2 * SCSI Zoned Block commands
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>
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.
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.
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,
25 #include <linux/blkdev.h>
27 #include <asm/unaligned.h>
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>
38 #include "scsi_priv.h"
41 ZBC_ZONE_TYPE_CONV = 0x1,
42 ZBC_ZONE_TYPE_SEQWRITE_REQ,
43 ZBC_ZONE_TYPE_SEQWRITE_PREF,
44 ZBC_ZONE_TYPE_RESERVED,
50 ZBC_ZONE_COND_IMP_OPEN,
51 ZBC_ZONE_COND_EXP_OPEN,
53 ZBC_ZONE_COND_READONLY = 0xd,
55 ZBC_ZONE_COND_OFFLINE,
59 * Convert a zone descriptor to a zone struct.
61 static void sd_zbc_parse_report(struct scsi_disk *sdkp,
63 struct blk_zone *zone)
65 struct scsi_device *sdp = sdkp->device;
67 memset(zone, 0, sizeof(struct blk_zone));
69 zone->type = buf[0] & 0x0f;
70 zone->cond = (buf[1] >> 4) & 0xf;
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;
85 * Issue a REPORT ZONES scsi command.
87 static int sd_zbc_report_zones(struct scsi_disk *sdkp, unsigned char *buf,
88 unsigned int buflen, sector_t lba)
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];
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);
104 result = scsi_execute_req(sdp, cmd, DMA_FROM_DEVICE,
106 timeout, SD_MAX_RETRIES, NULL);
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));
115 rep_len = get_unaligned_be32(&buf[0]);
117 sd_printk(KERN_ERR, sdkp,
118 "REPORT ZONES report invalid length %u\n",
126 int sd_zbc_setup_report_cmnd(struct scsi_cmnd *cmd)
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);
134 WARN_ON(nr_bytes == 0);
136 if (!sd_is_zoned(sdkp))
137 /* Not a zoned device */
140 ret = scsi_init_io(cmd);
141 if (ret != BLKPREP_OK)
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;
154 cmd->sc_data_direction = DMA_FROM_DEVICE;
155 cmd->sdb.length = nr_bytes;
156 cmd->transfersize = sdkp->device->sector_size;
160 * Report may return less bytes than requested. Make sure
161 * to report completion on the entire initial request.
163 rq->__data_len = nr_bytes;
168 static void sd_zbc_report_zones_complete(struct scsi_cmnd *scmd,
169 unsigned int good_bytes)
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;
183 memset(&hdr, 0, sizeof(struct blk_zone_report_hdr));
185 sg_miter_start(&miter, scsi_sglist(scmd), scsi_sg_count(scmd),
186 SG_MITER_TO_SG | SG_MITER_ATOMIC);
188 local_irq_save(flags);
189 while (sg_miter_next(&miter) && bytes < good_bytes) {
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));
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));
219 sg_miter_stop(&miter);
220 local_irq_restore(flags);
223 static inline sector_t sd_zbc_zone_sectors(struct scsi_disk *sdkp)
225 return logical_to_sectors(sdkp->device, sdkp->zone_blocks);
228 static inline unsigned int sd_zbc_zone_no(struct scsi_disk *sdkp,
231 return sectors_to_logical(sdkp->device, sector) >> sdkp->zone_shift;
234 int sd_zbc_setup_reset_cmnd(struct scsi_cmnd *cmd)
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);
241 if (!sd_is_zoned(sdkp))
242 /* Not a zoned device */
245 if (sdkp->device->changed)
248 if (sector & (sd_zbc_zone_sectors(sdkp) - 1))
249 /* Unaligned request */
253 memset(cmd->cmnd, 0, cmd->cmd_len);
254 cmd->cmnd[0] = ZBC_OUT;
255 cmd->cmnd[1] = ZO_RESET_WRITE_POINTER;
256 put_unaligned_be64(block, &cmd->cmnd[2]);
258 rq->timeout = SD_TIMEOUT;
259 cmd->sc_data_direction = DMA_NONE;
260 cmd->transfersize = 0;
266 int sd_zbc_write_lock_zone(struct scsi_cmnd *cmd)
268 struct request *rq = cmd->request;
269 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
270 sector_t sector = blk_rq_pos(rq);
271 sector_t zone_sectors = sd_zbc_zone_sectors(sdkp);
272 unsigned int zno = sd_zbc_zone_no(sdkp, sector);
275 * Note: Checks of the alignment of the write command on
276 * logical blocks is done in sd.c
279 /* Do not allow zone boundaries crossing on host-managed drives */
280 if (blk_queue_zoned_model(sdkp->disk->queue) == BLK_ZONED_HM &&
281 (sector & (zone_sectors - 1)) + blk_rq_sectors(rq) > zone_sectors)
285 * Do not issue more than one write at a time per
286 * zone. This solves write ordering problems due to
287 * the unlocking of the request queue in the dispatch
288 * path in the non scsi-mq case. For scsi-mq, this
289 * also avoids potential write reordering when multiple
290 * threads running on different CPUs write to the same
291 * zone (with a synchronized sequential pattern).
293 if (sdkp->zones_wlock &&
294 test_and_set_bit(zno, sdkp->zones_wlock))
295 return BLKPREP_DEFER;
297 WARN_ON_ONCE(cmd->flags & SCMD_ZONE_WRITE_LOCK);
298 cmd->flags |= SCMD_ZONE_WRITE_LOCK;
303 void sd_zbc_write_unlock_zone(struct scsi_cmnd *cmd)
305 struct request *rq = cmd->request;
306 struct scsi_disk *sdkp = scsi_disk(rq->rq_disk);
308 if (sdkp->zones_wlock && cmd->flags & SCMD_ZONE_WRITE_LOCK) {
309 unsigned int zno = sd_zbc_zone_no(sdkp, blk_rq_pos(rq));
310 WARN_ON_ONCE(!test_bit(zno, sdkp->zones_wlock));
311 cmd->flags &= ~SCMD_ZONE_WRITE_LOCK;
312 clear_bit_unlock(zno, sdkp->zones_wlock);
313 smp_mb__after_atomic();
317 void sd_zbc_complete(struct scsi_cmnd *cmd,
318 unsigned int good_bytes,
319 struct scsi_sense_hdr *sshdr)
321 int result = cmd->result;
322 struct request *rq = cmd->request;
324 switch (req_op(rq)) {
325 case REQ_OP_ZONE_RESET:
328 sshdr->sense_key == ILLEGAL_REQUEST &&
331 * INVALID FIELD IN CDB error: reset of a conventional
332 * zone was attempted. Nothing to worry about, so be
333 * quiet about the error.
335 rq->rq_flags |= RQF_QUIET;
339 case REQ_OP_WRITE_ZEROES:
340 case REQ_OP_WRITE_SAME:
343 sshdr->sense_key == ILLEGAL_REQUEST &&
346 * INVALID ADDRESS FOR WRITE error: It is unlikely that
347 * retrying write requests failed with any kind of
348 * alignement error will result in success. So don't.
353 case REQ_OP_ZONE_REPORT:
356 sd_zbc_report_zones_complete(cmd, good_bytes);
363 * Read zoned block device characteristics (VPD page B6).
365 static int sd_zbc_read_zoned_characteristics(struct scsi_disk *sdkp,
369 if (scsi_get_vpd_page(sdkp->device, 0xb6, buf, 64)) {
370 sd_printk(KERN_NOTICE, sdkp,
371 "Unconstrained-read check failed\n");
375 if (sdkp->device->type != TYPE_ZBC) {
378 sdkp->zones_optimal_open = get_unaligned_be32(&buf[8]);
379 sdkp->zones_optimal_nonseq = get_unaligned_be32(&buf[12]);
380 sdkp->zones_max_open = 0;
383 sdkp->urswrz = buf[4] & 1;
384 sdkp->zones_optimal_open = 0;
385 sdkp->zones_optimal_nonseq = 0;
386 sdkp->zones_max_open = get_unaligned_be32(&buf[16]);
393 * Check reported capacity.
395 static int sd_zbc_check_capacity(struct scsi_disk *sdkp,
401 if (sdkp->rc_basis != 0)
404 /* Do a report zone to get the maximum LBA to check capacity */
405 ret = sd_zbc_report_zones(sdkp, buf, SD_BUF_SIZE, 0);
409 /* The max_lba field is the capacity of this device */
410 lba = get_unaligned_be64(&buf[8]);
411 if (lba + 1 == sdkp->capacity)
414 if (sdkp->first_scan)
415 sd_printk(KERN_WARNING, sdkp,
416 "Changing capacity from %llu to max LBA+1 %llu\n",
417 (unsigned long long)sdkp->capacity,
418 (unsigned long long)lba + 1);
419 sdkp->capacity = lba + 1;
424 #define SD_ZBC_BUF_SIZE 131072
426 static int sd_zbc_check_zone_size(struct scsi_disk *sdkp)
432 unsigned int buf_len;
433 unsigned int list_length;
437 sdkp->zone_blocks = 0;
440 buf = kmalloc(SD_ZBC_BUF_SIZE, GFP_KERNEL);
444 /* Do a report zone to get the same field */
445 ret = sd_zbc_report_zones(sdkp, buf, SD_ZBC_BUF_SIZE, 0);
451 same = buf[4] & 0x0f;
454 zone_blocks = get_unaligned_be64(&rec[8]);
459 * Check the size of all zones: all zones must be of
460 * equal size, except the last zone which can be smaller
465 /* Parse REPORT ZONES header */
466 list_length = get_unaligned_be32(&buf[0]) + 64;
468 if (list_length < SD_ZBC_BUF_SIZE)
469 buf_len = list_length;
471 buf_len = SD_ZBC_BUF_SIZE;
473 /* Parse zone descriptors */
474 while (rec < buf + buf_len) {
475 zone_blocks = get_unaligned_be64(&rec[8]);
476 if (sdkp->zone_blocks == 0) {
477 sdkp->zone_blocks = zone_blocks;
478 } else if (zone_blocks != sdkp->zone_blocks &&
479 (block + zone_blocks < sdkp->capacity
480 || zone_blocks > sdkp->zone_blocks)) {
484 block += zone_blocks;
488 if (block < sdkp->capacity) {
489 ret = sd_zbc_report_zones(sdkp, buf,
490 SD_ZBC_BUF_SIZE, block);
495 } while (block < sdkp->capacity);
497 zone_blocks = sdkp->zone_blocks;
503 if (sdkp->first_scan)
504 sd_printk(KERN_NOTICE, sdkp,
505 "Devices with non constant zone "
506 "size are not supported\n");
510 if (!is_power_of_2(zone_blocks)) {
511 if (sdkp->first_scan)
512 sd_printk(KERN_NOTICE, sdkp,
513 "Devices with non power of 2 zone "
514 "size are not supported\n");
518 if (logical_to_sectors(sdkp->device, zone_blocks) > UINT_MAX) {
519 if (sdkp->first_scan)
520 sd_printk(KERN_NOTICE, sdkp,
521 "Zone size too large\n");
525 sdkp->zone_blocks = zone_blocks;
530 static int sd_zbc_setup(struct scsi_disk *sdkp)
533 /* chunk_sectors indicates the zone size */
534 blk_queue_chunk_sectors(sdkp->disk->queue,
535 logical_to_sectors(sdkp->device, sdkp->zone_blocks));
536 sdkp->zone_shift = ilog2(sdkp->zone_blocks);
537 sdkp->nr_zones = sdkp->capacity >> sdkp->zone_shift;
538 if (sdkp->capacity & (sdkp->zone_blocks - 1))
541 if (!sdkp->zones_wlock) {
542 sdkp->zones_wlock = kcalloc(BITS_TO_LONGS(sdkp->nr_zones),
543 sizeof(unsigned long),
545 if (!sdkp->zones_wlock)
552 int sd_zbc_read_zones(struct scsi_disk *sdkp,
557 if (!sd_is_zoned(sdkp))
559 * Device managed or normal SCSI disk,
560 * no special handling required
565 /* Get zoned block device characteristics */
566 ret = sd_zbc_read_zoned_characteristics(sdkp, buf);
571 * Check for unconstrained reads: host-managed devices with
572 * constrained reads (drives failing read after write pointer)
576 if (sdkp->first_scan)
577 sd_printk(KERN_NOTICE, sdkp,
578 "constrained reads devices are not supported\n");
584 ret = sd_zbc_check_capacity(sdkp, buf);
589 * Check zone size: only devices with a constant zone size (except
590 * an eventual last runt zone) that is a power of 2 are supported.
592 ret = sd_zbc_check_zone_size(sdkp);
596 /* The drive satisfies the kernel restrictions: set it up */
597 ret = sd_zbc_setup(sdkp);
601 /* READ16/WRITE16 is mandatory for ZBC disks */
602 sdkp->device->use_16_for_rw = 1;
603 sdkp->device->use_10_for_rw = 0;
613 void sd_zbc_remove(struct scsi_disk *sdkp)
615 kfree(sdkp->zones_wlock);
616 sdkp->zones_wlock = NULL;
619 void sd_zbc_print_zones(struct scsi_disk *sdkp)
621 if (!sd_is_zoned(sdkp) || !sdkp->capacity)
624 if (sdkp->capacity & (sdkp->zone_blocks - 1))
625 sd_printk(KERN_NOTICE, sdkp,
626 "%u zones of %u logical blocks + 1 runt zone\n",
630 sd_printk(KERN_NOTICE, sdkp,
631 "%u zones of %u logical blocks\n",