBA that controls a disk which is processing SCSI 186commands on a mounted file system is an interesting situation. Reference 187counting logic is being introduced into the mid level to cope with many 188of the issues involved. See the section on reference counting below. 189 190 191The hotplug concept may be extended to SCSI devices. Currently, when an 192HBA is added, the scsi_scan_host() function causes a scan for SCSI devices 193attached to the HBA's SCSI transport. On newer SCSI transports the HBA 194may become aware of a new SCSI device _after_ the scan has completed. 195An LLD can use this sequence to make the mid level aware of a SCSI device:: 196 197 SCSI DEVICE hotplug 198 LLD mid level LLD 199 ===-------------------=========--------------------===------ 200 scsi_add_device() ------+ 201 | 202 sdev_init() 203 sdev_configure() [--> scsi_change_queue_depth()] 204 205In a similar fashion, an LLD may become aware that a SCSI device has been 206removed (unplugged) or the connection to it has been interrupted. Some 207existing SCSI transports (e.g. SPI) may not become aware that a SCSI 208device has been removed until a subsequent SCSI command fails which will 209probably cause that device to be set offline by the mid level. An LLD that 210detects the removal of a SCSI device can instigate its removal from 211upper layers with this sequence:: 212 213 SCSI DEVICE hot unplug 214 LLD mid level LLD 215 ===----------------------=========-----------------===------ 216 scsi_remove_device() -------+ 217 | 218 sdev_destroy() 219 220It may be useful for an LLD to keep track of struct scsi_device instances 221(a pointer is passed as the parameter to sdev_init() and 222sdev_configure() callbacks). Such instances are "owned" by the mid-level. 223struct scsi_device instances are freed after sdev_destroy(). 224 225 226Reference Counting 227================== 228The Scsi_Host structure has had reference counting infrastructure added. 229This effectively spreads the ownership of struct Scsi_Host instances 230across the various SCSI layers which use them. Previously such instances 231were exclusively owned by the mid level. LLDs would not usually need to 232directly manipulate these reference counts but there may be some cases 233where they do. 234 235There are 3 reference counting functions of interest associated with 236struct Scsi_Host: 237 238 - scsi_host_alloc(): 239 returns a pointer to new instance of struct 240 Scsi_Host which has its reference count ^^ set to 1 241 242 - scsi_host_get(): 243 adds 1 to the reference count of the given instance 244 245 - scsi_host_put(): 246 decrements 1 from the reference count of the given 247 instance. If the reference count reaches 0 then the given instance 248 is freed 249 250The scsi_device structure has had reference counting infrastructure added. 251This effectively spreads the ownership of struct scsi_device instances 252across the various SCSI layers which use them. Previously such instances 253were exclusively owned by the mid level. See the access functions declared 254towards the end of include/scsi/scsi_device.h . If an LLD wants to keep 255a copy of a pointer to a scsi_device instance it should use scsi_device_get() 256to bump its reference count. When it is finished with the pointer it can 257use scsi_device_put() to decrement its reference count (and potentially 258delete it). 259 260.. Note:: 261 262 struct Scsi_Host actually has 2 reference counts which are manipulated 263 in parallel by these functions. 264 265 266Conventions 267=========== 268First, Linus Torvalds's thoughts on C coding style can be found in the 269Documentation/process/coding-style.rst file. 270 271Also, most C99 enhancements are encouraged to the extent they are supported 272by the relevant gcc compilers. So C99 style structure and array 273initializers are encouraged where appropriate. Don't go too far, 274VLAs are not properly supported yet. An exception to this is the use of 275``//`` style comments; ``/*...*/`` comments are still preferred in Linux. 276 277Well written, tested and documented code, need not be re-formatted to 278comply with the above conventions. For example, the aic7xxx driver 279comes to Linux from FreeBSD and Adaptec's own labs. No doubt FreeBSD 280and Adaptec have their own coding conventions. 281 282 283Mid level supplied functions 284============================ 285These functions are supplied by the SCSI mid level for use by LLDs. 286The names (i.e. entry points) of these functions are exported 287so an LLD that is a module can access them. The kernel will 288arrange for the SCSI mid level to be loaded and initialized before any LLD 289is initialized. The functions below are listed alphabetically and their 290names all start with ``scsi_``. 291 292Summary: 293 294 - scsi_add_device - creates new scsi device (lu) instance 295 - scsi_add_host - perform sysfs registration and set up transport class 296 - scsi_change_queue_depth - change the queue depth on a SCSI device 297 - scsi_bios_ptable - return copy of block device's partition table 298 - scsi_block_requests - prevent further commands being queued to given host 299 - scsi_host_alloc - return a new scsi_host instance whose refcount==1 300 - scsi_host_get - increments Scsi_Host instance's refcount 301 - scsi_host_put - decrements Scsi_Host instance's refcount (free if 0) 302 - scsi_remove_device - detach and remove a SCSI device 303 - scsi_remove_host - detach and remove all SCSI devices owned by host 304 - scsi_report_bus_reset - report scsi _bus_ reset observed 305 - scsi_scan_host - scan SCSI bus 306 - scsi_track_queue_full - track successive QUEUE_FULL events 307 - scsi_unblock_requests - allow further commands to be queued to given host 308 309 310Details:: 311 312 /** 313 * scsi_add_device - creates new scsi device (lu) instance 314 * @shost: pointer to scsi host instance 315 * @channel: channel number (rarely other than 0) 316 * @id: target id number 317 * @lun: logical unit number 318 * 319 * Returns pointer to new struct scsi_device instance or 320 * ERR_PTR(-ENODEV) (or some other bent pointer) if something is 321 * wrong (e.g. no lu responds at given address) 322 * 323 * Might block: yes 324 * 325 * Notes: This call is usually performed internally during a scsi 326 * bus scan when an HBA is added (i.e. scsi_scan_host()). So it 327 * should only be called if the HBA becomes aware of a new scsi 328 * device (lu) after scsi_scan_host() has completed. If successful 329 * this call can lead to sdev_init() and sdev_configure() callbacks 330 * into the LLD. 331 * 332 * Defined in: drivers/scsi/scsi_scan.c 333 **/ 334 struct scsi_device * scsi_add_device(struct Scsi_Host *shost, 335 unsigned int channel, 336 unsigned int id, unsigned int lun) 337 338 339 /** 340 * scsi_add_host - perform sysfs registration and set up transport class 341 * @shost: pointer to scsi host instance 342 * @dev: pointer to struct device of type scsi class 343 * 344 * Returns 0 on success, negative errno of failure (e.g. -ENOMEM) 345 * 346 * Might block: no 347 * 348 * Notes: Only required in "hotplug initialization model" after a 349 * successful call to scsi_host_alloc(). This function does not 350 * scan the bus; this can be done by calling scsi_scan_host() or 351 * in some other transport-specific way. The LLD must set up 352 * the transport template before calling this function and may only 353 * access the transport class data after this function has been called. 354 * 355 * Defined in: drivers/scsi/hosts.c 356 **/ 357 int scsi_add_host(struct Scsi_Host *shost, struct device * dev) 358 359 360 /** 361 * scsi_change_queue_depth - allow LLD to change queue depth on a SCSI device 362 * @sdev: pointer to SCSI device to change queue depth on 363 * @tags Number of tags allowed if tagged queuing enabled, 364 * or number of commands the LLD can queue up 365 * in non-tagged mode (as per cmd_per_lun). 366 * 367 * Returns nothing 368 * 369 * Might block: no 370 * 371 * Notes: Can be invoked any time on a SCSI device controlled by this 372 * LLD. [Specifically during and after sdev_configure() and prior to 373 * sdev_destroy().] Can safely be invoked from interrupt code. 374 * 375 * Defined in: drivers/scsi/scsi.c [see source code for more notes] 376 * 377 **/ 378 int scsi_change_queue_depth(struct scsi_device *sdev, int tags) 379 380 381 /** 382 * scsi_bios_ptable - return copy of block device's partition table 383 * @dev: pointer to block device 384 * 385 * Returns pointer to partition table, or NULL for failure 386 * 387 * Might block: yes 388 * 389 * Notes: Caller owns memory returned (free with kfree() ) 390 * 391 * Defined in: drivers/scsi/scsicam.c 392 **/ 393 unsigned char *scsi_bios_ptable(struct block_device *dev) 394 395 396 /** 397 * scsi_block_requests - prevent further commands being queued to given host 398 * 399 * @shost: pointer to host to block commands on 400 * 401 * Returns nothing 402 * 403 * Might block: no 404 * 405 * Notes: There is no timer nor any other means by which the requests 406 * get unblocked other than the LLD calling scsi_unblock_requests(). 407 * 408 * Defined in: drivers/scsi/scsi_lib.c 409 **/ 410 void scsi_block_requests(struct Scsi_Host * shost) 411 412 413 /** 414 * scsi_host_alloc - create a scsi host adapter instance and perform basic 415 * initialization. 416 * @sht: pointer to scsi host template 417 * @privsize: extra bytes to allocate in hostdata array (which is the 418 * last member of the returned Scsi_Host instance) 419 * 420 * Returns pointer to new Scsi_Host instance or NULL on failure 421 * 422 * Might block: yes 423 * 424 * Notes: When this call returns to the LLD, the SCSI bus scan on 425 * this host has _not_ yet been done. 426 * The hostdata array (by default zero length) is a per host scratch 427 * area for the LLD's exclusive use. 428 * Both associated refcounting objects have their refcount set to 1. 429 * Full registration (in sysfs) and a bus scan are performed later when 430 * scsi_add_host() and scsi_scan_host() are called. 431 * 432 * Defined in: drivers/scsi/hosts.c . 433 **/ 434 struct Scsi_Host * scsi_host_alloc(const struct scsi_host_template * sht, 435 int privsize) 436 437 438 /** 439 * scsi_host_get - increment Scsi_Host instance refcount 440 * @shost: pointer to struct Scsi_Host instance 441 * 442 * Returns nothing 443 * 444 * Might block: currently may block but may be changed to not block 445 * 446 * Notes: Actually increments the counts in two sub-objects 447 * 448 * Defined in: drivers/scsi/hosts.c 449 **/ 450 void scsi_host_get(struct Scsi_Host *shost) 451 452 453 /** 454 * scsi_host_put - decrement Scsi_Host instance refcount, free if 0 455 * @shost: pointer to struct Scsi_Host instance 456 * 457 * Returns nothing 458 * 459 * Might block: currently may block but may be changed to not block 460 * 461 * Notes: Actually decrements the counts in two sub-objects. If the 462 * latter refcount reaches 0, the Scsi_Host instance is freed. 463 * The LLD need not worry exactly when the Scsi_Host instance is 464 * freed, it just shouldn't access the instance after it has balanced 465 * out its refcount usage. 466 * 467 * Defined in: drivers/scsi/hosts.c 468 **/ 469 void scsi_host_put(struct Scsi_Host *shost) 470 471 472 /** 473 * scsi_remove_device - detach and remove a SCSI device 474 * @sdev: a pointer to a scsi device instance 475 * 476 * Returns value: 0 on success, -EINVAL if device not attached 477 * 478 * Might block: yes 479 * 480 * Notes: If an LLD becomes aware that a scsi device (lu) has 481 * been removed but its host is still present then it can request 482 * the removal of that scsi device. If successful this call will 483 * lead to the sdev_destroy() callback being invoked. sdev is an 484 * invalid pointer after this call. 485 * 486 * Defined in: drivers/scsi/scsi_sysfs.c . 487 **/ 488 int scsi_remove_device(struct scsi_device *sdev) 489 490 491 /** 492 * scsi_remove_host - detach and remove all SCSI devices owned by host 493 * @shost: a pointer to a scsi host instance 494 * 495 * Returns value: 0 on success, 1 on failure (e.g. LLD busy ??) 496 * 497 * Might block: yes 498 * 499 * Notes: Should only be invoked if the "hotplug initialization 500 * model" is being used. It should be called _prior_ to 501 * calling scsi_host_put(). 502 * 503 * Defined in: drivers/scsi/hosts.c . 504 **/ 505 int scsi_remove_host(struct Scsi_Host *shost) 506 507 508 /** 509 * scsi_report_bus_reset - report scsi _bus_ reset observed 510 * @shost: a pointer to a scsi host involved 511 * @channel: channel (within) host on which scsi bus reset occurred 512 * 513 * Returns nothing 514 * 515 * Might block: no 516 * 517 * Notes: This only needs to be called if the reset is one which 518 * originates from an unknown location. Resets originated by the 519 * mid level itself don't need to call this, but there should be 520 * no harm. The main purpose of this is to make sure that a 521 * CHECK_CONDITION is properly treated. 522 * 523 * Defined in: drivers/scsi/scsi_error.c . 524 **/ 525 void scsi_report_bus_reset(struct Scsi_Host * shost, int channel) 526 527 528 /** 529 * scsi_scan_host - scan SCSI bus 530 * @shost: a pointer to a scsi host instance 531 * 532 * Might block: yes 533 * 534 * Notes: Should be called after scsi_add_host() 535 * 536 * Defined in: drivers/scsi/scsi_scan.c 537 **/ 538 void scsi_scan_host(struct Scsi_Host *shost) 539 540 541 /** 542 * scsi_track_queue_full - track successive QUEUE_FULL events on given 543 * device to determine if and when there is a need 544 * to adjust the queue depth on the device. 545 * @sdev: pointer to SCSI device instance 546 * @depth: Current number of outstanding SCSI commands on this device, 547 * not counting the one returned as QUEUE_FULL. 548 * 549 * Returns 0 - no change needed 550 * >0 - adjust queue depth to this new depth 551 * -1 - drop back to untagged operation using host->cmd_per_lun 552 * as the untagged command depth 553 * 554 * Might block: no 555 * 556 * Notes: LLDs may call this at any time and we will do "The Right 557 * Thing"; interrupt context safe. 558 * 559 * Defined in: drivers/scsi/scsi.c . 560 **/ 561 int scsi_track_queue_full(struct scsi_device *sdev, int depth) 562 563 564 /** 565 * scsi_unblock_requests - allow further commands to be queued to given host 566 * 567 * @shost: pointer to host to unblock commands on 568 * 569 * Returns nothing 570 * 571 * Might block: no 572 * 573 * Defined in: drivers/scsi/scsi_lib.c . 574 **/ 575 void scsi_unblock_requests(struct Scsi_Host * shost) 576 577 578 579Interface Functions 580=================== 581Interface functions are supplied (defined) by LLDs and their function 582pointers are placed in an instance of struct scsi_host_template which 583is passed to scsi_host_alloc(). 584Some are mandatory. Interface functions should be declared static. The 585accepted convention is that driver "xyz" will declare its sdev_configure() 586function as:: 587 588 static int xyz_sdev_configure(struct scsi_device * sdev); 589 590and so forth for all interface functions listed below. 591 592A pointer to this function should be placed in the 'sdev_configure' member 593of a "struct scsi_host_template" instance. A pointer to such an instance 594should be passed to the mid level's scsi_host_alloc(). 595. 596 597The interface functions are also described in the include/scsi/scsi_host.h 598file immediately above their definition point in "struct scsi_host_template". 599In some cases more detail is given in scsi_host.h than below. 600 601The interface functions are listed below in alphabetical order. 602 603Summary: 604 605 - bios_param - fetch head, sector, cylinder info for a disk 606 - eh_timed_out - notify the host that a command timer expired 607 - eh_abort_handler - abort given command 608 - eh_bus_reset_handler - issue SCSI bus reset 609 - eh_device_reset_handler - issue SCSI device reset 610 - eh_host_reset_handler - reset host (host bus adapter) 611 - info - supply information about given host 612 - ioctl - driver can respond to ioctls 613 - proc_info - supports /proc/scsi/{driver_name}/{host_no} 614 - queuecommand - queue scsi command, invoke 'done' on completion 615 - sdev_init - prior to any commands being sent to a new device 616 - sdev_configure - driver fine tuning for given device after attach 617 - sdev_destroy - given device is about to be shut down 618 619 620Details:: 621 622 /** 623 * bios_param - fetch head, sector, cylinder info for a disk 624 * @sdev: pointer to scsi device context (defined in 625 * include/scsi/scsi_device.h) 626 * @bdev: pointer to block device context (defined in fs.h) 627 * @capacity: device size (in 512 byte sectors) 628 * @params: three element array to place output: 629 * params[0] number of heads (max 255) 630 * params[1] number of sectors (max 63) 631 * params[2] number of cylinders 632 * 633 * Return value is ignored 634 * 635 * Locks: none 636 * 637 * Calling context: process (sd) 638 * 639 * Notes: an arbitrary geometry (based on READ CAPACITY) is used 640 * if this function is not provided. The params array is 641 * pre-initialized with made up values just in case this function 642 * doesn't output anything. 643 * 644 * Optionally defined in: LLD 645 **/ 646 int bios_param(struct scsi_device * sdev, struct block_device *bdev, 647 sector_t capacity, int params[3]) 648 649 650 /** 651 * eh_timed_out - The timer for the command has just fired 652 * @scp: identifies command timing out 653 * 654 * Returns: 655 * 656 * EH_HANDLED: I fixed the error, please complete the command 657 * EH_RESET_TIMER: I need more time, reset the timer and 658 * begin counting again 659 * EH_NOT_HANDLED Begin normal error recovery 660 * 661 * 662 * Locks: None held 663 * 664 * Calling context: interrupt 665 * 666 * Notes: This is to give the LLD an opportunity to do local recovery. 667 * This recovery is limited to determining if the outstanding command 668 * will ever complete. You may not abort and restart the command from 669 * this callback. 670 * 671 * Optionally defined in: LLD 672 **/ 673 int eh_timed_out(struct scsi_cmnd * scp) 674 675 676 /** 677 * eh_abort_handler - abort command associated with scp 678 * @scp: identifies command to be aborted 679 * 680 * Returns SUCCESS if command aborted else FAILED 681 * 682 * Locks: None held 683 * 684 * Calling context: kernel thread 685 * 686 * Notes: This is called only for a command that has timed out. 687 * 688 * Optionally defined in: LLD 689 **/ 690 int eh_abort_handler(struct scsi_cmnd * scp) 691 692 693 /** 694 * eh_bus_reset_handler - issue SCSI bus reset 695 * @scp: SCSI bus that contains this device should be reset 696 * 697 * Returns SUCCESS if command aborted else FAILED 698 * 699 * Locks: None held 700 * 701 * Calling context: kernel thread 702 * 703 * Notes: Invoked from scsi_eh thread. No other commands will be 704 * queued on current host during eh. 705 * 706 * Optionally defined in: LLD 707 **/ 708 int eh_bus_reset_handler(struct scsi_cmnd * scp) 709 710 711 /** 712 * eh_device_reset_handler - issue SCSI device reset 713 * @scp: identifies SCSI device to be reset 714 * 715 * Returns SUCCESS if command aborted else FAILED 716 * 717 * Locks: None held 718 * 719 * Calling context: kernel thread 720 * 721 * Notes: Invoked from scsi_eh thread. No other commands will be 722 * queued on current host during eh. 723 * 724 * Optionally defined in: LLD 725 **/ 726 int eh_device_reset_handler(struct scsi_cmnd * scp) 727 728 729 /** 730 * eh_host_reset_handler - reset host (host bus adapter) 731 * @scp: SCSI host that contains this device should be reset 732 * 733 * Returns SUCCESS if command aborted else FAILED 734 * 735 * Locks: None held 736 * 737 * Calling context: kernel thread 738 * 739 * Notes: Invoked from scsi_eh thread. No other commands will be 740 * queued on current host during eh. 741 * With the default eh_strategy in place, if none of the _abort_, 742 * _device_reset_, _bus_reset_ or this eh handler function are 743 * defined (or they all return FAILED) then the device in question 744 * will be set offline whenever eh is invoked. 745 * 746 * Optionally defined in: LLD 747 **/ 748 int eh_host_reset_handler(struct scsi_cmnd * scp) 749 750 751 /** 752 * info - supply information about given host: driver name plus data 753 * to distinguish given host 754 * @shp: host to supply information about 755 * 756 * Return ASCII null terminated string. [This driver is assumed to 757 * manage the memory pointed to and maintain it, typically for the 758 * lifetime of this host.] 759 * 760 * Locks: none 761 * 762 * Calling context: process 763 * 764 * Notes: Often supplies PCI or ISA information such as IO addresses 765 * and interrupt numbers. If not supplied struct Scsi_Host::name used 766 * instead. It is assumed the returned information fits on one line 767 * (i.e. does not included embedded newlines). 768 * The SCSI_IOCTL_PROBE_HOST ioctl yields the string returned by this 769 * function (or struct Scsi_Host::name if this function is not 770 * available). 771 * 772 * Optionally defined in: LLD 773 **/ 774 const char * info(struct Scsi_Host * shp) 775 776 777 /** 778 * ioctl - driver can respond to ioctls 779 * @sdp: device that ioctl was issued for 780 * @cmd: ioctl number 781 * @arg: pointer to read or write data from. Since it points to 782 * user space, should use appropriate kernel functions 783 * (e.g. copy_from_user() ). In the Unix style this argument 784 * can also be viewed as an unsigned long. 785 * 786 * Returns negative "errno" value when there is a problem. 0 or a 787 * positive value indicates success and is returned to the user space. 788 * 789 * Locks: none 790 * 791 * Calling context: process 792 * 793 * Notes: The SCSI subsystem uses a "trickle down" ioctl model. 794 * The user issues an ioctl() against an upper level driver 795 * (e.g. /dev/sdc) and if the upper level driver doesn't recognize 796 * the 'cmd' then it is passed to the SCSI mid level. If the SCSI 797 * mid level does not recognize it, then the LLD that controls 798 * the device receives the ioctl. According to recent Unix standards 799 * unsupported ioctl() 'cmd' numbers should return -ENOTTY. 800 * 801 * Optionally defined in: LLD 802 **/ 803 int ioctl(struct scsi_device *sdp, int cmd, void *arg) 804 805 806 /** 807 * proc_info - supports /proc/scsi/{driver_name}/{host_no} 808 * @buffer: anchor point to output to (0==writeto1_read0) or fetch from 809 * (1==writeto1_read0). 810 * @start: where "interesting" data is written to. Ignored when 811 * 1==writeto1_read0. 812 * @offset: offset within buffer 0==writeto1_read0 is actually 813 * interested in. Ignored when 1==writeto1_read0 . 814 * @length: maximum (or actual) extent of buffer 815 * @host_no: host number of interest (struct Scsi_Host::host_no) 816 * @writeto1_read0: 1 -> data coming from user space towards driver 817 * (e.g. "echo some_string > /proc/scsi/xyz/2") 818 * 0 -> user what data from this driver 819 * (e.g. "cat /proc/scsi/xyz/2") 820 * 821 * Returns length when 1==writeto1_read0. Otherwise number of chars 822 * output to buffer past offset. 823 * 824 * Locks: none held 825 * 826 * Calling context: process 827 * 828 * Notes: Driven from scsi_proc.c which interfaces to proc_fs. proc_fs 829 * support can now be configured out of the scsi subsystem. 830 * 831 * Optionally defined in: LLD 832 **/ 833 int proc_info(char * buffer, char ** start, off_t offset, 834 int length, int host_no, int writeto1_read0) 835 836 837 /** 838 * queuecommand - queue scsi command, invoke scp->scsi_done on completion 839 * @shost: pointer to the scsi host object 840 * @scp: pointer to scsi command object 841 * 842 * Returns 0 on success. 843 * 844 * If there's a failure, return either: 845 * 846 * SCSI_MLQUEUE_DEVICE_BUSY if the device queue is full, or 847 * SCSI_MLQUEUE_HOST_BUSY if the entire host queue is full 848 * 849 * On both of these returns, the mid-layer will requeue the I/O 850 * 851 * - if the return is SCSI_MLQUEUE_DEVICE_BUSY, only that particular 852 * device will be paused, and it will be unpaused when a command to 853 * the device returns (or after a brief delay if there are no more 854 * outstanding commands to it). Commands to other devices continue 855 * to be processed normally. 856 * 857 * - if the return is SCSI_MLQUEUE_HOST_BUSY, all I/O to the host 858 * is paused and will be unpaused when any command returns from 859 * the host (or after a brief delay if there are no outstanding 860 * commands to the host). 861 * 862 * For compatibility with earlier versions of queuecommand, any 863 * other return value is treated the same as 864 * SCSI_MLQUEUE_HOST_BUSY. 865 * 866 * Other types of errors that are detected immediately may be 867 * flagged by setting scp->result to an appropriate value, 868 * invoking the scp->scsi_done callback, and then returning 0 869 * from this function. If the command is not performed 870 * immediately (and the LLD is starting (or will start) the given 871 * command) then this function should place 0 in scp->result and 872 * return 0. 873 * 874 * Command ownership. If the driver returns zero, it owns the 875 * command and must take responsibility for ensuring the 876 * scp->scsi_done callback is executed. Note: the driver may 877 * call scp->scsi_done before returning zero, but after it has 878 * called scp->scsi_done, it may not return any value other than 879 * zero. If the driver makes a non-zero return, it must not 880 * execute the command's scsi_done callback at any time. 881 * 882 * Locks: up to and including 2.6.36, struct Scsi_Host::host_lock 883 * held on entry (with "irqsave") and is expected to be 884 * held on return. From 2.6.37 onwards, queuecommand is 885 * called without any locks held. 886 * 887 * Calling context: in interrupt (soft irq) or process context 888 * 889 * Notes: This function should be relatively fast. Normally it 890 * will not wait for IO to complete. Hence the scp->scsi_done 891 * callback is invoked (often directly from an interrupt service 892 * routine) some time after this function has returned. In some 893 * cases (e.g. pseudo adapter drivers that manufacture the 894 * response to a SCSI INQUIRY) the scp->scsi_done callback may be 895 * invoked before this function returns. If the scp->scsi_done 896 * callback is not invoked within a certain period the SCSI mid 897 * level will commence error processing. If a status of CHECK 898 * CONDITION is placed in "result" when the scp->scsi_done 899 * callback is invoked, then the LLD driver should perform 900 * autosense and fill in the struct scsi_cmnd::sense_buffer 901 * array. The scsi_cmnd::sense_buffer array is zeroed prior to 902 * the mid level queuing a command to an LLD. 903 * 904 * Defined in: LLD 905 **/ 906 int queuecommand(struct Scsi_Host *shost, struct scsi_cmnd * scp) 907 908 909 /** 910 * sdev_init - prior to any commands being sent to a new device 911 * (i.e. just prior to scan) this call is made 912 * @sdp: pointer to new device (about to be scanned) 913 * 914 * Returns 0 if ok. Any other return is assumed to be an error and 915 * the device is ignored. 916 * 917 * Locks: none 918 * 919 * Calling context: process 920 * 921 * Notes: Allows the driver to allocate any resources for a device 922 * prior to its initial scan. The corresponding scsi device may not 923 * exist but the mid level is just about to scan for it (i.e. send 924 * and INQUIRY command plus ...). If a device is found then 925 * sdev_configure() will be called while if a device is not found 926 * sdev_destroy() is called. 927 * For more details see the include/scsi/scsi_host.h file. 928 * 929 * Optionally defined in: LLD 930 **/ 931 int sdev_init(struct scsi_device *sdp) 932 933 934 /** 935 * sdev_configure - driver fine tuning for given device just after it 936 * has been first scanned (i.e. it responded to an 937 * INQUIRY) 938 * @sdp: device that has just been attached 939 * 940 * Returns 0 if ok. Any other return is assumed to be an error and 941 * the device is taken offline. [offline devices will _not_ have 942 * sdev_destroy() called on them so clean up resources.] 943 * 944 * Locks: none 945 * 946 * Calling context: process 947 * 948 * Notes: Allows the driver to inspect the response to the initial 949 * INQUIRY done by the scanning code and take appropriate action. 950 * For more details see the include/scsi/scsi_host.h file. 951 * 952 * Optionally defined in: LLD 953 **/ 954 int sdev_configure(struct scsi_device *sdp) 955 956 957 /** 958 * sdev_destroy - given device is about to be shut down. All 959 * activity has ceased on this device. 960 * @sdp: device that is about to be shut down 961 * 962 * Returns nothing 963 * 964 * Locks: none 965 * 966 * Calling context: process 967 * 968 * Notes: Mid level structures for given device are still in place 969 * but are about to be torn down. Any per device resources allocated 970 * by this driver for given device should be freed now. No further 971 * commands will be sent for this sdp instance. [However the device 972 * could be re-attached in the future in which case a new instance 973 * of struct scsi_device would be supplied by future sdev_init() 974 * and sdev_configure() calls.] 975 * 976 * Optionally defined in: LLD 977 **/ 978 void sdev_destroy(struct scsi_device *sdp) 979 980 981 982Data Structures 983=============== 984struct scsi_host_template 985------------------------- 986There is one "struct scsi_host_template" instance per LLD [#]_. It is 987typically initialized as a file scope static in a driver's header file. That 988way members that are not explicitly initialized will be set to 0 or NULL. 989Members of interest: 990 991 name 992 - name of driver (may contain spaces, please limit to 993 less than 80 characters) 994 995 proc_name 996 - name used in "/proc/scsi/<proc_name>/<host_no>" and 997 by sysfs in one of its "drivers" directories. Hence 998 "proc_name" should only contain characters acceptable 999 to a Unix file name. 1000 1001 ``(*queuecommand)()`` 1002 - primary callback that the mid level uses to inject 1003 SCSI commands into an LLD. 1004 1005 vendor_id 1006 - a unique value that identifies the vendor supplying 1007 the LLD for the Scsi_Host. Used most often in validating 1008 vendor-specific message requests. Value consists of an 1009 identifier type and a vendor-specific value. 1010 See scsi_netlink.h for a description of valid formats. 1011 1012The structure is defined and commented in include/scsi/scsi_host.h 1013 1014.. [#] In extreme situations a single driver may have several instances 1015 if it controls several different classes of hardware (e.g. an LLD 1016 that handles both ISA and PCI cards and has a separate instance of 1017 struct scsi_host_template for each class). 1018 1019struct Scsi_Host 1020---------------- 1021There is one struct Scsi_Host instance per host (HBA) that an LLD 1022controls. The struct Scsi_Host structure has many members in common 1023with "struct scsi_host_template". When a new struct Scsi_Host instance 1024is created (in scsi_host_alloc() in hosts.c) those common members are 1025initialized from the driver's struct scsi_host_template instance. Members 1026of interest: 1027 1028 host_no 1029 - system wide unique number that is used for identifying 1030 this host. Issued in ascending order from 0. 1031 can_queue 1032 - must be greater than 0; do not send more than can_queue 1033 commands to the adapter. 1034 this_id 1035 - scsi id of host (scsi initiator) or -1 if not known 1036 sg_tablesize 1037 - maximum scatter gather elements allowed by host. 1038 Set this to SG_ALL or less to avoid chained SG lists. 1039 Must be at least 1. 1040 max_sectors 1041 - maximum number of sectors (usually 512 bytes) allowed 1042 in a single SCSI command. The default value of 0 leads 1043 to a setting of SCSI_DEFAULT_MAX_SECTORS (defined in 1044 scsi_host.h) which is currently set to 1024. So for a 1045 disk the maximum transfer size is 512 KB when max_sectors 1046 is not defined. Note that this size may not be sufficient 1047 for disk firmware uploads. 1048 cmd_per_lun 1049 - maximum number of commands that can be queued on devices 1050 controlled by the host. Overridden by LLD calls to 1051 scsi_change_queue_depth(). 1052 hostt 1053 - pointer to driver's struct scsi_host_template from which 1054 this struct Scsi_Host instance was spawned 1055 hostt->proc_name 1056 - name of LLD. This is the driver name that sysfs uses 1057 transportt 1058 - pointer to driver's struct scsi_transport_template instance 1059 (if any). FC and SPI transports currently supported. 1060 hostdata[0] 1061 - area reserved for LLD at end of struct Scsi_Host. Size 1062 is set by the second argument (named 'privsize') to 1063 scsi_host_alloc(). 1064 1065The scsi_host structure is defined in include/scsi/scsi_host.h 1066 1067struct scsi_device 1068------------------ 1069Generally, there is one instance of this structure for each SCSI logical unit 1070on a host. Scsi devices connected to a host are uniquely identified by a 1071channel number, target id and logical unit number (lun). 1072The structure is defined in include/scsi/scsi_device.h 1073 1074struct scsi_cmnd 1075---------------- 1076Instances of this structure convey SCSI commands to the LLD and responses 1077back to the mid level. The SCSI mid level will ensure that no more SCSI 1078commands become queued against the LLD than are indicated by 1079scsi_change_queue_depth() (or struct Scsi_Host::cmd_per_lun). There will 1080be at least one instance of struct scsi_cmnd available for each SCSI device. 1081Members of interest: 1082 1083 cmnd 1084 - array containing SCSI command 1085 cmd_len 1086 - length (in bytes) of SCSI command 1087 sc_data_direction 1088 - direction of data transfer in data phase. See 1089 "enum dma_data_direction" in include/linux/dma-mapping.h 1090 result 1091 - should be set by LLD prior to calling 'done'. A value 1092 of 0 implies a successfully completed command (and all 1093 data (if any) has been transferred to or from the SCSI 1094 target device). 'result' is a 32 bit unsigned integer that 1095 can be viewed as 2 related bytes. The SCSI status value is 1096 in the LSB. See include/scsi/scsi.h status_byte() and 1097 host_byte() macros and related constants. 1098 sense_buffer 1099 - an array (maximum size: SCSI_SENSE_BUFFERSIZE bytes) that 1100 should be written when the SCSI status (LSB of 'result') 1101 is set to CHECK_CONDITION (2). When CHECK_CONDITION is 1102 set, if the top nibble of sense_buffer[0] has the value 7 1103 then the mid level will assume the sense_buffer array 1104 contains a valid SCSI sense buffer; otherwise the mid 1105 level will issue a REQUEST_SENSE SCSI command to 1106 retrieve the sense buffer. The latter strategy is error 1107 prone in the presence of command queuing so the LLD should 1108 always "auto-sense". 1109 device 1110 - pointer to scsi_device object that this command is 1111 associated with. 1112 resid_len (access by calling scsi_set_resid() / scsi_get_resid()) 1113 - an LLD should set this unsigned integer to the requested 1114 transfer length (i.e. 'request_bufflen') less the number 1115 of bytes that are actually transferred. 'resid_len' is 1116 preset to 0 so an LLD can ignore it if it cannot detect 1117 underruns (overruns should not be reported). An LLD 1118 should set 'resid_len' prior to invoking 'done'. The most 1119 interesting case is data transfers from a SCSI target 1120 device (e.g. READs) that underrun. 1121 underflow 1122 - LLD should place (DID_ERROR << 16) in 'result' if 1123 actual number of bytes transferred is less than this 1124 figure. Not many LLDs implement this check and some that 1125 do just output an error message to the log rather than 1126 report a DID_ERROR. Better for an LLD to implement 1127 'resid_len'. 1128 1129It is recommended that a LLD set 'resid_len' on data transfers from a SCSI 1130target device (e.g. READs). It is especially important that 'resid_len' is set 1131when such data transfers have sense keys of MEDIUM ERROR and HARDWARE ERROR 1132(and possibly RECOVERED ERROR). In these cases if a LLD is in doubt how much 1133data has been received then the safest approach is to indicate no bytes have 1134been received. For example: to indicate that no valid data has been received 1135a LLD might use these helpers:: 1136 1137 scsi_set_resid(SCpnt, scsi_bufflen(SCpnt)); 1138 1139where 'SCpnt' is a pointer to a scsi_cmnd object. To indicate only three 512 1140bytes blocks have been received 'resid_len' could be set like this:: 1141 1142 scsi_set_resid(SCpnt, scsi_bufflen(SCpnt) - (3 * 512)); 1143 1144The scsi_cmnd structure is defined in include/scsi/scsi_cmnd.h 1145 1146 1147Locks 1148===== 1149Each struct Scsi_Host instance has a spin_lock called struct 1150Scsi_Host::default_lock which is initialized in scsi_host_alloc() [found in 1151hosts.c]. Within the same function the struct Scsi_Host::host_lock pointer 1152is initialized to point at default_lock. Thereafter lock and unlock 1153operations performed by the mid level use the struct Scsi_Host::host_lock 1154pointer. Previously drivers could override the host_lock pointer but 1155this is not allowed anymore. 1156 1157 1158Autosense 1159========= 1160Autosense (or auto-sense) is defined in the SAM-2 document as "the 1161automatic return of sense data to the application client coincident 1162with the completion of a SCSI command" when a status of CHECK CONDITION 1163occurs. LLDs should perform autosense. This should be done when the LLD 1164detects a CHECK CONDITION status by either: 1165 1166 a) instructing the SCSI protocol (e.g. SCSI Parallel Interface (SPI)) 1167 to perform an extra data in phase on such responses 1168 b) or, the LLD issuing a REQUEST SENSE command itself 1169 1170Either way, when a status of CHECK CONDITION is detected, the mid level 1171decides whether the LLD has performed autosense by checking struct 1172scsi_cmnd::sense_buffer[0] . If this byte has an upper nibble of 7 (or 0xf) 1173then autosense is assumed to have taken place. If it has another value (and 1174this byte is initialized to 0 before each command) then the mid level will 1175issue a REQUEST SENSE command. 1176 1177In the presence of queued commands the "nexus" that maintains sense 1178buffer data from the command that failed until a following REQUEST SENSE 1179may get out of synchronization. This is why it is best for the LLD 1180to perform autosense. 1181 1182 1183Changes since lk 2.4 series 1184=========================== 1185io_request_lock has been replaced by several finer grained locks. The lock 1186relevant to LLDs is struct Scsi_Host::host_lock and there is 1187one per SCSI host. 1188 1189The older error handling mechanism has been removed. This means the 1190LLD interface functions abort() and reset() have been removed. 1191The struct scsi_host_template::use_new_eh_code flag has been removed. 1192 1193In the 2.4 series the SCSI subsystem configuration descriptions were 1194aggregated with the configuration descriptions from all other Linux 1195subsystems in the Documentation/Configure.help file. In the 2.6 series, 1196the SCSI subsystem now has its own (much smaller) drivers/scsi/Kconfig 1197file that contains both configuration and help information. 1198 1199struct SHT has been renamed to struct scsi_host_template. 1200 1201Addition of the "hotplug initialization model" and many extra functions 1202to support it. 1203 1204 1205Credits 1206======= 1207The following people have contributed to this document: 1208 1209 - Mike Anderson <andmike at us dot ibm dot com> 1210 - James Bottomley <James dot Bottomley at hansenpartnership dot com> 1211 - Patrick Mansfield <patmans at us dot ibm dot com> 1212 - Christoph Hellwig <hch at infradead dot org> 1213 - Doug Ledford <dledford at redhat dot com> 1214 - Andries Brouwer <Andries dot Brouwer at cwi dot nl> 1215 - Randy Dunlap <rdunlap at xenotime dot net> 1216 - Alan Stern <stern at rowland dot harvard dot edu> 1217 1218 1219Douglas Gilbert 1220dgilbert at interlog dot com 1221 122221st September 2004 1223