aboutsummaryrefslogtreecommitdiffstats
path: root/drivers/block
diff options
context:
space:
mode:
authorJonathan Herman <hermanjl@cs.unc.edu>2013-01-22 10:38:37 -0500
committerJonathan Herman <hermanjl@cs.unc.edu>2013-01-22 10:38:37 -0500
commitfcc9d2e5a6c89d22b8b773a64fb4ad21ac318446 (patch)
treea57612d1888735a2ec7972891b68c1ac5ec8faea /drivers/block
parent8dea78da5cee153b8af9c07a2745f6c55057fe12 (diff)
Added missing tegra files.HEADmaster
Diffstat (limited to 'drivers/block')
-rw-r--r--drivers/block/ub.c2510
-rw-r--r--drivers/block/viodasd.c809
2 files changed, 3319 insertions, 0 deletions
diff --git a/drivers/block/ub.c b/drivers/block/ub.c
new file mode 100644
index 00000000000..0e376d46bdd
--- /dev/null
+++ b/drivers/block/ub.c
@@ -0,0 +1,2510 @@
1/*
2 * The low performance USB storage driver (ub).
3 *
4 * Copyright (c) 1999, 2000 Matthew Dharm (mdharm-usb@one-eyed-alien.net)
5 * Copyright (C) 2004 Pete Zaitcev (zaitcev@yahoo.com)
6 *
7 * This work is a part of Linux kernel, is derived from it,
8 * and is not licensed separately. See file COPYING for details.
9 *
10 * TODO (sorted by decreasing priority)
11 * -- Return sense now that rq allows it (we always auto-sense anyway).
12 * -- set readonly flag for CDs, set removable flag for CF readers
13 * -- do inquiry and verify we got a disk and not a tape (for LUN mismatch)
14 * -- verify the 13 conditions and do bulk resets
15 * -- highmem
16 * -- move top_sense and work_bcs into separate allocations (if they survive)
17 * for cache purists and esoteric architectures.
18 * -- Allocate structure for LUN 0 before the first ub_sync_tur, avoid NULL. ?
19 * -- prune comments, they are too volumnous
20 * -- Resove XXX's
21 * -- CLEAR, CLR2STS, CLRRS seem to be ripe for refactoring.
22 */
23#include <linux/kernel.h>
24#include <linux/module.h>
25#include <linux/usb.h>
26#include <linux/usb_usual.h>
27#include <linux/blkdev.h>
28#include <linux/timer.h>
29#include <linux/scatterlist.h>
30#include <linux/slab.h>
31#include <linux/mutex.h>
32#include <scsi/scsi.h>
33
34#define DRV_NAME "ub"
35
36#define UB_MAJOR 180
37
38/*
39 * The command state machine is the key model for understanding of this driver.
40 *
41 * The general rule is that all transitions are done towards the bottom
42 * of the diagram, thus preventing any loops.
43 *
44 * An exception to that is how the STAT state is handled. A counter allows it
45 * to be re-entered along the path marked with [C].
46 *
47 * +--------+
48 * ! INIT !
49 * +--------+
50 * !
51 * ub_scsi_cmd_start fails ->--------------------------------------\
52 * ! !
53 * V !
54 * +--------+ !
55 * ! CMD ! !
56 * +--------+ !
57 * ! +--------+ !
58 * was -EPIPE -->-------------------------------->! CLEAR ! !
59 * ! +--------+ !
60 * ! ! !
61 * was error -->------------------------------------- ! --------->\
62 * ! ! !
63 * /--<-- cmd->dir == NONE ? ! !
64 * ! ! ! !
65 * ! V ! !
66 * ! +--------+ ! !
67 * ! ! DATA ! ! !
68 * ! +--------+ ! !
69 * ! ! +---------+ ! !
70 * ! was -EPIPE -->--------------->! CLR2STS ! ! !
71 * ! ! +---------+ ! !
72 * ! ! ! ! !
73 * ! ! was error -->---- ! --------->\
74 * ! was error -->--------------------- ! ------------- ! --------->\
75 * ! ! ! ! !
76 * ! V ! ! !
77 * \--->+--------+ ! ! !
78 * ! STAT !<--------------------------/ ! !
79 * /--->+--------+ ! !
80 * ! ! ! !
81 * [C] was -EPIPE -->-----------\ ! !
82 * ! ! ! ! !
83 * +<---- len == 0 ! ! !
84 * ! ! ! ! !
85 * ! was error -->--------------------------------------!---------->\
86 * ! ! ! ! !
87 * +<---- bad CSW ! ! !
88 * +<---- bad tag ! ! !
89 * ! ! V ! !
90 * ! ! +--------+ ! !
91 * ! ! ! CLRRS ! ! !
92 * ! ! +--------+ ! !
93 * ! ! ! ! !
94 * \------- ! --------------------[C]--------\ ! !
95 * ! ! ! !
96 * cmd->error---\ +--------+ ! !
97 * ! +--------------->! SENSE !<----------/ !
98 * STAT_FAIL----/ +--------+ !
99 * ! ! V
100 * ! V +--------+
101 * \--------------------------------\--------------------->! DONE !
102 * +--------+
103 */
104
105/*
106 * This many LUNs per USB device.
107 * Every one of them takes a host, see UB_MAX_HOSTS.
108 */
109#define UB_MAX_LUNS 9
110
111/*
112 */
113
114#define UB_PARTS_PER_LUN 8
115
116#define UB_MAX_CDB_SIZE 16 /* Corresponds to Bulk */
117
118#define UB_SENSE_SIZE 18
119
120/*
121 */
122
123/* command block wrapper */
124struct bulk_cb_wrap {
125 __le32 Signature; /* contains 'USBC' */
126 u32 Tag; /* unique per command id */
127 __le32 DataTransferLength; /* size of data */
128 u8 Flags; /* direction in bit 0 */
129 u8 Lun; /* LUN */
130 u8 Length; /* of of the CDB */
131 u8 CDB[UB_MAX_CDB_SIZE]; /* max command */
132};
133
134#define US_BULK_CB_WRAP_LEN 31
135#define US_BULK_CB_SIGN 0x43425355 /*spells out USBC */
136#define US_BULK_FLAG_IN 1
137#define US_BULK_FLAG_OUT 0
138
139/* command status wrapper */
140struct bulk_cs_wrap {
141 __le32 Signature; /* should = 'USBS' */
142 u32 Tag; /* same as original command */
143 __le32 Residue; /* amount not transferred */
144 u8 Status; /* see below */
145};
146
147#define US_BULK_CS_WRAP_LEN 13
148#define US_BULK_CS_SIGN 0x53425355 /* spells out 'USBS' */
149#define US_BULK_STAT_OK 0
150#define US_BULK_STAT_FAIL 1
151#define US_BULK_STAT_PHASE 2
152
153/* bulk-only class specific requests */
154#define US_BULK_RESET_REQUEST 0xff
155#define US_BULK_GET_MAX_LUN 0xfe
156
157/*
158 */
159struct ub_dev;
160
161#define UB_MAX_REQ_SG 9 /* cdrecord requires 32KB and maybe a header */
162#define UB_MAX_SECTORS 64
163
164/*
165 * A second is more than enough for a 32K transfer (UB_MAX_SECTORS)
166 * even if a webcam hogs the bus, but some devices need time to spin up.
167 */
168#define UB_URB_TIMEOUT (HZ*2)
169#define UB_DATA_TIMEOUT (HZ*5) /* ZIP does spin-ups in the data phase */
170#define UB_STAT_TIMEOUT (HZ*5) /* Same spinups and eject for a dataless cmd. */
171#define UB_CTRL_TIMEOUT (HZ/2) /* 500ms ought to be enough to clear a stall */
172
173/*
174 * An instance of a SCSI command in transit.
175 */
176#define UB_DIR_NONE 0
177#define UB_DIR_READ 1
178#define UB_DIR_ILLEGAL2 2
179#define UB_DIR_WRITE 3
180
181#define UB_DIR_CHAR(c) (((c)==UB_DIR_WRITE)? 'w': \
182 (((c)==UB_DIR_READ)? 'r': 'n'))
183
184enum ub_scsi_cmd_state {
185 UB_CMDST_INIT, /* Initial state */
186 UB_CMDST_CMD, /* Command submitted */
187 UB_CMDST_DATA, /* Data phase */
188 UB_CMDST_CLR2STS, /* Clearing before requesting status */
189 UB_CMDST_STAT, /* Status phase */
190 UB_CMDST_CLEAR, /* Clearing a stall (halt, actually) */
191 UB_CMDST_CLRRS, /* Clearing before retrying status */
192 UB_CMDST_SENSE, /* Sending Request Sense */
193 UB_CMDST_DONE /* Final state */
194};
195
196struct ub_scsi_cmd {
197 unsigned char cdb[UB_MAX_CDB_SIZE];
198 unsigned char cdb_len;
199
200 unsigned char dir; /* 0 - none, 1 - read, 3 - write. */
201 enum ub_scsi_cmd_state state;
202 unsigned int tag;
203 struct ub_scsi_cmd *next;
204
205 int error; /* Return code - valid upon done */
206 unsigned int act_len; /* Return size */
207 unsigned char key, asc, ascq; /* May be valid if error==-EIO */
208
209 int stat_count; /* Retries getting status. */
210 unsigned int timeo; /* jiffies until rq->timeout changes */
211
212 unsigned int len; /* Requested length */
213 unsigned int current_sg;
214 unsigned int nsg; /* sgv[nsg] */
215 struct scatterlist sgv[UB_MAX_REQ_SG];
216
217 struct ub_lun *lun;
218 void (*done)(struct ub_dev *, struct ub_scsi_cmd *);
219 void *back;
220};
221
222struct ub_request {
223 struct request *rq;
224 unsigned int current_try;
225 unsigned int nsg; /* sgv[nsg] */
226 struct scatterlist sgv[UB_MAX_REQ_SG];
227};
228
229/*
230 */
231struct ub_capacity {
232 unsigned long nsec; /* Linux size - 512 byte sectors */
233 unsigned int bsize; /* Linux hardsect_size */
234 unsigned int bshift; /* Shift between 512 and hard sects */
235};
236
237/*
238 * This is a direct take-off from linux/include/completion.h
239 * The difference is that I do not wait on this thing, just poll.
240 * When I want to wait (ub_probe), I just use the stock completion.
241 *
242 * Note that INIT_COMPLETION takes no lock. It is correct. But why
243 * in the bloody hell that thing takes struct instead of pointer to struct
244 * is quite beyond me. I just copied it from the stock completion.
245 */
246struct ub_completion {
247 unsigned int done;
248 spinlock_t lock;
249};
250
251static DEFINE_MUTEX(ub_mutex);
252static inline void ub_init_completion(struct ub_completion *x)
253{
254 x->done = 0;
255 spin_lock_init(&x->lock);
256}
257
258#define UB_INIT_COMPLETION(x) ((x).done = 0)
259
260static void ub_complete(struct ub_completion *x)
261{
262 unsigned long flags;
263
264 spin_lock_irqsave(&x->lock, flags);
265 x->done++;
266 spin_unlock_irqrestore(&x->lock, flags);
267}
268
269static int ub_is_completed(struct ub_completion *x)
270{
271 unsigned long flags;
272 int ret;
273
274 spin_lock_irqsave(&x->lock, flags);
275 ret = x->done;
276 spin_unlock_irqrestore(&x->lock, flags);
277 return ret;
278}
279
280/*
281 */
282struct ub_scsi_cmd_queue {
283 int qlen, qmax;
284 struct ub_scsi_cmd *head, *tail;
285};
286
287/*
288 * The block device instance (one per LUN).
289 */
290struct ub_lun {
291 struct ub_dev *udev;
292 struct list_head link;
293 struct gendisk *disk;
294 int id; /* Host index */
295 int num; /* LUN number */
296 char name[16];
297
298 int changed; /* Media was changed */
299 int removable;
300 int readonly;
301
302 struct ub_request urq;
303
304 /* Use Ingo's mempool if or when we have more than one command. */
305 /*
306 * Currently we never need more than one command for the whole device.
307 * However, giving every LUN a command is a cheap and automatic way
308 * to enforce fairness between them.
309 */
310 int cmda[1];
311 struct ub_scsi_cmd cmdv[1];
312
313 struct ub_capacity capacity;
314};
315
316/*
317 * The USB device instance.
318 */
319struct ub_dev {
320 spinlock_t *lock;
321 atomic_t poison; /* The USB device is disconnected */
322 int openc; /* protected by ub_lock! */
323 /* kref is too implicit for our taste */
324 int reset; /* Reset is running */
325 int bad_resid;
326 unsigned int tagcnt;
327 char name[12];
328 struct usb_device *dev;
329 struct usb_interface *intf;
330
331 struct list_head luns;
332
333 unsigned int send_bulk_pipe; /* cached pipe values */
334 unsigned int recv_bulk_pipe;
335 unsigned int send_ctrl_pipe;
336 unsigned int recv_ctrl_pipe;
337
338 struct tasklet_struct tasklet;
339
340 struct ub_scsi_cmd_queue cmd_queue;
341 struct ub_scsi_cmd top_rqs_cmd; /* REQUEST SENSE */
342 unsigned char top_sense[UB_SENSE_SIZE];
343
344 struct ub_completion work_done;
345 struct urb work_urb;
346 struct timer_list work_timer;
347 int last_pipe; /* What might need clearing */
348 __le32 signature; /* Learned signature */
349 struct bulk_cb_wrap work_bcb;
350 struct bulk_cs_wrap work_bcs;
351 struct usb_ctrlrequest work_cr;
352
353 struct work_struct reset_work;
354 wait_queue_head_t reset_wait;
355};
356
357/*
358 */
359static void ub_cleanup(struct ub_dev *sc);
360static int ub_request_fn_1(struct ub_lun *lun, struct request *rq);
361static void ub_cmd_build_block(struct ub_dev *sc, struct ub_lun *lun,
362 struct ub_scsi_cmd *cmd, struct ub_request *urq);
363static void ub_cmd_build_packet(struct ub_dev *sc, struct ub_lun *lun,
364 struct ub_scsi_cmd *cmd, struct ub_request *urq);
365static void ub_rw_cmd_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
366static void ub_end_rq(struct request *rq, unsigned int status);
367static int ub_rw_cmd_retry(struct ub_dev *sc, struct ub_lun *lun,
368 struct ub_request *urq, struct ub_scsi_cmd *cmd);
369static int ub_submit_scsi(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
370static void ub_urb_complete(struct urb *urb);
371static void ub_scsi_action(unsigned long _dev);
372static void ub_scsi_dispatch(struct ub_dev *sc);
373static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
374static void ub_data_start(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
375static void ub_state_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd, int rc);
376static int __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
377static void ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
378static void ub_state_stat_counted(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
379static void ub_state_sense(struct ub_dev *sc, struct ub_scsi_cmd *cmd);
380static int ub_submit_clear_stall(struct ub_dev *sc, struct ub_scsi_cmd *cmd,
381 int stalled_pipe);
382static void ub_top_sense_done(struct ub_dev *sc, struct ub_scsi_cmd *scmd);
383static void ub_reset_enter(struct ub_dev *sc, int try);
384static void ub_reset_task(struct work_struct *work);
385static int ub_sync_tur(struct ub_dev *sc, struct ub_lun *lun);
386static int ub_sync_read_cap(struct ub_dev *sc, struct ub_lun *lun,
387 struct ub_capacity *ret);
388static int ub_sync_reset(struct ub_dev *sc);
389static int ub_probe_clear_stall(struct ub_dev *sc, int stalled_pipe);
390static int ub_probe_lun(struct ub_dev *sc, int lnum);
391
392/*
393 */
394#ifdef CONFIG_USB_LIBUSUAL
395
396#define ub_usb_ids usb_storage_usb_ids
397#else
398
399static const struct usb_device_id ub_usb_ids[] = {
400 { USB_INTERFACE_INFO(USB_CLASS_MASS_STORAGE, USB_SC_SCSI, USB_PR_BULK) },
401 { }
402};
403
404MODULE_DEVICE_TABLE(usb, ub_usb_ids);
405#endif /* CONFIG_USB_LIBUSUAL */
406
407/*
408 * Find me a way to identify "next free minor" for add_disk(),
409 * and the array disappears the next day. However, the number of
410 * hosts has something to do with the naming and /proc/partitions.
411 * This has to be thought out in detail before changing.
412 * If UB_MAX_HOST was 1000, we'd use a bitmap. Or a better data structure.
413 */
414#define UB_MAX_HOSTS 26
415static char ub_hostv[UB_MAX_HOSTS];
416
417#define UB_QLOCK_NUM 5
418static spinlock_t ub_qlockv[UB_QLOCK_NUM];
419static int ub_qlock_next = 0;
420
421static DEFINE_SPINLOCK(ub_lock); /* Locks globals and ->openc */
422
423/*
424 * The id allocator.
425 *
426 * This also stores the host for indexing by minor, which is somewhat dirty.
427 */
428static int ub_id_get(void)
429{
430 unsigned long flags;
431 int i;
432
433 spin_lock_irqsave(&ub_lock, flags);
434 for (i = 0; i < UB_MAX_HOSTS; i++) {
435 if (ub_hostv[i] == 0) {
436 ub_hostv[i] = 1;
437 spin_unlock_irqrestore(&ub_lock, flags);
438 return i;
439 }
440 }
441 spin_unlock_irqrestore(&ub_lock, flags);
442 return -1;
443}
444
445static void ub_id_put(int id)
446{
447 unsigned long flags;
448
449 if (id < 0 || id >= UB_MAX_HOSTS) {
450 printk(KERN_ERR DRV_NAME ": bad host ID %d\n", id);
451 return;
452 }
453
454 spin_lock_irqsave(&ub_lock, flags);
455 if (ub_hostv[id] == 0) {
456 spin_unlock_irqrestore(&ub_lock, flags);
457 printk(KERN_ERR DRV_NAME ": freeing free host ID %d\n", id);
458 return;
459 }
460 ub_hostv[id] = 0;
461 spin_unlock_irqrestore(&ub_lock, flags);
462}
463
464/*
465 * This is necessitated by the fact that blk_cleanup_queue does not
466 * necesserily destroy the queue. Instead, it may merely decrease q->refcnt.
467 * Since our blk_init_queue() passes a spinlock common with ub_dev,
468 * we have life time issues when ub_cleanup frees ub_dev.
469 */
470static spinlock_t *ub_next_lock(void)
471{
472 unsigned long flags;
473 spinlock_t *ret;
474
475 spin_lock_irqsave(&ub_lock, flags);
476 ret = &ub_qlockv[ub_qlock_next];
477 ub_qlock_next = (ub_qlock_next + 1) % UB_QLOCK_NUM;
478 spin_unlock_irqrestore(&ub_lock, flags);
479 return ret;
480}
481
482/*
483 * Downcount for deallocation. This rides on two assumptions:
484 * - once something is poisoned, its refcount cannot grow
485 * - opens cannot happen at this time (del_gendisk was done)
486 * If the above is true, we can drop the lock, which we need for
487 * blk_cleanup_queue(): the silly thing may attempt to sleep.
488 * [Actually, it never needs to sleep for us, but it calls might_sleep()]
489 */
490static void ub_put(struct ub_dev *sc)
491{
492 unsigned long flags;
493
494 spin_lock_irqsave(&ub_lock, flags);
495 --sc->openc;
496 if (sc->openc == 0 && atomic_read(&sc->poison)) {
497 spin_unlock_irqrestore(&ub_lock, flags);
498 ub_cleanup(sc);
499 } else {
500 spin_unlock_irqrestore(&ub_lock, flags);
501 }
502}
503
504/*
505 * Final cleanup and deallocation.
506 */
507static void ub_cleanup(struct ub_dev *sc)
508{
509 struct list_head *p;
510 struct ub_lun *lun;
511 struct request_queue *q;
512
513 while (!list_empty(&sc->luns)) {
514 p = sc->luns.next;
515 lun = list_entry(p, struct ub_lun, link);
516 list_del(p);
517
518 /* I don't think queue can be NULL. But... Stolen from sx8.c */
519 if ((q = lun->disk->queue) != NULL)
520 blk_cleanup_queue(q);
521 /*
522 * If we zero disk->private_data BEFORE put_disk, we have
523 * to check for NULL all over the place in open, release,
524 * check_media and revalidate, because the block level
525 * semaphore is well inside the put_disk.
526 * But we cannot zero after the call, because *disk is gone.
527 * The sd.c is blatantly racy in this area.
528 */
529 /* disk->private_data = NULL; */
530 put_disk(lun->disk);
531 lun->disk = NULL;
532
533 ub_id_put(lun->id);
534 kfree(lun);
535 }
536
537 usb_set_intfdata(sc->intf, NULL);
538 usb_put_intf(sc->intf);
539 usb_put_dev(sc->dev);
540 kfree(sc);
541}
542
543/*
544 * The "command allocator".
545 */
546static struct ub_scsi_cmd *ub_get_cmd(struct ub_lun *lun)
547{
548 struct ub_scsi_cmd *ret;
549
550 if (lun->cmda[0])
551 return NULL;
552 ret = &lun->cmdv[0];
553 lun->cmda[0] = 1;
554 return ret;
555}
556
557static void ub_put_cmd(struct ub_lun *lun, struct ub_scsi_cmd *cmd)
558{
559 if (cmd != &lun->cmdv[0]) {
560 printk(KERN_WARNING "%s: releasing a foreign cmd %p\n",
561 lun->name, cmd);
562 return;
563 }
564 if (!lun->cmda[0]) {
565 printk(KERN_WARNING "%s: releasing a free cmd\n", lun->name);
566 return;
567 }
568 lun->cmda[0] = 0;
569}
570
571/*
572 * The command queue.
573 */
574static void ub_cmdq_add(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
575{
576 struct ub_scsi_cmd_queue *t = &sc->cmd_queue;
577
578 if (t->qlen++ == 0) {
579 t->head = cmd;
580 t->tail = cmd;
581 } else {
582 t->tail->next = cmd;
583 t->tail = cmd;
584 }
585
586 if (t->qlen > t->qmax)
587 t->qmax = t->qlen;
588}
589
590static void ub_cmdq_insert(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
591{
592 struct ub_scsi_cmd_queue *t = &sc->cmd_queue;
593
594 if (t->qlen++ == 0) {
595 t->head = cmd;
596 t->tail = cmd;
597 } else {
598 cmd->next = t->head;
599 t->head = cmd;
600 }
601
602 if (t->qlen > t->qmax)
603 t->qmax = t->qlen;
604}
605
606static struct ub_scsi_cmd *ub_cmdq_pop(struct ub_dev *sc)
607{
608 struct ub_scsi_cmd_queue *t = &sc->cmd_queue;
609 struct ub_scsi_cmd *cmd;
610
611 if (t->qlen == 0)
612 return NULL;
613 if (--t->qlen == 0)
614 t->tail = NULL;
615 cmd = t->head;
616 t->head = cmd->next;
617 cmd->next = NULL;
618 return cmd;
619}
620
621#define ub_cmdq_peek(sc) ((sc)->cmd_queue.head)
622
623/*
624 * The request function is our main entry point
625 */
626
627static void ub_request_fn(struct request_queue *q)
628{
629 struct ub_lun *lun = q->queuedata;
630 struct request *rq;
631
632 while ((rq = blk_peek_request(q)) != NULL) {
633 if (ub_request_fn_1(lun, rq) != 0) {
634 blk_stop_queue(q);
635 break;
636 }
637 }
638}
639
640static int ub_request_fn_1(struct ub_lun *lun, struct request *rq)
641{
642 struct ub_dev *sc = lun->udev;
643 struct ub_scsi_cmd *cmd;
644 struct ub_request *urq;
645 int n_elem;
646
647 if (atomic_read(&sc->poison)) {
648 blk_start_request(rq);
649 ub_end_rq(rq, DID_NO_CONNECT << 16);
650 return 0;
651 }
652
653 if (lun->changed && rq->cmd_type != REQ_TYPE_BLOCK_PC) {
654 blk_start_request(rq);
655 ub_end_rq(rq, SAM_STAT_CHECK_CONDITION);
656 return 0;
657 }
658
659 if (lun->urq.rq != NULL)
660 return -1;
661 if ((cmd = ub_get_cmd(lun)) == NULL)
662 return -1;
663 memset(cmd, 0, sizeof(struct ub_scsi_cmd));
664
665 blk_start_request(rq);
666
667 urq = &lun->urq;
668 memset(urq, 0, sizeof(struct ub_request));
669 urq->rq = rq;
670
671 /*
672 * get scatterlist from block layer
673 */
674 sg_init_table(&urq->sgv[0], UB_MAX_REQ_SG);
675 n_elem = blk_rq_map_sg(lun->disk->queue, rq, &urq->sgv[0]);
676 if (n_elem < 0) {
677 /* Impossible, because blk_rq_map_sg should not hit ENOMEM. */
678 printk(KERN_INFO "%s: failed request map (%d)\n",
679 lun->name, n_elem);
680 goto drop;
681 }
682 if (n_elem > UB_MAX_REQ_SG) { /* Paranoia */
683 printk(KERN_WARNING "%s: request with %d segments\n",
684 lun->name, n_elem);
685 goto drop;
686 }
687 urq->nsg = n_elem;
688
689 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
690 ub_cmd_build_packet(sc, lun, cmd, urq);
691 } else {
692 ub_cmd_build_block(sc, lun, cmd, urq);
693 }
694 cmd->state = UB_CMDST_INIT;
695 cmd->lun = lun;
696 cmd->done = ub_rw_cmd_done;
697 cmd->back = urq;
698
699 cmd->tag = sc->tagcnt++;
700 if (ub_submit_scsi(sc, cmd) != 0)
701 goto drop;
702
703 return 0;
704
705drop:
706 ub_put_cmd(lun, cmd);
707 ub_end_rq(rq, DID_ERROR << 16);
708 return 0;
709}
710
711static void ub_cmd_build_block(struct ub_dev *sc, struct ub_lun *lun,
712 struct ub_scsi_cmd *cmd, struct ub_request *urq)
713{
714 struct request *rq = urq->rq;
715 unsigned int block, nblks;
716
717 if (rq_data_dir(rq) == WRITE)
718 cmd->dir = UB_DIR_WRITE;
719 else
720 cmd->dir = UB_DIR_READ;
721
722 cmd->nsg = urq->nsg;
723 memcpy(cmd->sgv, urq->sgv, sizeof(struct scatterlist) * cmd->nsg);
724
725 /*
726 * build the command
727 *
728 * The call to blk_queue_logical_block_size() guarantees that request
729 * is aligned, but it is given in terms of 512 byte units, always.
730 */
731 block = blk_rq_pos(rq) >> lun->capacity.bshift;
732 nblks = blk_rq_sectors(rq) >> lun->capacity.bshift;
733
734 cmd->cdb[0] = (cmd->dir == UB_DIR_READ)? READ_10: WRITE_10;
735 /* 10-byte uses 4 bytes of LBA: 2147483648KB, 2097152MB, 2048GB */
736 cmd->cdb[2] = block >> 24;
737 cmd->cdb[3] = block >> 16;
738 cmd->cdb[4] = block >> 8;
739 cmd->cdb[5] = block;
740 cmd->cdb[7] = nblks >> 8;
741 cmd->cdb[8] = nblks;
742 cmd->cdb_len = 10;
743
744 cmd->len = blk_rq_bytes(rq);
745}
746
747static void ub_cmd_build_packet(struct ub_dev *sc, struct ub_lun *lun,
748 struct ub_scsi_cmd *cmd, struct ub_request *urq)
749{
750 struct request *rq = urq->rq;
751
752 if (blk_rq_bytes(rq) == 0) {
753 cmd->dir = UB_DIR_NONE;
754 } else {
755 if (rq_data_dir(rq) == WRITE)
756 cmd->dir = UB_DIR_WRITE;
757 else
758 cmd->dir = UB_DIR_READ;
759 }
760
761 cmd->nsg = urq->nsg;
762 memcpy(cmd->sgv, urq->sgv, sizeof(struct scatterlist) * cmd->nsg);
763
764 memcpy(&cmd->cdb, rq->cmd, rq->cmd_len);
765 cmd->cdb_len = rq->cmd_len;
766
767 cmd->len = blk_rq_bytes(rq);
768
769 /*
770 * To reapply this to every URB is not as incorrect as it looks.
771 * In return, we avoid any complicated tracking calculations.
772 */
773 cmd->timeo = rq->timeout;
774}
775
776static void ub_rw_cmd_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
777{
778 struct ub_lun *lun = cmd->lun;
779 struct ub_request *urq = cmd->back;
780 struct request *rq;
781 unsigned int scsi_status;
782
783 rq = urq->rq;
784
785 if (cmd->error == 0) {
786 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
787 if (cmd->act_len >= rq->resid_len)
788 rq->resid_len = 0;
789 else
790 rq->resid_len -= cmd->act_len;
791 scsi_status = 0;
792 } else {
793 if (cmd->act_len != cmd->len) {
794 scsi_status = SAM_STAT_CHECK_CONDITION;
795 } else {
796 scsi_status = 0;
797 }
798 }
799 } else {
800 if (rq->cmd_type == REQ_TYPE_BLOCK_PC) {
801 /* UB_SENSE_SIZE is smaller than SCSI_SENSE_BUFFERSIZE */
802 memcpy(rq->sense, sc->top_sense, UB_SENSE_SIZE);
803 rq->sense_len = UB_SENSE_SIZE;
804 if (sc->top_sense[0] != 0)
805 scsi_status = SAM_STAT_CHECK_CONDITION;
806 else
807 scsi_status = DID_ERROR << 16;
808 } else {
809 if (cmd->error == -EIO &&
810 (cmd->key == 0 ||
811 cmd->key == MEDIUM_ERROR ||
812 cmd->key == UNIT_ATTENTION)) {
813 if (ub_rw_cmd_retry(sc, lun, urq, cmd) == 0)
814 return;
815 }
816 scsi_status = SAM_STAT_CHECK_CONDITION;
817 }
818 }
819
820 urq->rq = NULL;
821
822 ub_put_cmd(lun, cmd);
823 ub_end_rq(rq, scsi_status);
824 blk_start_queue(lun->disk->queue);
825}
826
827static void ub_end_rq(struct request *rq, unsigned int scsi_status)
828{
829 int error;
830
831 if (scsi_status == 0) {
832 error = 0;
833 } else {
834 error = -EIO;
835 rq->errors = scsi_status;
836 }
837 __blk_end_request_all(rq, error);
838}
839
840static int ub_rw_cmd_retry(struct ub_dev *sc, struct ub_lun *lun,
841 struct ub_request *urq, struct ub_scsi_cmd *cmd)
842{
843
844 if (atomic_read(&sc->poison))
845 return -ENXIO;
846
847 ub_reset_enter(sc, urq->current_try);
848
849 if (urq->current_try >= 3)
850 return -EIO;
851 urq->current_try++;
852
853 /* Remove this if anyone complains of flooding. */
854 printk(KERN_DEBUG "%s: dir %c len/act %d/%d "
855 "[sense %x %02x %02x] retry %d\n",
856 sc->name, UB_DIR_CHAR(cmd->dir), cmd->len, cmd->act_len,
857 cmd->key, cmd->asc, cmd->ascq, urq->current_try);
858
859 memset(cmd, 0, sizeof(struct ub_scsi_cmd));
860 ub_cmd_build_block(sc, lun, cmd, urq);
861
862 cmd->state = UB_CMDST_INIT;
863 cmd->lun = lun;
864 cmd->done = ub_rw_cmd_done;
865 cmd->back = urq;
866
867 cmd->tag = sc->tagcnt++;
868
869#if 0 /* Wasteful */
870 return ub_submit_scsi(sc, cmd);
871#else
872 ub_cmdq_add(sc, cmd);
873 return 0;
874#endif
875}
876
877/*
878 * Submit a regular SCSI operation (not an auto-sense).
879 *
880 * The Iron Law of Good Submit Routine is:
881 * Zero return - callback is done, Nonzero return - callback is not done.
882 * No exceptions.
883 *
884 * Host is assumed locked.
885 */
886static int ub_submit_scsi(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
887{
888
889 if (cmd->state != UB_CMDST_INIT ||
890 (cmd->dir != UB_DIR_NONE && cmd->len == 0)) {
891 return -EINVAL;
892 }
893
894 ub_cmdq_add(sc, cmd);
895 /*
896 * We can call ub_scsi_dispatch(sc) right away here, but it's a little
897 * safer to jump to a tasklet, in case upper layers do something silly.
898 */
899 tasklet_schedule(&sc->tasklet);
900 return 0;
901}
902
903/*
904 * Submit the first URB for the queued command.
905 * This function does not deal with queueing in any way.
906 */
907static int ub_scsi_cmd_start(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
908{
909 struct bulk_cb_wrap *bcb;
910 int rc;
911
912 bcb = &sc->work_bcb;
913
914 /*
915 * ``If the allocation length is eighteen or greater, and a device
916 * server returns less than eithteen bytes of data, the application
917 * client should assume that the bytes not transferred would have been
918 * zeroes had the device server returned those bytes.''
919 *
920 * We zero sense for all commands so that when a packet request
921 * fails it does not return a stale sense.
922 */
923 memset(&sc->top_sense, 0, UB_SENSE_SIZE);
924
925 /* set up the command wrapper */
926 bcb->Signature = cpu_to_le32(US_BULK_CB_SIGN);
927 bcb->Tag = cmd->tag; /* Endianness is not important */
928 bcb->DataTransferLength = cpu_to_le32(cmd->len);
929 bcb->Flags = (cmd->dir == UB_DIR_READ) ? 0x80 : 0;
930 bcb->Lun = (cmd->lun != NULL) ? cmd->lun->num : 0;
931 bcb->Length = cmd->cdb_len;
932
933 /* copy the command payload */
934 memcpy(bcb->CDB, cmd->cdb, UB_MAX_CDB_SIZE);
935
936 UB_INIT_COMPLETION(sc->work_done);
937
938 sc->last_pipe = sc->send_bulk_pipe;
939 usb_fill_bulk_urb(&sc->work_urb, sc->dev, sc->send_bulk_pipe,
940 bcb, US_BULK_CB_WRAP_LEN, ub_urb_complete, sc);
941
942 if ((rc = usb_submit_urb(&sc->work_urb, GFP_ATOMIC)) != 0) {
943 /* XXX Clear stalls */
944 ub_complete(&sc->work_done);
945 return rc;
946 }
947
948 sc->work_timer.expires = jiffies + UB_URB_TIMEOUT;
949 add_timer(&sc->work_timer);
950
951 cmd->state = UB_CMDST_CMD;
952 return 0;
953}
954
955/*
956 * Timeout handler.
957 */
958static void ub_urb_timeout(unsigned long arg)
959{
960 struct ub_dev *sc = (struct ub_dev *) arg;
961 unsigned long flags;
962
963 spin_lock_irqsave(sc->lock, flags);
964 if (!ub_is_completed(&sc->work_done))
965 usb_unlink_urb(&sc->work_urb);
966 spin_unlock_irqrestore(sc->lock, flags);
967}
968
969/*
970 * Completion routine for the work URB.
971 *
972 * This can be called directly from usb_submit_urb (while we have
973 * the sc->lock taken) and from an interrupt (while we do NOT have
974 * the sc->lock taken). Therefore, bounce this off to a tasklet.
975 */
976static void ub_urb_complete(struct urb *urb)
977{
978 struct ub_dev *sc = urb->context;
979
980 ub_complete(&sc->work_done);
981 tasklet_schedule(&sc->tasklet);
982}
983
984static void ub_scsi_action(unsigned long _dev)
985{
986 struct ub_dev *sc = (struct ub_dev *) _dev;
987 unsigned long flags;
988
989 spin_lock_irqsave(sc->lock, flags);
990 ub_scsi_dispatch(sc);
991 spin_unlock_irqrestore(sc->lock, flags);
992}
993
994static void ub_scsi_dispatch(struct ub_dev *sc)
995{
996 struct ub_scsi_cmd *cmd;
997 int rc;
998
999 while (!sc->reset && (cmd = ub_cmdq_peek(sc)) != NULL) {
1000 if (cmd->state == UB_CMDST_DONE) {
1001 ub_cmdq_pop(sc);
1002 (*cmd->done)(sc, cmd);
1003 } else if (cmd->state == UB_CMDST_INIT) {
1004 if ((rc = ub_scsi_cmd_start(sc, cmd)) == 0)
1005 break;
1006 cmd->error = rc;
1007 cmd->state = UB_CMDST_DONE;
1008 } else {
1009 if (!ub_is_completed(&sc->work_done))
1010 break;
1011 del_timer(&sc->work_timer);
1012 ub_scsi_urb_compl(sc, cmd);
1013 }
1014 }
1015}
1016
1017static void ub_scsi_urb_compl(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1018{
1019 struct urb *urb = &sc->work_urb;
1020 struct bulk_cs_wrap *bcs;
1021 int endp;
1022 int len;
1023 int rc;
1024
1025 if (atomic_read(&sc->poison)) {
1026 ub_state_done(sc, cmd, -ENODEV);
1027 return;
1028 }
1029
1030 endp = usb_pipeendpoint(sc->last_pipe);
1031 if (usb_pipein(sc->last_pipe))
1032 endp |= USB_DIR_IN;
1033
1034 if (cmd->state == UB_CMDST_CLEAR) {
1035 if (urb->status == -EPIPE) {
1036 /*
1037 * STALL while clearning STALL.
1038 * The control pipe clears itself - nothing to do.
1039 */
1040 printk(KERN_NOTICE "%s: stall on control pipe\n",
1041 sc->name);
1042 goto Bad_End;
1043 }
1044
1045 /*
1046 * We ignore the result for the halt clear.
1047 */
1048
1049 usb_reset_endpoint(sc->dev, endp);
1050
1051 ub_state_sense(sc, cmd);
1052
1053 } else if (cmd->state == UB_CMDST_CLR2STS) {
1054 if (urb->status == -EPIPE) {
1055 printk(KERN_NOTICE "%s: stall on control pipe\n",
1056 sc->name);
1057 goto Bad_End;
1058 }
1059
1060 /*
1061 * We ignore the result for the halt clear.
1062 */
1063
1064 usb_reset_endpoint(sc->dev, endp);
1065
1066 ub_state_stat(sc, cmd);
1067
1068 } else if (cmd->state == UB_CMDST_CLRRS) {
1069 if (urb->status == -EPIPE) {
1070 printk(KERN_NOTICE "%s: stall on control pipe\n",
1071 sc->name);
1072 goto Bad_End;
1073 }
1074
1075 /*
1076 * We ignore the result for the halt clear.
1077 */
1078
1079 usb_reset_endpoint(sc->dev, endp);
1080
1081 ub_state_stat_counted(sc, cmd);
1082
1083 } else if (cmd->state == UB_CMDST_CMD) {
1084 switch (urb->status) {
1085 case 0:
1086 break;
1087 case -EOVERFLOW:
1088 goto Bad_End;
1089 case -EPIPE:
1090 rc = ub_submit_clear_stall(sc, cmd, sc->last_pipe);
1091 if (rc != 0) {
1092 printk(KERN_NOTICE "%s: "
1093 "unable to submit clear (%d)\n",
1094 sc->name, rc);
1095 /*
1096 * This is typically ENOMEM or some other such shit.
1097 * Retrying is pointless. Just do Bad End on it...
1098 */
1099 ub_state_done(sc, cmd, rc);
1100 return;
1101 }
1102 cmd->state = UB_CMDST_CLEAR;
1103 return;
1104 case -ESHUTDOWN: /* unplug */
1105 case -EILSEQ: /* unplug timeout on uhci */
1106 ub_state_done(sc, cmd, -ENODEV);
1107 return;
1108 default:
1109 goto Bad_End;
1110 }
1111 if (urb->actual_length != US_BULK_CB_WRAP_LEN) {
1112 goto Bad_End;
1113 }
1114
1115 if (cmd->dir == UB_DIR_NONE || cmd->nsg < 1) {
1116 ub_state_stat(sc, cmd);
1117 return;
1118 }
1119
1120 // udelay(125); // usb-storage has this
1121 ub_data_start(sc, cmd);
1122
1123 } else if (cmd->state == UB_CMDST_DATA) {
1124 if (urb->status == -EPIPE) {
1125 rc = ub_submit_clear_stall(sc, cmd, sc->last_pipe);
1126 if (rc != 0) {
1127 printk(KERN_NOTICE "%s: "
1128 "unable to submit clear (%d)\n",
1129 sc->name, rc);
1130 ub_state_done(sc, cmd, rc);
1131 return;
1132 }
1133 cmd->state = UB_CMDST_CLR2STS;
1134 return;
1135 }
1136 if (urb->status == -EOVERFLOW) {
1137 /*
1138 * A babble? Failure, but we must transfer CSW now.
1139 */
1140 cmd->error = -EOVERFLOW; /* A cheap trick... */
1141 ub_state_stat(sc, cmd);
1142 return;
1143 }
1144
1145 if (cmd->dir == UB_DIR_WRITE) {
1146 /*
1147 * Do not continue writes in case of a failure.
1148 * Doing so would cause sectors to be mixed up,
1149 * which is worse than sectors lost.
1150 *
1151 * We must try to read the CSW, or many devices
1152 * get confused.
1153 */
1154 len = urb->actual_length;
1155 if (urb->status != 0 ||
1156 len != cmd->sgv[cmd->current_sg].length) {
1157 cmd->act_len += len;
1158
1159 cmd->error = -EIO;
1160 ub_state_stat(sc, cmd);
1161 return;
1162 }
1163
1164 } else {
1165 /*
1166 * If an error occurs on read, we record it, and
1167 * continue to fetch data in order to avoid bubble.
1168 *
1169 * As a small shortcut, we stop if we detect that
1170 * a CSW mixed into data.
1171 */
1172 if (urb->status != 0)
1173 cmd->error = -EIO;
1174
1175 len = urb->actual_length;
1176 if (urb->status != 0 ||
1177 len != cmd->sgv[cmd->current_sg].length) {
1178 if ((len & 0x1FF) == US_BULK_CS_WRAP_LEN)
1179 goto Bad_End;
1180 }
1181 }
1182
1183 cmd->act_len += urb->actual_length;
1184
1185 if (++cmd->current_sg < cmd->nsg) {
1186 ub_data_start(sc, cmd);
1187 return;
1188 }
1189 ub_state_stat(sc, cmd);
1190
1191 } else if (cmd->state == UB_CMDST_STAT) {
1192 if (urb->status == -EPIPE) {
1193 rc = ub_submit_clear_stall(sc, cmd, sc->last_pipe);
1194 if (rc != 0) {
1195 printk(KERN_NOTICE "%s: "
1196 "unable to submit clear (%d)\n",
1197 sc->name, rc);
1198 ub_state_done(sc, cmd, rc);
1199 return;
1200 }
1201
1202 /*
1203 * Having a stall when getting CSW is an error, so
1204 * make sure uppper levels are not oblivious to it.
1205 */
1206 cmd->error = -EIO; /* A cheap trick... */
1207
1208 cmd->state = UB_CMDST_CLRRS;
1209 return;
1210 }
1211
1212 /* Catch everything, including -EOVERFLOW and other nasties. */
1213 if (urb->status != 0)
1214 goto Bad_End;
1215
1216 if (urb->actual_length == 0) {
1217 ub_state_stat_counted(sc, cmd);
1218 return;
1219 }
1220
1221 /*
1222 * Check the returned Bulk protocol status.
1223 * The status block has to be validated first.
1224 */
1225
1226 bcs = &sc->work_bcs;
1227
1228 if (sc->signature == cpu_to_le32(0)) {
1229 /*
1230 * This is the first reply, so do not perform the check.
1231 * Instead, remember the signature the device uses
1232 * for future checks. But do not allow a nul.
1233 */
1234 sc->signature = bcs->Signature;
1235 if (sc->signature == cpu_to_le32(0)) {
1236 ub_state_stat_counted(sc, cmd);
1237 return;
1238 }
1239 } else {
1240 if (bcs->Signature != sc->signature) {
1241 ub_state_stat_counted(sc, cmd);
1242 return;
1243 }
1244 }
1245
1246 if (bcs->Tag != cmd->tag) {
1247 /*
1248 * This usually happens when we disagree with the
1249 * device's microcode about something. For instance,
1250 * a few of them throw this after timeouts. They buffer
1251 * commands and reply at commands we timed out before.
1252 * Without flushing these replies we loop forever.
1253 */
1254 ub_state_stat_counted(sc, cmd);
1255 return;
1256 }
1257
1258 if (!sc->bad_resid) {
1259 len = le32_to_cpu(bcs->Residue);
1260 if (len != cmd->len - cmd->act_len) {
1261 /*
1262 * Only start ignoring if this cmd ended well.
1263 */
1264 if (cmd->len == cmd->act_len) {
1265 printk(KERN_NOTICE "%s: "
1266 "bad residual %d of %d, ignoring\n",
1267 sc->name, len, cmd->len);
1268 sc->bad_resid = 1;
1269 }
1270 }
1271 }
1272
1273 switch (bcs->Status) {
1274 case US_BULK_STAT_OK:
1275 break;
1276 case US_BULK_STAT_FAIL:
1277 ub_state_sense(sc, cmd);
1278 return;
1279 case US_BULK_STAT_PHASE:
1280 goto Bad_End;
1281 default:
1282 printk(KERN_INFO "%s: unknown CSW status 0x%x\n",
1283 sc->name, bcs->Status);
1284 ub_state_done(sc, cmd, -EINVAL);
1285 return;
1286 }
1287
1288 /* Not zeroing error to preserve a babble indicator */
1289 if (cmd->error != 0) {
1290 ub_state_sense(sc, cmd);
1291 return;
1292 }
1293 cmd->state = UB_CMDST_DONE;
1294 ub_cmdq_pop(sc);
1295 (*cmd->done)(sc, cmd);
1296
1297 } else if (cmd->state == UB_CMDST_SENSE) {
1298 ub_state_done(sc, cmd, -EIO);
1299
1300 } else {
1301 printk(KERN_WARNING "%s: wrong command state %d\n",
1302 sc->name, cmd->state);
1303 ub_state_done(sc, cmd, -EINVAL);
1304 return;
1305 }
1306 return;
1307
1308Bad_End: /* Little Excel is dead */
1309 ub_state_done(sc, cmd, -EIO);
1310}
1311
1312/*
1313 * Factorization helper for the command state machine:
1314 * Initiate a data segment transfer.
1315 */
1316static void ub_data_start(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1317{
1318 struct scatterlist *sg = &cmd->sgv[cmd->current_sg];
1319 int pipe;
1320 int rc;
1321
1322 UB_INIT_COMPLETION(sc->work_done);
1323
1324 if (cmd->dir == UB_DIR_READ)
1325 pipe = sc->recv_bulk_pipe;
1326 else
1327 pipe = sc->send_bulk_pipe;
1328 sc->last_pipe = pipe;
1329 usb_fill_bulk_urb(&sc->work_urb, sc->dev, pipe, sg_virt(sg),
1330 sg->length, ub_urb_complete, sc);
1331
1332 if ((rc = usb_submit_urb(&sc->work_urb, GFP_ATOMIC)) != 0) {
1333 /* XXX Clear stalls */
1334 ub_complete(&sc->work_done);
1335 ub_state_done(sc, cmd, rc);
1336 return;
1337 }
1338
1339 if (cmd->timeo)
1340 sc->work_timer.expires = jiffies + cmd->timeo;
1341 else
1342 sc->work_timer.expires = jiffies + UB_DATA_TIMEOUT;
1343 add_timer(&sc->work_timer);
1344
1345 cmd->state = UB_CMDST_DATA;
1346}
1347
1348/*
1349 * Factorization helper for the command state machine:
1350 * Finish the command.
1351 */
1352static void ub_state_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd, int rc)
1353{
1354
1355 cmd->error = rc;
1356 cmd->state = UB_CMDST_DONE;
1357 ub_cmdq_pop(sc);
1358 (*cmd->done)(sc, cmd);
1359}
1360
1361/*
1362 * Factorization helper for the command state machine:
1363 * Submit a CSW read.
1364 */
1365static int __ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1366{
1367 int rc;
1368
1369 UB_INIT_COMPLETION(sc->work_done);
1370
1371 sc->last_pipe = sc->recv_bulk_pipe;
1372 usb_fill_bulk_urb(&sc->work_urb, sc->dev, sc->recv_bulk_pipe,
1373 &sc->work_bcs, US_BULK_CS_WRAP_LEN, ub_urb_complete, sc);
1374
1375 if ((rc = usb_submit_urb(&sc->work_urb, GFP_ATOMIC)) != 0) {
1376 /* XXX Clear stalls */
1377 ub_complete(&sc->work_done);
1378 ub_state_done(sc, cmd, rc);
1379 return -1;
1380 }
1381
1382 if (cmd->timeo)
1383 sc->work_timer.expires = jiffies + cmd->timeo;
1384 else
1385 sc->work_timer.expires = jiffies + UB_STAT_TIMEOUT;
1386 add_timer(&sc->work_timer);
1387 return 0;
1388}
1389
1390/*
1391 * Factorization helper for the command state machine:
1392 * Submit a CSW read and go to STAT state.
1393 */
1394static void ub_state_stat(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1395{
1396
1397 if (__ub_state_stat(sc, cmd) != 0)
1398 return;
1399
1400 cmd->stat_count = 0;
1401 cmd->state = UB_CMDST_STAT;
1402}
1403
1404/*
1405 * Factorization helper for the command state machine:
1406 * Submit a CSW read and go to STAT state with counter (along [C] path).
1407 */
1408static void ub_state_stat_counted(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1409{
1410
1411 if (++cmd->stat_count >= 4) {
1412 ub_state_sense(sc, cmd);
1413 return;
1414 }
1415
1416 if (__ub_state_stat(sc, cmd) != 0)
1417 return;
1418
1419 cmd->state = UB_CMDST_STAT;
1420}
1421
1422/*
1423 * Factorization helper for the command state machine:
1424 * Submit a REQUEST SENSE and go to SENSE state.
1425 */
1426static void ub_state_sense(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1427{
1428 struct ub_scsi_cmd *scmd;
1429 struct scatterlist *sg;
1430 int rc;
1431
1432 if (cmd->cdb[0] == REQUEST_SENSE) {
1433 rc = -EPIPE;
1434 goto error;
1435 }
1436
1437 scmd = &sc->top_rqs_cmd;
1438 memset(scmd, 0, sizeof(struct ub_scsi_cmd));
1439 scmd->cdb[0] = REQUEST_SENSE;
1440 scmd->cdb[4] = UB_SENSE_SIZE;
1441 scmd->cdb_len = 6;
1442 scmd->dir = UB_DIR_READ;
1443 scmd->state = UB_CMDST_INIT;
1444 scmd->nsg = 1;
1445 sg = &scmd->sgv[0];
1446 sg_init_table(sg, UB_MAX_REQ_SG);
1447 sg_set_page(sg, virt_to_page(sc->top_sense), UB_SENSE_SIZE,
1448 (unsigned long)sc->top_sense & (PAGE_SIZE-1));
1449 scmd->len = UB_SENSE_SIZE;
1450 scmd->lun = cmd->lun;
1451 scmd->done = ub_top_sense_done;
1452 scmd->back = cmd;
1453
1454 scmd->tag = sc->tagcnt++;
1455
1456 cmd->state = UB_CMDST_SENSE;
1457
1458 ub_cmdq_insert(sc, scmd);
1459 return;
1460
1461error:
1462 ub_state_done(sc, cmd, rc);
1463}
1464
1465/*
1466 * A helper for the command's state machine:
1467 * Submit a stall clear.
1468 */
1469static int ub_submit_clear_stall(struct ub_dev *sc, struct ub_scsi_cmd *cmd,
1470 int stalled_pipe)
1471{
1472 int endp;
1473 struct usb_ctrlrequest *cr;
1474 int rc;
1475
1476 endp = usb_pipeendpoint(stalled_pipe);
1477 if (usb_pipein (stalled_pipe))
1478 endp |= USB_DIR_IN;
1479
1480 cr = &sc->work_cr;
1481 cr->bRequestType = USB_RECIP_ENDPOINT;
1482 cr->bRequest = USB_REQ_CLEAR_FEATURE;
1483 cr->wValue = cpu_to_le16(USB_ENDPOINT_HALT);
1484 cr->wIndex = cpu_to_le16(endp);
1485 cr->wLength = cpu_to_le16(0);
1486
1487 UB_INIT_COMPLETION(sc->work_done);
1488
1489 usb_fill_control_urb(&sc->work_urb, sc->dev, sc->send_ctrl_pipe,
1490 (unsigned char*) cr, NULL, 0, ub_urb_complete, sc);
1491
1492 if ((rc = usb_submit_urb(&sc->work_urb, GFP_ATOMIC)) != 0) {
1493 ub_complete(&sc->work_done);
1494 return rc;
1495 }
1496
1497 sc->work_timer.expires = jiffies + UB_CTRL_TIMEOUT;
1498 add_timer(&sc->work_timer);
1499 return 0;
1500}
1501
1502/*
1503 */
1504static void ub_top_sense_done(struct ub_dev *sc, struct ub_scsi_cmd *scmd)
1505{
1506 unsigned char *sense = sc->top_sense;
1507 struct ub_scsi_cmd *cmd;
1508
1509 /*
1510 * Find the command which triggered the unit attention or a check,
1511 * save the sense into it, and advance its state machine.
1512 */
1513 if ((cmd = ub_cmdq_peek(sc)) == NULL) {
1514 printk(KERN_WARNING "%s: sense done while idle\n", sc->name);
1515 return;
1516 }
1517 if (cmd != scmd->back) {
1518 printk(KERN_WARNING "%s: "
1519 "sense done for wrong command 0x%x\n",
1520 sc->name, cmd->tag);
1521 return;
1522 }
1523 if (cmd->state != UB_CMDST_SENSE) {
1524 printk(KERN_WARNING "%s: sense done with bad cmd state %d\n",
1525 sc->name, cmd->state);
1526 return;
1527 }
1528
1529 /*
1530 * Ignoring scmd->act_len, because the buffer was pre-zeroed.
1531 */
1532 cmd->key = sense[2] & 0x0F;
1533 cmd->asc = sense[12];
1534 cmd->ascq = sense[13];
1535
1536 ub_scsi_urb_compl(sc, cmd);
1537}
1538
1539/*
1540 * Reset management
1541 */
1542
1543static void ub_reset_enter(struct ub_dev *sc, int try)
1544{
1545
1546 if (sc->reset) {
1547 /* This happens often on multi-LUN devices. */
1548 return;
1549 }
1550 sc->reset = try + 1;
1551
1552#if 0 /* Not needed because the disconnect waits for us. */
1553 unsigned long flags;
1554 spin_lock_irqsave(&ub_lock, flags);
1555 sc->openc++;
1556 spin_unlock_irqrestore(&ub_lock, flags);
1557#endif
1558
1559#if 0 /* We let them stop themselves. */
1560 struct ub_lun *lun;
1561 list_for_each_entry(lun, &sc->luns, link) {
1562 blk_stop_queue(lun->disk->queue);
1563 }
1564#endif
1565
1566 schedule_work(&sc->reset_work);
1567}
1568
1569static void ub_reset_task(struct work_struct *work)
1570{
1571 struct ub_dev *sc = container_of(work, struct ub_dev, reset_work);
1572 unsigned long flags;
1573 struct ub_lun *lun;
1574 int rc;
1575
1576 if (!sc->reset) {
1577 printk(KERN_WARNING "%s: Running reset unrequested\n",
1578 sc->name);
1579 return;
1580 }
1581
1582 if (atomic_read(&sc->poison)) {
1583 ;
1584 } else if ((sc->reset & 1) == 0) {
1585 ub_sync_reset(sc);
1586 msleep(700); /* usb-storage sleeps 6s (!) */
1587 ub_probe_clear_stall(sc, sc->recv_bulk_pipe);
1588 ub_probe_clear_stall(sc, sc->send_bulk_pipe);
1589 } else if (sc->dev->actconfig->desc.bNumInterfaces != 1) {
1590 ;
1591 } else {
1592 rc = usb_lock_device_for_reset(sc->dev, sc->intf);
1593 if (rc < 0) {
1594 printk(KERN_NOTICE
1595 "%s: usb_lock_device_for_reset failed (%d)\n",
1596 sc->name, rc);
1597 } else {
1598 rc = usb_reset_device(sc->dev);
1599 if (rc < 0) {
1600 printk(KERN_NOTICE "%s: "
1601 "usb_lock_device_for_reset failed (%d)\n",
1602 sc->name, rc);
1603 }
1604 usb_unlock_device(sc->dev);
1605 }
1606 }
1607
1608 /*
1609 * In theory, no commands can be running while reset is active,
1610 * so nobody can ask for another reset, and so we do not need any
1611 * queues of resets or anything. We do need a spinlock though,
1612 * to interact with block layer.
1613 */
1614 spin_lock_irqsave(sc->lock, flags);
1615 sc->reset = 0;
1616 tasklet_schedule(&sc->tasklet);
1617 list_for_each_entry(lun, &sc->luns, link) {
1618 blk_start_queue(lun->disk->queue);
1619 }
1620 wake_up(&sc->reset_wait);
1621 spin_unlock_irqrestore(sc->lock, flags);
1622}
1623
1624/*
1625 * XXX Reset brackets are too much hassle to implement, so just stub them
1626 * in order to prevent forced unbinding (which deadlocks solid when our
1627 * ->disconnect method waits for the reset to complete and this kills keventd).
1628 *
1629 * XXX Tell Alan to move usb_unlock_device inside of usb_reset_device,
1630 * or else the post_reset is invoked, and restats I/O on a locked device.
1631 */
1632static int ub_pre_reset(struct usb_interface *iface) {
1633 return 0;
1634}
1635
1636static int ub_post_reset(struct usb_interface *iface) {
1637 return 0;
1638}
1639
1640/*
1641 * This is called from a process context.
1642 */
1643static void ub_revalidate(struct ub_dev *sc, struct ub_lun *lun)
1644{
1645
1646 lun->readonly = 0; /* XXX Query this from the device */
1647
1648 lun->capacity.nsec = 0;
1649 lun->capacity.bsize = 512;
1650 lun->capacity.bshift = 0;
1651
1652 if (ub_sync_tur(sc, lun) != 0)
1653 return; /* Not ready */
1654 lun->changed = 0;
1655
1656 if (ub_sync_read_cap(sc, lun, &lun->capacity) != 0) {
1657 /*
1658 * The retry here means something is wrong, either with the
1659 * device, with the transport, or with our code.
1660 * We keep this because sd.c has retries for capacity.
1661 */
1662 if (ub_sync_read_cap(sc, lun, &lun->capacity) != 0) {
1663 lun->capacity.nsec = 0;
1664 lun->capacity.bsize = 512;
1665 lun->capacity.bshift = 0;
1666 }
1667 }
1668}
1669
1670/*
1671 * The open funcion.
1672 * This is mostly needed to keep refcounting, but also to support
1673 * media checks on removable media drives.
1674 */
1675static int ub_bd_open(struct block_device *bdev, fmode_t mode)
1676{
1677 struct ub_lun *lun = bdev->bd_disk->private_data;
1678 struct ub_dev *sc = lun->udev;
1679 unsigned long flags;
1680 int rc;
1681
1682 spin_lock_irqsave(&ub_lock, flags);
1683 if (atomic_read(&sc->poison)) {
1684 spin_unlock_irqrestore(&ub_lock, flags);
1685 return -ENXIO;
1686 }
1687 sc->openc++;
1688 spin_unlock_irqrestore(&ub_lock, flags);
1689
1690 if (lun->removable || lun->readonly)
1691 check_disk_change(bdev);
1692
1693 /*
1694 * The sd.c considers ->media_present and ->changed not equivalent,
1695 * under some pretty murky conditions (a failure of READ CAPACITY).
1696 * We may need it one day.
1697 */
1698 if (lun->removable && lun->changed && !(mode & FMODE_NDELAY)) {
1699 rc = -ENOMEDIUM;
1700 goto err_open;
1701 }
1702
1703 if (lun->readonly && (mode & FMODE_WRITE)) {
1704 rc = -EROFS;
1705 goto err_open;
1706 }
1707
1708 return 0;
1709
1710err_open:
1711 ub_put(sc);
1712 return rc;
1713}
1714
1715static int ub_bd_unlocked_open(struct block_device *bdev, fmode_t mode)
1716{
1717 int ret;
1718
1719 mutex_lock(&ub_mutex);
1720 ret = ub_bd_open(bdev, mode);
1721 mutex_unlock(&ub_mutex);
1722
1723 return ret;
1724}
1725
1726
1727/*
1728 */
1729static int ub_bd_release(struct gendisk *disk, fmode_t mode)
1730{
1731 struct ub_lun *lun = disk->private_data;
1732 struct ub_dev *sc = lun->udev;
1733
1734 mutex_lock(&ub_mutex);
1735 ub_put(sc);
1736 mutex_unlock(&ub_mutex);
1737
1738 return 0;
1739}
1740
1741/*
1742 * The ioctl interface.
1743 */
1744static int ub_bd_ioctl(struct block_device *bdev, fmode_t mode,
1745 unsigned int cmd, unsigned long arg)
1746{
1747 struct gendisk *disk = bdev->bd_disk;
1748 void __user *usermem = (void __user *) arg;
1749 int ret;
1750
1751 mutex_lock(&ub_mutex);
1752 ret = scsi_cmd_ioctl(disk->queue, disk, mode, cmd, usermem);
1753 mutex_unlock(&ub_mutex);
1754
1755 return ret;
1756}
1757
1758/*
1759 * This is called by check_disk_change if we reported a media change.
1760 * The main onjective here is to discover the features of the media such as
1761 * the capacity, read-only status, etc. USB storage generally does not
1762 * need to be spun up, but if we needed it, this would be the place.
1763 *
1764 * This call can sleep.
1765 *
1766 * The return code is not used.
1767 */
1768static int ub_bd_revalidate(struct gendisk *disk)
1769{
1770 struct ub_lun *lun = disk->private_data;
1771
1772 ub_revalidate(lun->udev, lun);
1773
1774 /* XXX Support sector size switching like in sr.c */
1775 blk_queue_logical_block_size(disk->queue, lun->capacity.bsize);
1776 set_capacity(disk, lun->capacity.nsec);
1777 // set_disk_ro(sdkp->disk, lun->readonly);
1778
1779 return 0;
1780}
1781
1782/*
1783 * The check is called by the block layer to verify if the media
1784 * is still available. It is supposed to be harmless, lightweight and
1785 * non-intrusive in case the media was not changed.
1786 *
1787 * This call can sleep.
1788 *
1789 * The return code is bool!
1790 */
1791static unsigned int ub_bd_check_events(struct gendisk *disk,
1792 unsigned int clearing)
1793{
1794 struct ub_lun *lun = disk->private_data;
1795
1796 if (!lun->removable)
1797 return 0;
1798
1799 /*
1800 * We clean checks always after every command, so this is not
1801 * as dangerous as it looks. If the TEST_UNIT_READY fails here,
1802 * the device is actually not ready with operator or software
1803 * intervention required. One dangerous item might be a drive which
1804 * spins itself down, and come the time to write dirty pages, this
1805 * will fail, then block layer discards the data. Since we never
1806 * spin drives up, such devices simply cannot be used with ub anyway.
1807 */
1808 if (ub_sync_tur(lun->udev, lun) != 0) {
1809 lun->changed = 1;
1810 return DISK_EVENT_MEDIA_CHANGE;
1811 }
1812
1813 return lun->changed ? DISK_EVENT_MEDIA_CHANGE : 0;
1814}
1815
1816static const struct block_device_operations ub_bd_fops = {
1817 .owner = THIS_MODULE,
1818 .open = ub_bd_unlocked_open,
1819 .release = ub_bd_release,
1820 .ioctl = ub_bd_ioctl,
1821 .check_events = ub_bd_check_events,
1822 .revalidate_disk = ub_bd_revalidate,
1823};
1824
1825/*
1826 * Common ->done routine for commands executed synchronously.
1827 */
1828static void ub_probe_done(struct ub_dev *sc, struct ub_scsi_cmd *cmd)
1829{
1830 struct completion *cop = cmd->back;
1831 complete(cop);
1832}
1833
1834/*
1835 * Test if the device has a check condition on it, synchronously.
1836 */
1837static int ub_sync_tur(struct ub_dev *sc, struct ub_lun *lun)
1838{
1839 struct ub_scsi_cmd *cmd;
1840 enum { ALLOC_SIZE = sizeof(struct ub_scsi_cmd) };
1841 unsigned long flags;
1842 struct completion compl;
1843 int rc;
1844
1845 init_completion(&compl);
1846
1847 rc = -ENOMEM;
1848 if ((cmd = kzalloc(ALLOC_SIZE, GFP_KERNEL)) == NULL)
1849 goto err_alloc;
1850
1851 cmd->cdb[0] = TEST_UNIT_READY;
1852 cmd->cdb_len = 6;
1853 cmd->dir = UB_DIR_NONE;
1854 cmd->state = UB_CMDST_INIT;
1855 cmd->lun = lun; /* This may be NULL, but that's ok */
1856 cmd->done = ub_probe_done;
1857 cmd->back = &compl;
1858
1859 spin_lock_irqsave(sc->lock, flags);
1860 cmd->tag = sc->tagcnt++;
1861
1862 rc = ub_submit_scsi(sc, cmd);
1863 spin_unlock_irqrestore(sc->lock, flags);
1864
1865 if (rc != 0)
1866 goto err_submit;
1867
1868 wait_for_completion(&compl);
1869
1870 rc = cmd->error;
1871
1872 if (rc == -EIO && cmd->key != 0) /* Retries for benh's key */
1873 rc = cmd->key;
1874
1875err_submit:
1876 kfree(cmd);
1877err_alloc:
1878 return rc;
1879}
1880
1881/*
1882 * Read the SCSI capacity synchronously (for probing).
1883 */
1884static int ub_sync_read_cap(struct ub_dev *sc, struct ub_lun *lun,
1885 struct ub_capacity *ret)
1886{
1887 struct ub_scsi_cmd *cmd;
1888 struct scatterlist *sg;
1889 char *p;
1890 enum { ALLOC_SIZE = sizeof(struct ub_scsi_cmd) + 8 };
1891 unsigned long flags;
1892 unsigned int bsize, shift;
1893 unsigned long nsec;
1894 struct completion compl;
1895 int rc;
1896
1897 init_completion(&compl);
1898
1899 rc = -ENOMEM;
1900 if ((cmd = kzalloc(ALLOC_SIZE, GFP_KERNEL)) == NULL)
1901 goto err_alloc;
1902 p = (char *)cmd + sizeof(struct ub_scsi_cmd);
1903
1904 cmd->cdb[0] = 0x25;
1905 cmd->cdb_len = 10;
1906 cmd->dir = UB_DIR_READ;
1907 cmd->state = UB_CMDST_INIT;
1908 cmd->nsg = 1;
1909 sg = &cmd->sgv[0];
1910 sg_init_table(sg, UB_MAX_REQ_SG);
1911 sg_set_page(sg, virt_to_page(p), 8, (unsigned long)p & (PAGE_SIZE-1));
1912 cmd->len = 8;
1913 cmd->lun = lun;
1914 cmd->done = ub_probe_done;
1915 cmd->back = &compl;
1916
1917 spin_lock_irqsave(sc->lock, flags);
1918 cmd->tag = sc->tagcnt++;
1919
1920 rc = ub_submit_scsi(sc, cmd);
1921 spin_unlock_irqrestore(sc->lock, flags);
1922
1923 if (rc != 0)
1924 goto err_submit;
1925
1926 wait_for_completion(&compl);
1927
1928 if (cmd->error != 0) {
1929 rc = -EIO;
1930 goto err_read;
1931 }
1932 if (cmd->act_len != 8) {
1933 rc = -EIO;
1934 goto err_read;
1935 }
1936
1937 /* sd.c special-cases sector size of 0 to mean 512. Needed? Safe? */
1938 nsec = be32_to_cpu(*(__be32 *)p) + 1;
1939 bsize = be32_to_cpu(*(__be32 *)(p + 4));
1940 switch (bsize) {
1941 case 512: shift = 0; break;
1942 case 1024: shift = 1; break;
1943 case 2048: shift = 2; break;
1944 case 4096: shift = 3; break;
1945 default:
1946 rc = -EDOM;
1947 goto err_inv_bsize;
1948 }
1949
1950 ret->bsize = bsize;
1951 ret->bshift = shift;
1952 ret->nsec = nsec << shift;
1953 rc = 0;
1954
1955err_inv_bsize:
1956err_read:
1957err_submit:
1958 kfree(cmd);
1959err_alloc:
1960 return rc;
1961}
1962
1963/*
1964 */
1965static void ub_probe_urb_complete(struct urb *urb)
1966{
1967 struct completion *cop = urb->context;
1968 complete(cop);
1969}
1970
1971static void ub_probe_timeout(unsigned long arg)
1972{
1973 struct completion *cop = (struct completion *) arg;
1974 complete(cop);
1975}
1976
1977/*
1978 * Reset with a Bulk reset.
1979 */
1980static int ub_sync_reset(struct ub_dev *sc)
1981{
1982 int ifnum = sc->intf->cur_altsetting->desc.bInterfaceNumber;
1983 struct usb_ctrlrequest *cr;
1984 struct completion compl;
1985 struct timer_list timer;
1986 int rc;
1987
1988 init_completion(&compl);
1989
1990 cr = &sc->work_cr;
1991 cr->bRequestType = USB_TYPE_CLASS | USB_RECIP_INTERFACE;
1992 cr->bRequest = US_BULK_RESET_REQUEST;
1993 cr->wValue = cpu_to_le16(0);
1994 cr->wIndex = cpu_to_le16(ifnum);
1995 cr->wLength = cpu_to_le16(0);
1996
1997 usb_fill_control_urb(&sc->work_urb, sc->dev, sc->send_ctrl_pipe,
1998 (unsigned char*) cr, NULL, 0, ub_probe_urb_complete, &compl);
1999
2000 if ((rc = usb_submit_urb(&sc->work_urb, GFP_KERNEL)) != 0) {
2001 printk(KERN_WARNING
2002 "%s: Unable to submit a bulk reset (%d)\n", sc->name, rc);
2003 return rc;
2004 }
2005
2006 init_timer(&timer);
2007 timer.function = ub_probe_timeout;
2008 timer.data = (unsigned long) &compl;
2009 timer.expires = jiffies + UB_CTRL_TIMEOUT;
2010 add_timer(&timer);
2011
2012 wait_for_completion(&compl);
2013
2014 del_timer_sync(&timer);
2015 usb_kill_urb(&sc->work_urb);
2016
2017 return sc->work_urb.status;
2018}
2019
2020/*
2021 * Get number of LUNs by the way of Bulk GetMaxLUN command.
2022 */
2023static int ub_sync_getmaxlun(struct ub_dev *sc)
2024{
2025 int ifnum = sc->intf->cur_altsetting->desc.bInterfaceNumber;
2026 unsigned char *p;
2027 enum { ALLOC_SIZE = 1 };
2028 struct usb_ctrlrequest *cr;
2029 struct completion compl;
2030 struct timer_list timer;
2031 int nluns;
2032 int rc;
2033
2034 init_completion(&compl);
2035
2036 rc = -ENOMEM;
2037 if ((p = kmalloc(ALLOC_SIZE, GFP_KERNEL)) == NULL)
2038 goto err_alloc;
2039 *p = 55;
2040
2041 cr = &sc->work_cr;
2042 cr->bRequestType = USB_DIR_IN | USB_TYPE_CLASS | USB_RECIP_INTERFACE;
2043 cr->bRequest = US_BULK_GET_MAX_LUN;
2044 cr->wValue = cpu_to_le16(0);
2045 cr->wIndex = cpu_to_le16(ifnum);
2046 cr->wLength = cpu_to_le16(1);
2047
2048 usb_fill_control_urb(&sc->work_urb, sc->dev, sc->recv_ctrl_pipe,
2049 (unsigned char*) cr, p, 1, ub_probe_urb_complete, &compl);
2050
2051 if ((rc = usb_submit_urb(&sc->work_urb, GFP_KERNEL)) != 0)
2052 goto err_submit;
2053
2054 init_timer(&timer);
2055 timer.function = ub_probe_timeout;
2056 timer.data = (unsigned long) &compl;
2057 timer.expires = jiffies + UB_CTRL_TIMEOUT;
2058 add_timer(&timer);
2059
2060 wait_for_completion(&compl);
2061
2062 del_timer_sync(&timer);
2063 usb_kill_urb(&sc->work_urb);
2064
2065 if ((rc = sc->work_urb.status) < 0)
2066 goto err_io;
2067
2068 if (sc->work_urb.actual_length != 1) {
2069 nluns = 0;
2070 } else {
2071 if ((nluns = *p) == 55) {
2072 nluns = 0;
2073 } else {
2074 /* GetMaxLUN returns the maximum LUN number */
2075 nluns += 1;
2076 if (nluns > UB_MAX_LUNS)
2077 nluns = UB_MAX_LUNS;
2078 }
2079 }
2080
2081 kfree(p);
2082 return nluns;
2083
2084err_io:
2085err_submit:
2086 kfree(p);
2087err_alloc:
2088 return rc;
2089}
2090
2091/*
2092 * Clear initial stalls.
2093 */
2094static int ub_probe_clear_stall(struct ub_dev *sc, int stalled_pipe)
2095{
2096 int endp;
2097 struct usb_ctrlrequest *cr;
2098 struct completion compl;
2099 struct timer_list timer;
2100 int rc;
2101
2102 init_completion(&compl);
2103
2104 endp = usb_pipeendpoint(stalled_pipe);
2105 if (usb_pipein (stalled_pipe))
2106 endp |= USB_DIR_IN;
2107
2108 cr = &sc->work_cr;
2109 cr->bRequestType = USB_RECIP_ENDPOINT;
2110 cr->bRequest = USB_REQ_CLEAR_FEATURE;
2111 cr->wValue = cpu_to_le16(USB_ENDPOINT_HALT);
2112 cr->wIndex = cpu_to_le16(endp);
2113 cr->wLength = cpu_to_le16(0);
2114
2115 usb_fill_control_urb(&sc->work_urb, sc->dev, sc->send_ctrl_pipe,
2116 (unsigned char*) cr, NULL, 0, ub_probe_urb_complete, &compl);
2117
2118 if ((rc = usb_submit_urb(&sc->work_urb, GFP_KERNEL)) != 0) {
2119 printk(KERN_WARNING
2120 "%s: Unable to submit a probe clear (%d)\n", sc->name, rc);
2121 return rc;
2122 }
2123
2124 init_timer(&timer);
2125 timer.function = ub_probe_timeout;
2126 timer.data = (unsigned long) &compl;
2127 timer.expires = jiffies + UB_CTRL_TIMEOUT;
2128 add_timer(&timer);
2129
2130 wait_for_completion(&compl);
2131
2132 del_timer_sync(&timer);
2133 usb_kill_urb(&sc->work_urb);
2134
2135 usb_reset_endpoint(sc->dev, endp);
2136
2137 return 0;
2138}
2139
2140/*
2141 * Get the pipe settings.
2142 */
2143static int ub_get_pipes(struct ub_dev *sc, struct usb_device *dev,
2144 struct usb_interface *intf)
2145{
2146 struct usb_host_interface *altsetting = intf->cur_altsetting;
2147 struct usb_endpoint_descriptor *ep_in = NULL;
2148 struct usb_endpoint_descriptor *ep_out = NULL;
2149 struct usb_endpoint_descriptor *ep;
2150 int i;
2151
2152 /*
2153 * Find the endpoints we need.
2154 * We are expecting a minimum of 2 endpoints - in and out (bulk).
2155 * We will ignore any others.
2156 */
2157 for (i = 0; i < altsetting->desc.bNumEndpoints; i++) {
2158 ep = &altsetting->endpoint[i].desc;
2159
2160 /* Is it a BULK endpoint? */
2161 if (usb_endpoint_xfer_bulk(ep)) {
2162 /* BULK in or out? */
2163 if (usb_endpoint_dir_in(ep)) {
2164 if (ep_in == NULL)
2165 ep_in = ep;
2166 } else {
2167 if (ep_out == NULL)
2168 ep_out = ep;
2169 }
2170 }
2171 }
2172
2173 if (ep_in == NULL || ep_out == NULL) {
2174 printk(KERN_NOTICE "%s: failed endpoint check\n", sc->name);
2175 return -ENODEV;
2176 }
2177
2178 /* Calculate and store the pipe values */
2179 sc->send_ctrl_pipe = usb_sndctrlpipe(dev, 0);
2180 sc->recv_ctrl_pipe = usb_rcvctrlpipe(dev, 0);
2181 sc->send_bulk_pipe = usb_sndbulkpipe(dev,
2182 usb_endpoint_num(ep_out));
2183 sc->recv_bulk_pipe = usb_rcvbulkpipe(dev,
2184 usb_endpoint_num(ep_in));
2185
2186 return 0;
2187}
2188
2189/*
2190 * Probing is done in the process context, which allows us to cheat
2191 * and not to build a state machine for the discovery.
2192 */
2193static int ub_probe(struct usb_interface *intf,
2194 const struct usb_device_id *dev_id)
2195{
2196 struct ub_dev *sc;
2197 int nluns;
2198 int rc;
2199 int i;
2200
2201 if (usb_usual_check_type(dev_id, USB_US_TYPE_UB))
2202 return -ENXIO;
2203
2204 rc = -ENOMEM;
2205 if ((sc = kzalloc(sizeof(struct ub_dev), GFP_KERNEL)) == NULL)
2206 goto err_core;
2207 sc->lock = ub_next_lock();
2208 INIT_LIST_HEAD(&sc->luns);
2209 usb_init_urb(&sc->work_urb);
2210 tasklet_init(&sc->tasklet, ub_scsi_action, (unsigned long)sc);
2211 atomic_set(&sc->poison, 0);
2212 INIT_WORK(&sc->reset_work, ub_reset_task);
2213 init_waitqueue_head(&sc->reset_wait);
2214
2215 init_timer(&sc->work_timer);
2216 sc->work_timer.data = (unsigned long) sc;
2217 sc->work_timer.function = ub_urb_timeout;
2218
2219 ub_init_completion(&sc->work_done);
2220 sc->work_done.done = 1; /* A little yuk, but oh well... */
2221
2222 sc->dev = interface_to_usbdev(intf);
2223 sc->intf = intf;
2224 // sc->ifnum = intf->cur_altsetting->desc.bInterfaceNumber;
2225 usb_set_intfdata(intf, sc);
2226 usb_get_dev(sc->dev);
2227 /*
2228 * Since we give the interface struct to the block level through
2229 * disk->driverfs_dev, we have to pin it. Otherwise, block_uevent
2230 * oopses on close after a disconnect (kernels 2.6.16 and up).
2231 */
2232 usb_get_intf(sc->intf);
2233
2234 snprintf(sc->name, 12, DRV_NAME "(%d.%d)",
2235 sc->dev->bus->busnum, sc->dev->devnum);
2236
2237 /* XXX Verify that we can handle the device (from descriptors) */
2238
2239 if (ub_get_pipes(sc, sc->dev, intf) != 0)
2240 goto err_dev_desc;
2241
2242 /*
2243 * At this point, all USB initialization is done, do upper layer.
2244 * We really hate halfway initialized structures, so from the
2245 * invariants perspective, this ub_dev is fully constructed at
2246 * this point.
2247 */
2248
2249 /*
2250 * This is needed to clear toggles. It is a problem only if we do
2251 * `rmmod ub && modprobe ub` without disconnects, but we like that.
2252 */
2253#if 0 /* iPod Mini fails if we do this (big white iPod works) */
2254 ub_probe_clear_stall(sc, sc->recv_bulk_pipe);
2255 ub_probe_clear_stall(sc, sc->send_bulk_pipe);
2256#endif
2257
2258 /*
2259 * The way this is used by the startup code is a little specific.
2260 * A SCSI check causes a USB stall. Our common case code sees it
2261 * and clears the check, after which the device is ready for use.
2262 * But if a check was not present, any command other than
2263 * TEST_UNIT_READY ends with a lockup (including REQUEST_SENSE).
2264 *
2265 * If we neglect to clear the SCSI check, the first real command fails
2266 * (which is the capacity readout). We clear that and retry, but why
2267 * causing spurious retries for no reason.
2268 *
2269 * Revalidation may start with its own TEST_UNIT_READY, but that one
2270 * has to succeed, so we clear checks with an additional one here.
2271 * In any case it's not our business how revaliadation is implemented.
2272 */
2273 for (i = 0; i < 3; i++) { /* Retries for the schwag key from KS'04 */
2274 if ((rc = ub_sync_tur(sc, NULL)) <= 0) break;
2275 if (rc != 0x6) break;
2276 msleep(10);
2277 }
2278
2279 nluns = 1;
2280 for (i = 0; i < 3; i++) {
2281 if ((rc = ub_sync_getmaxlun(sc)) < 0)
2282 break;
2283 if (rc != 0) {
2284 nluns = rc;
2285 break;
2286 }
2287 msleep(100);
2288 }
2289
2290 for (i = 0; i < nluns; i++) {
2291 ub_probe_lun(sc, i);
2292 }
2293 return 0;
2294
2295err_dev_desc:
2296 usb_set_intfdata(intf, NULL);
2297 usb_put_intf(sc->intf);
2298 usb_put_dev(sc->dev);
2299 kfree(sc);
2300err_core:
2301 return rc;
2302}
2303
2304static int ub_probe_lun(struct ub_dev *sc, int lnum)
2305{
2306 struct ub_lun *lun;
2307 struct request_queue *q;
2308 struct gendisk *disk;
2309 int rc;
2310
2311 rc = -ENOMEM;
2312 if ((lun = kzalloc(sizeof(struct ub_lun), GFP_KERNEL)) == NULL)
2313 goto err_alloc;
2314 lun->num = lnum;
2315
2316 rc = -ENOSR;
2317 if ((lun->id = ub_id_get()) == -1)
2318 goto err_id;
2319
2320 lun->udev = sc;
2321
2322 snprintf(lun->name, 16, DRV_NAME "%c(%d.%d.%d)",
2323 lun->id + 'a', sc->dev->bus->busnum, sc->dev->devnum, lun->num);
2324
2325 lun->removable = 1; /* XXX Query this from the device */
2326 lun->changed = 1; /* ub_revalidate clears only */
2327 ub_revalidate(sc, lun);
2328
2329 rc = -ENOMEM;
2330 if ((disk = alloc_disk(UB_PARTS_PER_LUN)) == NULL)
2331 goto err_diskalloc;
2332
2333 sprintf(disk->disk_name, DRV_NAME "%c", lun->id + 'a');
2334 disk->major = UB_MAJOR;
2335 disk->first_minor = lun->id * UB_PARTS_PER_LUN;
2336 disk->fops = &ub_bd_fops;
2337 disk->private_data = lun;
2338 disk->driverfs_dev = &sc->intf->dev;
2339
2340 rc = -ENOMEM;
2341 if ((q = blk_init_queue(ub_request_fn, sc->lock)) == NULL)
2342 goto err_blkqinit;
2343
2344 disk->queue = q;
2345
2346 blk_queue_bounce_limit(q, BLK_BOUNCE_HIGH);
2347 blk_queue_max_segments(q, UB_MAX_REQ_SG);
2348 blk_queue_segment_boundary(q, 0xffffffff); /* Dubious. */
2349 blk_queue_max_hw_sectors(q, UB_MAX_SECTORS);
2350 blk_queue_logical_block_size(q, lun->capacity.bsize);
2351
2352 lun->disk = disk;
2353 q->queuedata = lun;
2354 list_add(&lun->link, &sc->luns);
2355
2356 set_capacity(disk, lun->capacity.nsec);
2357 if (lun->removable)
2358 disk->flags |= GENHD_FL_REMOVABLE;
2359
2360 add_disk(disk);
2361
2362 return 0;
2363
2364err_blkqinit:
2365 put_disk(disk);
2366err_diskalloc:
2367 ub_id_put(lun->id);
2368err_id:
2369 kfree(lun);
2370err_alloc:
2371 return rc;
2372}
2373
2374static void ub_disconnect(struct usb_interface *intf)
2375{
2376 struct ub_dev *sc = usb_get_intfdata(intf);
2377 struct ub_lun *lun;
2378 unsigned long flags;
2379
2380 /*
2381 * Prevent ub_bd_release from pulling the rug from under us.
2382 * XXX This is starting to look like a kref.
2383 * XXX Why not to take this ref at probe time?
2384 */
2385 spin_lock_irqsave(&ub_lock, flags);
2386 sc->openc++;
2387 spin_unlock_irqrestore(&ub_lock, flags);
2388
2389 /*
2390 * Fence stall clearings, operations triggered by unlinkings and so on.
2391 * We do not attempt to unlink any URBs, because we do not trust the
2392 * unlink paths in HC drivers. Also, we get -84 upon disconnect anyway.
2393 */
2394 atomic_set(&sc->poison, 1);
2395
2396 /*
2397 * Wait for reset to end, if any.
2398 */
2399 wait_event(sc->reset_wait, !sc->reset);
2400
2401 /*
2402 * Blow away queued commands.
2403 *
2404 * Actually, this never works, because before we get here
2405 * the HCD terminates outstanding URB(s). It causes our
2406 * SCSI command queue to advance, commands fail to submit,
2407 * and the whole queue drains. So, we just use this code to
2408 * print warnings.
2409 */
2410 spin_lock_irqsave(sc->lock, flags);
2411 {
2412 struct ub_scsi_cmd *cmd;
2413 int cnt = 0;
2414 while ((cmd = ub_cmdq_peek(sc)) != NULL) {
2415 cmd->error = -ENOTCONN;
2416 cmd->state = UB_CMDST_DONE;
2417 ub_cmdq_pop(sc);
2418 (*cmd->done)(sc, cmd);
2419 cnt++;
2420 }
2421 if (cnt != 0) {
2422 printk(KERN_WARNING "%s: "
2423 "%d was queued after shutdown\n", sc->name, cnt);
2424 }
2425 }
2426 spin_unlock_irqrestore(sc->lock, flags);
2427
2428 /*
2429 * Unregister the upper layer.
2430 */
2431 list_for_each_entry(lun, &sc->luns, link) {
2432 del_gendisk(lun->disk);
2433 /*
2434 * I wish I could do:
2435 * queue_flag_set(QUEUE_FLAG_DEAD, q);
2436 * As it is, we rely on our internal poisoning and let
2437 * the upper levels to spin furiously failing all the I/O.
2438 */
2439 }
2440
2441 /*
2442 * Testing for -EINPROGRESS is always a bug, so we are bending
2443 * the rules a little.
2444 */
2445 spin_lock_irqsave(sc->lock, flags);
2446 if (sc->work_urb.status == -EINPROGRESS) { /* janitors: ignore */
2447 printk(KERN_WARNING "%s: "
2448 "URB is active after disconnect\n", sc->name);
2449 }
2450 spin_unlock_irqrestore(sc->lock, flags);
2451
2452 /*
2453 * There is virtually no chance that other CPU runs a timeout so long
2454 * after ub_urb_complete should have called del_timer, but only if HCD
2455 * didn't forget to deliver a callback on unlink.
2456 */
2457 del_timer_sync(&sc->work_timer);
2458
2459 /*
2460 * At this point there must be no commands coming from anyone
2461 * and no URBs left in transit.
2462 */
2463
2464 ub_put(sc);
2465}
2466
2467static struct usb_driver ub_driver = {
2468 .name = "ub",
2469 .probe = ub_probe,
2470 .disconnect = ub_disconnect,
2471 .id_table = ub_usb_ids,
2472 .pre_reset = ub_pre_reset,
2473 .post_reset = ub_post_reset,
2474};
2475
2476static int __init ub_init(void)
2477{
2478 int rc;
2479 int i;
2480
2481 for (i = 0; i < UB_QLOCK_NUM; i++)
2482 spin_lock_init(&ub_qlockv[i]);
2483
2484 if ((rc = register_blkdev(UB_MAJOR, DRV_NAME)) != 0)
2485 goto err_regblkdev;
2486
2487 if ((rc = usb_register(&ub_driver)) != 0)
2488 goto err_register;
2489
2490 usb_usual_set_present(USB_US_TYPE_UB);
2491 return 0;
2492
2493err_register:
2494 unregister_blkdev(UB_MAJOR, DRV_NAME);
2495err_regblkdev:
2496 return rc;
2497}
2498
2499static void __exit ub_exit(void)
2500{
2501 usb_deregister(&ub_driver);
2502
2503 unregister_blkdev(UB_MAJOR, DRV_NAME);
2504 usb_usual_clear_present(USB_US_TYPE_UB);
2505}
2506
2507module_init(ub_init);
2508module_exit(ub_exit);
2509
2510MODULE_LICENSE("GPL");
diff --git a/drivers/block/viodasd.c b/drivers/block/viodasd.c
new file mode 100644
index 00000000000..9a5b2a2d616
--- /dev/null
+++ b/drivers/block/viodasd.c
@@ -0,0 +1,809 @@
1/* -*- linux-c -*-
2 * viodasd.c
3 * Authors: Dave Boutcher <boutcher@us.ibm.com>
4 * Ryan Arnold <ryanarn@us.ibm.com>
5 * Colin Devilbiss <devilbis@us.ibm.com>
6 * Stephen Rothwell
7 *
8 * (C) Copyright 2000-2004 IBM Corporation
9 *
10 * This program is free software; you can redistribute it and/or
11 * modify it under the terms of the GNU General Public License as
12 * published by the Free Software Foundation; either version 2 of the
13 * License, or (at your option) any later version.
14 *
15 * This program is distributed in the hope that it will be useful,
16 * but WITHOUT ANY WARRANTY; without even the implied warranty of
17 * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE. See the
18 * GNU General Public License for more details.
19 *
20 * You should have received a copy of the GNU General Public License
21 * along with this program; if not, write to the Free Software
22 * Foundation, Inc., 59 Temple Place, Suite 330, Boston, MA 02111-1307 USA
23 *
24 * This routine provides access to disk space (termed "DASD" in historical
25 * IBM terms) owned and managed by an OS/400 partition running on the
26 * same box as this Linux partition.
27 *
28 * All disk operations are performed by sending messages back and forth to
29 * the OS/400 partition.
30 */
31
32#define pr_fmt(fmt) "viod: " fmt
33
34#include <linux/major.h>
35#include <linux/fs.h>
36#include <linux/module.h>
37#include <linux/kernel.h>
38#include <linux/blkdev.h>
39#include <linux/genhd.h>
40#include <linux/hdreg.h>
41#include <linux/errno.h>
42#include <linux/init.h>
43#include <linux/string.h>
44#include <linux/mutex.h>
45#include <linux/dma-mapping.h>
46#include <linux/completion.h>
47#include <linux/device.h>
48#include <linux/scatterlist.h>
49
50#include <asm/uaccess.h>
51#include <asm/vio.h>
52#include <asm/iseries/hv_types.h>
53#include <asm/iseries/hv_lp_event.h>
54#include <asm/iseries/hv_lp_config.h>
55#include <asm/iseries/vio.h>
56#include <asm/firmware.h>
57
58MODULE_DESCRIPTION("iSeries Virtual DASD");
59MODULE_AUTHOR("Dave Boutcher");
60MODULE_LICENSE("GPL");
61
62/*
63 * We only support 7 partitions per physical disk....so with minor
64 * numbers 0-255 we get a maximum of 32 disks.
65 */
66#define VIOD_GENHD_NAME "iseries/vd"
67
68#define VIOD_VERS "1.64"
69
70enum {
71 PARTITION_SHIFT = 3,
72 MAX_DISKNO = HVMAXARCHITECTEDVIRTUALDISKS,
73 MAX_DISK_NAME = FIELD_SIZEOF(struct gendisk, disk_name)
74};
75
76static DEFINE_MUTEX(viodasd_mutex);
77static DEFINE_SPINLOCK(viodasd_spinlock);
78
79#define VIOMAXREQ 16
80
81#define DEVICE_NO(cell) ((struct viodasd_device *)(cell) - &viodasd_devices[0])
82
83struct viodasd_waitevent {
84 struct completion com;
85 int rc;
86 u16 sub_result;
87 int max_disk; /* open */
88};
89
90static const struct vio_error_entry viodasd_err_table[] = {
91 { 0x0201, EINVAL, "Invalid Range" },
92 { 0x0202, EINVAL, "Invalid Token" },
93 { 0x0203, EIO, "DMA Error" },
94 { 0x0204, EIO, "Use Error" },
95 { 0x0205, EIO, "Release Error" },
96 { 0x0206, EINVAL, "Invalid Disk" },
97 { 0x0207, EBUSY, "Can't Lock" },
98 { 0x0208, EIO, "Already Locked" },
99 { 0x0209, EIO, "Already Unlocked" },
100 { 0x020A, EIO, "Invalid Arg" },
101 { 0x020B, EIO, "Bad IFS File" },
102 { 0x020C, EROFS, "Read Only Device" },
103 { 0x02FF, EIO, "Internal Error" },
104 { 0x0000, 0, NULL },
105};
106
107/*
108 * Figure out the biggest I/O request (in sectors) we can accept
109 */
110#define VIODASD_MAXSECTORS (4096 / 512 * VIOMAXBLOCKDMA)
111
112/*
113 * Number of disk I/O requests we've sent to OS/400
114 */
115static int num_req_outstanding;
116
117/*
118 * This is our internal structure for keeping track of disk devices
119 */
120struct viodasd_device {
121 u16 cylinders;
122 u16 tracks;
123 u16 sectors;
124 u16 bytes_per_sector;
125 u64 size;
126 int read_only;
127 spinlock_t q_lock;
128 struct gendisk *disk;
129 struct device *dev;
130} viodasd_devices[MAX_DISKNO];
131
132/*
133 * External open entry point.
134 */
135static int viodasd_open(struct block_device *bdev, fmode_t mode)
136{
137 struct viodasd_device *d = bdev->bd_disk->private_data;
138 HvLpEvent_Rc hvrc;
139 struct viodasd_waitevent we;
140 u16 flags = 0;
141
142 if (d->read_only) {
143 if (mode & FMODE_WRITE)
144 return -EROFS;
145 flags = vioblockflags_ro;
146 }
147
148 init_completion(&we.com);
149
150 /* Send the open event to OS/400 */
151 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
152 HvLpEvent_Type_VirtualIo,
153 viomajorsubtype_blockio | vioblockopen,
154 HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
155 viopath_sourceinst(viopath_hostLp),
156 viopath_targetinst(viopath_hostLp),
157 (u64)(unsigned long)&we, VIOVERSION << 16,
158 ((u64)DEVICE_NO(d) << 48) | ((u64)flags << 32),
159 0, 0, 0);
160 if (hvrc != 0) {
161 pr_warning("HV open failed %d\n", (int)hvrc);
162 return -EIO;
163 }
164
165 wait_for_completion(&we.com);
166
167 /* Check the return code */
168 if (we.rc != 0) {
169 const struct vio_error_entry *err =
170 vio_lookup_rc(viodasd_err_table, we.sub_result);
171
172 pr_warning("bad rc opening disk: %d:0x%04x (%s)\n",
173 (int)we.rc, we.sub_result, err->msg);
174 return -EIO;
175 }
176
177 return 0;
178}
179
180static int viodasd_unlocked_open(struct block_device *bdev, fmode_t mode)
181{
182 int ret;
183
184 mutex_lock(&viodasd_mutex);
185 ret = viodasd_open(bdev, mode);
186 mutex_unlock(&viodasd_mutex);
187
188 return ret;
189}
190
191
192/*
193 * External release entry point.
194 */
195static int viodasd_release(struct gendisk *disk, fmode_t mode)
196{
197 struct viodasd_device *d = disk->private_data;
198 HvLpEvent_Rc hvrc;
199
200 mutex_lock(&viodasd_mutex);
201 /* Send the event to OS/400. We DON'T expect a response */
202 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
203 HvLpEvent_Type_VirtualIo,
204 viomajorsubtype_blockio | vioblockclose,
205 HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
206 viopath_sourceinst(viopath_hostLp),
207 viopath_targetinst(viopath_hostLp),
208 0, VIOVERSION << 16,
209 ((u64)DEVICE_NO(d) << 48) /* | ((u64)flags << 32) */,
210 0, 0, 0);
211 if (hvrc != 0)
212 pr_warning("HV close call failed %d\n", (int)hvrc);
213
214 mutex_unlock(&viodasd_mutex);
215
216 return 0;
217}
218
219
220/* External ioctl entry point.
221 */
222static int viodasd_getgeo(struct block_device *bdev, struct hd_geometry *geo)
223{
224 struct gendisk *disk = bdev->bd_disk;
225 struct viodasd_device *d = disk->private_data;
226
227 geo->sectors = d->sectors ? d->sectors : 32;
228 geo->heads = d->tracks ? d->tracks : 64;
229 geo->cylinders = d->cylinders ? d->cylinders :
230 get_capacity(disk) / (geo->sectors * geo->heads);
231
232 return 0;
233}
234
235/*
236 * Our file operations table
237 */
238static const struct block_device_operations viodasd_fops = {
239 .owner = THIS_MODULE,
240 .open = viodasd_unlocked_open,
241 .release = viodasd_release,
242 .getgeo = viodasd_getgeo,
243};
244
245/*
246 * End a request
247 */
248static void viodasd_end_request(struct request *req, int error,
249 int num_sectors)
250{
251 __blk_end_request(req, error, num_sectors << 9);
252}
253
254/*
255 * Send an actual I/O request to OS/400
256 */
257static int send_request(struct request *req)
258{
259 u64 start;
260 int direction;
261 int nsg;
262 u16 viocmd;
263 HvLpEvent_Rc hvrc;
264 struct vioblocklpevent *bevent;
265 struct HvLpEvent *hev;
266 struct scatterlist sg[VIOMAXBLOCKDMA];
267 int sgindex;
268 struct viodasd_device *d;
269 unsigned long flags;
270
271 start = (u64)blk_rq_pos(req) << 9;
272
273 if (rq_data_dir(req) == READ) {
274 direction = DMA_FROM_DEVICE;
275 viocmd = viomajorsubtype_blockio | vioblockread;
276 } else {
277 direction = DMA_TO_DEVICE;
278 viocmd = viomajorsubtype_blockio | vioblockwrite;
279 }
280
281 d = req->rq_disk->private_data;
282
283 /* Now build the scatter-gather list */
284 sg_init_table(sg, VIOMAXBLOCKDMA);
285 nsg = blk_rq_map_sg(req->q, req, sg);
286 nsg = dma_map_sg(d->dev, sg, nsg, direction);
287
288 spin_lock_irqsave(&viodasd_spinlock, flags);
289 num_req_outstanding++;
290
291 /* This optimization handles a single DMA block */
292 if (nsg == 1)
293 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
294 HvLpEvent_Type_VirtualIo, viocmd,
295 HvLpEvent_AckInd_DoAck,
296 HvLpEvent_AckType_ImmediateAck,
297 viopath_sourceinst(viopath_hostLp),
298 viopath_targetinst(viopath_hostLp),
299 (u64)(unsigned long)req, VIOVERSION << 16,
300 ((u64)DEVICE_NO(d) << 48), start,
301 ((u64)sg_dma_address(&sg[0])) << 32,
302 sg_dma_len(&sg[0]));
303 else {
304 bevent = (struct vioblocklpevent *)
305 vio_get_event_buffer(viomajorsubtype_blockio);
306 if (bevent == NULL) {
307 pr_warning("error allocating disk event buffer\n");
308 goto error_ret;
309 }
310
311 /*
312 * Now build up the actual request. Note that we store
313 * the pointer to the request in the correlation
314 * token so we can match the response up later
315 */
316 memset(bevent, 0, sizeof(struct vioblocklpevent));
317 hev = &bevent->event;
318 hev->flags = HV_LP_EVENT_VALID | HV_LP_EVENT_DO_ACK |
319 HV_LP_EVENT_INT;
320 hev->xType = HvLpEvent_Type_VirtualIo;
321 hev->xSubtype = viocmd;
322 hev->xSourceLp = HvLpConfig_getLpIndex();
323 hev->xTargetLp = viopath_hostLp;
324 hev->xSizeMinus1 =
325 offsetof(struct vioblocklpevent, u.rw_data.dma_info) +
326 (sizeof(bevent->u.rw_data.dma_info[0]) * nsg) - 1;
327 hev->xSourceInstanceId = viopath_sourceinst(viopath_hostLp);
328 hev->xTargetInstanceId = viopath_targetinst(viopath_hostLp);
329 hev->xCorrelationToken = (u64)req;
330 bevent->version = VIOVERSION;
331 bevent->disk = DEVICE_NO(d);
332 bevent->u.rw_data.offset = start;
333
334 /*
335 * Copy just the dma information from the sg list
336 * into the request
337 */
338 for (sgindex = 0; sgindex < nsg; sgindex++) {
339 bevent->u.rw_data.dma_info[sgindex].token =
340 sg_dma_address(&sg[sgindex]);
341 bevent->u.rw_data.dma_info[sgindex].len =
342 sg_dma_len(&sg[sgindex]);
343 }
344
345 /* Send the request */
346 hvrc = HvCallEvent_signalLpEvent(&bevent->event);
347 vio_free_event_buffer(viomajorsubtype_blockio, bevent);
348 }
349
350 if (hvrc != HvLpEvent_Rc_Good) {
351 pr_warning("error sending disk event to OS/400 (rc %d)\n",
352 (int)hvrc);
353 goto error_ret;
354 }
355 spin_unlock_irqrestore(&viodasd_spinlock, flags);
356 return 0;
357
358error_ret:
359 num_req_outstanding--;
360 spin_unlock_irqrestore(&viodasd_spinlock, flags);
361 dma_unmap_sg(d->dev, sg, nsg, direction);
362 return -1;
363}
364
365/*
366 * This is the external request processing routine
367 */
368static void do_viodasd_request(struct request_queue *q)
369{
370 struct request *req;
371
372 /*
373 * If we already have the maximum number of requests
374 * outstanding to OS/400 just bail out. We'll come
375 * back later.
376 */
377 while (num_req_outstanding < VIOMAXREQ) {
378 req = blk_fetch_request(q);
379 if (req == NULL)
380 return;
381 /* check that request contains a valid command */
382 if (req->cmd_type != REQ_TYPE_FS) {
383 viodasd_end_request(req, -EIO, blk_rq_sectors(req));
384 continue;
385 }
386 /* Try sending the request */
387 if (send_request(req) != 0)
388 viodasd_end_request(req, -EIO, blk_rq_sectors(req));
389 }
390}
391
392/*
393 * Probe a single disk and fill in the viodasd_device structure
394 * for it.
395 */
396static int probe_disk(struct viodasd_device *d)
397{
398 HvLpEvent_Rc hvrc;
399 struct viodasd_waitevent we;
400 int dev_no = DEVICE_NO(d);
401 struct gendisk *g;
402 struct request_queue *q;
403 u16 flags = 0;
404
405retry:
406 init_completion(&we.com);
407
408 /* Send the open event to OS/400 */
409 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
410 HvLpEvent_Type_VirtualIo,
411 viomajorsubtype_blockio | vioblockopen,
412 HvLpEvent_AckInd_DoAck, HvLpEvent_AckType_ImmediateAck,
413 viopath_sourceinst(viopath_hostLp),
414 viopath_targetinst(viopath_hostLp),
415 (u64)(unsigned long)&we, VIOVERSION << 16,
416 ((u64)dev_no << 48) | ((u64)flags<< 32),
417 0, 0, 0);
418 if (hvrc != 0) {
419 pr_warning("bad rc on HV open %d\n", (int)hvrc);
420 return 0;
421 }
422
423 wait_for_completion(&we.com);
424
425 if (we.rc != 0) {
426 if (flags != 0)
427 return 0;
428 /* try again with read only flag set */
429 flags = vioblockflags_ro;
430 goto retry;
431 }
432 if (we.max_disk > (MAX_DISKNO - 1)) {
433 printk_once(KERN_INFO pr_fmt("Only examining the first %d of %d disks connected\n"),
434 MAX_DISKNO, we.max_disk + 1);
435 }
436
437 /* Send the close event to OS/400. We DON'T expect a response */
438 hvrc = HvCallEvent_signalLpEventFast(viopath_hostLp,
439 HvLpEvent_Type_VirtualIo,
440 viomajorsubtype_blockio | vioblockclose,
441 HvLpEvent_AckInd_NoAck, HvLpEvent_AckType_ImmediateAck,
442 viopath_sourceinst(viopath_hostLp),
443 viopath_targetinst(viopath_hostLp),
444 0, VIOVERSION << 16,
445 ((u64)dev_no << 48) | ((u64)flags << 32),
446 0, 0, 0);
447 if (hvrc != 0) {
448 pr_warning("bad rc sending event to OS/400 %d\n", (int)hvrc);
449 return 0;
450 }
451
452 if (d->dev == NULL) {
453 /* this is when we reprobe for new disks */
454 if (vio_create_viodasd(dev_no) == NULL) {
455 pr_warning("cannot allocate virtual device for disk %d\n",
456 dev_no);
457 return 0;
458 }
459 /*
460 * The vio_create_viodasd will have recursed into this
461 * routine with d->dev set to the new vio device and
462 * will finish the setup of the disk below.
463 */
464 return 1;
465 }
466
467 /* create the request queue for the disk */
468 spin_lock_init(&d->q_lock);
469 q = blk_init_queue(do_viodasd_request, &d->q_lock);
470 if (q == NULL) {
471 pr_warning("cannot allocate queue for disk %d\n", dev_no);
472 return 0;
473 }
474 g = alloc_disk(1 << PARTITION_SHIFT);
475 if (g == NULL) {
476 pr_warning("cannot allocate disk structure for disk %d\n",
477 dev_no);
478 blk_cleanup_queue(q);
479 return 0;
480 }
481
482 d->disk = g;
483 blk_queue_max_segments(q, VIOMAXBLOCKDMA);
484 blk_queue_max_hw_sectors(q, VIODASD_MAXSECTORS);
485 g->major = VIODASD_MAJOR;
486 g->first_minor = dev_no << PARTITION_SHIFT;
487 if (dev_no >= 26)
488 snprintf(g->disk_name, sizeof(g->disk_name),
489 VIOD_GENHD_NAME "%c%c",
490 'a' + (dev_no / 26) - 1, 'a' + (dev_no % 26));
491 else
492 snprintf(g->disk_name, sizeof(g->disk_name),
493 VIOD_GENHD_NAME "%c", 'a' + (dev_no % 26));
494 g->fops = &viodasd_fops;
495 g->queue = q;
496 g->private_data = d;
497 g->driverfs_dev = d->dev;
498 set_capacity(g, d->size >> 9);
499
500 pr_info("disk %d: %lu sectors (%lu MB) CHS=%d/%d/%d sector size %d%s\n",
501 dev_no, (unsigned long)(d->size >> 9),
502 (unsigned long)(d->size >> 20),
503 (int)d->cylinders, (int)d->tracks,
504 (int)d->sectors, (int)d->bytes_per_sector,
505 d->read_only ? " (RO)" : "");
506
507 /* register us in the global list */
508 add_disk(g);
509 return 1;
510}
511
512/* returns the total number of scatterlist elements converted */
513static int block_event_to_scatterlist(const struct vioblocklpevent *bevent,
514 struct scatterlist *sg, int *total_len)
515{
516 int i, numsg;
517 const struct rw_data *rw_data = &bevent->u.rw_data;
518 static const int offset =
519 offsetof(struct vioblocklpevent, u.rw_data.dma_info);
520 static const int element_size = sizeof(rw_data->dma_info[0]);
521
522 numsg = ((bevent->event.xSizeMinus1 + 1) - offset) / element_size;
523 if (numsg > VIOMAXBLOCKDMA)
524 numsg = VIOMAXBLOCKDMA;
525
526 *total_len = 0;
527 sg_init_table(sg, VIOMAXBLOCKDMA);
528 for (i = 0; (i < numsg) && (rw_data->dma_info[i].len > 0); ++i) {
529 sg_dma_address(&sg[i]) = rw_data->dma_info[i].token;
530 sg_dma_len(&sg[i]) = rw_data->dma_info[i].len;
531 *total_len += rw_data->dma_info[i].len;
532 }
533 return i;
534}
535
536/*
537 * Restart all queues, starting with the one _after_ the disk given,
538 * thus reducing the chance of starvation of higher numbered disks.
539 */
540static void viodasd_restart_all_queues_starting_from(int first_index)
541{
542 int i;
543
544 for (i = first_index + 1; i < MAX_DISKNO; ++i)
545 if (viodasd_devices[i].disk)
546 blk_run_queue(viodasd_devices[i].disk->queue);
547 for (i = 0; i <= first_index; ++i)
548 if (viodasd_devices[i].disk)
549 blk_run_queue(viodasd_devices[i].disk->queue);
550}
551
552/*
553 * For read and write requests, decrement the number of outstanding requests,
554 * Free the DMA buffers we allocated.
555 */
556static int viodasd_handle_read_write(struct vioblocklpevent *bevent)
557{
558 int num_sg, num_sect, pci_direction, total_len;
559 struct request *req;
560 struct scatterlist sg[VIOMAXBLOCKDMA];
561 struct HvLpEvent *event = &bevent->event;
562 unsigned long irq_flags;
563 struct viodasd_device *d;
564 int error;
565 spinlock_t *qlock;
566
567 num_sg = block_event_to_scatterlist(bevent, sg, &total_len);
568 num_sect = total_len >> 9;
569 if (event->xSubtype == (viomajorsubtype_blockio | vioblockread))
570 pci_direction = DMA_FROM_DEVICE;
571 else
572 pci_direction = DMA_TO_DEVICE;
573 req = (struct request *)bevent->event.xCorrelationToken;
574 d = req->rq_disk->private_data;
575
576 dma_unmap_sg(d->dev, sg, num_sg, pci_direction);
577
578 /*
579 * Since this is running in interrupt mode, we need to make sure
580 * we're not stepping on any global I/O operations
581 */
582 spin_lock_irqsave(&viodasd_spinlock, irq_flags);
583 num_req_outstanding--;
584 spin_unlock_irqrestore(&viodasd_spinlock, irq_flags);
585
586 error = (event->xRc == HvLpEvent_Rc_Good) ? 0 : -EIO;
587 if (error) {
588 const struct vio_error_entry *err;
589 err = vio_lookup_rc(viodasd_err_table, bevent->sub_result);
590 pr_warning("read/write error %d:0x%04x (%s)\n",
591 event->xRc, bevent->sub_result, err->msg);
592 num_sect = blk_rq_sectors(req);
593 }
594 qlock = req->q->queue_lock;
595 spin_lock_irqsave(qlock, irq_flags);
596 viodasd_end_request(req, error, num_sect);
597 spin_unlock_irqrestore(qlock, irq_flags);
598
599 /* Finally, try to get more requests off of this device's queue */
600 viodasd_restart_all_queues_starting_from(DEVICE_NO(d));
601
602 return 0;
603}
604
605/* This routine handles incoming block LP events */
606static void handle_block_event(struct HvLpEvent *event)
607{
608 struct vioblocklpevent *bevent = (struct vioblocklpevent *)event;
609 struct viodasd_waitevent *pwe;
610
611 if (event == NULL)
612 /* Notification that a partition went away! */
613 return;
614 /* First, we should NEVER get an int here...only acks */
615 if (hvlpevent_is_int(event)) {
616 pr_warning("Yikes! got an int in viodasd event handler!\n");
617 if (hvlpevent_need_ack(event)) {
618 event->xRc = HvLpEvent_Rc_InvalidSubtype;
619 HvCallEvent_ackLpEvent(event);
620 }
621 }
622
623 switch (event->xSubtype & VIOMINOR_SUBTYPE_MASK) {
624 case vioblockopen:
625 /*
626 * Handle a response to an open request. We get all the
627 * disk information in the response, so update it. The
628 * correlation token contains a pointer to a waitevent
629 * structure that has a completion in it. update the
630 * return code in the waitevent structure and post the
631 * completion to wake up the guy who sent the request
632 */
633 pwe = (struct viodasd_waitevent *)event->xCorrelationToken;
634 pwe->rc = event->xRc;
635 pwe->sub_result = bevent->sub_result;
636 if (event->xRc == HvLpEvent_Rc_Good) {
637 const struct open_data *data = &bevent->u.open_data;
638 struct viodasd_device *device =
639 &viodasd_devices[bevent->disk];
640 device->read_only =
641 bevent->flags & vioblockflags_ro;
642 device->size = data->disk_size;
643 device->cylinders = data->cylinders;
644 device->tracks = data->tracks;
645 device->sectors = data->sectors;
646 device->bytes_per_sector = data->bytes_per_sector;
647 pwe->max_disk = data->max_disk;
648 }
649 complete(&pwe->com);
650 break;
651 case vioblockclose:
652 break;
653 case vioblockread:
654 case vioblockwrite:
655 viodasd_handle_read_write(bevent);
656 break;
657
658 default:
659 pr_warning("invalid subtype!");
660 if (hvlpevent_need_ack(event)) {
661 event->xRc = HvLpEvent_Rc_InvalidSubtype;
662 HvCallEvent_ackLpEvent(event);
663 }
664 }
665}
666
667/*
668 * Get the driver to reprobe for more disks.
669 */
670static ssize_t probe_disks(struct device_driver *drv, const char *buf,
671 size_t count)
672{
673 struct viodasd_device *d;
674
675 for (d = viodasd_devices; d < &viodasd_devices[MAX_DISKNO]; d++) {
676 if (d->disk == NULL)
677 probe_disk(d);
678 }
679 return count;
680}
681static DRIVER_ATTR(probe, S_IWUSR, NULL, probe_disks);
682
683static int viodasd_probe(struct vio_dev *vdev, const struct vio_device_id *id)
684{
685 struct viodasd_device *d = &viodasd_devices[vdev->unit_address];
686
687 d->dev = &vdev->dev;
688 if (!probe_disk(d))
689 return -ENODEV;
690 return 0;
691}
692
693static int viodasd_remove(struct vio_dev *vdev)
694{
695 struct viodasd_device *d;
696
697 d = &viodasd_devices[vdev->unit_address];
698 if (d->disk) {
699 del_gendisk(d->disk);
700 blk_cleanup_queue(d->disk->queue);
701 put_disk(d->disk);
702 d->disk = NULL;
703 }
704 d->dev = NULL;
705 return 0;
706}
707
708/**
709 * viodasd_device_table: Used by vio.c to match devices that we
710 * support.
711 */
712static struct vio_device_id viodasd_device_table[] __devinitdata = {
713 { "block", "IBM,iSeries-viodasd" },
714 { "", "" }
715};
716MODULE_DEVICE_TABLE(vio, viodasd_device_table);
717
718static struct vio_driver viodasd_driver = {
719 .id_table = viodasd_device_table,
720 .probe = viodasd_probe,
721 .remove = viodasd_remove,
722 .driver = {
723 .name = "viodasd",
724 .owner = THIS_MODULE,
725 }
726};
727
728static int need_delete_probe;
729
730/*
731 * Initialize the whole device driver. Handle module and non-module
732 * versions
733 */
734static int __init viodasd_init(void)
735{
736 int rc;
737
738 if (!firmware_has_feature(FW_FEATURE_ISERIES)) {
739 rc = -ENODEV;
740 goto early_fail;
741 }
742
743 /* Try to open to our host lp */
744 if (viopath_hostLp == HvLpIndexInvalid)
745 vio_set_hostlp();
746
747 if (viopath_hostLp == HvLpIndexInvalid) {
748 pr_warning("invalid hosting partition\n");
749 rc = -EIO;
750 goto early_fail;
751 }
752
753 pr_info("vers " VIOD_VERS ", hosting partition %d\n", viopath_hostLp);
754
755 /* register the block device */
756 rc = register_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
757 if (rc) {
758 pr_warning("Unable to get major number %d for %s\n",
759 VIODASD_MAJOR, VIOD_GENHD_NAME);
760 goto early_fail;
761 }
762 /* Actually open the path to the hosting partition */
763 rc = viopath_open(viopath_hostLp, viomajorsubtype_blockio,
764 VIOMAXREQ + 2);
765 if (rc) {
766 pr_warning("error opening path to host partition %d\n",
767 viopath_hostLp);
768 goto unregister_blk;
769 }
770
771 /* Initialize our request handler */
772 vio_setHandler(viomajorsubtype_blockio, handle_block_event);
773
774 rc = vio_register_driver(&viodasd_driver);
775 if (rc) {
776 pr_warning("vio_register_driver failed\n");
777 goto unset_handler;
778 }
779
780 /*
781 * If this call fails, it just means that we cannot dynamically
782 * add virtual disks, but the driver will still work fine for
783 * all existing disk, so ignore the failure.
784 */
785 if (!driver_create_file(&viodasd_driver.driver, &driver_attr_probe))
786 need_delete_probe = 1;
787
788 return 0;
789
790unset_handler:
791 vio_clearHandler(viomajorsubtype_blockio);
792 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
793unregister_blk:
794 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
795early_fail:
796 return rc;
797}
798module_init(viodasd_init);
799
800void __exit viodasd_exit(void)
801{
802 if (need_delete_probe)
803 driver_remove_file(&viodasd_driver.driver, &driver_attr_probe);
804 vio_unregister_driver(&viodasd_driver);
805 vio_clearHandler(viomajorsubtype_blockio);
806 viopath_close(viopath_hostLp, viomajorsubtype_blockio, VIOMAXREQ + 2);
807 unregister_blkdev(VIODASD_MAJOR, VIOD_GENHD_NAME);
808}
809module_exit(viodasd_exit);
generated by cgit v1.2.2 (git 2.25.0) at 2026-01-12 09:13:34 -0500