Merge branch 'mailbox-for-next' of git://git.linaro.org/landing-teams/working/fujitsu...
[linux-2.6-block.git] / drivers / scsi / sd_zbc.c
1 /*
2  * SCSI Zoned Block commands
3  *
4  * Copyright (C) 2014-2015 SUSE Linux GmbH
5  * Written by: Hannes Reinecke <hare@suse.de>
6  * Modified by: Damien Le Moal <damien.lemoal@hgst.com>
7  * Modified by: Shaun Tancheff <shaun.tancheff@seagate.com>
8  *
9  * This program is free software; you can redistribute it and/or
10  * modify it under the terms of the GNU General Public License version
11  * 2 as published by the Free Software Foundation.
12  *
13  * This program is distributed in the hope that it will be useful, but
14  * WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the GNU
16  * General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program; see the file COPYING.  If not, write to
20  * the Free Software Foundation, 675 Mass Ave, Cambridge, MA 02139,
21  * USA.
22  *
23  */
24
25 #include <linux/blkdev.h>
26
27 #include <asm/unaligned.h>
28
29 #include <scsi/scsi.h>
30 #include <scsi/scsi_cmnd.h>
31 #include <scsi/scsi_dbg.h>
32 #include <scsi/scsi_device.h>
33 #include <scsi/scsi_driver.h>
34 #include <scsi/scsi_host.h>
35 #include <scsi/scsi_eh.h>
36
37 #include "sd.h"
38 #include "scsi_priv.h"
39
40 enum zbc_zone_type {
41         ZBC_ZONE_TYPE_CONV = 0x1,
42         ZBC_ZONE_TYPE_SEQWRITE_REQ,
43         ZBC_ZONE_TYPE_SEQWRITE_PREF,
44         ZBC_ZONE_TYPE_RESERVED,
45 };
46
47 enum zbc_zone_cond {
48         ZBC_ZONE_COND_NO_WP,
49         ZBC_ZONE_COND_EMPTY,
50         ZBC_ZONE_COND_IMP_OPEN,
51         ZBC_ZONE_COND_EXP_OPEN,
52         ZBC_ZONE_COND_CLOSED,
53         ZBC_ZONE_COND_READONLY = 0xd,
54         ZBC_ZONE_COND_FULL,
55         ZBC_ZONE_COND_OFFLINE,
56 };
57
58 /**
59  * Convert a zone descriptor to a zone struct.
60  */
61 static void sd_zbc_parse_report(struct scsi_disk *sdkp,
62                                 u8 *buf,
63                                 struct blk_zone *zone)
64 {
65         struct scsi_device *sdp = sdkp->device;
66
67         memset(zone, 0, sizeof(struct blk_zone));
68
69         zone->type = buf[0] & 0x0f;
70         zone->cond = (buf[1] >> 4) & 0xf;
71         if (buf[1] & 0x01)
72                 zone->reset = 1;
73         if (buf[1] & 0x02)
74                 zone->non_seq = 1;
75
76         zone->len = logical_to_sectors(sdp, get_unaligned_be64(&buf[8]));
77         zone->start = logical_to_sectors(sdp, get_unaligned_be64(&buf[16]));
78         zone->wp = logical_to_sectors(sdp, get_unaligned_be64(&buf[24]));
79         if (zone->type != ZBC_ZONE_TYPE_CONV &&
80             zone->cond == ZBC_ZONE_COND_FULL)
81                 zone->wp = zone->start + zone->len;
82 }
83
84 /**
85  * Issue a REPORT ZONES scsi command.
86  */
87 static int sd_zbc_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
88                                unsigned int buflen, sector_t lba)
89 {
90         struct scsi_device *sdp = sdkp->device;
91         const int timeout = sdp->request_queue->rq_timeout;
92         struct scsi_sense_hdr sshdr;
93         unsigned char cmd[16];
94         unsigned int rep_len;
95         int result;
96
97         memset(cmd, 0, 16);
98         cmd[0] = ZBC_IN;
99         cmd[1] = ZI_REPORT_ZONES;
100         put_unaligned_be64(lba, &cmd[2]);
101         put_unaligned_be32(buflen, &cmd[10]);
102         memset(buf, 0, buflen);
103
104         result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
105                                   buf, buflen, &sshdr,
106                                   timeout, SD_MAX_RETRIES, NULL);
107         if (result) {
108                 sd_printk(KERN_ERR, sdkp,
109                           "REPORT ZONES lba %llu failed with %d/%d\n",
110                           (unsigned long long)lba,
111                           host_byte(result), driver_byte(result));
112                 return -EIO;
113         }
114
115         rep_len = get_unaligned_be32(&buf[0]);
116         if (rep_len < 64) {
117                 sd_printk(KERN_ERR, sdkp,
118                           "REPORT ZONES report invalid length %u\n",
119                           rep_len);
120                 return -EIO;
121         }
122
123         return 0;
124 }
125
126 int sd_zbc_setup_report_cmnd(struct scsi_cmnd *cmd)
127 {
128         struct request *rq = cmd->request;
129         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
130         sector_t lba, sector = blk_rq_pos(rq);
131         unsigned int nr_bytes = blk_rq_bytes(rq);
132         int ret;
133
134         WARN_ON(nr_bytes == 0);
135
136         if (!sd_is_zoned(sdkp))
137                 /* Not a zoned device */
138                 return BLKPREP_KILL;
139
140         ret = scsi_init_io(cmd);
141         if (ret != BLKPREP_OK)
142                 return ret;
143
144         cmd->cmd_len = 16;
145         memset(cmd->cmnd, 0, cmd->cmd_len);
146         cmd->cmnd[0] = ZBC_IN;
147         cmd->cmnd[1] = ZI_REPORT_ZONES;
148         lba = sectors_to_logical(sdkp->device, sector);
149         put_unaligned_be64(lba, &cmd->cmnd[2]);
150         put_unaligned_be32(nr_bytes, &cmd->cmnd[10]);
151         /* Do partial report for speeding things up */
152         cmd->cmnd[14] = ZBC_REPORT_ZONE_PARTIAL;
153
154         cmd->sc_data_direction = DMA_FROM_DEVICE;
155         cmd->sdb.length = nr_bytes;
156         cmd->transfersize = sdkp->device->sector_size;
157         cmd->allowed = 0;
158
159         /*
160          * Report may return less bytes than requested. Make sure
161          * to report completion on the entire initial request.
162          */
163         rq->__data_len = nr_bytes;
164
165         return BLKPREP_OK;
166 }
167
168 static void sd_zbc_report_zones_complete(struct scsi_cmnd *scmd,
169                                          unsigned int good_bytes)
170 {
171         struct request *rq = scmd->request;
172         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
173         struct sg_mapping_iter miter;
174         struct blk_zone_report_hdr hdr;
175         struct blk_zone zone;
176         unsigned int offset, bytes = 0;
177         unsigned long flags;
178         u8 *buf;
179
180         if (good_bytes < 64)
181                 return;
182
183         memset(&hdr, 0, sizeof(struct blk_zone_report_hdr));
184
185         sg_miter_start(&miter, scsi_sglist(scmd), scsi_sg_count(scmd),
186                        SG_MITER_TO_SG | SG_MITER_ATOMIC);
187
188         local_irq_save(flags);
189         while (sg_miter_next(&miter) && bytes < good_bytes) {
190
191                 buf = miter.addr;
192                 offset = 0;
193
194                 if (bytes == 0) {
195                         /* Set the report header */
196                         hdr.nr_zones = min_t(unsigned int,
197                                          (good_bytes - 64) / 64,
198                                          get_unaligned_be32(&buf[0]) / 64);
199                         memcpy(buf, &hdr, sizeof(struct blk_zone_report_hdr));
200                         offset += 64;
201                         bytes += 64;
202                 }
203
204                 /* Parse zone descriptors */
205                 while (offset < miter.length && hdr.nr_zones) {
206                         WARN_ON(offset > miter.length);
207                         buf = miter.addr + offset;
208                         sd_zbc_parse_report(sdkp, buf, &zone);
209                         memcpy(buf, &zone, sizeof(struct blk_zone));
210                         offset += 64;
211                         bytes += 64;
212                         hdr.nr_zones--;
213                 }
214
215                 if (!hdr.nr_zones)
216                         break;
217
218         }
219         sg_miter_stop(&miter);
220         local_irq_restore(flags);
221 }
222
223 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
224 {
225         return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
226 }
227
228 static inline unsigned int sd_zbc_zone_no(struct scsi_disk *sdkp,
229                                           sector_t sector)
230 {
231         return sectors_to_logical(sdkp->device, sector) >> sdkp->zone_shift;
232 }
233
234 int sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd)
235 {
236         struct request *rq = cmd->request;
237         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
238         sector_t sector = blk_rq_pos(rq);
239         sector_t block = sectors_to_logical(sdkp->device, sector);
240         unsigned int zno = block >> sdkp->zone_shift;
241
242         if (!sd_is_zoned(sdkp))
243                 /* Not a zoned device */
244                 return BLKPREP_KILL;
245
246         if (sdkp->device->changed)
247                 return BLKPREP_KILL;
248
249         if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
250                 /* Unaligned request */
251                 return BLKPREP_KILL;
252
253         /* Do not allow concurrent reset and writes */
254         if (sdkp->zones_wlock &&
255             test_and_set_bit(zno, sdkp->zones_wlock))
256                 return BLKPREP_DEFER;
257
258         cmd->cmd_len = 16;
259         memset(cmd->cmnd, 0, cmd->cmd_len);
260         cmd->cmnd[0] = ZBC_OUT;
261         cmd->cmnd[1] = ZO_RESET_WRITE_POINTER;
262         put_unaligned_be64(block, &cmd->cmnd[2]);
263
264         rq->timeout = SD_TIMEOUT;
265         cmd->sc_data_direction = DMA_NONE;
266         cmd->transfersize = 0;
267         cmd->allowed = 0;
268
269         return BLKPREP_OK;
270 }
271
272 int sd_zbc_setup_write_cmnd(struct scsi_cmnd *cmd)
273 {
274         struct request *rq = cmd->request;
275         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
276         sector_t sector = blk_rq_pos(rq);
277         sector_t zone_sectors = sd_zbc_zone_sectors(sdkp);
278         unsigned int zno = sd_zbc_zone_no(sdkp, sector);
279
280         /*
281          * Note: Checks of the alignment of the write command on
282          * logical blocks is done in sd.c
283          */
284
285         /* Do not allow zone boundaries crossing on host-managed drives */
286         if (blk_queue_zoned_model(sdkp->disk->queue) == BLK_ZONED_HM &&
287             (sector & (zone_sectors - 1)) + blk_rq_sectors(rq) > zone_sectors)
288                 return BLKPREP_KILL;
289
290         /*
291          * Do not issue more than one write at a time per
292          * zone. This solves write ordering problems due to
293          * the unlocking of the request queue in the dispatch
294          * path in the non scsi-mq case. For scsi-mq, this
295          * also avoids potential write reordering when multiple
296          * threads running on different CPUs write to the same
297          * zone (with a synchronized sequential pattern).
298          */
299         if (sdkp->zones_wlock &&
300             test_and_set_bit(zno, sdkp->zones_wlock))
301                 return BLKPREP_DEFER;
302
303         return BLKPREP_OK;
304 }
305
306 static void sd_zbc_unlock_zone(struct request *rq)
307 {
308         struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
309
310         if (sdkp->zones_wlock) {
311                 unsigned int zno = sd_zbc_zone_no(sdkp, blk_rq_pos(rq));
312                 WARN_ON_ONCE(!test_bit(zno, sdkp->zones_wlock));
313                 clear_bit_unlock(zno, sdkp->zones_wlock);
314                 smp_mb__after_atomic();
315         }
316 }
317
318 void sd_zbc_cancel_write_cmnd(struct scsi_cmnd *cmd)
319 {
320         sd_zbc_unlock_zone(cmd->request);
321 }
322
323 void sd_zbc_complete(struct scsi_cmnd *cmd,
324                      unsigned int good_bytes,
325                      struct scsi_sense_hdr *sshdr)
326 {
327         int result = cmd->result;
328         struct request *rq = cmd->request;
329
330         switch (req_op(rq)) {
331         case REQ_OP_WRITE:
332         case REQ_OP_WRITE_ZEROES:
333         case REQ_OP_WRITE_SAME:
334         case REQ_OP_ZONE_RESET:
335
336                 /* Unlock the zone */
337                 sd_zbc_unlock_zone(rq);
338
339                 if (!result ||
340                     sshdr->sense_key != ILLEGAL_REQUEST)
341                         break;
342
343                 switch (sshdr->asc) {
344                 case 0x24:
345                         /*
346                          * INVALID FIELD IN CDB error: For a zone reset,
347                          * this means that a reset of a conventional
348                          * zone was attempted. Nothing to worry about in
349                          * this case, so be quiet about the error.
350                          */
351                         if (req_op(rq) == REQ_OP_ZONE_RESET)
352                                 rq->rq_flags |= RQF_QUIET;
353                         break;
354                 case 0x21:
355                         /*
356                          * INVALID ADDRESS FOR WRITE error: It is unlikely that
357                          * retrying write requests failed with any kind of
358                          * alignement error will result in success. So don't.
359                          */
360                         cmd->allowed = 0;
361                         break;
362                 }
363
364                 break;
365
366         case REQ_OP_ZONE_REPORT:
367
368                 if (!result)
369                         sd_zbc_report_zones_complete(cmd, good_bytes);
370                 break;
371
372         }
373 }
374
375 /**
376  * Read zoned block device characteristics (VPD page B6).
377  */
378 static int sd_zbc_read_zoned_characteristics(struct scsi_disk *sdkp,
379                                              unsigned char *buf)
380 {
381
382         if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
383                 sd_printk(KERN_NOTICE, sdkp,
384                           "Unconstrained-read check failed\n");
385                 return -ENODEV;
386         }
387
388         if (sdkp->device->type != TYPE_ZBC) {
389                 /* Host-aware */
390                 sdkp->urswrz = 1;
391                 sdkp->zones_optimal_open = get_unaligned_be64(&buf[8]);
392                 sdkp->zones_optimal_nonseq = get_unaligned_be64(&buf[12]);
393                 sdkp->zones_max_open = 0;
394         } else {
395                 /* Host-managed */
396                 sdkp->urswrz = buf[4] & 1;
397                 sdkp->zones_optimal_open = 0;
398                 sdkp->zones_optimal_nonseq = 0;
399                 sdkp->zones_max_open = get_unaligned_be64(&buf[16]);
400         }
401
402         return 0;
403 }
404
405 /**
406  * Check reported capacity.
407  */
408 static int sd_zbc_check_capacity(struct scsi_disk *sdkp,
409                                  unsigned char *buf)
410 {
411         sector_t lba;
412         int ret;
413
414         if (sdkp->rc_basis != 0)
415                 return 0;
416
417         /* Do a report zone to get the maximum LBA to check capacity */
418         ret = sd_zbc_report_zones(sdkp, buf, SD_BUF_SIZE, 0);
419         if (ret)
420                 return ret;
421
422         /* The max_lba field is the capacity of this device */
423         lba = get_unaligned_be64(&buf[8]);
424         if (lba + 1 == sdkp->capacity)
425                 return 0;
426
427         if (sdkp->first_scan)
428                 sd_printk(KERN_WARNING, sdkp,
429                           "Changing capacity from %llu to max LBA+1 %llu\n",
430                           (unsigned long long)sdkp->capacity,
431                           (unsigned long long)lba + 1);
432         sdkp->capacity = lba + 1;
433
434         return 0;
435 }
436
437 #define SD_ZBC_BUF_SIZE 131072
438
439 static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
440 {
441         u64 zone_blocks;
442         sector_t block = 0;
443         unsigned char *buf;
444         unsigned char *rec;
445         unsigned int buf_len;
446         unsigned int list_length;
447         int ret;
448         u8 same;
449
450         sdkp->zone_blocks = 0;
451
452         /* Get a buffer */
453         buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
454         if (!buf)
455                 return -ENOMEM;
456
457         /* Do a report zone to get the same field */
458         ret = sd_zbc_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE, 0);
459         if (ret) {
460                 zone_blocks = 0;
461                 goto out;
462         }
463
464         same = buf[4] & 0x0f;
465         if (same > 0) {
466                 rec = &buf[64];
467                 zone_blocks = get_unaligned_be64(&rec[8]);
468                 goto out;
469         }
470
471         /*
472          * Check the size of all zones: all zones must be of
473          * equal size, except the last zone which can be smaller
474          * than other zones.
475          */
476         do {
477
478                 /* Parse REPORT ZONES header */
479                 list_length = get_unaligned_be32(&buf[0]) + 64;
480                 rec = buf + 64;
481                 if (list_length < SD_ZBC_BUF_SIZE)
482                         buf_len = list_length;
483                 else
484                         buf_len = SD_ZBC_BUF_SIZE;
485
486                 /* Parse zone descriptors */
487                 while (rec < buf + buf_len) {
488                         zone_blocks = get_unaligned_be64(&rec[8]);
489                         if (sdkp->zone_blocks == 0) {
490                                 sdkp->zone_blocks = zone_blocks;
491                         } else if (zone_blocks != sdkp->zone_blocks &&
492                                    (block + zone_blocks < sdkp->capacity
493                                     || zone_blocks > sdkp->zone_blocks)) {
494                                 zone_blocks = 0;
495                                 goto out;
496                         }
497                         block += zone_blocks;
498                         rec += 64;
499                 }
500
501                 if (block < sdkp->capacity) {
502                         ret = sd_zbc_report_zones(sdkp, buf,
503                                                   SD_ZBC_BUF_SIZE, block);
504                         if (ret)
505                                 return ret;
506                 }
507
508         } while (block < sdkp->capacity);
509
510         zone_blocks = sdkp->zone_blocks;
511
512 out:
513         kfree(buf);
514
515         if (!zone_blocks) {
516                 if (sdkp->first_scan)
517                         sd_printk(KERN_NOTICE, sdkp,
518                                   "Devices with non constant zone "
519                                   "size are not supported\n");
520                 return -ENODEV;
521         }
522
523         if (!is_power_of_2(zone_blocks)) {
524                 if (sdkp->first_scan)
525                         sd_printk(KERN_NOTICE, sdkp,
526                                   "Devices with non power of 2 zone "
527                                   "size are not supported\n");
528                 return -ENODEV;
529         }
530
531         if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
532                 if (sdkp->first_scan)
533                         sd_printk(KERN_NOTICE, sdkp,
534                                   "Zone size too large\n");
535                 return -ENODEV;
536         }
537
538         sdkp->zone_blocks = zone_blocks;
539
540         return 0;
541 }
542
543 static int sd_zbc_setup(struct scsi_disk *sdkp)
544 {
545
546         /* chunk_sectors indicates the zone size */
547         blk_queue_chunk_sectors(sdkp->disk->queue,
548                         logical_to_sectors(sdkp->device, sdkp->zone_blocks));
549         sdkp->zone_shift = ilog2(sdkp->zone_blocks);
550         sdkp->nr_zones = sdkp->capacity >> sdkp->zone_shift;
551         if (sdkp->capacity & (sdkp->zone_blocks - 1))
552                 sdkp->nr_zones++;
553
554         if (!sdkp->zones_wlock) {
555                 sdkp->zones_wlock = kcalloc(BITS_TO_LONGS(sdkp->nr_zones),
556                                             sizeof(unsigned long),
557                                             GFP_KERNEL);
558                 if (!sdkp->zones_wlock)
559                         return -ENOMEM;
560         }
561
562         return 0;
563 }
564
565 int sd_zbc_read_zones(struct scsi_disk *sdkp,
566                       unsigned char *buf)
567 {
568         sector_t capacity;
569         int ret = 0;
570
571         if (!sd_is_zoned(sdkp))
572                 /*
573                  * Device managed or normal SCSI disk,
574                  * no special handling required
575                  */
576                 return 0;
577
578
579         /* Get zoned block device characteristics */
580         ret = sd_zbc_read_zoned_characteristics(sdkp, buf);
581         if (ret)
582                 goto err;
583
584         /*
585          * Check for unconstrained reads: host-managed devices with
586          * constrained reads (drives failing read after write pointer)
587          * are not supported.
588          */
589         if (!sdkp->urswrz) {
590                 if (sdkp->first_scan)
591                         sd_printk(KERN_NOTICE, sdkp,
592                           "constrained reads devices are not supported\n");
593                 ret = -ENODEV;
594                 goto err;
595         }
596
597         /* Check capacity */
598         ret = sd_zbc_check_capacity(sdkp, buf);
599         if (ret)
600                 goto err;
601         capacity = logical_to_sectors(sdkp->device, sdkp->capacity);
602
603         /*
604          * Check zone size: only devices with a constant zone size (except
605          * an eventual last runt zone) that is a power of 2 are supported.
606          */
607         ret = sd_zbc_check_zone_size(sdkp);
608         if (ret)
609                 goto err;
610
611         /* The drive satisfies the kernel restrictions: set it up */
612         ret = sd_zbc_setup(sdkp);
613         if (ret)
614                 goto err;
615
616         /* READ16/WRITE16 is mandatory for ZBC disks */
617         sdkp->device->use_16_for_rw = 1;
618         sdkp->device->use_10_for_rw = 0;
619
620         return 0;
621
622 err:
623         sdkp->capacity = 0;
624
625         return ret;
626 }
627
628 void sd_zbc_remove(struct scsi_disk *sdkp)
629 {
630         kfree(sdkp->zones_wlock);
631         sdkp->zones_wlock = NULL;
632 }
633
634 void sd_zbc_print_zones(struct scsi_disk *sdkp)
635 {
636         if (!sd_is_zoned(sdkp) || !sdkp->capacity)
637                 return;
638
639         if (sdkp->capacity & (sdkp->zone_blocks - 1))
640                 sd_printk(KERN_NOTICE, sdkp,
641                           "%u zones of %u logical blocks + 1 runt zone\n",
642                           sdkp->nr_zones - 1,
643                           sdkp->zone_blocks);
644         else
645                 sd_printk(KERN_NOTICE, sdkp,
646                           "%u zones of %u logical blocks\n",
647                           sdkp->nr_zones,
648                           sdkp->zone_blocks);
649 }