1========================
2libATA Developer's Guide
3========================
4
5:Author: Jeff Garzik
6
7Introduction
8============
9
10libATA is a library used inside the Linux kernel to support ATA host
11controllers and devices. libATA provides an ATA driver API, class
12transports for ATA and ATAPI devices, and SCSI<->ATA translation for ATA
13devices according to the T10 SAT specification.
14
15This Guide documents the libATA driver API, library functions, library
16internals, and a couple sample ATA low-level drivers.
17
18libata Driver API
19=================
20
21:c:type:`struct ata_port_operations <ata_port_operations>`
22is defined for every low-level libata
23hardware driver, and it controls how the low-level driver interfaces
24with the ATA and SCSI layers.
25
26FIS-based drivers will hook into the system with ``->qc_prep()`` and
27``->qc_issue()`` high-level hooks. Hardware which behaves in a manner
28similar to PCI IDE hardware may utilize several generic helpers,
29defining at a bare minimum the bus I/O addresses of the ATA shadow
30register blocks.
31
32:c:type:`struct ata_port_operations <ata_port_operations>`
33----------------------------------------------------------
34
35Disable ATA port
36~~~~~~~~~~~~~~~~
37
38::
39
40    void (*port_disable) (struct ata_port *);
41
42
43Called from :c:func:`ata_bus_probe` error path, as well as when unregistering
44from the SCSI module (rmmod, hot unplug). This function should do
45whatever needs to be done to take the port out of use. In most cases,
46:c:func:`ata_port_disable` can be used as this hook.
47
48Called from :c:func:`ata_bus_probe` on a failed probe. Called from
49:c:func:`ata_scsi_release`.
50
51Post-IDENTIFY device configuration
52~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
53
54::
55
56    void (*dev_config) (struct ata_port *, struct ata_device *);
57
58
59Called after IDENTIFY [PACKET] DEVICE is issued to each device found.
60Typically used to apply device-specific fixups prior to issue of SET
61FEATURES - XFER MODE, and prior to operation.
62
63This entry may be specified as NULL in ata_port_operations.
64
65Set PIO/DMA mode
66~~~~~~~~~~~~~~~~
67
68::
69
70    void (*set_piomode) (struct ata_port *, struct ata_device *);
71    void (*set_dmamode) (struct ata_port *, struct ata_device *);
72    void (*post_set_mode) (struct ata_port *);
73    unsigned int (*mode_filter) (struct ata_port *, struct ata_device *, unsigned int);
74
75
76Hooks called prior to the issue of SET FEATURES - XFER MODE command. The
77optional ``->mode_filter()`` hook is called when libata has built a mask of
78the possible modes. This is passed to the ``->mode_filter()`` function
79which should return a mask of valid modes after filtering those
80unsuitable due to hardware limits. It is not valid to use this interface
81to add modes.
82
83``dev->pio_mode`` and ``dev->dma_mode`` are guaranteed to be valid when
84``->set_piomode()`` and when ``->set_dmamode()`` is called. The timings for
85any other drive sharing the cable will also be valid at this point. That
86is the library records the decisions for the modes of each drive on a
87channel before it attempts to set any of them.
88
89``->post_set_mode()`` is called unconditionally, after the SET FEATURES -
90XFER MODE command completes successfully.
91
92``->set_piomode()`` is always called (if present), but ``->set_dma_mode()``
93is only called if DMA is possible.
94
95Taskfile read/write
96~~~~~~~~~~~~~~~~~~~
97
98::
99
100    void (*sff_tf_load) (struct ata_port *ap, struct ata_taskfile *tf);
101    void (*sff_tf_read) (struct ata_port *ap, struct ata_taskfile *tf);
102
103
104``->tf_load()`` is called to load the given taskfile into hardware
105registers / DMA buffers. ``->tf_read()`` is called to read the hardware
106registers / DMA buffers, to obtain the current set of taskfile register
107values. Most drivers for taskfile-based hardware (PIO or MMIO) use
108:c:func:`ata_sff_tf_load` and :c:func:`ata_sff_tf_read` for these hooks.
109
110PIO data read/write
111~~~~~~~~~~~~~~~~~~~
112
113::
114
115    void (*sff_data_xfer) (struct ata_device *, unsigned char *, unsigned int, int);
116
117
118All bmdma-style drivers must implement this hook. This is the low-level
119operation that actually copies the data bytes during a PIO data
120transfer. Typically the driver will choose one of
121:c:func:`ata_sff_data_xfer`, or :c:func:`ata_sff_data_xfer32`.
122
123ATA command execute
124~~~~~~~~~~~~~~~~~~~
125
126::
127
128    void (*sff_exec_command)(struct ata_port *ap, struct ata_taskfile *tf);
129
130
131causes an ATA command, previously loaded with ``->tf_load()``, to be
132initiated in hardware. Most drivers for taskfile-based hardware use
133:c:func:`ata_sff_exec_command` for this hook.
134
135Per-cmd ATAPI DMA capabilities filter
136~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
137
138::
139
140    int (*check_atapi_dma) (struct ata_queued_cmd *qc);
141
142
143Allow low-level driver to filter ATA PACKET commands, returning a status
144indicating whether or not it is OK to use DMA for the supplied PACKET
145command.
146
147This hook may be specified as NULL, in which case libata will assume
148that atapi dma can be supported.
149
150Read specific ATA shadow registers
151~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
152
153::
154
155    u8   (*sff_check_status)(struct ata_port *ap);
156    u8   (*sff_check_altstatus)(struct ata_port *ap);
157
158
159Reads the Status/AltStatus ATA shadow register from hardware. On some
160hardware, reading the Status register has the side effect of clearing
161the interrupt condition. Most drivers for taskfile-based hardware use
162:c:func:`ata_sff_check_status` for this hook.
163
164Write specific ATA shadow register
165~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
166
167::
168
169    void (*sff_set_devctl)(struct ata_port *ap, u8 ctl);
170
171
172Write the device control ATA shadow register to the hardware. Most
173drivers don't need to define this.
174
175Select ATA device on bus
176~~~~~~~~~~~~~~~~~~~~~~~~
177
178::
179
180    void (*sff_dev_select)(struct ata_port *ap, unsigned int device);
181
182
183Issues the low-level hardware command(s) that causes one of N hardware
184devices to be considered 'selected' (active and available for use) on
185the ATA bus. This generally has no meaning on FIS-based devices.
186
187Most drivers for taskfile-based hardware use :c:func:`ata_sff_dev_select` for
188this hook.
189
190Private tuning method
191~~~~~~~~~~~~~~~~~~~~~
192
193::
194
195    void (*set_mode) (struct ata_port *ap);
196
197
198By default libata performs drive and controller tuning in accordance
199with the ATA timing rules and also applies blacklists and cable limits.
200Some controllers need special handling and have custom tuning rules,
201typically raid controllers that use ATA commands but do not actually do
202drive timing.
203
204    **Warning**
205
206    This hook should not be used to replace the standard controller
207    tuning logic when a controller has quirks. Replacing the default
208    tuning logic in that case would bypass handling for drive and bridge
209    quirks that may be important to data reliability. If a controller
210    needs to filter the mode selection it should use the mode_filter
211    hook instead.
212
213Control PCI IDE BMDMA engine
214~~~~~~~~~~~~~~~~~~~~~~~~~~~~
215
216::
217
218    void (*bmdma_setup) (struct ata_queued_cmd *qc);
219    void (*bmdma_start) (struct ata_queued_cmd *qc);
220    void (*bmdma_stop) (struct ata_port *ap);
221    u8   (*bmdma_status) (struct ata_port *ap);
222
223
224When setting up an IDE BMDMA transaction, these hooks arm
225(``->bmdma_setup``), fire (``->bmdma_start``), and halt (``->bmdma_stop``) the
226hardware's DMA engine. ``->bmdma_status`` is used to read the standard PCI
227IDE DMA Status register.
228
229These hooks are typically either no-ops, or simply not implemented, in
230FIS-based drivers.
231
232Most legacy IDE drivers use :c:func:`ata_bmdma_setup` for the
233:c:func:`bmdma_setup` hook. :c:func:`ata_bmdma_setup` will write the pointer
234to the PRD table to the IDE PRD Table Address register, enable DMA in the DMA
235Command register, and call :c:func:`exec_command` to begin the transfer.
236
237Most legacy IDE drivers use :c:func:`ata_bmdma_start` for the
238:c:func:`bmdma_start` hook. :c:func:`ata_bmdma_start` will write the
239ATA_DMA_START flag to the DMA Command register.
240
241Many legacy IDE drivers use :c:func:`ata_bmdma_stop` for the
242:c:func:`bmdma_stop` hook. :c:func:`ata_bmdma_stop` clears the ATA_DMA_START
243flag in the DMA command register.
244
245Many legacy IDE drivers use :c:func:`ata_bmdma_status` as the
246:c:func:`bmdma_status` hook.
247
248High-level taskfile hooks
249~~~~~~~~~~~~~~~~~~~~~~~~~
250
251::
252
253    enum ata_completion_errors (*qc_prep) (struct ata_queued_cmd *qc);
254    int (*qc_issue) (struct ata_queued_cmd *qc);
255
256
257Higher-level hooks, these two hooks can potentially supersede several of
258the above taskfile/DMA engine hooks. ``->qc_prep`` is called after the
259buffers have been DMA-mapped, and is typically used to populate the
260hardware's DMA scatter-gather table. Some drivers use the standard
261:c:func:`ata_bmdma_qc_prep` and :c:func:`ata_bmdma_dumb_qc_prep` helper
262functions, but more advanced drivers roll their own.
263
264``->qc_issue`` is used to make a command active, once the hardware and S/G
265tables have been prepared. IDE BMDMA drivers use the helper function
266:c:func:`ata_sff_qc_issue` for taskfile protocol-based dispatch. More
267advanced drivers implement their own ``->qc_issue``.
268
269:c:func:`ata_sff_qc_issue` calls ``->sff_tf_load()``, ``->bmdma_setup()``, and
270``->bmdma_start()`` as necessary to initiate a transfer.
271
272Exception and probe handling (EH)
273~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
274
275::
276
277    void (*eng_timeout) (struct ata_port *ap);
278    void (*phy_reset) (struct ata_port *ap);
279
280
281Deprecated. Use ``->error_handler()`` instead.
282
283::
284
285    void (*freeze) (struct ata_port *ap);
286    void (*thaw) (struct ata_port *ap);
287
288
289:c:func:`ata_port_freeze` is called when HSM violations or some other
290condition disrupts normal operation of the port. A frozen port is not
291allowed to perform any operation until the port is thawed, which usually
292follows a successful reset.
293
294The optional ``->freeze()`` callback can be used for freezing the port
295hardware-wise (e.g. mask interrupt and stop DMA engine). If a port
296cannot be frozen hardware-wise, the interrupt handler must ack and clear
297interrupts unconditionally while the port is frozen.
298
299The optional ``->thaw()`` callback is called to perform the opposite of
300``->freeze()``: prepare the port for normal operation once again. Unmask
301interrupts, start DMA engine, etc.
302
303::
304
305    void (*error_handler) (struct ata_port *ap);
306
307
308``->error_handler()`` is a driver's hook into probe, hotplug, and recovery
309and other exceptional conditions. The primary responsibility of an
310implementation is to call :c:func:`ata_do_eh` or :c:func:`ata_bmdma_drive_eh`
311with a set of EH hooks as arguments:
312
313'prereset' hook (may be NULL) is called during an EH reset, before any
314other actions are taken.
315
316'postreset' hook (may be NULL) is called after the EH reset is
317performed. Based on existing conditions, severity of the problem, and
318hardware capabilities,
319
320Either 'softreset' (may be NULL) or 'hardreset' (may be NULL) will be
321called to perform the low-level EH reset.
322
323::
324
325    void (*post_internal_cmd) (struct ata_queued_cmd *qc);
326
327
328Perform any hardware-specific actions necessary to finish processing
329after executing a probe-time or EH-time command via
330:c:func:`ata_exec_internal`.
331
332Hardware interrupt handling
333~~~~~~~~~~~~~~~~~~~~~~~~~~~
334
335::
336
337    irqreturn_t (*irq_handler)(int, void *, struct pt_regs *);
338    void (*irq_clear) (struct ata_port *);
339
340
341``->irq_handler`` is the interrupt handling routine registered with the
342system, by libata. ``->irq_clear`` is called during probe just before the
343interrupt handler is registered, to be sure hardware is quiet.
344
345The second argument, dev_instance, should be cast to a pointer to
346:c:type:`struct ata_host_set <ata_host_set>`.
347
348Most legacy IDE drivers use :c:func:`ata_sff_interrupt` for the irq_handler
349hook, which scans all ports in the host_set, determines which queued
350command was active (if any), and calls ata_sff_host_intr(ap,qc).
351
352Most legacy IDE drivers use :c:func:`ata_sff_irq_clear` for the
353:c:func:`irq_clear` hook, which simply clears the interrupt and error flags
354in the DMA status register.
355
356SATA phy read/write
357~~~~~~~~~~~~~~~~~~~
358
359::
360
361    int (*scr_read) (struct ata_port *ap, unsigned int sc_reg,
362             u32 *val);
363    int (*scr_write) (struct ata_port *ap, unsigned int sc_reg,
364                       u32 val);
365
366
367Read and write standard SATA phy registers. Currently only used if
368``->phy_reset`` hook called the :c:func:`sata_phy_reset` helper function.
369sc_reg is one of SCR_STATUS, SCR_CONTROL, SCR_ERROR, or SCR_ACTIVE.
370
371Init and shutdown
372~~~~~~~~~~~~~~~~~
373
374::
375
376    int (*port_start) (struct ata_port *ap);
377    void (*port_stop) (struct ata_port *ap);
378    void (*host_stop) (struct ata_host_set *host_set);
379
380
381``->port_start()`` is called just after the data structures for each port
382are initialized. Typically this is used to alloc per-port DMA buffers /
383tables / rings, enable DMA engines, and similar tasks. Some drivers also
384use this entry point as a chance to allocate driver-private memory for
385``ap->private_data``.
386
387Many drivers use :c:func:`ata_port_start` as this hook or call it from their
388own :c:func:`port_start` hooks. :c:func:`ata_port_start` allocates space for
389a legacy IDE PRD table and returns.
390
391``->port_stop()`` is called after ``->host_stop()``. Its sole function is to
392release DMA/memory resources, now that they are no longer actively being
393used. Many drivers also free driver-private data from port at this time.
394
395``->host_stop()`` is called after all ``->port_stop()`` calls have completed.
396The hook must finalize hardware shutdown, release DMA and other
397resources, etc. This hook may be specified as NULL, in which case it is
398not called.
399
400Error handling
401==============
402
403This chapter describes how errors are handled under libata. Readers are
404advised to read SCSI EH (Documentation/scsi/scsi_eh.rst) and ATA
405exceptions doc first.
406
407Origins of commands
408-------------------
409
410In libata, a command is represented with
411:c:type:`struct ata_queued_cmd <ata_queued_cmd>` or qc.
412qc's are preallocated during port initialization and repetitively used
413for command executions. Currently only one qc is allocated per port but
414yet-to-be-merged NCQ branch allocates one for each tag and maps each qc
415to NCQ tag 1-to-1.
416
417libata commands can originate from two sources - libata itself and SCSI
418midlayer. libata internal commands are used for initialization and error
419handling. All normal blk requests and commands for SCSI emulation are
420passed as SCSI commands through queuecommand callback of SCSI host
421template.
422
423How commands are issued
424-----------------------
425
426Internal commands
427    First, qc is allocated and initialized using :c:func:`ata_qc_new_init`.
428    Although :c:func:`ata_qc_new_init` doesn't implement any wait or retry
429    mechanism when qc is not available, internal commands are currently
430    issued only during initialization and error recovery, so no other
431    command is active and allocation is guaranteed to succeed.
432
433    Once allocated qc's taskfile is initialized for the command to be
434    executed. qc currently has two mechanisms to notify completion. One
435    is via ``qc->complete_fn()`` callback and the other is completion
436    ``qc->waiting``. ``qc->complete_fn()`` callback is the asynchronous path
437    used by normal SCSI translated commands and ``qc->waiting`` is the
438    synchronous (issuer sleeps in process context) path used by internal
439    commands.
440
441    Once initialization is complete, host_set lock is acquired and the
442    qc is issued.
443
444SCSI commands
445    All libata drivers use :c:func:`ata_scsi_queuecmd` as
446    ``hostt->queuecommand`` callback. scmds can either be simulated or
447    translated. No qc is involved in processing a simulated scmd. The
448    result is computed right away and the scmd is completed.
449
450    For a translated scmd, :c:func:`ata_qc_new_init` is invoked to allocate a
451    qc and the scmd is translated into the qc. SCSI midlayer's
452    completion notification function pointer is stored into
453    ``qc->scsidone``.
454
455    ``qc->complete_fn()`` callback is used for completion notification. ATA
456    commands use :c:func:`ata_scsi_qc_complete` while ATAPI commands use
457    :c:func:`atapi_qc_complete`. Both functions end up calling ``qc->scsidone``
458    to notify upper layer when the qc is finished. After translation is
459    completed, the qc is issued with :c:func:`ata_qc_issue`.
460
461    Note that SCSI midlayer invokes hostt->queuecommand while holding
462    host_set lock, so all above occur while holding host_set lock.
463
464How commands are processed
465--------------------------
466
467Depending on which protocol and which controller are used, commands are
468processed differently. For the purpose of discussion, a controller which
469uses taskfile interface and all standard callbacks is assumed.
470
471Currently 6 ATA command protocols are used. They can be sorted into the
472following four categories according to how they are processed.
473
474ATA NO DATA or DMA
475    ATA_PROT_NODATA and ATA_PROT_DMA fall into this category. These
476    types of commands don't require any software intervention once
477    issued. Device will raise interrupt on completion.
478
479ATA PIO
480    ATA_PROT_PIO is in this category. libata currently implements PIO
481    with polling. ATA_NIEN bit is set to turn off interrupt and
482    pio_task on ata_wq performs polling and IO.
483
484ATAPI NODATA or DMA
485    ATA_PROT_ATAPI_NODATA and ATA_PROT_ATAPI_DMA are in this
486    category. packet_task is used to poll BSY bit after issuing PACKET
487    command. Once BSY is turned off by the device, packet_task
488    transfers CDB and hands off processing to interrupt handler.
489
490ATAPI PIO
491    ATA_PROT_ATAPI is in this category. ATA_NIEN bit is set and, as
492    in ATAPI NODATA or DMA, packet_task submits cdb. However, after
493    submitting cdb, further processing (data transfer) is handed off to
494    pio_task.
495
496How commands are completed
497--------------------------
498
499Once issued, all qc's are either completed with :c:func:`ata_qc_complete` or
500time out. For commands which are handled by interrupts,
501:c:func:`ata_host_intr` invokes :c:func:`ata_qc_complete`, and, for PIO tasks,
502pio_task invokes :c:func:`ata_qc_complete`. In error cases, packet_task may
503also complete commands.
504
505:c:func:`ata_qc_complete` does the following.
506
5071. DMA memory is unmapped.
508
5092. ATA_QCFLAG_ACTIVE is cleared from qc->flags.
510
5113. :c:func:`qc->complete_fn` callback is invoked. If the return value of the
512   callback is not zero. Completion is short circuited and
513   :c:func:`ata_qc_complete` returns.
514
5154. :c:func:`__ata_qc_complete` is called, which does
516
517   1. ``qc->flags`` is cleared to zero.
518
519   2. ``ap->active_tag`` and ``qc->tag`` are poisoned.
520
521   3. ``qc->waiting`` is cleared & completed (in that order).
522
523   4. qc is deallocated by clearing appropriate bit in ``ap->qactive``.
524
525So, it basically notifies upper layer and deallocates qc. One exception
526is short-circuit path in #3 which is used by :c:func:`atapi_qc_complete`.
527
528For all non-ATAPI commands, whether it fails or not, almost the same
529code path is taken and very little error handling takes place. A qc is
530completed with success status if it succeeded, with failed status
531otherwise.
532
533However, failed ATAPI commands require more handling as REQUEST SENSE is
534needed to acquire sense data. If an ATAPI command fails,
535:c:func:`ata_qc_complete` is invoked with error status, which in turn invokes
536:c:func:`atapi_qc_complete` via ``qc->complete_fn()`` callback.
537
538This makes :c:func:`atapi_qc_complete` set ``scmd->result`` to
539SAM_STAT_CHECK_CONDITION, complete the scmd and return 1. As the
540sense data is empty but ``scmd->result`` is CHECK CONDITION, SCSI midlayer
541will invoke EH for the scmd, and returning 1 makes :c:func:`ata_qc_complete`
542to return without deallocating the qc. This leads us to
543:c:func:`ata_scsi_error` with partially completed qc.
544
545:c:func:`ata_scsi_error`
546------------------------
547
548:c:func:`ata_scsi_error` is the current ``transportt->eh_strategy_handler()``
549for libata. As discussed above, this will be entered in two cases -
550timeout and ATAPI error completion. This function calls low level libata
551driver's :c:func:`eng_timeout` callback, the standard callback for which is
552:c:func:`ata_eng_timeout`. It checks if a qc is active and calls
553:c:func:`ata_qc_timeout` on the qc if so. Actual error handling occurs in
554:c:func:`ata_qc_timeout`.
555
556If EH is invoked for timeout, :c:func:`ata_qc_timeout` stops BMDMA and
557completes the qc. Note that as we're currently in EH, we cannot call
558scsi_done. As described in SCSI EH doc, a recovered scmd should be
559either retried with :c:func:`scsi_queue_insert` or finished with
560:c:func:`scsi_finish_command`. Here, we override ``qc->scsidone`` with
561:c:func:`scsi_finish_command` and calls :c:func:`ata_qc_complete`.
562
563If EH is invoked due to a failed ATAPI qc, the qc here is completed but
564not deallocated. The purpose of this half-completion is to use the qc as
565place holder to make EH code reach this place. This is a bit hackish,
566but it works.
567
568Once control reaches here, the qc is deallocated by invoking
569:c:func:`__ata_qc_complete` explicitly. Then, internal qc for REQUEST SENSE
570is issued. Once sense data is acquired, scmd is finished by directly
571invoking :c:func:`scsi_finish_command` on the scmd. Note that as we already
572have completed and deallocated the qc which was associated with the
573scmd, we don't need to/cannot call :c:func:`ata_qc_complete` again.
574
575Problems with the current EH
576----------------------------
577
578-  Error representation is too crude. Currently any and all error
579   conditions are represented with ATA STATUS and ERROR registers.
580   Errors which aren't ATA device errors are treated as ATA device
581   errors by setting ATA_ERR bit. Better error descriptor which can
582   properly represent ATA and other errors/exceptions is needed.
583
584-  When handling timeouts, no action is taken to make device forget
585   about the timed out command and ready for new commands.
586
587-  EH handling via :c:func:`ata_scsi_error` is not properly protected from
588   usual command processing. On EH entrance, the device is not in
589   quiescent state. Timed out commands may succeed or fail any time.
590   pio_task and atapi_task may still be running.
591
592-  Too weak error recovery. Devices / controllers causing HSM mismatch
593   errors and other errors quite often require reset to return to known
594   state. Also, advanced error handling is necessary to support features
595   like NCQ and hotplug.
596
597-  ATA errors are directly handled in the interrupt handler and PIO
598   errors in pio_task. This is problematic for advanced error handling
599   for the following reasons.
600
601   First, advanced error handling often requires context and internal qc
602   execution.
603
604   Second, even a simple failure (say, CRC error) needs information
605   gathering and could trigger complex error handling (say, resetting &
606   reconfiguring). Having multiple code paths to gather information,
607   enter EH and trigger actions makes life painful.
608
609   Third, scattered EH code makes implementing low level drivers
610   difficult. Low level drivers override libata callbacks. If EH is
611   scattered over several places, each affected callbacks should perform
612   its part of error handling. This can be error prone and painful.
613
614libata Library
615==============
616
617.. kernel-doc:: drivers/ata/libata-core.c
618   :export:
619
620libata Core Internals
621=====================
622
623.. kernel-doc:: drivers/ata/libata-core.c
624   :internal:
625
626.. kernel-doc:: drivers/ata/libata-eh.c
627
628libata SCSI translation/emulation
629=================================
630
631.. kernel-doc:: drivers/ata/libata-scsi.c
632   :export:
633
634.. kernel-doc:: drivers/ata/libata-scsi.c
635   :internal:
636
637ATA errors and exceptions
638=========================
639
640This chapter tries to identify what error/exception conditions exist for
641ATA/ATAPI devices and describe how they should be handled in
642implementation-neutral way.
643
644The term 'error' is used to describe conditions where either an explicit
645error condition is reported from device or a command has timed out.
646
647The term 'exception' is either used to describe exceptional conditions
648which are not errors (say, power or hotplug events), or to describe both
649errors and non-error exceptional conditions. Where explicit distinction
650between error and exception is necessary, the term 'non-error exception'
651is used.
652
653Exception categories
654--------------------
655
656Exceptions are described primarily with respect to legacy taskfile + bus
657master IDE interface. If a controller provides other better mechanism
658for error reporting, mapping those into categories described below
659shouldn't be difficult.
660
661In the following sections, two recovery actions - reset and
662reconfiguring transport - are mentioned. These are described further in
663`EH recovery actions <#exrec>`__.
664
665HSM violation
666~~~~~~~~~~~~~
667
668This error is indicated when STATUS value doesn't match HSM requirement
669during issuing or execution any ATA/ATAPI command.
670
671-  ATA_STATUS doesn't contain !BSY && DRDY && !DRQ while trying to
672   issue a command.
673
674-  !BSY && !DRQ during PIO data transfer.
675
676-  DRQ on command completion.
677
678-  !BSY && ERR after CDB transfer starts but before the last byte of CDB
679   is transferred. ATA/ATAPI standard states that "The device shall not
680   terminate the PACKET command with an error before the last byte of
681   the command packet has been written" in the error outputs description
682   of PACKET command and the state diagram doesn't include such
683   transitions.
684
685In these cases, HSM is violated and not much information regarding the
686error can be acquired from STATUS or ERROR register. IOW, this error can
687be anything - driver bug, faulty device, controller and/or cable.
688
689As HSM is violated, reset is necessary to restore known state.
690Reconfiguring transport for lower speed might be helpful too as
691transmission errors sometimes cause this kind of errors.
692
693ATA/ATAPI device error (non-NCQ / non-CHECK CONDITION)
694~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
695
696These are errors detected and reported by ATA/ATAPI devices indicating
697device problems. For this type of errors, STATUS and ERROR register
698values are valid and describe error condition. Note that some of ATA bus
699errors are detected by ATA/ATAPI devices and reported using the same
700mechanism as device errors. Those cases are described later in this
701section.
702
703For ATA commands, this type of errors are indicated by !BSY && ERR
704during command execution and on completion.
705
706For ATAPI commands,
707
708-  !BSY && ERR && ABRT right after issuing PACKET indicates that PACKET
709   command is not supported and falls in this category.
710
711-  !BSY && ERR(==CHK) && !ABRT after the last byte of CDB is transferred
712   indicates CHECK CONDITION and doesn't fall in this category.
713
714-  !BSY && ERR(==CHK) && ABRT after the last byte of CDB is transferred
715   \*probably\* indicates CHECK CONDITION and doesn't fall in this
716   category.
717
718Of errors detected as above, the following are not ATA/ATAPI device
719errors but ATA bus errors and should be handled according to
720`ATA bus error <#excatATAbusErr>`__.
721
722CRC error during data transfer
723    This is indicated by ICRC bit in the ERROR register and means that
724    corruption occurred during data transfer. Up to ATA/ATAPI-7, the
725    standard specifies that this bit is only applicable to UDMA
726    transfers but ATA/ATAPI-8 draft revision 1f says that the bit may be
727    applicable to multiword DMA and PIO.
728
729ABRT error during data transfer or on completion
730    Up to ATA/ATAPI-7, the standard specifies that ABRT could be set on
731    ICRC errors and on cases where a device is not able to complete a
732    command. Combined with the fact that MWDMA and PIO transfer errors
733    aren't allowed to use ICRC bit up to ATA/ATAPI-7, it seems to imply
734    that ABRT bit alone could indicate transfer errors.
735
736    However, ATA/ATAPI-8 draft revision 1f removes the part that ICRC
737    errors can turn on ABRT. So, this is kind of gray area. Some
738    heuristics are needed here.
739
740ATA/ATAPI device errors can be further categorized as follows.
741
742Media errors
743    This is indicated by UNC bit in the ERROR register. ATA devices
744    reports UNC error only after certain number of retries cannot
745    recover the data, so there's nothing much else to do other than
746    notifying upper layer.
747
748    READ and WRITE commands report CHS or LBA of the first failed sector
749    but ATA/ATAPI standard specifies that the amount of transferred data
750    on error completion is indeterminate, so we cannot assume that
751    sectors preceding the failed sector have been transferred and thus
752    cannot complete those sectors successfully as SCSI does.
753
754Media changed / media change requested error
755    <<TODO: fill here>>
756
757Address error
758    This is indicated by IDNF bit in the ERROR register. Report to upper
759    layer.
760
761Other errors
762    This can be invalid command or parameter indicated by ABRT ERROR bit
763    or some other error condition. Note that ABRT bit can indicate a lot
764    of things including ICRC and Address errors. Heuristics needed.
765
766Depending on commands, not all STATUS/ERROR bits are applicable. These
767non-applicable bits are marked with "na" in the output descriptions but
768up to ATA/ATAPI-7 no definition of "na" can be found. However,
769ATA/ATAPI-8 draft revision 1f describes "N/A" as follows.
770
771    3.2.3.3a N/A
772        A keyword the indicates a field has no defined value in this
773        standard and should not be checked by the host or device. N/A
774        fields should be cleared to zero.
775
776So, it seems reasonable to assume that "na" bits are cleared to zero by
777devices and thus need no explicit masking.
778
779ATAPI device CHECK CONDITION
780~~~~~~~~~~~~~~~~~~~~~~~~~~~~
781
782ATAPI device CHECK CONDITION error is indicated by set CHK bit (ERR bit)
783in the STATUS register after the last byte of CDB is transferred for a
784PACKET command. For this kind of errors, sense data should be acquired
785to gather information regarding the errors. REQUEST SENSE packet command
786should be used to acquire sense data.
787
788Once sense data is acquired, this type of errors can be handled
789similarly to other SCSI errors. Note that sense data may indicate ATA
790bus error (e.g. Sense Key 04h HARDWARE ERROR && ASC/ASCQ 47h/00h SCSI
791PARITY ERROR). In such cases, the error should be considered as an ATA
792bus error and handled according to `ATA bus error <#excatATAbusErr>`__.
793
794ATA device error (NCQ)
795~~~~~~~~~~~~~~~~~~~~~~
796
797NCQ command error is indicated by cleared BSY and set ERR bit during NCQ
798command phase (one or more NCQ commands outstanding). Although STATUS
799and ERROR registers will contain valid values describing the error, READ
800LOG EXT is required to clear the error condition, determine which
801command has failed and acquire more information.
802
803READ LOG EXT Log Page 10h reports which tag has failed and taskfile
804register values describing the error. With this information the failed
805command can be handled as a normal ATA command error as in
806`ATA/ATAPI device error (non-NCQ / non-CHECK CONDITION) <#excatDevErr>`__
807and all other in-flight commands must be retried. Note that this retry
808should not be counted - it's likely that commands retried this way would
809have completed normally if it were not for the failed command.
810
811Note that ATA bus errors can be reported as ATA device NCQ errors. This
812should be handled as described in `ATA bus error <#excatATAbusErr>`__.
813
814If READ LOG EXT Log Page 10h fails or reports NQ, we're thoroughly
815screwed. This condition should be treated according to
816`HSM violation <#excatHSMviolation>`__.
817
818ATA bus error
819~~~~~~~~~~~~~
820
821ATA bus error means that data corruption occurred during transmission
822over ATA bus (SATA or PATA). This type of errors can be indicated by
823
824-  ICRC or ABRT error as described in
825   `ATA/ATAPI device error (non-NCQ / non-CHECK CONDITION) <#excatDevErr>`__.
826
827-  Controller-specific error completion with error information
828   indicating transmission error.
829
830-  On some controllers, command timeout. In this case, there may be a
831   mechanism to determine that the timeout is due to transmission error.
832
833-  Unknown/random errors, timeouts and all sorts of weirdities.
834
835As described above, transmission errors can cause wide variety of
836symptoms ranging from device ICRC error to random device lockup, and,
837for many cases, there is no way to tell if an error condition is due to
838transmission error or not; therefore, it's necessary to employ some kind
839of heuristic when dealing with errors and timeouts. For example,
840encountering repetitive ABRT errors for known supported command is
841likely to indicate ATA bus error.
842
843Once it's determined that ATA bus errors have possibly occurred,
844lowering ATA bus transmission speed is one of actions which may
845alleviate the problem. See `Reconfigure transport <#exrecReconf>`__ for
846more information.
847
848PCI bus error
849~~~~~~~~~~~~~
850
851Data corruption or other failures during transmission over PCI (or other
852system bus). For standard BMDMA, this is indicated by Error bit in the
853BMDMA Status register. This type of errors must be logged as it
854indicates something is very wrong with the system. Resetting host
855controller is recommended.
856
857Late completion
858~~~~~~~~~~~~~~~
859
860This occurs when timeout occurs and the timeout handler finds out that
861the timed out command has completed successfully or with error. This is
862usually caused by lost interrupts. This type of errors must be logged.
863Resetting host controller is recommended.
864
865Unknown error (timeout)
866~~~~~~~~~~~~~~~~~~~~~~~
867
868This is when timeout occurs and the command is still processing or the
869host and device are in unknown state. When this occurs, HSM could be in
870any valid or invalid state. To bring the device to known state and make
871it forget about the timed out command, resetting is necessary. The timed
872out command may be retried.
873
874Timeouts can also be caused by transmission errors. Refer to
875`ATA bus error <#excatATAbusErr>`__ for more details.
876
877Hotplug and power management exceptions
878~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~~
879
880<<TODO: fill here>>
881
882EH recovery actions
883-------------------
884
885This section discusses several important recovery actions.
886
887Clearing error condition
888~~~~~~~~~~~~~~~~~~~~~~~~
889
890Many controllers require its error registers to be cleared by error
891handler. Different controllers may have different requirements.
892
893For SATA, it's strongly recommended to clear at least SError register
894during error handling.
895
896Reset
897~~~~~
898
899During EH, resetting is necessary in the following cases.
900
901-  HSM is in unknown or invalid state
902
903-  HBA is in unknown or invalid state
904
905-  EH needs to make HBA/device forget about in-flight commands
906
907-  HBA/device behaves weirdly
908
909Resetting during EH might be a good idea regardless of error condition
910to improve EH robustness. Whether to reset both or either one of HBA and
911device depends on situation but the following scheme is recommended.
912
913-  When it's known that HBA is in ready state but ATA/ATAPI device is in
914   unknown state, reset only device.
915
916-  If HBA is in unknown state, reset both HBA and device.
917
918HBA resetting is implementation specific. For a controller complying to
919taskfile/BMDMA PCI IDE, stopping active DMA transaction may be
920sufficient iff BMDMA state is the only HBA context. But even mostly
921taskfile/BMDMA PCI IDE complying controllers may have implementation
922specific requirements and mechanism to reset themselves. This must be
923addressed by specific drivers.
924
925OTOH, ATA/ATAPI standard describes in detail ways to reset ATA/ATAPI
926devices.
927
928PATA hardware reset
929    This is hardware initiated device reset signalled with asserted PATA
930    RESET- signal. There is no standard way to initiate hardware reset
931    from software although some hardware provides registers that allow
932    driver to directly tweak the RESET- signal.
933
934Software reset
935    This is achieved by turning CONTROL SRST bit on for at least 5us.
936    Both PATA and SATA support it but, in case of SATA, this may require
937    controller-specific support as the second Register FIS to clear SRST
938    should be transmitted while BSY bit is still set. Note that on PATA,
939    this resets both master and slave devices on a channel.
940
941EXECUTE DEVICE DIAGNOSTIC command
942    Although ATA/ATAPI standard doesn't describe exactly, EDD implies
943    some level of resetting, possibly similar level with software reset.
944    Host-side EDD protocol can be handled with normal command processing
945    and most SATA controllers should be able to handle EDD's just like
946    other commands. As in software reset, EDD affects both devices on a
947    PATA bus.
948
949    Although EDD does reset devices, this doesn't suit error handling as
950    EDD cannot be issued while BSY is set and it's unclear how it will
951    act when device is in unknown/weird state.
952
953ATAPI DEVICE RESET command
954    This is very similar to software reset except that reset can be
955    restricted to the selected device without affecting the other device
956    sharing the cable.
957
958SATA phy reset
959    This is the preferred way of resetting a SATA device. In effect,
960    it's identical to PATA hardware reset. Note that this can be done
961    with the standard SCR Control register. As such, it's usually easier
962    to implement than software reset.
963
964One more thing to consider when resetting devices is that resetting
965clears certain configuration parameters and they need to be set to their
966previous or newly adjusted values after reset.
967
968Parameters affected are.
969
970-  CHS set up with INITIALIZE DEVICE PARAMETERS (seldom used)
971
972-  Parameters set with SET FEATURES including transfer mode setting
973
974-  Block count set with SET MULTIPLE MODE
975
976-  Other parameters (SET MAX, MEDIA LOCK...)
977
978ATA/ATAPI standard specifies that some parameters must be maintained
979across hardware or software reset, but doesn't strictly specify all of
980them. Always reconfiguring needed parameters after reset is required for
981robustness. Note that this also applies when resuming from deep sleep
982(power-off).
983
984Also, ATA/ATAPI standard requires that IDENTIFY DEVICE / IDENTIFY PACKET
985DEVICE is issued after any configuration parameter is updated or a
986hardware reset and the result used for further operation. OS driver is
987required to implement revalidation mechanism to support this.
988
989Reconfigure transport
990~~~~~~~~~~~~~~~~~~~~~
991
992For both PATA and SATA, a lot of corners are cut for cheap connectors,
993cables or controllers and it's quite common to see high transmission
994error rate. This can be mitigated by lowering transmission speed.
995
996The following is a possible scheme Jeff Garzik suggested.
997
998    If more than $N (3?) transmission errors happen in 15 minutes,
999
1000    -  if SATA, decrease SATA PHY speed. if speed cannot be decreased,
1001
1002    -  decrease UDMA xfer speed. if at UDMA0, switch to PIO4,
1003
1004    -  decrease PIO xfer speed. if at PIO3, complain, but continue
1005
1006ata_piix Internals
1007===================
1008
1009.. kernel-doc:: drivers/ata/ata_piix.c
1010   :internal:
1011
1012sata_sil Internals
1013===================
1014
1015.. kernel-doc:: drivers/ata/sata_sil.c
1016   :internal:
1017
1018Thanks
1019======
1020
1021The bulk of the ATA knowledge comes thanks to long conversations with
1022Andre Hedrick (www.linux-ide.org), and long hours pondering the ATA and
1023SCSI specifications.
1024
1025Thanks to Alan Cox for pointing out similarities between SATA and SCSI,
1026and in general for motivation to hack on libata.
1027
1028libata's device detection method, ata_pio_devchk, and in general all
1029the early probing was based on extensive study of Hale Landis's
1030probe/reset code in his ATADRVR driver (www.ata-atapi.com).
1031