1 Linux Kernel 2.6 series
2 SCSI mid_level - lower_level driver interface
3 =============================================
7 This document outlines the interface between the Linux SCSI mid level and
8 SCSI lower level drivers. Lower level drivers (LLDs) are variously called
9 host bus adapter (HBA) drivers and host drivers (HD). A "host" in this
10 context is a bridge between a computer IO bus (e.g. PCI or ISA) and a
11 single SCSI initiator device on a SCSI transport. An "initiator" device
12 (SCSI terminology) sends SCSI commands to "target" SCSI devices (e.g.
13 disks). There can be many LLDs in a running system, but only one per
14 hardware type. Most LLDs can control one or more SCSI HBAs. Some HBAs
15 contain multiple hosts.
17 In some cases the SCSI transport is an external bus that already has
18 its own subsystem in Linux (e.g. USB and ieee1394). In such cases the
19 SCSI subsystem LLD is a software bridge to the other driver subsystem.
20 Examples are the usb-storage driver (found in the drivers/usb/storage
21 directory) and the ieee1394/sbp2 driver (found in the drivers/ieee1394
24 For example, the aic7xxx LLD controls Adaptec SCSI parallel interface
25 (SPI) controllers based on that company's 7xxx chip series. The aic7xxx
26 LLD can be built into the kernel or loaded as a module. There can only be
27 one aic7xxx LLD running in a Linux system but it may be controlling many
28 HBAs. These HBAs might be either on PCI daughter-boards or built into
29 the motherboard (or both). Some aic7xxx based HBAs are dual controllers
30 and thus represent two hosts. Like most modern HBAs, each aic7xxx host
31 has its own PCI device address. [The one-to-one correspondence between
32 a SCSI host and a PCI device is common but not required (e.g. with
33 ISA or MCA adapters).]
35 The SCSI mid level isolates an LLD from other layers such as the SCSI
36 upper layer drivers and the block layer.
38 This version of the document roughly matches linux kernel version 2.6.0-test4.
42 There is a SCSI documentation directory within the kernel source tree,
43 typically /usr/src/linux/Documentation/scsi . Most documents are in plain
44 (i.e. ASCII) text. This file is named scsi_mid_low_api.txt and can be
45 found in that directory. A more recent copy of this document may be found
46 at http://www.torque.net/scsi/scsi_mid_low_api.txt.gz .
47 Many LLDs are documented there (e.g. aic7xxx.txt). The SCSI mid-level is
48 briefly described in scsi.txt which contains a url to a document
49 describing the SCSI subsystem in the lk 2.4 series. Two upper level
50 drivers have documents in that directory: st.txt (SCSI tape driver) and
51 scsi-generic.txt (for the sg driver).
53 Some documentation (or urls) for LLDs may be found in the C source code
54 or in the same directory as the C source code. For example to find a url
55 about the USB mass storage driver see the
56 /usr/src/linux/drivers/usb/storage directory.
58 The Linux kernel source Documentation/DocBook/scsidrivers.tmpl file
59 refers to this file. With the appropriate DocBook tool-set, this permits
60 users to generate html, ps and pdf renderings of information within this
61 file (e.g. the interface functions).
65 Traditionally an LLD for the SCSI subsystem has been at least two files in
66 the drivers/scsi directory. For example, a driver called "xyz" has a header
67 file "xyz.h" and a source file "xyz.c". [Actually there is no good reason
68 why this couldn't all be in one file; the header file is superfluous.] Some
69 drivers that have been ported to several operating systems have more than
70 two files. For example the aic7xxx driver has separate files for generic
71 and OS-specific code (e.g. FreeBSD and Linux). Such drivers tend to have
72 their own directory under the drivers/scsi directory.
74 When a new LLD is being added to Linux, the following files (found in the
75 drivers/scsi directory) will need some attention: Makefile and Kconfig .
76 It is probably best to study how existing LLDs are organized.
78 As the 2.5 series development kernels evolve into the 2.6 series
79 production series, changes are being introduced into this interface. An
80 example of this is driver initialization code where there are now 2 models
81 available. The older one, similar to what was found in the lk 2.4 series,
82 is based on hosts that are detected at HBA driver load time. This will be
83 referred to the "passive" initialization model. The newer model allows HBAs
84 to be hot plugged (and unplugged) during the lifetime of the LLD and will
85 be referred to as the "hotplug" initialization model. The newer model is
86 preferred as it can handle both traditional SCSI equipment that is
87 permanently connected as well as modern "SCSI" devices (e.g. USB or
88 IEEE 1394 connected digital cameras) that are hotplugged. Both
89 initialization models are discussed in the following sections.
91 An LLD interfaces to the SCSI subsystem several ways:
92 a) directly invoking functions supplied by the mid level
93 b) passing a set of function pointers to a registration function
94 supplied by the mid level. The mid level will then invoke these
95 functions at some point in the future. The LLD will supply
96 implementations of these functions.
97 c) direct access to instances of well known data structures maintained
100 Those functions in group a) are listed in a section entitled "Mid level
101 supplied functions" below.
103 Those functions in group b) are listed in a section entitled "Interface
104 functions" below. Their function pointers are placed in the members of
105 "struct scsi_host_template", an instance of which is passed to
106 scsi_host_alloc() ** . Those interface functions that the LLD does not
107 wish to supply should have NULL placed in the corresponding member of
108 struct scsi_host_template. Defining an instance of struct
109 scsi_host_template at file scope will cause NULL to be placed in function
110 pointer members not explicitly initialized.
112 Those usages in group c) should be handled with care, especially in a
113 "hotplug" environment. LLDs should be aware of the lifetime of instances
114 that are shared with the mid level and other layers.
116 All functions defined within an LLD and all data defined at file scope
117 should be static. For example the slave_alloc() function in an LLD
118 called "xxx" could be defined as
119 "static int xxx_slave_alloc(struct scsi_device * sdev) { /* code */ }"
121 ** the scsi_host_alloc() function is a replacement for the rather vaguely
122 named scsi_register() function in most situations. The scsi_register()
123 and scsi_unregister() functions remain to support legacy LLDs that use
124 the passive initialization model.
127 Hotplug initialization model
128 ============================
129 In this model an LLD controls when SCSI hosts are introduced and removed
130 from the SCSI subsystem. Hosts can be introduced as early as driver
131 initialization and removed as late as driver shutdown. Typically a driver
132 will respond to a sysfs probe() callback that indicates an HBA has been
133 detected. After confirming that the new device is one that the LLD wants
134 to control, the LLD will initialize the HBA and then register a new host
135 with the SCSI mid level.
137 During LLD initialization the driver should register itself with the
138 appropriate IO bus on which it expects to find HBA(s) (e.g. the PCI bus).
139 This can probably be done via sysfs. Any driver parameters (especially
140 those that are writable after the driver is loaded) could also be
141 registered with sysfs at this point. The SCSI mid level first becomes
142 aware of an LLD when that LLD registers its first HBA.
144 At some later time, the LLD becomes aware of an HBA and what follows
145 is a typical sequence of calls between the LLD and the mid level.
146 This example shows the mid level scanning the newly introduced HBA for 3
147 scsi devices of which only the first 2 respond:
149 HBA PROBE: assume 2 SCSI devices found in scan
151 ===-------------------=========--------------------===------
152 scsi_host_alloc() -->
153 scsi_add_host() --------+
156 slave_configure() --> scsi_adjust_queue_depth()
163 ------------------------------------------------------------
165 If the LLD wants to adjust the default queue settings, it can invoke
166 scsi_adjust_queue_depth() in its slave_configure() routine.
168 *** For scsi devices that the mid level tries to scan but do not
169 respond, a slave_alloc(), slave_destroy() pair is called.
171 When an HBA is being removed it could be as part of an orderly shutdown
172 associated with the LLD module being unloaded (e.g. with the "rmmod"
173 command) or in response to a "hot unplug" indicated by sysfs()'s
174 remove() callback being invoked. In either case, the sequence is the
177 HBA REMOVE: assume 2 SCSI devices attached
179 ===----------------------=========-----------------===------
180 scsi_remove_host() ---------+
185 ------------------------------------------------------------
187 It may be useful for a LLD to keep track of struct Scsi_Host instances
188 (a pointer is returned by scsi_host_alloc()). Such instances are "owned"
189 by the mid-level. struct Scsi_Host instances are freed from
190 scsi_host_put() when the reference count hits zero.
192 Hot unplugging an HBA that controls a disk which is processing SCSI
193 commands on a mounted file system is an interesting situation. Reference
194 counting logic is being introduced into the mid level to cope with many
195 of the issues involved. See the section on reference counting below.
198 The hotplug concept may be extended to SCSI devices. Currently, when an
199 HBA is added, the scsi_add_host() function causes a scan for SCSI devices
200 attached to the HBA's SCSI transport. On newer SCSI transports the HBA
201 may become aware of a new SCSI device _after_ the scan has completed.
202 An LLD can use this sequence to make the mid level aware of a SCSI device:
206 ===-------------------=========--------------------===------
207 scsi_add_device() ------+
210 slave_configure() [--> scsi_adjust_queue_depth()]
211 ------------------------------------------------------------
213 In a similar fashion, an LLD may become aware that a SCSI device has been
214 removed (unplugged) or the connection to it has been interrupted. Some
215 existing SCSI transports (e.g. SPI) may not become aware that a SCSI
216 device has been removed until a subsequent SCSI command fails which will
217 probably cause that device to be set offline by the mid level. An LLD that
218 detects the removal of a SCSI device can instigate its removal from
219 upper layers with this sequence:
221 SCSI DEVICE hot unplug
223 ===----------------------=========-----------------===------
224 scsi_remove_device() -------+
227 ------------------------------------------------------------
229 It may be useful for an LLD to keep track of struct scsi_device instances
230 (a pointer is passed as the parameter to slave_alloc() and
231 slave_configure() callbacks). Such instances are "owned" by the mid-level.
232 struct scsi_device instances are freed after slave_destroy().
235 Passive initialization model
236 ============================
237 These older LLDs include a file called "scsi_module.c" [yes the ".c" is a
238 little surprising] in their source code. For that file to work an
239 instance of struct scsi_host_template with the name "driver_template"
240 needs to be defined. Here is a typical code sequence used in this model:
241 static struct scsi_host_template driver_template = {
244 #include "scsi_module.c"
246 The scsi_module.c file contains two functions:
247 - init_this_scsi_driver() which is executed when the LLD is
248 initialized (i.e. boot time or module load time)
249 - exit_this_scsi_driver() which is executed when the LLD is shut
250 down (i.e. module unload time)
251 Note: since these functions are tagged with __init and __exit qualifiers
252 an LLD should not call them explicitly (since the kernel does that).
254 Here is an example of an initialization sequence when two hosts are
255 detected (so detect() returns 2) and the SCSI bus scan on each host
256 finds 1 SCSI device (and a second device does not respond).
259 ===----------------------=========-----------------===------
260 init_this_scsi_driver() ----+
262 detect() -----------------+
268 slave_configure() --> scsi_adjust_queue_depth()
276 ------------------------------------------------------------
278 The mid level invokes scsi_adjust_queue_depth() with tagged queuing off and
279 "cmd_per_lun" for that host as the queue length. These settings can be
280 overridden by a slave_configure() supplied by the LLD.
282 *** For scsi devices that the mid level tries to scan but do not
283 respond, a slave_alloc(), slave_destroy() pair is called.
285 Here is an LLD shutdown sequence:
288 ===----------------------=========-----------------===------
289 exit_this_scsi_driver() ----+
292 release() --> scsi_unregister()
295 release() --> scsi_unregister()
296 ------------------------------------------------------------
298 An LLD need not define slave_destroy() (i.e. it is optional).
300 The shortcoming of the "passive initialization model" is that host
301 registration and de-registration are (typically) tied to LLD initialization
302 and shutdown. Once the LLD is initialized then a new host that appears
303 (e.g. via hotplugging) cannot easily be added without a redundant
304 driver shutdown and re-initialization. It may be possible to write an LLD
305 that uses both initialization models.
310 The Scsi_Host structure has had reference counting infrastructure added.
311 This effectively spreads the ownership of struct Scsi_Host instances
312 across the various SCSI layers which use them. Previously such instances
313 were exclusively owned by the mid level. LLDs would not usually need to
314 directly manipulate these reference counts but there may be some cases
317 There are 3 reference counting functions of interest associated with
319 - scsi_host_alloc(): returns a pointer to new instance of struct
320 Scsi_Host which has its reference count ^^ set to 1
321 - scsi_host_get(): adds 1 to the reference count of the given instance
322 - scsi_host_put(): decrements 1 from the reference count of the given
323 instance. If the reference count reaches 0 then the given instance
326 ^^ struct Scsi_Host actually has 2 reference counts which are manipulated
327 in parallel by these functions.
332 First, Linus Torvalds's thoughts on C coding style can be found in the
333 Documentation/CodingStyle file.
335 Next, there is a movement to "outlaw" typedefs introducing synonyms for
336 struct tags. Both can be still found in the SCSI subsystem, but
337 the typedefs have been moved to a single file, scsi_typedefs.h to
338 make their future removal easier, for example:
339 "typedef struct scsi_host_template Scsi_Host_Template;"
341 Also, most C99 enhancements are encouraged to the extent they are supported
342 by the relevant gcc compilers. So C99 style structure and array
343 initializers are encouraged where appropriate. Don't go too far,
344 VLAs are not properly supported yet. An exception to this is the use of
345 "//" style comments; /*...*/ comments are still preferred in Linux.
347 Well written, tested and documented code, need not be re-formatted to
348 comply with the above conventions. For example, the aic7xxx driver
349 comes to Linux from FreeBSD and Adaptec's own labs. No doubt FreeBSD
350 and Adaptec have their own coding conventions.
353 Mid level supplied functions
354 ============================
355 These functions are supplied by the SCSI mid level for use by LLDs.
356 The names (i.e. entry points) of these functions are exported (mainly in
357 scsi_syms.c) so an LLD that is a module can access them. The kernel will
358 arrange for the SCSI mid level to be loaded and initialized before any LLD
359 is initialized. The functions below are listed alphabetically and their
360 names all start with "scsi_".
363 scsi_add_device - creates new scsi device (lu) instance
364 scsi_add_host - perform sysfs registration and SCSI bus scan.
365 scsi_add_timer - (re-)start timer on a SCSI command.
366 scsi_adjust_queue_depth - change the queue depth on a SCSI device
367 scsi_assign_lock - replace default host_lock with given lock
368 scsi_bios_ptable - return copy of block device's partition table
369 scsi_block_requests - prevent further commands being queued to given host
370 scsi_delete_timer - cancel timer on a SCSI command.
371 scsi_host_alloc - return a new scsi_host instance whose refcount==1
372 scsi_host_get - increments Scsi_Host instance's refcount
373 scsi_host_put - decrements Scsi_Host instance's refcount (free if 0)
374 scsi_partsize - parse partition table into cylinders, heads + sectors
375 scsi_register - create and register a scsi host adapter instance.
376 scsi_remove_device - detach and remove a SCSI device
377 scsi_remove_host - detach and remove all SCSI devices owned by host
378 scsi_report_bus_reset - report scsi _bus_ reset observed
379 scsi_set_device - place device reference in host structure
380 scsi_to_pci_dma_dir - convert SCSI subsystem direction flag to PCI
381 scsi_to_sbus_dma_dir - convert SCSI subsystem direction flag to SBUS
382 scsi_track_queue_full - track successive QUEUE_FULL events
383 scsi_unblock_requests - allow further commands to be queued to given host
384 scsi_unregister - [calls scsi_host_put()]
390 * scsi_add_device - creates new scsi device (lu) instance
391 * @shost: pointer to scsi host instance
392 * @channel: channel number (rarely other than 0)
393 * @id: target id number
394 * @lun: logical unit number
396 * Returns pointer to new struct scsi_device instance or
397 * ERR_PTR(-ENODEV) (or some other bent pointer) if something is
398 * wrong (e.g. no lu responds at given address)
402 * Notes: This call is usually performed internally during a scsi
403 * bus scan when an HBA is added (i.e. scsi_add_host()). So it
404 * should only be called if the HBA becomes aware of a new scsi
405 * device (lu) after scsi_add_host() has completed. If successful
406 * this call we lead to slave_alloc() and slave_configure() callbacks
409 * Defined in: drivers/scsi/scsi_scan.c
411 struct scsi_device * scsi_add_device(struct Scsi_Host *shost,
412 unsigned int channel,
413 unsigned int id, unsigned int lun)
417 * scsi_add_host - perform sysfs registration and SCSI bus scan.
418 * @shost: pointer to scsi host instance
419 * @dev: pointer to struct device host instance
421 * Returns 0 on success, negative errno of failure (e.g. -ENOMEM)
425 * Notes: Only required in "hotplug initialization model" after a
426 * successful call to scsi_host_alloc().
428 * Defined in: drivers/scsi/hosts.c
430 int scsi_add_host(struct Scsi_Host *shost, struct device * dev)
434 * scsi_add_timer - (re-)start timer on a SCSI command.
435 * @scmd: pointer to scsi command instance
436 * @timeout: duration of timeout in "jiffies"
437 * @complete: pointer to function to call if timeout expires
443 * Notes: Each scsi command has its own timer, and as it is added
444 * to the queue, we set up the timer. When the command completes,
445 * we cancel the timer. An LLD can use this function to change
446 * the existing timeout value.
448 * Defined in: drivers/scsi/scsi_error.c
450 void scsi_add_timer(struct scsi_cmnd *scmd, int timeout,
451 void (*complete)(struct scsi_cmnd *))
455 * scsi_adjust_queue_depth - change the queue depth on a SCSI device
456 * @SDpnt: pointer to SCSI device to change queue depth on
457 * @tagged: 0 - no tagged queuing
458 * MSG_SIMPLE_TAG - simple (unordered) tagged queuing
459 * MSG_ORDERED_TAG - ordered tagged queuing
460 * @tags Number of tags allowed if tagged queuing enabled,
461 * or number of commands the LLD can queue up
462 * in non-tagged mode (as per cmd_per_lun).
468 * Notes: Can be invoked any time on a SCSI device controlled by this
469 * LLD. [Specifically during and after slave_configure() and prior to
470 * slave_destroy().] Can safely be invoked from interrupt code. Actual
471 * queue depth change may be delayed until the next command is being
474 * Defined in: drivers/scsi/scsi.c [see source code for more notes]
477 void scsi_adjust_queue_depth(struct scsi_device * SDpnt, int tagged,
482 * scsi_assign_lock - replace default host_lock with given lock
483 * @shost: a pointer to a scsi host instance
484 * @lock: pointer to lock to replace host_lock for this host
490 * Defined in: include/scsi/scsi_host.h .
492 void scsi_assign_lock(struct Scsi_Host *shost, spinlock_t *lock)
496 * scsi_bios_ptable - return copy of block device's partition table
497 * @dev: pointer to block device
499 * Returns pointer to partition table, or NULL for failure
503 * Notes: Caller owns memory returned (free with kfree() )
505 * Defined in: drivers/scsi/scsicam.c
507 unsigned char *scsi_bios_ptable(struct block_device *dev)
511 * scsi_block_requests - prevent further commands being queued to given host
513 * @SHpnt: pointer to host to block commands on
519 * Notes: There is no timer nor any other means by which the requests
520 * get unblocked other than the LLD calling scsi_unblock_requests().
522 * Defined in: drivers/scsi/scsi_lib.c
524 void scsi_block_requests(struct Scsi_Host * SHpnt)
528 * scsi_delete_timer - cancel timer on a SCSI command.
529 * @scmd: pointer to scsi command instance
531 * Returns 1 if able to cancel timer else 0 (i.e. too late or already
534 * Might block: no [may in the future if it invokes del_timer_sync()]
536 * Notes: All commands issued by upper levels already have a timeout
537 * associated with them. An LLD can use this function to cancel the
540 * Defined in: drivers/scsi/scsi_error.c
542 int scsi_delete_timer(struct scsi_cmnd *scmd)
546 * scsi_host_alloc - create and register a scsi host adapter instance.
547 * @shost_tp: pointer to scsi host template
548 * @xtr_bytes: extra bytes to allocate in hostdata array (which is the
549 * last member of the returned Scsi_Host instance)
551 * Returns pointer to new Scsi_Host instance or NULL on failure
555 * Notes: When this call returns to the LLD, the SCSI bus scan on
556 * this host has _not_ yet been done.
557 * The hostdata array (by default zero length) is a per host scratch
559 * Both associated refcounting objects have their refcount set to 1.
561 * Defined in: drivers/scsi/hosts.c .
563 struct Scsi_Host * scsi_host_alloc(struct scsi_host_template *, int xtr_bytes)
567 * scsi_host_get - increment Scsi_Host instance refcount
568 * @shost: pointer to struct Scsi_Host instance
572 * Might block: currently may block but may be changed to not block
574 * Notes: Actually increments the counts in two sub-objects
576 * Defined in: drivers/scsi/hosts.c
578 void scsi_host_get(struct Scsi_Host *shost)
582 * scsi_host_put - decrement Scsi_Host instance refcount, free if 0
583 * @shost: pointer to struct Scsi_Host instance
587 * Might block: currently may block but may be changed to not block
589 * Notes: Actually decrements the counts in two sub-objects. If the
590 * latter refcount reaches 0, the Scsi_Host instance is freed.
591 * The LLD need not worry exactly when the Scsi_Host instance is
592 * freed, it just shouldn't access the instance after it has balanced
593 * out its refcount usage.
595 * Defined in: drivers/scsi/hosts.c
597 void scsi_host_put(struct Scsi_Host *shost)
601 * scsi_partsize - parse partition table into cylinders, heads + sectors
602 * @buf: pointer to partition table
603 * @capacity: size of (total) disk in 512 byte sectors
604 * @cyls: outputs number of cylinders calculated via this pointer
605 * @hds: outputs number of heads calculated via this pointer
606 * @secs: outputs number of sectors calculated via this pointer
608 * Returns 0 on success, -1 on failure
612 * Notes: Caller owns memory returned (free with kfree() )
614 * Defined in: drivers/scsi/scsicam.c
616 int scsi_partsize(unsigned char *buf, unsigned long capacity,
617 unsigned int *cyls, unsigned int *hds, unsigned int *secs)
621 * scsi_register - create and register a scsi host adapter instance.
622 * @shost_tp: pointer to scsi host template
623 * @xtr_bytes: extra bytes to allocate in hostdata array (which is the
624 * last member of the returned Scsi_Host instance)
626 * Returns pointer to new Scsi_Host instance or NULL on failure
630 * Notes: When this call returns to the LLD, the SCSI bus scan on
631 * this host has _not_ yet been done.
632 * The hostdata array (by default zero length) is a per host scratch
635 * Defined in: drivers/scsi/hosts.c .
637 struct Scsi_Host * scsi_register(struct scsi_host_template *, int xtr_bytes)
641 * scsi_remove_device - detach and remove a SCSI device
642 * @sdev: a pointer to a scsi device instance
644 * Returns value: 0 on success, -EINVAL if device not attached
648 * Notes: If an LLD becomes aware that a scsi device (lu) has
649 * been removed but its host is still present then it can request
650 * the removal of that scsi device. If successful this call will
651 * lead to the slave_destroy() callback being invoked. sdev is an
652 * invalid pointer after this call.
654 * Defined in: drivers/scsi/scsi_sysfs.c .
656 int scsi_remove_device(struct scsi_device *sdev)
660 * scsi_remove_host - detach and remove all SCSI devices owned by host
661 * @shost: a pointer to a scsi host instance
663 * Returns value: 0 on success, 1 on failure (e.g. LLD busy ??)
667 * Notes: Should only be invoked if the "hotplug initialization
668 * model" is being used. It should be called _prior_ to
671 * Defined in: drivers/scsi/hosts.c .
673 int scsi_remove_host(struct Scsi_Host *shost)
677 * scsi_report_bus_reset - report scsi _bus_ reset observed
678 * @shost: a pointer to a scsi host involved
679 * @channel: channel (within) host on which scsi bus reset occurred
685 * Notes: This only needs to be called if the reset is one which
686 * originates from an unknown location. Resets originated by the
687 * mid level itself don't need to call this, but there should be
688 * no harm. The main purpose of this is to make sure that a
689 * CHECK_CONDITION is properly treated.
691 * Defined in: drivers/scsi/scsi_error.c .
693 void scsi_report_bus_reset(struct Scsi_Host * shost, int channel)
697 * scsi_set_device - place device reference in host structure
698 * @shost: a pointer to a scsi host instance
699 * @pdev: pointer to device instance to assign
705 * Defined in: include/scsi/scsi_host.h .
707 void scsi_set_device(struct Scsi_Host * shost, struct device * dev)
711 * scsi_to_pci_dma_dir - convert SCSI subsystem direction flag to PCI
712 * @scsi_data_direction: SCSI subsystem direction flag
714 * Returns PCI_DMA_TODEVICE given SCSI_DATA_WRITE,
715 * PCI_DMA_FROMDEVICE given SCSI_DATA_READ
716 * PCI_DMA_BIDIRECTIONAL given SCSI_DATA_UNKNOWN
717 * else returns PCI_DMA_NONE
721 * Notes: The SCSI subsystem now uses the same values for these
722 * constants as the PCI subsystem so this function is a nop.
724 * Defined in: drivers/scsi/scsi.h .
726 int scsi_to_pci_dma_dir(unsigned char scsi_data_direction)
730 * scsi_to_sbus_dma_dir - convert SCSI subsystem direction flag to SBUS
731 * @scsi_data_direction: SCSI subsystem direction flag
733 * Returns SBUS_DMA_TODEVICE given SCSI_DATA_WRITE,
734 * SBUS_DMA_FROMDEVICE given SCSI_DATA_READ
735 * SBUS_DMA_BIDIRECTIONAL given SCSI_DATA_UNKNOWN
736 * else returns SBUS_DMA_NONE
738 * Notes: The SCSI subsystem now uses the same values for these
739 * constants as the SBUS subsystem so this function is a nop.
743 * Defined in: drivers/scsi/scsi.h .
745 int scsi_to_sbus_dma_dir(unsigned char scsi_data_direction)
749 * scsi_track_queue_full - track successive QUEUE_FULL events on given
750 * device to determine if and when there is a need
751 * to adjust the queue depth on the device.
752 * @SDptr: pointer to SCSI device instance
753 * @depth: Current number of outstanding SCSI commands on this device,
754 * not counting the one returned as QUEUE_FULL.
756 * Returns 0 - no change needed
757 * >0 - adjust queue depth to this new depth
758 * -1 - drop back to untagged operation using host->cmd_per_lun
759 * as the untagged command depth
763 * Notes: LLDs may call this at any time and we will do "The Right
764 * Thing"; interrupt context safe.
766 * Defined in: drivers/scsi/scsi.c .
768 int scsi_track_queue_full(Scsi_Device *SDptr, int depth)
772 * scsi_unblock_requests - allow further commands to be queued to given host
774 * @SHpnt: pointer to host to unblock commands on
780 * Defined in: drivers/scsi/scsi_lib.c .
782 void scsi_unblock_requests(struct Scsi_Host * SHpnt)
786 * scsi_unregister - unregister and free memory used by host instance
787 * @shp: pointer to scsi host instance to unregister.
793 * Notes: Should not be invoked if the "hotplug initialization
794 * model" is being used. Called internally by exit_this_scsi_driver()
795 * in the "passive initialization model". Hence a LLD has no need to
796 * call this function directly.
798 * Defined in: drivers/scsi/hosts.c .
800 void scsi_unregister(struct Scsi_Host * shp)
807 Interface functions are supplied (defined) by LLDs and their function
808 pointers are placed in an instance of struct scsi_host_template which
809 is passed to scsi_host_alloc() [or scsi_register() / init_this_scsi_driver()].
810 Some are mandatory. Interface functions should be declared static. The
811 accepted convention is that driver "xyz" will declare its slave_configure()
813 static int xyz_slave_configure(struct scsi_device * sdev);
814 and so forth for all interface functions listed below.
816 A pointer to this function should be placed in the 'slave_configure' member
817 of a "struct scsi_host_template" instance. A pointer to such an instance
818 should be passed to the mid level's scsi_host_alloc() [or scsi_register() /
819 init_this_scsi_driver()].
821 The interface functions are also described in the include/scsi/scsi_host.h
822 file immediately above their definition point in "struct scsi_host_template".
823 In some cases more detail is given in scsi_host.h than below.
825 The interface functions are listed below in alphabetical order.
828 bios_param - fetch head, sector, cylinder info for a disk
829 detect - detects HBAs this driver wants to control
830 eh_abort_handler - abort given command
831 eh_bus_reset_handler - issue SCSI bus reset
832 eh_device_reset_handler - issue SCSI device reset
833 eh_host_reset_handler - reset host (host bus adapter)
834 eh_strategy_handler - driver supplied alternate to scsi_unjam_host()
835 info - supply information about given host
836 ioctl - driver can respond to ioctls
837 proc_info - supports /proc/scsi/{driver_name}/{host_no}
838 queuecommand - queue scsi command, invoke 'done' on completion
839 release - release all resources associated with given host
840 slave_alloc - prior to any commands being sent to a new device
841 slave_configure - driver fine tuning for given device after attach
842 slave_destroy - given device is about to be shut down
848 * bios_param - fetch head, sector, cylinder info for a disk
849 * @sdev: pointer to scsi device context (defined in
850 * include/scsi/scsi_device.h)
851 * @bdev: pointer to block device context (defined in fs.h)
852 * @capacity: device size (in 512 byte sectors)
853 * @params: three element array to place output:
854 * params[0] number of heads (max 255)
855 * params[1] number of sectors (max 63)
856 * params[2] number of cylinders
858 * Return value is ignored
862 * Calling context: process (sd)
864 * Notes: an arbitrary geometry (based on READ CAPACITY) is used
865 * if this function is not provided. The params array is
866 * pre-initialized with made up values just in case this function
867 * doesn't output anything.
869 * Optionally defined in: LLD
871 int bios_param(struct scsi_device * sdev, struct block_device *bdev,
872 sector_t capacity, int params[3])
876 * detect - detects HBAs this driver wants to control
877 * @shtp: host template for this driver.
879 * Returns number of hosts this driver wants to control. 0 means no
880 * suitable hosts found.
884 * Calling context: process [invoked from init_this_scsi_driver()]
886 * Notes: First function called from the SCSI mid level on this
887 * driver. Upper level drivers (e.g. sd) may not (yet) be present.
888 * For each host found, this method should call scsi_register()
891 * Defined in: LLD (required if "passive initialization mode" is used,
892 * not invoked in "hotplug initialization mode")
894 int detect(struct scsi_host_template * shtp)
898 * eh_abort_handler - abort command associated with scp
899 * @scp: identifies command to be aborted
901 * Returns SUCCESS if command aborted else FAILED
903 * Locks: struct Scsi_Host::host_lock held (with irqsave) on entry
904 * and assumed to be held on return.
906 * Calling context: kernel thread
908 * Notes: Invoked from scsi_eh thread. No other commands will be
909 * queued on current host during eh.
911 * Optionally defined in: LLD
913 int eh_abort_handler(struct scsi_cmnd * scp)
917 * eh_bus_reset_handler - issue SCSI bus reset
918 * @scp: SCSI bus that contains this device should be reset
920 * Returns SUCCESS if command aborted else FAILED
922 * Locks: struct Scsi_Host::host_lock held (with irqsave) on entry
923 * and assumed to be held on return.
925 * Calling context: kernel thread
927 * Notes: Invoked from scsi_eh thread. No other commands will be
928 * queued on current host during eh.
930 * Optionally defined in: LLD
932 int eh_bus_reset_handler(struct scsi_cmnd * scp)
936 * eh_device_reset_handler - issue SCSI device reset
937 * @scp: identifies SCSI device to be reset
939 * Returns SUCCESS if command aborted else FAILED
941 * Locks: struct Scsi_Host::host_lock held (with irqsave) on entry
942 * and assumed to be held on return.
944 * Calling context: kernel thread
946 * Notes: Invoked from scsi_eh thread. No other commands will be
947 * queued on current host during eh.
949 * Optionally defined in: LLD
951 int eh_device_reset_handler(struct scsi_cmnd * scp)
955 * eh_host_reset_handler - reset host (host bus adapter)
956 * @scp: SCSI host that contains this device should be reset
958 * Returns SUCCESS if command aborted else FAILED
960 * Locks: struct Scsi_Host::host_lock held (with irqsave) on entry
961 * and assumed to be held on return.
963 * Calling context: kernel thread
965 * Notes: Invoked from scsi_eh thread. No other commands will be
966 * queued on current host during eh.
967 * With the default eh_strategy in place, if none of the _abort_,
968 * _device_reset_, _bus_reset_ or this eh handler function are
969 * defined (or they all return FAILED) then the device in question
970 * will be set offline whenever eh is invoked.
972 * Optionally defined in: LLD
974 int eh_host_reset_handler(struct scsi_cmnd * scp)
978 * eh_strategy_handler - driver supplied alternate to scsi_unjam_host()
979 * @shp: host on which error has occurred
981 * Returns TRUE if host unjammed, else FALSE.
985 * Calling context: kernel thread
987 * Notes: Invoked from scsi_eh thread. LLD supplied alternate to
988 * scsi_unjam_host() found in scsi_error.c
990 * Optionally defined in: LLD
992 int eh_strategy_handler(struct Scsi_Host * shp)
996 * info - supply information about given host: driver name plus data
997 * to distinguish given host
998 * @shp: host to supply information about
1000 * Return ASCII null terminated string. [This driver is assumed to
1001 * manage the memory pointed to and maintain it, typically for the
1002 * lifetime of this host.]
1006 * Calling context: process
1008 * Notes: Often supplies PCI or ISA information such as IO addresses
1009 * and interrupt numbers. If not supplied struct Scsi_Host::name used
1010 * instead. It is assumed the returned information fits on one line
1011 * (i.e. does not included embedded newlines).
1012 * The SCSI_IOCTL_PROBE_HOST ioctl yields the string returned by this
1013 * function (or struct Scsi_Host::name if this function is not
1015 * In a similar manner, init_this_scsi_driver() outputs to the console
1016 * each host's "info" (or name) for the driver it is registering.
1017 * Also if proc_info() is not supplied, the output of this function
1020 * Optionally defined in: LLD
1022 const char * info(struct Scsi_Host * shp)
1026 * ioctl - driver can respond to ioctls
1027 * @sdp: device that ioctl was issued for
1028 * @cmd: ioctl number
1029 * @arg: pointer to read or write data from. Since it points to
1030 * user space, should use appropriate kernel functions
1031 * (e.g. copy_from_user() ). In the Unix style this argument
1032 * can also be viewed as an unsigned long.
1034 * Returns negative "errno" value when there is a problem. 0 or a
1035 * positive value indicates success and is returned to the user space.
1039 * Calling context: process
1041 * Notes: The SCSI subsystem uses a "trickle down" ioctl model.
1042 * The user issues an ioctl() against an upper level driver
1043 * (e.g. /dev/sdc) and if the upper level driver doesn't recognize
1044 * the 'cmd' then it is passed to the SCSI mid level. If the SCSI
1045 * mid level does not recognize it, then the LLD that controls
1046 * the device receives the ioctl. According to recent Unix standards
1047 * unsupported ioctl() 'cmd' numbers should return -ENOTTY.
1048 * However the mid level returns -EINVAL for unrecognized 'cmd'
1049 * numbers when this function is not supplied by the driver.
1050 * Unfortunately some applications expect -EINVAL and react badly
1051 * when -ENOTTY is returned; stick with -EINVAL.
1053 * Optionally defined in: LLD
1055 int ioctl(struct scsi_device *sdp, int cmd, void *arg)
1059 * proc_info - supports /proc/scsi/{driver_name}/{host_no}
1060 * @buffer: anchor point to output to (0==writeto1_read0) or fetch from
1061 * (1==writeto1_read0).
1062 * @start: where "interesting" data is written to. Ignored when
1063 * 1==writeto1_read0.
1064 * @offset: offset within buffer 0==writeto1_read0 is actually
1065 * interested in. Ignored when 1==writeto1_read0 .
1066 * @length: maximum (or actual) extent of buffer
1067 * @host_no: host number of interest (struct Scsi_Host::host_no)
1068 * @writeto1_read0: 1 -> data coming from user space towards driver
1069 * (e.g. "echo some_string > /proc/scsi/xyz/2")
1070 * 0 -> user what data from this driver
1071 * (e.g. "cat /proc/scsi/xyz/2")
1073 * Returns length when 1==writeto1_read0. Otherwise number of chars
1074 * output to buffer past offset.
1078 * Calling context: process
1080 * Notes: Driven from scsi_proc.c which interfaces to proc_fs. proc_fs
1081 * support can now be configured out of the scsi subsystem.
1083 * Optionally defined in: LLD
1085 int proc_info(char * buffer, char ** start, off_t offset,
1086 int length, int hostno, int writeto1_read0)
1090 * queuecommand - queue scsi command, invoke 'done' on completion
1091 * @scp: pointer to scsi command object
1092 * @done: function pointer to be invoked on completion
1094 * Returns 0 on success.
1096 * If there's a failure, return either:
1098 * SCSI_MLQUEUE_DEVICE_BUSY if the device queue is full, or
1099 * SCSI_MLQUEUE_HOST_BUSY if the entire host queue is full
1101 * On both of these returns, the mid-layer will requeue the I/O
1103 * - if the return is SCSI_MLQUEUE_DEVICE_BUSY, only that particular
1104 * device will be paused, and it will be unpaused when a command to
1105 * the device returns (or after a brief delay if there are no more
1106 * outstanding commands to it). Commands to other devices continue
1107 * to be processed normally.
1109 * - if the return is SCSI_MLQUEUE_HOST_BUSY, all I/O to the host
1110 * is paused and will be unpaused when any command returns from
1111 * the host (or after a brief delay if there are no outstanding
1112 * commands to the host).
1114 * For compatibility with earlier versions of queuecommand, any
1115 * other return value is treated the same as
1116 * SCSI_MLQUEUE_HOST_BUSY.
1118 * Other types of errors that are detected immediately may be
1119 * flagged by setting scp->result to an appropriate value,
1120 * invoking the 'done' callback, and then returning 0 from this
1121 * function. If the command is not performed immediately (and the
1122 * LLD is starting (or will start) the given command) then this
1123 * function should place 0 in scp->result and return 0.
1125 * Command ownership. If the driver returns zero, it owns the
1126 * command and must take responsibility for ensuring the 'done'
1127 * callback is executed. Note: the driver may call done before
1128 * returning zero, but after it has called done, it may not
1129 * return any value other than zero. If the driver makes a
1130 * non-zero return, it must not execute the command's done
1131 * callback at any time.
1133 * Locks: struct Scsi_Host::host_lock held on entry (with "irqsave")
1134 * and is expected to be held on return.
1136 * Calling context: in interrupt (soft irq) or process context
1138 * Notes: This function should be relatively fast. Normally it will
1139 * not wait for IO to complete. Hence the 'done' callback is invoked
1140 * (often directly from an interrupt service routine) some time after
1141 * this function has returned. In some cases (e.g. pseudo adapter
1142 * drivers that manufacture the response to a SCSI INQUIRY)
1143 * the 'done' callback may be invoked before this function returns.
1144 * If the 'done' callback is not invoked within a certain period
1145 * the SCSI mid level will commence error processing.
1146 * If a status of CHECK CONDITION is placed in "result" when the
1147 * 'done' callback is invoked, then the LLD driver should
1148 * perform autosense and fill in the struct scsi_cmnd::sense_buffer
1149 * array. The scsi_cmnd::sense_buffer array is zeroed prior to
1150 * the mid level queuing a command to an LLD.
1154 int queuecommand(struct scsi_cmnd * scp,
1155 void (*done)(struct scsi_cmnd *))
1159 * release - release all resources associated with given host
1160 * @shp: host to be released.
1162 * Return value ignored (could soon be a function returning void).
1166 * Calling context: process
1168 * Notes: Invoked from scsi_module.c's exit_this_scsi_driver().
1169 * LLD's implementation of this function should call
1170 * scsi_unregister(shp) prior to returning.
1171 * Only needed for old-style host templates.
1173 * Defined in: LLD (required in "passive initialization model",
1174 * should not be defined in hotplug model)
1176 int release(struct Scsi_Host * shp)
1180 * slave_alloc - prior to any commands being sent to a new device
1181 * (i.e. just prior to scan) this call is made
1182 * @sdp: pointer to new device (about to be scanned)
1184 * Returns 0 if ok. Any other return is assumed to be an error and
1185 * the device is ignored.
1189 * Calling context: process
1191 * Notes: Allows the driver to allocate any resources for a device
1192 * prior to its initial scan. The corresponding scsi device may not
1193 * exist but the mid level is just about to scan for it (i.e. send
1194 * and INQUIRY command plus ...). If a device is found then
1195 * slave_configure() will be called while if a device is not found
1196 * slave_destroy() is called.
1197 * For more details see the include/scsi/scsi_host.h file.
1199 * Optionally defined in: LLD
1201 int slave_alloc(struct scsi_device *sdp)
1205 * slave_configure - driver fine tuning for given device just after it
1206 * has been first scanned (i.e. it responded to an
1208 * @sdp: device that has just been attached
1210 * Returns 0 if ok. Any other return is assumed to be an error and
1211 * the device is taken offline. [offline devices will _not_ have
1212 * slave_destroy() called on them so clean up resources.]
1216 * Calling context: process
1218 * Notes: Allows the driver to inspect the response to the initial
1219 * INQUIRY done by the scanning code and take appropriate action.
1220 * For more details see the include/scsi/scsi_host.h file.
1222 * Optionally defined in: LLD
1224 int slave_configure(struct scsi_device *sdp)
1228 * slave_destroy - given device is about to be shut down. All
1229 * activity has ceased on this device.
1230 * @sdp: device that is about to be shut down
1236 * Calling context: process
1238 * Notes: Mid level structures for given device are still in place
1239 * but are about to be torn down. Any per device resources allocated
1240 * by this driver for given device should be freed now. No further
1241 * commands will be sent for this sdp instance. [However the device
1242 * could be re-attached in the future in which case a new instance
1243 * of struct scsi_device would be supplied by future slave_alloc()
1244 * and slave_configure() calls.]
1246 * Optionally defined in: LLD
1248 void slave_destroy(struct scsi_device *sdp)
1254 struct scsi_host_template
1255 -------------------------
1256 There is one "struct scsi_host_template" instance per LLD ***. It is
1257 typically initialized as a file scope static in a driver's header file. That
1258 way members that are not explicitly initialized will be set to 0 or NULL.
1260 name - name of driver (may contain spaces, please limit to
1261 less than 80 characters)
1262 proc_name - name used in "/proc/scsi/<proc_name>/<host_no>" and
1263 by sysfs in one of its "drivers" directories. Hence
1264 "proc_name" should only contain characters acceptable
1265 to a Unix file name.
1266 (*queuecommand)() - primary callback that the mid level uses to inject
1267 SCSI commands into an LLD.
1268 The structure is defined and commented in include/scsi/scsi_host.h
1270 *** In extreme situations a single driver may have several instances
1271 if it controls several different classes of hardware (e.g. an LLD
1272 that handles both ISA and PCI cards and has a separate instance of
1273 struct scsi_host_template for each class).
1277 There is one struct Scsi_Host instance per host (HBA) that an LLD
1278 controls. The struct Scsi_Host structure has many members in common
1279 with "struct scsi_host_template". When a new struct Scsi_Host instance
1280 is created (in scsi_host_alloc() in hosts.c) those common members are
1281 initialized from the driver's struct scsi_host_template instance. Members
1283 host_no - system wide unique number that is used for identifying
1284 this host. Issued in ascending order from 0 (and the
1285 positioning can be influenced by the scsihosts
1286 kernel boot (or module) parameter)
1287 can_queue - must be greater than 0; do not send more than can_queue
1288 commands to the adapter.
1289 this_id - scsi id of host (scsi initiator) or -1 if not known
1290 sg_tablesize - maximum scatter gather elements allowed by host.
1291 0 implies scatter gather not supported by host
1292 max_sectors - maximum number of sectors (usually 512 bytes) allowed
1293 in a single SCSI command. 0 implies no maximum.
1294 cmd_per_lun - maximum number of commands that can be queued on devices
1295 controlled by the host. Overridden by LLD calls to
1296 scsi_adjust_queue_depth().
1297 unchecked_isa_dma - 1=>only use bottom 16 MB of ram (ISA DMA addressing
1298 restriction), 0=>can use full 32 bit (or better) DMA
1300 use_clustering - 1=>SCSI commands in mid level's queue can be merged,
1301 0=>disallow SCSI command merging
1302 hostt - pointer to driver's struct scsi_host_template from which
1303 this struct Scsi_Host instance was spawned
1304 sh_list - a double linked list of pointers to all struct Scsi_Host
1305 instances (currently ordered by ascending host_no)
1306 my_devices - a double linked list of pointers to struct scsi_device
1307 instances that belong to this host.
1308 hostdata[0] - area reserved for LLD at end of struct Scsi_Host. Size
1309 is set by the second argument (named 'xtr_bytes') to
1310 scsi_host_alloc() or scsi_register().
1311 The structure is defined in include/scsi/scsi_host.h
1315 Generally, there is one instance of this structure for each SCSI logical unit
1316 on a host. Scsi devices connected to a host are uniquely identified by a
1317 channel number, target id and logical unit number (lun).
1318 The structure is defined in include/scsi/scsi_device.h
1322 Instances of this structure convey SCSI commands to the LLD and responses
1323 back to the mid level. The SCSI mid level will ensure that no more SCSI
1324 commands become queued against the LLD than are indicated by
1325 scsi_adjust_queue_depth() (or struct Scsi_Host::cmd_per_lun). There will
1326 be at least one instance of struct scsi_cmnd available for each SCSI device.
1327 The structure is defined in include/scsi/scsi_cmnd.h
1332 Each struct Scsi_Host instance has a spin_lock called struct
1333 Scsi_Host::default_lock which is initialized in scsi_host_alloc() [found in
1334 hosts.c]. Within the same function the struct Scsi_Host::host_lock pointer
1335 is initialized to point at default_lock with the scsi_assign_lock() function.
1336 Thereafter lock and unlock operations performed by the mid level use the
1337 struct Scsi_Host::host_lock pointer.
1339 LLDs can override the use of struct Scsi_Host::default_lock by
1340 using scsi_assign_lock(). The earliest opportunity to do this would
1341 be in the detect() function after it has invoked scsi_register(). It
1342 could be replaced by a coarser grain lock (e.g. per driver) or a
1343 lock of equal granularity (i.e. per host). Using finer grain locks
1344 (e.g. per SCSI device) may be possible by juggling locks in
1349 Autosense (or auto-sense) is defined in the SAM-2 document as "the
1350 automatic return of sense data to the application client coincident
1351 with the completion of a SCSI command" when a status of CHECK CONDITION
1352 occurs. LLDs should perform autosense. This should be done when the LLD
1353 detects a CHECK CONDITION status by either:
1354 a) instructing the SCSI protocol (e.g. SCSI Parallel Interface (SPI))
1355 to perform an extra data in phase on such responses
1356 b) or, the LLD issuing a REQUEST SENSE command itself
1358 Either way, when a status of CHECK CONDITION is detected, the mid level
1359 decides whether the LLD has performed autosense by checking struct
1360 scsi_cmnd::sense_buffer[0] . If this byte has an upper nibble of 7 (or 0xf)
1361 then autosense is assumed to have taken place. If it has another value (and
1362 this byte is initialized to 0 before each command) then the mid level will
1363 issue a REQUEST SENSE command.
1365 In the presence of queued commands the "nexus" that maintains sense
1366 buffer data from the command that failed until a following REQUEST SENSE
1367 may get out of synchronization. This is why it is best for the LLD
1368 to perform autosense.
1371 Changes since lk 2.4 series
1372 ===========================
1373 io_request_lock has been replaced by several finer grained locks. The lock
1374 relevant to LLDs is struct Scsi_Host::host_lock and there is
1377 The older error handling mechanism has been removed. This means the
1378 LLD interface functions abort() and reset() have been removed.
1379 The struct scsi_host_template::use_new_eh_code flag has been removed.
1381 In the 2.4 series the SCSI subsystem configuration descriptions were
1382 aggregated with the configuration descriptions from all other Linux
1383 subsystems in the Documentation/Configure.help file. In the 2.6 series,
1384 the SCSI subsystem now has its own (much smaller) drivers/scsi/Kconfig
1385 file that contains both configuration and help information.
1387 struct SHT has been renamed to struct scsi_host_template.
1389 Addition of the "hotplug initialization model" and many extra functions
1395 The following people have contributed to this document:
1396 Mike Anderson <andmike@us.ibm.com>
1397 James Bottomley <James.Bottomley@steeleye.com>
1398 Patrick Mansfield <patmans@us.ibm.com>
1399 Christoph Hellwig <hch@infradead.org>
1400 Doug Ledford <dledford@redhat.com>
1401 Andries Brouwer <Andries.Brouwer@cwi.nl>
1402 Randy Dunlap <rddunlap@osdl.org>
1406 dgilbert@interlog.com