1 // SPDX-License-Identifier: GPL-2.0
3 * Block driver for media (i.e., flash cards)
5 * Copyright 2002 Hewlett-Packard Company
6 * Copyright 2005-2008 Pierre Ossman
8 * Use consistent with the GNU GPL is permitted,
9 * provided that this copyright notice is
10 * preserved in its entirety in all copies and derived works.
12 * HEWLETT-PACKARD COMPANY MAKES NO WARRANTIES, EXPRESSED OR IMPLIED,
13 * AS TO THE USEFULNESS OR CORRECTNESS OF THIS CODE OR ITS
14 * FITNESS FOR ANY PARTICULAR PURPOSE.
16 * Many thanks to Alessandro Rubini and Jonathan Corbet!
18 * Author: Andrew Christian
21 #include <linux/moduleparam.h>
22 #include <linux/module.h>
23 #include <linux/init.h>
25 #include <linux/kernel.h>
27 #include <linux/slab.h>
28 #include <linux/errno.h>
29 #include <linux/hdreg.h>
30 #include <linux/kdev_t.h>
31 #include <linux/kref.h>
32 #include <linux/blkdev.h>
33 #include <linux/cdev.h>
34 #include <linux/mutex.h>
35 #include <linux/scatterlist.h>
36 #include <linux/string_helpers.h>
37 #include <linux/delay.h>
38 #include <linux/capability.h>
39 #include <linux/compat.h>
40 #include <linux/pm_runtime.h>
41 #include <linux/idr.h>
42 #include <linux/debugfs.h>
44 #include <linux/mmc/ioctl.h>
45 #include <linux/mmc/card.h>
46 #include <linux/mmc/host.h>
47 #include <linux/mmc/mmc.h>
48 #include <linux/mmc/sd.h>
50 #include <linux/uaccess.h>
63 MODULE_ALIAS("mmc:block");
64 #ifdef MODULE_PARAM_PREFIX
65 #undef MODULE_PARAM_PREFIX
67 #define MODULE_PARAM_PREFIX "mmcblk."
70 * Set a 10 second timeout for polling write request busy state. Note, mmc core
71 * is setting a 3 second timeout for SD cards, and SDHCI has long had a 10
72 * second software timer to timeout the whole request, so 10 seconds should be
75 #define MMC_BLK_TIMEOUT_MS (10 * 1000)
76 #define MMC_EXTRACT_INDEX_FROM_ARG(x) ((x & 0x00FF0000) >> 16)
77 #define MMC_EXTRACT_VALUE_FROM_ARG(x) ((x & 0x0000FF00) >> 8)
79 static DEFINE_MUTEX(block_mutex);
82 * The defaults come from config options but can be overriden by module
85 static int perdev_minors = CONFIG_MMC_BLOCK_MINORS;
88 * We've only got one major, so number of mmcblk devices is
89 * limited to (1 << 20) / number of minors per device. It is also
90 * limited by the MAX_DEVICES below.
92 static int max_devices;
94 #define MAX_DEVICES 256
96 static DEFINE_IDA(mmc_blk_ida);
97 static DEFINE_IDA(mmc_rpmb_ida);
99 struct mmc_blk_busy_data {
100 struct mmc_card *card;
105 * There is one mmc_blk_data per slot.
107 struct mmc_blk_data {
108 struct device *parent;
109 struct gendisk *disk;
110 struct mmc_queue queue;
111 struct list_head part;
112 struct list_head rpmbs;
115 #define MMC_BLK_CMD23 (1 << 0) /* Can do SET_BLOCK_COUNT for multiblock */
116 #define MMC_BLK_REL_WR (1 << 1) /* MMC Reliable write support */
119 unsigned int read_only;
120 unsigned int part_type;
121 unsigned int reset_done;
122 #define MMC_BLK_READ BIT(0)
123 #define MMC_BLK_WRITE BIT(1)
124 #define MMC_BLK_DISCARD BIT(2)
125 #define MMC_BLK_SECDISCARD BIT(3)
126 #define MMC_BLK_CQE_RECOVERY BIT(4)
127 #define MMC_BLK_TRIM BIT(5)
130 * Only set in main mmc_blk_data associated
131 * with mmc_card with dev_set_drvdata, and keeps
132 * track of the current selected device partition.
134 unsigned int part_curr;
135 #define MMC_BLK_PART_INVALID UINT_MAX /* Unknown partition active */
138 /* debugfs files (only in main mmc_blk_data) */
139 struct dentry *status_dentry;
140 struct dentry *ext_csd_dentry;
143 /* Device type for RPMB character devices */
144 static dev_t mmc_rpmb_devt;
146 /* Bus type for RPMB character devices */
147 static struct bus_type mmc_rpmb_bus_type = {
152 * struct mmc_rpmb_data - special RPMB device type for these areas
153 * @dev: the device for the RPMB area
154 * @chrdev: character device for the RPMB area
155 * @id: unique device ID number
156 * @part_index: partition index (0 on first)
157 * @md: parent MMC block device
158 * @node: list item, so we can put this device on a list
160 struct mmc_rpmb_data {
164 unsigned int part_index;
165 struct mmc_blk_data *md;
166 struct list_head node;
169 static DEFINE_MUTEX(open_lock);
171 module_param(perdev_minors, int, 0444);
172 MODULE_PARM_DESC(perdev_minors, "Minors numbers to allocate per device");
174 static inline int mmc_blk_part_switch(struct mmc_card *card,
175 unsigned int part_type);
176 static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
177 struct mmc_card *card,
179 struct mmc_queue *mq);
180 static void mmc_blk_hsq_req_done(struct mmc_request *mrq);
181 static int mmc_spi_err_check(struct mmc_card *card);
183 static struct mmc_blk_data *mmc_blk_get(struct gendisk *disk)
185 struct mmc_blk_data *md;
187 mutex_lock(&open_lock);
188 md = disk->private_data;
189 if (md && !kref_get_unless_zero(&md->kref))
191 mutex_unlock(&open_lock);
196 static inline int mmc_get_devidx(struct gendisk *disk)
198 int devidx = disk->first_minor / perdev_minors;
202 static void mmc_blk_kref_release(struct kref *ref)
204 struct mmc_blk_data *md = container_of(ref, struct mmc_blk_data, kref);
207 devidx = mmc_get_devidx(md->disk);
208 ida_simple_remove(&mmc_blk_ida, devidx);
210 mutex_lock(&open_lock);
211 md->disk->private_data = NULL;
212 mutex_unlock(&open_lock);
218 static void mmc_blk_put(struct mmc_blk_data *md)
220 kref_put(&md->kref, mmc_blk_kref_release);
223 static ssize_t power_ro_lock_show(struct device *dev,
224 struct device_attribute *attr, char *buf)
227 struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
228 struct mmc_card *card = md->queue.card;
231 if (card->ext_csd.boot_ro_lock & EXT_CSD_BOOT_WP_B_PERM_WP_EN)
233 else if (card->ext_csd.boot_ro_lock & EXT_CSD_BOOT_WP_B_PWR_WP_EN)
236 ret = snprintf(buf, PAGE_SIZE, "%d\n", locked);
243 static ssize_t power_ro_lock_store(struct device *dev,
244 struct device_attribute *attr, const char *buf, size_t count)
247 struct mmc_blk_data *md, *part_md;
248 struct mmc_queue *mq;
252 if (kstrtoul(buf, 0, &set))
258 md = mmc_blk_get(dev_to_disk(dev));
261 /* Dispatch locking to the block layer */
262 req = blk_mq_alloc_request(mq->queue, REQ_OP_DRV_OUT, 0);
264 count = PTR_ERR(req);
267 req_to_mmc_queue_req(req)->drv_op = MMC_DRV_OP_BOOT_WP;
268 req_to_mmc_queue_req(req)->drv_op_result = -EIO;
269 blk_execute_rq(req, false);
270 ret = req_to_mmc_queue_req(req)->drv_op_result;
271 blk_mq_free_request(req);
274 pr_info("%s: Locking boot partition ro until next power on\n",
275 md->disk->disk_name);
276 set_disk_ro(md->disk, 1);
278 list_for_each_entry(part_md, &md->part, part)
279 if (part_md->area_type == MMC_BLK_DATA_AREA_BOOT) {
280 pr_info("%s: Locking boot partition ro until next power on\n", part_md->disk->disk_name);
281 set_disk_ro(part_md->disk, 1);
289 static DEVICE_ATTR(ro_lock_until_next_power_on, 0,
290 power_ro_lock_show, power_ro_lock_store);
292 static ssize_t force_ro_show(struct device *dev, struct device_attribute *attr,
296 struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
298 ret = snprintf(buf, PAGE_SIZE, "%d\n",
299 get_disk_ro(dev_to_disk(dev)) ^
305 static ssize_t force_ro_store(struct device *dev, struct device_attribute *attr,
306 const char *buf, size_t count)
310 struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
311 unsigned long set = simple_strtoul(buf, &end, 0);
317 set_disk_ro(dev_to_disk(dev), set || md->read_only);
324 static DEVICE_ATTR(force_ro, 0644, force_ro_show, force_ro_store);
326 static struct attribute *mmc_disk_attrs[] = {
327 &dev_attr_force_ro.attr,
328 &dev_attr_ro_lock_until_next_power_on.attr,
332 static umode_t mmc_disk_attrs_is_visible(struct kobject *kobj,
333 struct attribute *a, int n)
335 struct device *dev = kobj_to_dev(kobj);
336 struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
337 umode_t mode = a->mode;
339 if (a == &dev_attr_ro_lock_until_next_power_on.attr &&
340 (md->area_type & MMC_BLK_DATA_AREA_BOOT) &&
341 md->queue.card->ext_csd.boot_ro_lockable) {
343 if (!(md->queue.card->ext_csd.boot_ro_lock &
344 EXT_CSD_BOOT_WP_B_PWR_WP_DIS))
352 static const struct attribute_group mmc_disk_attr_group = {
353 .is_visible = mmc_disk_attrs_is_visible,
354 .attrs = mmc_disk_attrs,
357 static const struct attribute_group *mmc_disk_attr_groups[] = {
358 &mmc_disk_attr_group,
362 static int mmc_blk_open(struct gendisk *disk, blk_mode_t mode)
364 struct mmc_blk_data *md = mmc_blk_get(disk);
367 mutex_lock(&block_mutex);
370 if ((mode & BLK_OPEN_WRITE) && md->read_only) {
375 mutex_unlock(&block_mutex);
380 static void mmc_blk_release(struct gendisk *disk)
382 struct mmc_blk_data *md = disk->private_data;
384 mutex_lock(&block_mutex);
386 mutex_unlock(&block_mutex);
390 mmc_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
392 geo->cylinders = get_capacity(bdev->bd_disk) / (4 * 16);
398 struct mmc_blk_ioc_data {
399 struct mmc_ioc_cmd ic;
402 struct mmc_rpmb_data *rpmb;
405 static struct mmc_blk_ioc_data *mmc_blk_ioctl_copy_from_user(
406 struct mmc_ioc_cmd __user *user)
408 struct mmc_blk_ioc_data *idata;
411 idata = kmalloc(sizeof(*idata), GFP_KERNEL);
417 if (copy_from_user(&idata->ic, user, sizeof(idata->ic))) {
422 idata->buf_bytes = (u64) idata->ic.blksz * idata->ic.blocks;
423 if (idata->buf_bytes > MMC_IOC_MAX_BYTES) {
428 if (!idata->buf_bytes) {
433 idata->buf = memdup_user((void __user *)(unsigned long)
434 idata->ic.data_ptr, idata->buf_bytes);
435 if (IS_ERR(idata->buf)) {
436 err = PTR_ERR(idata->buf);
448 static int mmc_blk_ioctl_copy_to_user(struct mmc_ioc_cmd __user *ic_ptr,
449 struct mmc_blk_ioc_data *idata)
451 struct mmc_ioc_cmd *ic = &idata->ic;
453 if (copy_to_user(&(ic_ptr->response), ic->response,
454 sizeof(ic->response)))
457 if (!idata->ic.write_flag) {
458 if (copy_to_user((void __user *)(unsigned long)ic->data_ptr,
459 idata->buf, idata->buf_bytes))
466 static int __mmc_blk_ioctl_cmd(struct mmc_card *card, struct mmc_blk_data *md,
467 struct mmc_blk_ioc_data *idata)
469 struct mmc_command cmd = {}, sbc = {};
470 struct mmc_data data = {};
471 struct mmc_request mrq = {};
472 struct scatterlist sg;
473 bool r1b_resp, use_r1b_resp = false;
474 unsigned int busy_timeout_ms;
476 unsigned int target_part;
478 if (!card || !md || !idata)
482 * The RPMB accesses comes in from the character device, so we
483 * need to target these explicitly. Else we just target the
484 * partition type for the block device the ioctl() was issued
488 /* Support multiple RPMB partitions */
489 target_part = idata->rpmb->part_index;
490 target_part |= EXT_CSD_PART_CONFIG_ACC_RPMB;
492 target_part = md->part_type;
495 cmd.opcode = idata->ic.opcode;
496 cmd.arg = idata->ic.arg;
497 cmd.flags = idata->ic.flags;
499 if (idata->buf_bytes) {
502 data.blksz = idata->ic.blksz;
503 data.blocks = idata->ic.blocks;
505 sg_init_one(data.sg, idata->buf, idata->buf_bytes);
507 if (idata->ic.write_flag)
508 data.flags = MMC_DATA_WRITE;
510 data.flags = MMC_DATA_READ;
512 /* data.flags must already be set before doing this. */
513 mmc_set_data_timeout(&data, card);
515 /* Allow overriding the timeout_ns for empirical tuning. */
516 if (idata->ic.data_timeout_ns)
517 data.timeout_ns = idata->ic.data_timeout_ns;
524 err = mmc_blk_part_switch(card, target_part);
528 if (idata->ic.is_acmd) {
529 err = mmc_app_cmd(card->host, card);
535 sbc.opcode = MMC_SET_BLOCK_COUNT;
537 * We don't do any blockcount validation because the max size
538 * may be increased by a future standard. We just copy the
539 * 'Reliable Write' bit here.
541 sbc.arg = data.blocks | (idata->ic.write_flag & BIT(31));
542 sbc.flags = MMC_RSP_R1 | MMC_CMD_AC;
546 if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_SANITIZE_START) &&
547 (cmd.opcode == MMC_SWITCH))
548 return mmc_sanitize(card, idata->ic.cmd_timeout_ms);
550 /* If it's an R1B response we need some more preparations. */
551 busy_timeout_ms = idata->ic.cmd_timeout_ms ? : MMC_BLK_TIMEOUT_MS;
552 r1b_resp = (cmd.flags & MMC_RSP_R1B) == MMC_RSP_R1B;
554 use_r1b_resp = mmc_prepare_busy_cmd(card->host, &cmd,
557 mmc_wait_for_req(card->host, &mrq);
558 memcpy(&idata->ic.response, cmd.resp, sizeof(cmd.resp));
561 dev_err(mmc_dev(card->host), "%s: cmd error %d\n",
562 __func__, cmd.error);
566 dev_err(mmc_dev(card->host), "%s: data error %d\n",
567 __func__, data.error);
572 * Make sure the cache of the PARTITION_CONFIG register and
573 * PARTITION_ACCESS bits is updated in case the ioctl ext_csd write
574 * changed it successfully.
576 if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_PART_CONFIG) &&
577 (cmd.opcode == MMC_SWITCH)) {
578 struct mmc_blk_data *main_md = dev_get_drvdata(&card->dev);
579 u8 value = MMC_EXTRACT_VALUE_FROM_ARG(cmd.arg);
582 * Update cache so the next mmc_blk_part_switch call operates
583 * on up-to-date data.
585 card->ext_csd.part_config = value;
586 main_md->part_curr = value & EXT_CSD_PART_CONFIG_ACC_MASK;
590 * Make sure to update CACHE_CTRL in case it was changed. The cache
591 * will get turned back on if the card is re-initialized, e.g.
592 * suspend/resume or hw reset in recovery.
594 if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_CACHE_CTRL) &&
595 (cmd.opcode == MMC_SWITCH)) {
596 u8 value = MMC_EXTRACT_VALUE_FROM_ARG(cmd.arg) & 1;
598 card->ext_csd.cache_ctrl = value;
602 * According to the SD specs, some commands require a delay after
603 * issuing the command.
605 if (idata->ic.postsleep_min_us)
606 usleep_range(idata->ic.postsleep_min_us, idata->ic.postsleep_max_us);
608 /* No need to poll when using HW busy detection. */
609 if ((card->host->caps & MMC_CAP_WAIT_WHILE_BUSY) && use_r1b_resp)
612 if (mmc_host_is_spi(card->host)) {
613 if (idata->ic.write_flag || r1b_resp || cmd.flags & MMC_RSP_SPI_BUSY)
614 return mmc_spi_err_check(card);
617 /* Ensure RPMB/R1B command has completed by polling with CMD13. */
618 if (idata->rpmb || r1b_resp)
619 err = mmc_poll_for_busy(card, busy_timeout_ms, false,
625 static int mmc_blk_ioctl_cmd(struct mmc_blk_data *md,
626 struct mmc_ioc_cmd __user *ic_ptr,
627 struct mmc_rpmb_data *rpmb)
629 struct mmc_blk_ioc_data *idata;
630 struct mmc_blk_ioc_data *idatas[1];
631 struct mmc_queue *mq;
632 struct mmc_card *card;
633 int err = 0, ioc_err = 0;
636 idata = mmc_blk_ioctl_copy_from_user(ic_ptr);
638 return PTR_ERR(idata);
639 /* This will be NULL on non-RPMB ioctl():s */
642 card = md->queue.card;
649 * Dispatch the ioctl() into the block request queue.
652 req = blk_mq_alloc_request(mq->queue,
653 idata->ic.write_flag ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
659 req_to_mmc_queue_req(req)->drv_op =
660 rpmb ? MMC_DRV_OP_IOCTL_RPMB : MMC_DRV_OP_IOCTL;
661 req_to_mmc_queue_req(req)->drv_op_result = -EIO;
662 req_to_mmc_queue_req(req)->drv_op_data = idatas;
663 req_to_mmc_queue_req(req)->ioc_count = 1;
664 blk_execute_rq(req, false);
665 ioc_err = req_to_mmc_queue_req(req)->drv_op_result;
666 err = mmc_blk_ioctl_copy_to_user(ic_ptr, idata);
667 blk_mq_free_request(req);
672 return ioc_err ? ioc_err : err;
675 static int mmc_blk_ioctl_multi_cmd(struct mmc_blk_data *md,
676 struct mmc_ioc_multi_cmd __user *user,
677 struct mmc_rpmb_data *rpmb)
679 struct mmc_blk_ioc_data **idata = NULL;
680 struct mmc_ioc_cmd __user *cmds = user->cmds;
681 struct mmc_card *card;
682 struct mmc_queue *mq;
683 int err = 0, ioc_err = 0;
688 if (copy_from_user(&num_of_cmds, &user->num_of_cmds,
689 sizeof(num_of_cmds)))
695 if (num_of_cmds > MMC_IOC_MAX_CMDS)
699 idata = kcalloc(n, sizeof(*idata), GFP_KERNEL);
703 for (i = 0; i < n; i++) {
704 idata[i] = mmc_blk_ioctl_copy_from_user(&cmds[i]);
705 if (IS_ERR(idata[i])) {
706 err = PTR_ERR(idata[i]);
710 /* This will be NULL on non-RPMB ioctl():s */
711 idata[i]->rpmb = rpmb;
714 card = md->queue.card;
722 * Dispatch the ioctl()s into the block request queue.
725 req = blk_mq_alloc_request(mq->queue,
726 idata[0]->ic.write_flag ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
731 req_to_mmc_queue_req(req)->drv_op =
732 rpmb ? MMC_DRV_OP_IOCTL_RPMB : MMC_DRV_OP_IOCTL;
733 req_to_mmc_queue_req(req)->drv_op_result = -EIO;
734 req_to_mmc_queue_req(req)->drv_op_data = idata;
735 req_to_mmc_queue_req(req)->ioc_count = n;
736 blk_execute_rq(req, false);
737 ioc_err = req_to_mmc_queue_req(req)->drv_op_result;
739 /* copy to user if data and response */
740 for (i = 0; i < n && !err; i++)
741 err = mmc_blk_ioctl_copy_to_user(&cmds[i], idata[i]);
743 blk_mq_free_request(req);
746 for (i = 0; i < n; i++) {
747 kfree(idata[i]->buf);
751 return ioc_err ? ioc_err : err;
754 static int mmc_blk_check_blkdev(struct block_device *bdev)
757 * The caller must have CAP_SYS_RAWIO, and must be calling this on the
758 * whole block device, not on a partition. This prevents overspray
759 * between sibling partitions.
761 if (!capable(CAP_SYS_RAWIO) || bdev_is_partition(bdev))
766 static int mmc_blk_ioctl(struct block_device *bdev, blk_mode_t mode,
767 unsigned int cmd, unsigned long arg)
769 struct mmc_blk_data *md;
774 ret = mmc_blk_check_blkdev(bdev);
777 md = mmc_blk_get(bdev->bd_disk);
780 ret = mmc_blk_ioctl_cmd(md,
781 (struct mmc_ioc_cmd __user *)arg,
785 case MMC_IOC_MULTI_CMD:
786 ret = mmc_blk_check_blkdev(bdev);
789 md = mmc_blk_get(bdev->bd_disk);
792 ret = mmc_blk_ioctl_multi_cmd(md,
793 (struct mmc_ioc_multi_cmd __user *)arg,
803 static int mmc_blk_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
804 unsigned int cmd, unsigned long arg)
806 return mmc_blk_ioctl(bdev, mode, cmd, (unsigned long) compat_ptr(arg));
810 static int mmc_blk_alternative_gpt_sector(struct gendisk *disk,
813 struct mmc_blk_data *md;
816 md = mmc_blk_get(disk);
821 ret = mmc_card_alternative_gpt_sector(md->queue.card, sector);
830 static const struct block_device_operations mmc_bdops = {
831 .open = mmc_blk_open,
832 .release = mmc_blk_release,
833 .getgeo = mmc_blk_getgeo,
834 .owner = THIS_MODULE,
835 .ioctl = mmc_blk_ioctl,
837 .compat_ioctl = mmc_blk_compat_ioctl,
839 .alternative_gpt_sector = mmc_blk_alternative_gpt_sector,
842 static int mmc_blk_part_switch_pre(struct mmc_card *card,
843 unsigned int part_type)
847 if (part_type == EXT_CSD_PART_CONFIG_ACC_RPMB) {
848 if (card->ext_csd.cmdq_en) {
849 ret = mmc_cmdq_disable(card);
853 mmc_retune_pause(card->host);
859 static int mmc_blk_part_switch_post(struct mmc_card *card,
860 unsigned int part_type)
864 if (part_type == EXT_CSD_PART_CONFIG_ACC_RPMB) {
865 mmc_retune_unpause(card->host);
866 if (card->reenable_cmdq && !card->ext_csd.cmdq_en)
867 ret = mmc_cmdq_enable(card);
873 static inline int mmc_blk_part_switch(struct mmc_card *card,
874 unsigned int part_type)
877 struct mmc_blk_data *main_md = dev_get_drvdata(&card->dev);
879 if (main_md->part_curr == part_type)
882 if (mmc_card_mmc(card)) {
883 u8 part_config = card->ext_csd.part_config;
885 ret = mmc_blk_part_switch_pre(card, part_type);
889 part_config &= ~EXT_CSD_PART_CONFIG_ACC_MASK;
890 part_config |= part_type;
892 ret = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
893 EXT_CSD_PART_CONFIG, part_config,
894 card->ext_csd.part_time);
896 mmc_blk_part_switch_post(card, part_type);
900 card->ext_csd.part_config = part_config;
902 ret = mmc_blk_part_switch_post(card, main_md->part_curr);
905 main_md->part_curr = part_type;
909 static int mmc_sd_num_wr_blocks(struct mmc_card *card, u32 *written_blocks)
915 struct mmc_request mrq = {};
916 struct mmc_command cmd = {};
917 struct mmc_data data = {};
919 struct scatterlist sg;
921 err = mmc_app_cmd(card->host, card);
925 cmd.opcode = SD_APP_SEND_NUM_WR_BLKS;
927 cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC;
931 data.flags = MMC_DATA_READ;
934 mmc_set_data_timeout(&data, card);
939 blocks = kmalloc(4, GFP_KERNEL);
943 sg_init_one(&sg, blocks, 4);
945 mmc_wait_for_req(card->host, &mrq);
947 result = ntohl(*blocks);
950 if (cmd.error || data.error)
953 *written_blocks = result;
958 static unsigned int mmc_blk_clock_khz(struct mmc_host *host)
960 if (host->actual_clock)
961 return host->actual_clock / 1000;
963 /* Clock may be subject to a divisor, fudge it by a factor of 2. */
965 return host->ios.clock / 2000;
967 /* How can there be no clock */
969 return 100; /* 100 kHz is minimum possible value */
972 static unsigned int mmc_blk_data_timeout_ms(struct mmc_host *host,
973 struct mmc_data *data)
975 unsigned int ms = DIV_ROUND_UP(data->timeout_ns, 1000000);
978 if (data->timeout_clks) {
979 khz = mmc_blk_clock_khz(host);
980 ms += DIV_ROUND_UP(data->timeout_clks, khz);
987 * Attempts to reset the card and get back to the requested partition.
988 * Therefore any error here must result in cancelling the block layer
989 * request, it must not be reattempted without going through the mmc_blk
990 * partition sanity checks.
992 static int mmc_blk_reset(struct mmc_blk_data *md, struct mmc_host *host,
996 struct mmc_blk_data *main_md = dev_get_drvdata(&host->card->dev);
998 if (md->reset_done & type)
1001 md->reset_done |= type;
1002 err = mmc_hw_reset(host->card);
1004 * A successful reset will leave the card in the main partition, but
1005 * upon failure it might not be, so set it to MMC_BLK_PART_INVALID
1008 main_md->part_curr = err ? MMC_BLK_PART_INVALID : main_md->part_type;
1011 /* Ensure we switch back to the correct partition */
1012 if (mmc_blk_part_switch(host->card, md->part_type))
1014 * We have failed to get back into the correct
1015 * partition, so we need to abort the whole request.
1021 static inline void mmc_blk_reset_success(struct mmc_blk_data *md, int type)
1023 md->reset_done &= ~type;
1027 * The non-block commands come back from the block layer after it queued it and
1028 * processed it with all other requests and then they get issued in this
1031 static void mmc_blk_issue_drv_op(struct mmc_queue *mq, struct request *req)
1033 struct mmc_queue_req *mq_rq;
1034 struct mmc_card *card = mq->card;
1035 struct mmc_blk_data *md = mq->blkdata;
1036 struct mmc_blk_ioc_data **idata;
1043 mq_rq = req_to_mmc_queue_req(req);
1044 rpmb_ioctl = (mq_rq->drv_op == MMC_DRV_OP_IOCTL_RPMB);
1046 switch (mq_rq->drv_op) {
1047 case MMC_DRV_OP_IOCTL:
1048 if (card->ext_csd.cmdq_en) {
1049 ret = mmc_cmdq_disable(card);
1054 case MMC_DRV_OP_IOCTL_RPMB:
1055 idata = mq_rq->drv_op_data;
1056 for (i = 0, ret = 0; i < mq_rq->ioc_count; i++) {
1057 ret = __mmc_blk_ioctl_cmd(card, md, idata[i]);
1061 /* Always switch back to main area after RPMB access */
1063 mmc_blk_part_switch(card, 0);
1064 else if (card->reenable_cmdq && !card->ext_csd.cmdq_en)
1065 mmc_cmdq_enable(card);
1067 case MMC_DRV_OP_BOOT_WP:
1068 ret = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL, EXT_CSD_BOOT_WP,
1069 card->ext_csd.boot_ro_lock |
1070 EXT_CSD_BOOT_WP_B_PWR_WP_EN,
1071 card->ext_csd.part_time);
1073 pr_err("%s: Locking boot partition ro until next power on failed: %d\n",
1074 md->disk->disk_name, ret);
1076 card->ext_csd.boot_ro_lock |=
1077 EXT_CSD_BOOT_WP_B_PWR_WP_EN;
1079 case MMC_DRV_OP_GET_CARD_STATUS:
1080 ret = mmc_send_status(card, &status);
1084 case MMC_DRV_OP_GET_EXT_CSD:
1085 ext_csd = mq_rq->drv_op_data;
1086 ret = mmc_get_ext_csd(card, ext_csd);
1089 pr_err("%s: unknown driver specific operation\n",
1090 md->disk->disk_name);
1094 mq_rq->drv_op_result = ret;
1095 blk_mq_end_request(req, ret ? BLK_STS_IOERR : BLK_STS_OK);
1098 static void mmc_blk_issue_erase_rq(struct mmc_queue *mq, struct request *req,
1099 int type, unsigned int erase_arg)
1101 struct mmc_blk_data *md = mq->blkdata;
1102 struct mmc_card *card = md->queue.card;
1103 unsigned int from, nr;
1105 blk_status_t status = BLK_STS_OK;
1107 if (!mmc_can_erase(card)) {
1108 status = BLK_STS_NOTSUPP;
1112 from = blk_rq_pos(req);
1113 nr = blk_rq_sectors(req);
1117 if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1118 err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1119 INAND_CMD38_ARG_EXT_CSD,
1120 erase_arg == MMC_TRIM_ARG ?
1121 INAND_CMD38_ARG_TRIM :
1122 INAND_CMD38_ARG_ERASE,
1123 card->ext_csd.generic_cmd6_time);
1126 err = mmc_erase(card, from, nr, erase_arg);
1127 } while (err == -EIO && !mmc_blk_reset(md, card->host, type));
1129 status = BLK_STS_IOERR;
1131 mmc_blk_reset_success(md, type);
1133 blk_mq_end_request(req, status);
1136 static void mmc_blk_issue_trim_rq(struct mmc_queue *mq, struct request *req)
1138 mmc_blk_issue_erase_rq(mq, req, MMC_BLK_TRIM, MMC_TRIM_ARG);
1141 static void mmc_blk_issue_discard_rq(struct mmc_queue *mq, struct request *req)
1143 struct mmc_blk_data *md = mq->blkdata;
1144 struct mmc_card *card = md->queue.card;
1145 unsigned int arg = card->erase_arg;
1147 if (mmc_card_broken_sd_discard(card))
1150 mmc_blk_issue_erase_rq(mq, req, MMC_BLK_DISCARD, arg);
1153 static void mmc_blk_issue_secdiscard_rq(struct mmc_queue *mq,
1154 struct request *req)
1156 struct mmc_blk_data *md = mq->blkdata;
1157 struct mmc_card *card = md->queue.card;
1158 unsigned int from, nr, arg;
1159 int err = 0, type = MMC_BLK_SECDISCARD;
1160 blk_status_t status = BLK_STS_OK;
1162 if (!(mmc_can_secure_erase_trim(card))) {
1163 status = BLK_STS_NOTSUPP;
1167 from = blk_rq_pos(req);
1168 nr = blk_rq_sectors(req);
1170 if (mmc_can_trim(card) && !mmc_erase_group_aligned(card, from, nr))
1171 arg = MMC_SECURE_TRIM1_ARG;
1173 arg = MMC_SECURE_ERASE_ARG;
1176 if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1177 err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1178 INAND_CMD38_ARG_EXT_CSD,
1179 arg == MMC_SECURE_TRIM1_ARG ?
1180 INAND_CMD38_ARG_SECTRIM1 :
1181 INAND_CMD38_ARG_SECERASE,
1182 card->ext_csd.generic_cmd6_time);
1187 err = mmc_erase(card, from, nr, arg);
1191 status = BLK_STS_IOERR;
1195 if (arg == MMC_SECURE_TRIM1_ARG) {
1196 if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1197 err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1198 INAND_CMD38_ARG_EXT_CSD,
1199 INAND_CMD38_ARG_SECTRIM2,
1200 card->ext_csd.generic_cmd6_time);
1205 err = mmc_erase(card, from, nr, MMC_SECURE_TRIM2_ARG);
1209 status = BLK_STS_IOERR;
1215 if (err && !mmc_blk_reset(md, card->host, type))
1218 mmc_blk_reset_success(md, type);
1220 blk_mq_end_request(req, status);
1223 static void mmc_blk_issue_flush(struct mmc_queue *mq, struct request *req)
1225 struct mmc_blk_data *md = mq->blkdata;
1226 struct mmc_card *card = md->queue.card;
1229 ret = mmc_flush_cache(card->host);
1230 blk_mq_end_request(req, ret ? BLK_STS_IOERR : BLK_STS_OK);
1234 * Reformat current write as a reliable write, supporting
1235 * both legacy and the enhanced reliable write MMC cards.
1236 * In each transfer we'll handle only as much as a single
1237 * reliable write can handle, thus finish the request in
1238 * partial completions.
1240 static inline void mmc_apply_rel_rw(struct mmc_blk_request *brq,
1241 struct mmc_card *card,
1242 struct request *req)
1244 if (!(card->ext_csd.rel_param & EXT_CSD_WR_REL_PARAM_EN)) {
1245 /* Legacy mode imposes restrictions on transfers. */
1246 if (!IS_ALIGNED(blk_rq_pos(req), card->ext_csd.rel_sectors))
1247 brq->data.blocks = 1;
1249 if (brq->data.blocks > card->ext_csd.rel_sectors)
1250 brq->data.blocks = card->ext_csd.rel_sectors;
1251 else if (brq->data.blocks < card->ext_csd.rel_sectors)
1252 brq->data.blocks = 1;
1256 #define CMD_ERRORS_EXCL_OOR \
1257 (R1_ADDRESS_ERROR | /* Misaligned address */ \
1258 R1_BLOCK_LEN_ERROR | /* Transferred block length incorrect */\
1259 R1_WP_VIOLATION | /* Tried to write to protected block */ \
1260 R1_CARD_ECC_FAILED | /* Card ECC failed */ \
1261 R1_CC_ERROR | /* Card controller error */ \
1262 R1_ERROR) /* General/unknown error */
1264 #define CMD_ERRORS \
1265 (CMD_ERRORS_EXCL_OOR | \
1266 R1_OUT_OF_RANGE) /* Command argument out of range */ \
1268 static void mmc_blk_eval_resp_error(struct mmc_blk_request *brq)
1273 * Per the SD specification(physical layer version 4.10)[1],
1274 * section 4.3.3, it explicitly states that "When the last
1275 * block of user area is read using CMD18, the host should
1276 * ignore OUT_OF_RANGE error that may occur even the sequence
1277 * is correct". And JESD84-B51 for eMMC also has a similar
1278 * statement on section 6.8.3.
1280 * Multiple block read/write could be done by either predefined
1281 * method, namely CMD23, or open-ending mode. For open-ending mode,
1282 * we should ignore the OUT_OF_RANGE error as it's normal behaviour.
1284 * However the spec[1] doesn't tell us whether we should also
1285 * ignore that for predefined method. But per the spec[1], section
1286 * 4.15 Set Block Count Command, it says"If illegal block count
1287 * is set, out of range error will be indicated during read/write
1288 * operation (For example, data transfer is stopped at user area
1289 * boundary)." In another word, we could expect a out of range error
1290 * in the response for the following CMD18/25. And if argument of
1291 * CMD23 + the argument of CMD18/25 exceed the max number of blocks,
1292 * we could also expect to get a -ETIMEDOUT or any error number from
1293 * the host drivers due to missing data response(for write)/data(for
1294 * read), as the cards will stop the data transfer by itself per the
1295 * spec. So we only need to check R1_OUT_OF_RANGE for open-ending mode.
1298 if (!brq->stop.error) {
1299 bool oor_with_open_end;
1300 /* If there is no error yet, check R1 response */
1302 val = brq->stop.resp[0] & CMD_ERRORS;
1303 oor_with_open_end = val & R1_OUT_OF_RANGE && !brq->mrq.sbc;
1305 if (val && !oor_with_open_end)
1306 brq->stop.error = -EIO;
1310 static void mmc_blk_data_prep(struct mmc_queue *mq, struct mmc_queue_req *mqrq,
1311 int recovery_mode, bool *do_rel_wr_p,
1312 bool *do_data_tag_p)
1314 struct mmc_blk_data *md = mq->blkdata;
1315 struct mmc_card *card = md->queue.card;
1316 struct mmc_blk_request *brq = &mqrq->brq;
1317 struct request *req = mmc_queue_req_to_req(mqrq);
1318 bool do_rel_wr, do_data_tag;
1321 * Reliable writes are used to implement Forced Unit Access and
1322 * are supported only on MMCs.
1324 do_rel_wr = (req->cmd_flags & REQ_FUA) &&
1325 rq_data_dir(req) == WRITE &&
1326 (md->flags & MMC_BLK_REL_WR);
1328 memset(brq, 0, sizeof(struct mmc_blk_request));
1330 mmc_crypto_prepare_req(mqrq);
1332 brq->mrq.data = &brq->data;
1333 brq->mrq.tag = req->tag;
1335 brq->stop.opcode = MMC_STOP_TRANSMISSION;
1338 if (rq_data_dir(req) == READ) {
1339 brq->data.flags = MMC_DATA_READ;
1340 brq->stop.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC;
1342 brq->data.flags = MMC_DATA_WRITE;
1343 brq->stop.flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC;
1346 brq->data.blksz = 512;
1347 brq->data.blocks = blk_rq_sectors(req);
1348 brq->data.blk_addr = blk_rq_pos(req);
1351 * The command queue supports 2 priorities: "high" (1) and "simple" (0).
1352 * The eMMC will give "high" priority tasks priority over "simple"
1353 * priority tasks. Here we always set "simple" priority by not setting
1358 * The block layer doesn't support all sector count
1359 * restrictions, so we need to be prepared for too big
1362 if (brq->data.blocks > card->host->max_blk_count)
1363 brq->data.blocks = card->host->max_blk_count;
1365 if (brq->data.blocks > 1) {
1367 * Some SD cards in SPI mode return a CRC error or even lock up
1368 * completely when trying to read the last block using a
1369 * multiblock read command.
1371 if (mmc_host_is_spi(card->host) && (rq_data_dir(req) == READ) &&
1372 (blk_rq_pos(req) + blk_rq_sectors(req) ==
1373 get_capacity(md->disk)))
1377 * After a read error, we redo the request one (native) sector
1378 * at a time in order to accurately determine which
1379 * sectors can be read successfully.
1382 brq->data.blocks = queue_physical_block_size(mq->queue) >> 9;
1385 * Some controllers have HW issues while operating
1386 * in multiple I/O mode
1388 if (card->host->ops->multi_io_quirk)
1389 brq->data.blocks = card->host->ops->multi_io_quirk(card,
1390 (rq_data_dir(req) == READ) ?
1391 MMC_DATA_READ : MMC_DATA_WRITE,
1396 mmc_apply_rel_rw(brq, card, req);
1397 brq->data.flags |= MMC_DATA_REL_WR;
1401 * Data tag is used only during writing meta data to speed
1402 * up write and any subsequent read of this meta data
1404 do_data_tag = card->ext_csd.data_tag_unit_size &&
1405 (req->cmd_flags & REQ_META) &&
1406 (rq_data_dir(req) == WRITE) &&
1407 ((brq->data.blocks * brq->data.blksz) >=
1408 card->ext_csd.data_tag_unit_size);
1411 brq->data.flags |= MMC_DATA_DAT_TAG;
1413 mmc_set_data_timeout(&brq->data, card);
1415 brq->data.sg = mqrq->sg;
1416 brq->data.sg_len = mmc_queue_map_sg(mq, mqrq);
1419 * Adjust the sg list so it is the same size as the
1422 if (brq->data.blocks != blk_rq_sectors(req)) {
1423 int i, data_size = brq->data.blocks << 9;
1424 struct scatterlist *sg;
1426 for_each_sg(brq->data.sg, sg, brq->data.sg_len, i) {
1427 data_size -= sg->length;
1428 if (data_size <= 0) {
1429 sg->length += data_size;
1434 brq->data.sg_len = i;
1438 *do_rel_wr_p = do_rel_wr;
1441 *do_data_tag_p = do_data_tag;
1444 #define MMC_CQE_RETRIES 2
1446 static void mmc_blk_cqe_complete_rq(struct mmc_queue *mq, struct request *req)
1448 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1449 struct mmc_request *mrq = &mqrq->brq.mrq;
1450 struct request_queue *q = req->q;
1451 struct mmc_host *host = mq->card->host;
1452 enum mmc_issue_type issue_type = mmc_issue_type(mq, req);
1453 unsigned long flags;
1457 mmc_cqe_post_req(host, mrq);
1459 if (mrq->cmd && mrq->cmd->error)
1460 err = mrq->cmd->error;
1461 else if (mrq->data && mrq->data->error)
1462 err = mrq->data->error;
1467 if (mqrq->retries++ < MMC_CQE_RETRIES)
1468 blk_mq_requeue_request(req, true);
1470 blk_mq_end_request(req, BLK_STS_IOERR);
1471 } else if (mrq->data) {
1472 if (blk_update_request(req, BLK_STS_OK, mrq->data->bytes_xfered))
1473 blk_mq_requeue_request(req, true);
1475 __blk_mq_end_request(req, BLK_STS_OK);
1477 blk_mq_end_request(req, BLK_STS_OK);
1480 spin_lock_irqsave(&mq->lock, flags);
1482 mq->in_flight[issue_type] -= 1;
1484 put_card = (mmc_tot_in_flight(mq) == 0);
1486 mmc_cqe_check_busy(mq);
1488 spin_unlock_irqrestore(&mq->lock, flags);
1491 blk_mq_run_hw_queues(q, true);
1494 mmc_put_card(mq->card, &mq->ctx);
1497 void mmc_blk_cqe_recovery(struct mmc_queue *mq)
1499 struct mmc_card *card = mq->card;
1500 struct mmc_host *host = card->host;
1503 pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
1505 err = mmc_cqe_recovery(host);
1507 mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
1508 mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
1510 pr_debug("%s: CQE recovery done\n", mmc_hostname(host));
1513 static void mmc_blk_cqe_req_done(struct mmc_request *mrq)
1515 struct mmc_queue_req *mqrq = container_of(mrq, struct mmc_queue_req,
1517 struct request *req = mmc_queue_req_to_req(mqrq);
1518 struct request_queue *q = req->q;
1519 struct mmc_queue *mq = q->queuedata;
1522 * Block layer timeouts race with completions which means the normal
1523 * completion path cannot be used during recovery.
1525 if (mq->in_recovery)
1526 mmc_blk_cqe_complete_rq(mq, req);
1527 else if (likely(!blk_should_fake_timeout(req->q)))
1528 blk_mq_complete_request(req);
1531 static int mmc_blk_cqe_start_req(struct mmc_host *host, struct mmc_request *mrq)
1533 mrq->done = mmc_blk_cqe_req_done;
1534 mrq->recovery_notifier = mmc_cqe_recovery_notifier;
1536 return mmc_cqe_start_req(host, mrq);
1539 static struct mmc_request *mmc_blk_cqe_prep_dcmd(struct mmc_queue_req *mqrq,
1540 struct request *req)
1542 struct mmc_blk_request *brq = &mqrq->brq;
1544 memset(brq, 0, sizeof(*brq));
1546 brq->mrq.cmd = &brq->cmd;
1547 brq->mrq.tag = req->tag;
1552 static int mmc_blk_cqe_issue_flush(struct mmc_queue *mq, struct request *req)
1554 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1555 struct mmc_request *mrq = mmc_blk_cqe_prep_dcmd(mqrq, req);
1557 mrq->cmd->opcode = MMC_SWITCH;
1558 mrq->cmd->arg = (MMC_SWITCH_MODE_WRITE_BYTE << 24) |
1559 (EXT_CSD_FLUSH_CACHE << 16) |
1561 EXT_CSD_CMD_SET_NORMAL;
1562 mrq->cmd->flags = MMC_CMD_AC | MMC_RSP_R1B;
1564 return mmc_blk_cqe_start_req(mq->card->host, mrq);
1567 static int mmc_blk_hsq_issue_rw_rq(struct mmc_queue *mq, struct request *req)
1569 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1570 struct mmc_host *host = mq->card->host;
1573 mmc_blk_rw_rq_prep(mqrq, mq->card, 0, mq);
1574 mqrq->brq.mrq.done = mmc_blk_hsq_req_done;
1575 mmc_pre_req(host, &mqrq->brq.mrq);
1577 err = mmc_cqe_start_req(host, &mqrq->brq.mrq);
1579 mmc_post_req(host, &mqrq->brq.mrq, err);
1584 static int mmc_blk_cqe_issue_rw_rq(struct mmc_queue *mq, struct request *req)
1586 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1587 struct mmc_host *host = mq->card->host;
1589 if (host->hsq_enabled)
1590 return mmc_blk_hsq_issue_rw_rq(mq, req);
1592 mmc_blk_data_prep(mq, mqrq, 0, NULL, NULL);
1594 return mmc_blk_cqe_start_req(mq->card->host, &mqrq->brq.mrq);
1597 static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
1598 struct mmc_card *card,
1600 struct mmc_queue *mq)
1602 u32 readcmd, writecmd;
1603 struct mmc_blk_request *brq = &mqrq->brq;
1604 struct request *req = mmc_queue_req_to_req(mqrq);
1605 struct mmc_blk_data *md = mq->blkdata;
1606 bool do_rel_wr, do_data_tag;
1608 mmc_blk_data_prep(mq, mqrq, recovery_mode, &do_rel_wr, &do_data_tag);
1610 brq->mrq.cmd = &brq->cmd;
1612 brq->cmd.arg = blk_rq_pos(req);
1613 if (!mmc_card_blockaddr(card))
1615 brq->cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC;
1617 if (brq->data.blocks > 1 || do_rel_wr) {
1618 /* SPI multiblock writes terminate using a special
1619 * token, not a STOP_TRANSMISSION request.
1621 if (!mmc_host_is_spi(card->host) ||
1622 rq_data_dir(req) == READ)
1623 brq->mrq.stop = &brq->stop;
1624 readcmd = MMC_READ_MULTIPLE_BLOCK;
1625 writecmd = MMC_WRITE_MULTIPLE_BLOCK;
1627 brq->mrq.stop = NULL;
1628 readcmd = MMC_READ_SINGLE_BLOCK;
1629 writecmd = MMC_WRITE_BLOCK;
1631 brq->cmd.opcode = rq_data_dir(req) == READ ? readcmd : writecmd;
1634 * Pre-defined multi-block transfers are preferable to
1635 * open ended-ones (and necessary for reliable writes).
1636 * However, it is not sufficient to just send CMD23,
1637 * and avoid the final CMD12, as on an error condition
1638 * CMD12 (stop) needs to be sent anyway. This, coupled
1639 * with Auto-CMD23 enhancements provided by some
1640 * hosts, means that the complexity of dealing
1641 * with this is best left to the host. If CMD23 is
1642 * supported by card and host, we'll fill sbc in and let
1643 * the host deal with handling it correctly. This means
1644 * that for hosts that don't expose MMC_CAP_CMD23, no
1645 * change of behavior will be observed.
1647 * N.B: Some MMC cards experience perf degradation.
1648 * We'll avoid using CMD23-bounded multiblock writes for
1649 * these, while retaining features like reliable writes.
1651 if ((md->flags & MMC_BLK_CMD23) && mmc_op_multi(brq->cmd.opcode) &&
1652 (do_rel_wr || !(card->quirks & MMC_QUIRK_BLK_NO_CMD23) ||
1654 brq->sbc.opcode = MMC_SET_BLOCK_COUNT;
1655 brq->sbc.arg = brq->data.blocks |
1656 (do_rel_wr ? (1 << 31) : 0) |
1657 (do_data_tag ? (1 << 29) : 0);
1658 brq->sbc.flags = MMC_RSP_R1 | MMC_CMD_AC;
1659 brq->mrq.sbc = &brq->sbc;
1663 #define MMC_MAX_RETRIES 5
1664 #define MMC_DATA_RETRIES 2
1665 #define MMC_NO_RETRIES (MMC_MAX_RETRIES + 1)
1667 static int mmc_blk_send_stop(struct mmc_card *card, unsigned int timeout)
1669 struct mmc_command cmd = {
1670 .opcode = MMC_STOP_TRANSMISSION,
1671 .flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC,
1672 /* Some hosts wait for busy anyway, so provide a busy timeout */
1673 .busy_timeout = timeout,
1676 return mmc_wait_for_cmd(card->host, &cmd, 5);
1679 static int mmc_blk_fix_state(struct mmc_card *card, struct request *req)
1681 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1682 struct mmc_blk_request *brq = &mqrq->brq;
1683 unsigned int timeout = mmc_blk_data_timeout_ms(card->host, &brq->data);
1686 mmc_retune_hold_now(card->host);
1688 mmc_blk_send_stop(card, timeout);
1690 err = mmc_poll_for_busy(card, timeout, false, MMC_BUSY_IO);
1692 mmc_retune_release(card->host);
1697 #define MMC_READ_SINGLE_RETRIES 2
1699 /* Single (native) sector read during recovery */
1700 static void mmc_blk_read_single(struct mmc_queue *mq, struct request *req)
1702 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1703 struct mmc_request *mrq = &mqrq->brq.mrq;
1704 struct mmc_card *card = mq->card;
1705 struct mmc_host *host = card->host;
1706 blk_status_t error = BLK_STS_OK;
1707 size_t bytes_per_read = queue_physical_block_size(mq->queue);
1714 while (retries++ <= MMC_READ_SINGLE_RETRIES) {
1715 mmc_blk_rw_rq_prep(mqrq, card, 1, mq);
1717 mmc_wait_for_req(host, mrq);
1719 err = mmc_send_status(card, &status);
1723 if (!mmc_host_is_spi(host) &&
1724 !mmc_ready_for_data(status)) {
1725 err = mmc_blk_fix_state(card, req);
1730 if (!mrq->cmd->error)
1734 if (mrq->cmd->error ||
1736 (!mmc_host_is_spi(host) &&
1737 (mrq->cmd->resp[0] & CMD_ERRORS || status & CMD_ERRORS)))
1738 error = BLK_STS_IOERR;
1742 } while (blk_update_request(req, error, bytes_per_read));
1747 mrq->data->bytes_xfered = 0;
1748 blk_update_request(req, BLK_STS_IOERR, bytes_per_read);
1749 /* Let it try the remaining request again */
1750 if (mqrq->retries > MMC_MAX_RETRIES - 1)
1751 mqrq->retries = MMC_MAX_RETRIES - 1;
1754 static inline bool mmc_blk_oor_valid(struct mmc_blk_request *brq)
1756 return !!brq->mrq.sbc;
1759 static inline u32 mmc_blk_stop_err_bits(struct mmc_blk_request *brq)
1761 return mmc_blk_oor_valid(brq) ? CMD_ERRORS : CMD_ERRORS_EXCL_OOR;
1765 * Check for errors the host controller driver might not have seen such as
1766 * response mode errors or invalid card state.
1768 static bool mmc_blk_status_error(struct request *req, u32 status)
1770 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1771 struct mmc_blk_request *brq = &mqrq->brq;
1772 struct mmc_queue *mq = req->q->queuedata;
1775 if (mmc_host_is_spi(mq->card->host))
1778 stop_err_bits = mmc_blk_stop_err_bits(brq);
1780 return brq->cmd.resp[0] & CMD_ERRORS ||
1781 brq->stop.resp[0] & stop_err_bits ||
1782 status & stop_err_bits ||
1783 (rq_data_dir(req) == WRITE && !mmc_ready_for_data(status));
1786 static inline bool mmc_blk_cmd_started(struct mmc_blk_request *brq)
1788 return !brq->sbc.error && !brq->cmd.error &&
1789 !(brq->cmd.resp[0] & CMD_ERRORS);
1793 * Requests are completed by mmc_blk_mq_complete_rq() which sets simple
1795 * 1. A request that has transferred at least some data is considered
1796 * successful and will be requeued if there is remaining data to
1798 * 2. Otherwise the number of retries is incremented and the request
1799 * will be requeued if there are remaining retries.
1800 * 3. Otherwise the request will be errored out.
1801 * That means mmc_blk_mq_complete_rq() is controlled by bytes_xfered and
1802 * mqrq->retries. So there are only 4 possible actions here:
1803 * 1. do not accept the bytes_xfered value i.e. set it to zero
1804 * 2. change mqrq->retries to determine the number of retries
1805 * 3. try to reset the card
1806 * 4. read one sector at a time
1808 static void mmc_blk_mq_rw_recovery(struct mmc_queue *mq, struct request *req)
1810 int type = rq_data_dir(req) == READ ? MMC_BLK_READ : MMC_BLK_WRITE;
1811 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1812 struct mmc_blk_request *brq = &mqrq->brq;
1813 struct mmc_blk_data *md = mq->blkdata;
1814 struct mmc_card *card = mq->card;
1820 * Some errors the host driver might not have seen. Set the number of
1821 * bytes transferred to zero in that case.
1823 err = __mmc_send_status(card, &status, 0);
1824 if (err || mmc_blk_status_error(req, status))
1825 brq->data.bytes_xfered = 0;
1827 mmc_retune_release(card->host);
1830 * Try again to get the status. This also provides an opportunity for
1834 err = __mmc_send_status(card, &status, 0);
1837 * Nothing more to do after the number of bytes transferred has been
1838 * updated and there is no card.
1840 if (err && mmc_detect_card_removed(card->host))
1843 /* Try to get back to "tran" state */
1844 if (!mmc_host_is_spi(mq->card->host) &&
1845 (err || !mmc_ready_for_data(status)))
1846 err = mmc_blk_fix_state(mq->card, req);
1849 * Special case for SD cards where the card might record the number of
1852 if (!err && mmc_blk_cmd_started(brq) && mmc_card_sd(card) &&
1853 rq_data_dir(req) == WRITE) {
1854 if (mmc_sd_num_wr_blocks(card, &blocks))
1855 brq->data.bytes_xfered = 0;
1857 brq->data.bytes_xfered = blocks << 9;
1860 /* Reset if the card is in a bad state */
1861 if (!mmc_host_is_spi(mq->card->host) &&
1862 err && mmc_blk_reset(md, card->host, type)) {
1863 pr_err("%s: recovery failed!\n", req->q->disk->disk_name);
1864 mqrq->retries = MMC_NO_RETRIES;
1869 * If anything was done, just return and if there is anything remaining
1870 * on the request it will get requeued.
1872 if (brq->data.bytes_xfered)
1875 /* Reset before last retry */
1876 if (mqrq->retries + 1 == MMC_MAX_RETRIES &&
1877 mmc_blk_reset(md, card->host, type))
1880 /* Command errors fail fast, so use all MMC_MAX_RETRIES */
1881 if (brq->sbc.error || brq->cmd.error)
1884 /* Reduce the remaining retries for data errors */
1885 if (mqrq->retries < MMC_MAX_RETRIES - MMC_DATA_RETRIES) {
1886 mqrq->retries = MMC_MAX_RETRIES - MMC_DATA_RETRIES;
1890 if (rq_data_dir(req) == READ && brq->data.blocks >
1891 queue_physical_block_size(mq->queue) >> 9) {
1892 /* Read one (native) sector at a time */
1893 mmc_blk_read_single(mq, req);
1898 static inline bool mmc_blk_rq_error(struct mmc_blk_request *brq)
1900 mmc_blk_eval_resp_error(brq);
1902 return brq->sbc.error || brq->cmd.error || brq->stop.error ||
1903 brq->data.error || brq->cmd.resp[0] & CMD_ERRORS;
1906 static int mmc_spi_err_check(struct mmc_card *card)
1912 * SPI does not have a TRAN state we have to wait on, instead the
1913 * card is ready again when it no longer holds the line LOW.
1914 * We still have to ensure two things here before we know the write
1916 * 1. The card has not disconnected during busy and we actually read our
1917 * own pull-up, thinking it was still connected, so ensure it
1919 * 2. Check for any error bits, in particular R1_SPI_IDLE to catch a
1920 * just reconnected card after being disconnected during busy.
1922 err = __mmc_send_status(card, &status, 0);
1925 /* All R1 and R2 bits of SPI are errors in our case */
1931 static int mmc_blk_busy_cb(void *cb_data, bool *busy)
1933 struct mmc_blk_busy_data *data = cb_data;
1937 err = mmc_send_status(data->card, &status);
1941 /* Accumulate response error bits. */
1942 data->status |= status;
1944 *busy = !mmc_ready_for_data(status);
1948 static int mmc_blk_card_busy(struct mmc_card *card, struct request *req)
1950 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1951 struct mmc_blk_busy_data cb_data;
1954 if (rq_data_dir(req) == READ)
1957 if (mmc_host_is_spi(card->host)) {
1958 err = mmc_spi_err_check(card);
1960 mqrq->brq.data.bytes_xfered = 0;
1964 cb_data.card = card;
1966 err = __mmc_poll_for_busy(card->host, 0, MMC_BLK_TIMEOUT_MS,
1967 &mmc_blk_busy_cb, &cb_data);
1970 * Do not assume data transferred correctly if there are any error bits
1973 if (cb_data.status & mmc_blk_stop_err_bits(&mqrq->brq)) {
1974 mqrq->brq.data.bytes_xfered = 0;
1975 err = err ? err : -EIO;
1978 /* Copy the exception bit so it will be seen later on */
1979 if (mmc_card_mmc(card) && cb_data.status & R1_EXCEPTION_EVENT)
1980 mqrq->brq.cmd.resp[0] |= R1_EXCEPTION_EVENT;
1985 static inline void mmc_blk_rw_reset_success(struct mmc_queue *mq,
1986 struct request *req)
1988 int type = rq_data_dir(req) == READ ? MMC_BLK_READ : MMC_BLK_WRITE;
1990 mmc_blk_reset_success(mq->blkdata, type);
1993 static void mmc_blk_mq_complete_rq(struct mmc_queue *mq, struct request *req)
1995 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1996 unsigned int nr_bytes = mqrq->brq.data.bytes_xfered;
1999 if (blk_update_request(req, BLK_STS_OK, nr_bytes))
2000 blk_mq_requeue_request(req, true);
2002 __blk_mq_end_request(req, BLK_STS_OK);
2003 } else if (!blk_rq_bytes(req)) {
2004 __blk_mq_end_request(req, BLK_STS_IOERR);
2005 } else if (mqrq->retries++ < MMC_MAX_RETRIES) {
2006 blk_mq_requeue_request(req, true);
2008 if (mmc_card_removed(mq->card))
2009 req->rq_flags |= RQF_QUIET;
2010 blk_mq_end_request(req, BLK_STS_IOERR);
2014 static bool mmc_blk_urgent_bkops_needed(struct mmc_queue *mq,
2015 struct mmc_queue_req *mqrq)
2017 return mmc_card_mmc(mq->card) && !mmc_host_is_spi(mq->card->host) &&
2018 (mqrq->brq.cmd.resp[0] & R1_EXCEPTION_EVENT ||
2019 mqrq->brq.stop.resp[0] & R1_EXCEPTION_EVENT);
2022 static void mmc_blk_urgent_bkops(struct mmc_queue *mq,
2023 struct mmc_queue_req *mqrq)
2025 if (mmc_blk_urgent_bkops_needed(mq, mqrq))
2026 mmc_run_bkops(mq->card);
2029 static void mmc_blk_hsq_req_done(struct mmc_request *mrq)
2031 struct mmc_queue_req *mqrq =
2032 container_of(mrq, struct mmc_queue_req, brq.mrq);
2033 struct request *req = mmc_queue_req_to_req(mqrq);
2034 struct request_queue *q = req->q;
2035 struct mmc_queue *mq = q->queuedata;
2036 struct mmc_host *host = mq->card->host;
2037 unsigned long flags;
2039 if (mmc_blk_rq_error(&mqrq->brq) ||
2040 mmc_blk_urgent_bkops_needed(mq, mqrq)) {
2041 spin_lock_irqsave(&mq->lock, flags);
2042 mq->recovery_needed = true;
2043 mq->recovery_req = req;
2044 spin_unlock_irqrestore(&mq->lock, flags);
2046 host->cqe_ops->cqe_recovery_start(host);
2048 schedule_work(&mq->recovery_work);
2052 mmc_blk_rw_reset_success(mq, req);
2055 * Block layer timeouts race with completions which means the normal
2056 * completion path cannot be used during recovery.
2058 if (mq->in_recovery)
2059 mmc_blk_cqe_complete_rq(mq, req);
2060 else if (likely(!blk_should_fake_timeout(req->q)))
2061 blk_mq_complete_request(req);
2064 void mmc_blk_mq_complete(struct request *req)
2066 struct mmc_queue *mq = req->q->queuedata;
2067 struct mmc_host *host = mq->card->host;
2069 if (host->cqe_enabled)
2070 mmc_blk_cqe_complete_rq(mq, req);
2071 else if (likely(!blk_should_fake_timeout(req->q)))
2072 mmc_blk_mq_complete_rq(mq, req);
2075 static void mmc_blk_mq_poll_completion(struct mmc_queue *mq,
2076 struct request *req)
2078 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2079 struct mmc_host *host = mq->card->host;
2081 if (mmc_blk_rq_error(&mqrq->brq) ||
2082 mmc_blk_card_busy(mq->card, req)) {
2083 mmc_blk_mq_rw_recovery(mq, req);
2085 mmc_blk_rw_reset_success(mq, req);
2086 mmc_retune_release(host);
2089 mmc_blk_urgent_bkops(mq, mqrq);
2092 static void mmc_blk_mq_dec_in_flight(struct mmc_queue *mq, enum mmc_issue_type issue_type)
2094 unsigned long flags;
2097 spin_lock_irqsave(&mq->lock, flags);
2099 mq->in_flight[issue_type] -= 1;
2101 put_card = (mmc_tot_in_flight(mq) == 0);
2103 spin_unlock_irqrestore(&mq->lock, flags);
2106 mmc_put_card(mq->card, &mq->ctx);
2109 static void mmc_blk_mq_post_req(struct mmc_queue *mq, struct request *req,
2112 enum mmc_issue_type issue_type = mmc_issue_type(mq, req);
2113 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2114 struct mmc_request *mrq = &mqrq->brq.mrq;
2115 struct mmc_host *host = mq->card->host;
2117 mmc_post_req(host, mrq, 0);
2120 * Block layer timeouts race with completions which means the normal
2121 * completion path cannot be used during recovery.
2123 if (mq->in_recovery) {
2124 mmc_blk_mq_complete_rq(mq, req);
2125 } else if (likely(!blk_should_fake_timeout(req->q))) {
2127 blk_mq_complete_request_direct(req, mmc_blk_mq_complete);
2129 blk_mq_complete_request(req);
2132 mmc_blk_mq_dec_in_flight(mq, issue_type);
2135 void mmc_blk_mq_recovery(struct mmc_queue *mq)
2137 struct request *req = mq->recovery_req;
2138 struct mmc_host *host = mq->card->host;
2139 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2141 mq->recovery_req = NULL;
2142 mq->rw_wait = false;
2144 if (mmc_blk_rq_error(&mqrq->brq)) {
2145 mmc_retune_hold_now(host);
2146 mmc_blk_mq_rw_recovery(mq, req);
2149 mmc_blk_urgent_bkops(mq, mqrq);
2151 mmc_blk_mq_post_req(mq, req, true);
2154 static void mmc_blk_mq_complete_prev_req(struct mmc_queue *mq,
2155 struct request **prev_req)
2157 if (mmc_host_done_complete(mq->card->host))
2160 mutex_lock(&mq->complete_lock);
2162 if (!mq->complete_req)
2165 mmc_blk_mq_poll_completion(mq, mq->complete_req);
2168 *prev_req = mq->complete_req;
2170 mmc_blk_mq_post_req(mq, mq->complete_req, true);
2172 mq->complete_req = NULL;
2175 mutex_unlock(&mq->complete_lock);
2178 void mmc_blk_mq_complete_work(struct work_struct *work)
2180 struct mmc_queue *mq = container_of(work, struct mmc_queue,
2183 mmc_blk_mq_complete_prev_req(mq, NULL);
2186 static void mmc_blk_mq_req_done(struct mmc_request *mrq)
2188 struct mmc_queue_req *mqrq = container_of(mrq, struct mmc_queue_req,
2190 struct request *req = mmc_queue_req_to_req(mqrq);
2191 struct request_queue *q = req->q;
2192 struct mmc_queue *mq = q->queuedata;
2193 struct mmc_host *host = mq->card->host;
2194 unsigned long flags;
2196 if (!mmc_host_done_complete(host)) {
2200 * We cannot complete the request in this context, so record
2201 * that there is a request to complete, and that a following
2202 * request does not need to wait (although it does need to
2203 * complete complete_req first).
2205 spin_lock_irqsave(&mq->lock, flags);
2206 mq->complete_req = req;
2207 mq->rw_wait = false;
2208 waiting = mq->waiting;
2209 spin_unlock_irqrestore(&mq->lock, flags);
2212 * If 'waiting' then the waiting task will complete this
2213 * request, otherwise queue a work to do it. Note that
2214 * complete_work may still race with the dispatch of a following
2220 queue_work(mq->card->complete_wq, &mq->complete_work);
2225 /* Take the recovery path for errors or urgent background operations */
2226 if (mmc_blk_rq_error(&mqrq->brq) ||
2227 mmc_blk_urgent_bkops_needed(mq, mqrq)) {
2228 spin_lock_irqsave(&mq->lock, flags);
2229 mq->recovery_needed = true;
2230 mq->recovery_req = req;
2231 spin_unlock_irqrestore(&mq->lock, flags);
2233 schedule_work(&mq->recovery_work);
2237 mmc_blk_rw_reset_success(mq, req);
2239 mq->rw_wait = false;
2242 /* context unknown */
2243 mmc_blk_mq_post_req(mq, req, false);
2246 static bool mmc_blk_rw_wait_cond(struct mmc_queue *mq, int *err)
2248 unsigned long flags;
2252 * Wait while there is another request in progress, but not if recovery
2253 * is needed. Also indicate whether there is a request waiting to start.
2255 spin_lock_irqsave(&mq->lock, flags);
2256 if (mq->recovery_needed) {
2260 done = !mq->rw_wait;
2262 mq->waiting = !done;
2263 spin_unlock_irqrestore(&mq->lock, flags);
2268 static int mmc_blk_rw_wait(struct mmc_queue *mq, struct request **prev_req)
2272 wait_event(mq->wait, mmc_blk_rw_wait_cond(mq, &err));
2274 /* Always complete the previous request if there is one */
2275 mmc_blk_mq_complete_prev_req(mq, prev_req);
2280 static int mmc_blk_mq_issue_rw_rq(struct mmc_queue *mq,
2281 struct request *req)
2283 struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2284 struct mmc_host *host = mq->card->host;
2285 struct request *prev_req = NULL;
2288 mmc_blk_rw_rq_prep(mqrq, mq->card, 0, mq);
2290 mqrq->brq.mrq.done = mmc_blk_mq_req_done;
2292 mmc_pre_req(host, &mqrq->brq.mrq);
2294 err = mmc_blk_rw_wait(mq, &prev_req);
2300 err = mmc_start_request(host, &mqrq->brq.mrq);
2303 mmc_blk_mq_post_req(mq, prev_req, true);
2306 mq->rw_wait = false;
2308 /* Release re-tuning here where there is no synchronization required */
2309 if (err || mmc_host_done_complete(host))
2310 mmc_retune_release(host);
2314 mmc_post_req(host, &mqrq->brq.mrq, err);
2319 static int mmc_blk_wait_for_idle(struct mmc_queue *mq, struct mmc_host *host)
2321 if (host->cqe_enabled)
2322 return host->cqe_ops->cqe_wait_for_idle(host);
2324 return mmc_blk_rw_wait(mq, NULL);
2327 enum mmc_issued mmc_blk_mq_issue_rq(struct mmc_queue *mq, struct request *req)
2329 struct mmc_blk_data *md = mq->blkdata;
2330 struct mmc_card *card = md->queue.card;
2331 struct mmc_host *host = card->host;
2334 ret = mmc_blk_part_switch(card, md->part_type);
2336 return MMC_REQ_FAILED_TO_START;
2338 switch (mmc_issue_type(mq, req)) {
2339 case MMC_ISSUE_SYNC:
2340 ret = mmc_blk_wait_for_idle(mq, host);
2342 return MMC_REQ_BUSY;
2343 switch (req_op(req)) {
2345 case REQ_OP_DRV_OUT:
2346 mmc_blk_issue_drv_op(mq, req);
2348 case REQ_OP_DISCARD:
2349 mmc_blk_issue_discard_rq(mq, req);
2351 case REQ_OP_SECURE_ERASE:
2352 mmc_blk_issue_secdiscard_rq(mq, req);
2354 case REQ_OP_WRITE_ZEROES:
2355 mmc_blk_issue_trim_rq(mq, req);
2358 mmc_blk_issue_flush(mq, req);
2362 return MMC_REQ_FAILED_TO_START;
2364 return MMC_REQ_FINISHED;
2365 case MMC_ISSUE_DCMD:
2366 case MMC_ISSUE_ASYNC:
2367 switch (req_op(req)) {
2369 if (!mmc_cache_enabled(host)) {
2370 blk_mq_end_request(req, BLK_STS_OK);
2371 return MMC_REQ_FINISHED;
2373 ret = mmc_blk_cqe_issue_flush(mq, req);
2377 if (host->cqe_enabled)
2378 ret = mmc_blk_cqe_issue_rw_rq(mq, req);
2380 ret = mmc_blk_mq_issue_rw_rq(mq, req);
2387 return MMC_REQ_STARTED;
2388 return ret == -EBUSY ? MMC_REQ_BUSY : MMC_REQ_FAILED_TO_START;
2391 return MMC_REQ_FAILED_TO_START;
2395 static inline int mmc_blk_readonly(struct mmc_card *card)
2397 return mmc_card_readonly(card) ||
2398 !(card->csd.cmdclass & CCC_BLOCK_WRITE);
2401 static struct mmc_blk_data *mmc_blk_alloc_req(struct mmc_card *card,
2402 struct device *parent,
2405 const char *subname,
2407 unsigned int part_type)
2409 struct mmc_blk_data *md;
2412 bool cache_enabled = false;
2413 bool fua_enabled = false;
2415 devidx = ida_simple_get(&mmc_blk_ida, 0, max_devices, GFP_KERNEL);
2418 * We get -ENOSPC because there are no more any available
2419 * devidx. The reason may be that, either userspace haven't yet
2420 * unmounted the partitions, which postpones mmc_blk_release()
2421 * from being called, or the device has more partitions than
2424 if (devidx == -ENOSPC)
2425 dev_err(mmc_dev(card->host),
2426 "no more device IDs available\n");
2428 return ERR_PTR(devidx);
2431 md = kzalloc(sizeof(struct mmc_blk_data), GFP_KERNEL);
2437 md->area_type = area_type;
2440 * Set the read-only status based on the supported commands
2441 * and the write protect switch.
2443 md->read_only = mmc_blk_readonly(card);
2445 md->disk = mmc_init_queue(&md->queue, card);
2446 if (IS_ERR(md->disk)) {
2447 ret = PTR_ERR(md->disk);
2451 INIT_LIST_HEAD(&md->part);
2452 INIT_LIST_HEAD(&md->rpmbs);
2453 kref_init(&md->kref);
2455 md->queue.blkdata = md;
2456 md->part_type = part_type;
2458 md->disk->major = MMC_BLOCK_MAJOR;
2459 md->disk->minors = perdev_minors;
2460 md->disk->first_minor = devidx * perdev_minors;
2461 md->disk->fops = &mmc_bdops;
2462 md->disk->private_data = md;
2463 md->parent = parent;
2464 set_disk_ro(md->disk, md->read_only || default_ro);
2465 if (area_type & (MMC_BLK_DATA_AREA_RPMB | MMC_BLK_DATA_AREA_BOOT))
2466 md->disk->flags |= GENHD_FL_NO_PART;
2469 * As discussed on lkml, GENHD_FL_REMOVABLE should:
2471 * - be set for removable media with permanent block devices
2472 * - be unset for removable block devices with permanent media
2474 * Since MMC block devices clearly fall under the second
2475 * case, we do not set GENHD_FL_REMOVABLE. Userspace
2476 * should use the block device creation/destruction hotplug
2477 * messages to tell when the card is present.
2480 snprintf(md->disk->disk_name, sizeof(md->disk->disk_name),
2481 "mmcblk%u%s", card->host->index, subname ? subname : "");
2483 set_capacity(md->disk, size);
2485 if (mmc_host_cmd23(card->host)) {
2486 if ((mmc_card_mmc(card) &&
2487 card->csd.mmca_vsn >= CSD_SPEC_VER_3) ||
2488 (mmc_card_sd(card) &&
2489 card->scr.cmds & SD_SCR_CMD23_SUPPORT))
2490 md->flags |= MMC_BLK_CMD23;
2493 if (md->flags & MMC_BLK_CMD23 &&
2494 ((card->ext_csd.rel_param & EXT_CSD_WR_REL_PARAM_EN) ||
2495 card->ext_csd.rel_sectors)) {
2496 md->flags |= MMC_BLK_REL_WR;
2498 cache_enabled = true;
2500 if (mmc_cache_enabled(card->host))
2501 cache_enabled = true;
2503 blk_queue_write_cache(md->queue.queue, cache_enabled, fua_enabled);
2505 string_get_size((u64)size, 512, STRING_UNITS_2,
2506 cap_str, sizeof(cap_str));
2507 pr_info("%s: %s %s %s%s\n",
2508 md->disk->disk_name, mmc_card_id(card), mmc_card_name(card),
2509 cap_str, md->read_only ? " (ro)" : "");
2511 /* used in ->open, must be set before add_disk: */
2512 if (area_type == MMC_BLK_DATA_AREA_MAIN)
2513 dev_set_drvdata(&card->dev, md);
2514 ret = device_add_disk(md->parent, md->disk, mmc_disk_attr_groups);
2521 blk_mq_free_tag_set(&md->queue.tag_set);
2525 ida_simple_remove(&mmc_blk_ida, devidx);
2526 return ERR_PTR(ret);
2529 static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
2533 if (!mmc_card_sd(card) && mmc_card_blockaddr(card)) {
2535 * The EXT_CSD sector count is in number or 512 byte
2538 size = card->ext_csd.sectors;
2541 * The CSD capacity field is in units of read_blkbits.
2542 * set_capacity takes units of 512 bytes.
2544 size = (typeof(sector_t))card->csd.capacity
2545 << (card->csd.read_blkbits - 9);
2548 return mmc_blk_alloc_req(card, &card->dev, size, false, NULL,
2549 MMC_BLK_DATA_AREA_MAIN, 0);
2552 static int mmc_blk_alloc_part(struct mmc_card *card,
2553 struct mmc_blk_data *md,
2554 unsigned int part_type,
2557 const char *subname,
2560 struct mmc_blk_data *part_md;
2562 part_md = mmc_blk_alloc_req(card, disk_to_dev(md->disk), size, default_ro,
2563 subname, area_type, part_type);
2564 if (IS_ERR(part_md))
2565 return PTR_ERR(part_md);
2566 list_add(&part_md->part, &md->part);
2572 * mmc_rpmb_ioctl() - ioctl handler for the RPMB chardev
2573 * @filp: the character device file
2574 * @cmd: the ioctl() command
2575 * @arg: the argument from userspace
2577 * This will essentially just redirect the ioctl()s coming in over to
2578 * the main block device spawning the RPMB character device.
2580 static long mmc_rpmb_ioctl(struct file *filp, unsigned int cmd,
2583 struct mmc_rpmb_data *rpmb = filp->private_data;
2588 ret = mmc_blk_ioctl_cmd(rpmb->md,
2589 (struct mmc_ioc_cmd __user *)arg,
2592 case MMC_IOC_MULTI_CMD:
2593 ret = mmc_blk_ioctl_multi_cmd(rpmb->md,
2594 (struct mmc_ioc_multi_cmd __user *)arg,
2605 #ifdef CONFIG_COMPAT
2606 static long mmc_rpmb_ioctl_compat(struct file *filp, unsigned int cmd,
2609 return mmc_rpmb_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
2613 static int mmc_rpmb_chrdev_open(struct inode *inode, struct file *filp)
2615 struct mmc_rpmb_data *rpmb = container_of(inode->i_cdev,
2616 struct mmc_rpmb_data, chrdev);
2618 get_device(&rpmb->dev);
2619 filp->private_data = rpmb;
2620 mmc_blk_get(rpmb->md->disk);
2622 return nonseekable_open(inode, filp);
2625 static int mmc_rpmb_chrdev_release(struct inode *inode, struct file *filp)
2627 struct mmc_rpmb_data *rpmb = container_of(inode->i_cdev,
2628 struct mmc_rpmb_data, chrdev);
2630 mmc_blk_put(rpmb->md);
2631 put_device(&rpmb->dev);
2636 static const struct file_operations mmc_rpmb_fileops = {
2637 .release = mmc_rpmb_chrdev_release,
2638 .open = mmc_rpmb_chrdev_open,
2639 .owner = THIS_MODULE,
2640 .llseek = no_llseek,
2641 .unlocked_ioctl = mmc_rpmb_ioctl,
2642 #ifdef CONFIG_COMPAT
2643 .compat_ioctl = mmc_rpmb_ioctl_compat,
2647 static void mmc_blk_rpmb_device_release(struct device *dev)
2649 struct mmc_rpmb_data *rpmb = dev_get_drvdata(dev);
2651 ida_simple_remove(&mmc_rpmb_ida, rpmb->id);
2655 static int mmc_blk_alloc_rpmb_part(struct mmc_card *card,
2656 struct mmc_blk_data *md,
2657 unsigned int part_index,
2659 const char *subname)
2662 char rpmb_name[DISK_NAME_LEN];
2664 struct mmc_rpmb_data *rpmb;
2666 /* This creates the minor number for the RPMB char device */
2667 devidx = ida_simple_get(&mmc_rpmb_ida, 0, max_devices, GFP_KERNEL);
2671 rpmb = kzalloc(sizeof(*rpmb), GFP_KERNEL);
2673 ida_simple_remove(&mmc_rpmb_ida, devidx);
2677 snprintf(rpmb_name, sizeof(rpmb_name),
2678 "mmcblk%u%s", card->host->index, subname ? subname : "");
2681 rpmb->part_index = part_index;
2682 rpmb->dev.init_name = rpmb_name;
2683 rpmb->dev.bus = &mmc_rpmb_bus_type;
2684 rpmb->dev.devt = MKDEV(MAJOR(mmc_rpmb_devt), rpmb->id);
2685 rpmb->dev.parent = &card->dev;
2686 rpmb->dev.release = mmc_blk_rpmb_device_release;
2687 device_initialize(&rpmb->dev);
2688 dev_set_drvdata(&rpmb->dev, rpmb);
2691 cdev_init(&rpmb->chrdev, &mmc_rpmb_fileops);
2692 rpmb->chrdev.owner = THIS_MODULE;
2693 ret = cdev_device_add(&rpmb->chrdev, &rpmb->dev);
2695 pr_err("%s: could not add character device\n", rpmb_name);
2696 goto out_put_device;
2699 list_add(&rpmb->node, &md->rpmbs);
2701 string_get_size((u64)size, 512, STRING_UNITS_2,
2702 cap_str, sizeof(cap_str));
2704 pr_info("%s: %s %s %s, chardev (%d:%d)\n",
2705 rpmb_name, mmc_card_id(card), mmc_card_name(card), cap_str,
2706 MAJOR(mmc_rpmb_devt), rpmb->id);
2711 put_device(&rpmb->dev);
2715 static void mmc_blk_remove_rpmb_part(struct mmc_rpmb_data *rpmb)
2718 cdev_device_del(&rpmb->chrdev, &rpmb->dev);
2719 put_device(&rpmb->dev);
2722 /* MMC Physical partitions consist of two boot partitions and
2723 * up to four general purpose partitions.
2724 * For each partition enabled in EXT_CSD a block device will be allocatedi
2725 * to provide access to the partition.
2728 static int mmc_blk_alloc_parts(struct mmc_card *card, struct mmc_blk_data *md)
2732 if (!mmc_card_mmc(card))
2735 for (idx = 0; idx < card->nr_parts; idx++) {
2736 if (card->part[idx].area_type & MMC_BLK_DATA_AREA_RPMB) {
2738 * RPMB partitions does not provide block access, they
2739 * are only accessed using ioctl():s. Thus create
2740 * special RPMB block devices that do not have a
2741 * backing block queue for these.
2743 ret = mmc_blk_alloc_rpmb_part(card, md,
2744 card->part[idx].part_cfg,
2745 card->part[idx].size >> 9,
2746 card->part[idx].name);
2749 } else if (card->part[idx].size) {
2750 ret = mmc_blk_alloc_part(card, md,
2751 card->part[idx].part_cfg,
2752 card->part[idx].size >> 9,
2753 card->part[idx].force_ro,
2754 card->part[idx].name,
2755 card->part[idx].area_type);
2764 static void mmc_blk_remove_req(struct mmc_blk_data *md)
2767 * Flush remaining requests and free queues. It is freeing the queue
2768 * that stops new requests from being accepted.
2770 del_gendisk(md->disk);
2771 mmc_cleanup_queue(&md->queue);
2775 static void mmc_blk_remove_parts(struct mmc_card *card,
2776 struct mmc_blk_data *md)
2778 struct list_head *pos, *q;
2779 struct mmc_blk_data *part_md;
2780 struct mmc_rpmb_data *rpmb;
2782 /* Remove RPMB partitions */
2783 list_for_each_safe(pos, q, &md->rpmbs) {
2784 rpmb = list_entry(pos, struct mmc_rpmb_data, node);
2786 mmc_blk_remove_rpmb_part(rpmb);
2788 /* Remove block partitions */
2789 list_for_each_safe(pos, q, &md->part) {
2790 part_md = list_entry(pos, struct mmc_blk_data, part);
2792 mmc_blk_remove_req(part_md);
2796 #ifdef CONFIG_DEBUG_FS
2798 static int mmc_dbg_card_status_get(void *data, u64 *val)
2800 struct mmc_card *card = data;
2801 struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
2802 struct mmc_queue *mq = &md->queue;
2803 struct request *req;
2806 /* Ask the block layer about the card status */
2807 req = blk_mq_alloc_request(mq->queue, REQ_OP_DRV_IN, 0);
2809 return PTR_ERR(req);
2810 req_to_mmc_queue_req(req)->drv_op = MMC_DRV_OP_GET_CARD_STATUS;
2811 req_to_mmc_queue_req(req)->drv_op_result = -EIO;
2812 blk_execute_rq(req, false);
2813 ret = req_to_mmc_queue_req(req)->drv_op_result;
2818 blk_mq_free_request(req);
2822 DEFINE_DEBUGFS_ATTRIBUTE(mmc_dbg_card_status_fops, mmc_dbg_card_status_get,
2825 /* That is two digits * 512 + 1 for newline */
2826 #define EXT_CSD_STR_LEN 1025
2828 static int mmc_ext_csd_open(struct inode *inode, struct file *filp)
2830 struct mmc_card *card = inode->i_private;
2831 struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
2832 struct mmc_queue *mq = &md->queue;
2833 struct request *req;
2839 buf = kmalloc(EXT_CSD_STR_LEN + 1, GFP_KERNEL);
2843 /* Ask the block layer for the EXT CSD */
2844 req = blk_mq_alloc_request(mq->queue, REQ_OP_DRV_IN, 0);
2849 req_to_mmc_queue_req(req)->drv_op = MMC_DRV_OP_GET_EXT_CSD;
2850 req_to_mmc_queue_req(req)->drv_op_result = -EIO;
2851 req_to_mmc_queue_req(req)->drv_op_data = &ext_csd;
2852 blk_execute_rq(req, false);
2853 err = req_to_mmc_queue_req(req)->drv_op_result;
2854 blk_mq_free_request(req);
2856 pr_err("FAILED %d\n", err);
2860 for (i = 0; i < 512; i++)
2861 n += sprintf(buf + n, "%02x", ext_csd[i]);
2862 n += sprintf(buf + n, "\n");
2864 if (n != EXT_CSD_STR_LEN) {
2870 filp->private_data = buf;
2879 static ssize_t mmc_ext_csd_read(struct file *filp, char __user *ubuf,
2880 size_t cnt, loff_t *ppos)
2882 char *buf = filp->private_data;
2884 return simple_read_from_buffer(ubuf, cnt, ppos,
2885 buf, EXT_CSD_STR_LEN);
2888 static int mmc_ext_csd_release(struct inode *inode, struct file *file)
2890 kfree(file->private_data);
2894 static const struct file_operations mmc_dbg_ext_csd_fops = {
2895 .open = mmc_ext_csd_open,
2896 .read = mmc_ext_csd_read,
2897 .release = mmc_ext_csd_release,
2898 .llseek = default_llseek,
2901 static void mmc_blk_add_debugfs(struct mmc_card *card, struct mmc_blk_data *md)
2903 struct dentry *root;
2905 if (!card->debugfs_root)
2908 root = card->debugfs_root;
2910 if (mmc_card_mmc(card) || mmc_card_sd(card)) {
2912 debugfs_create_file_unsafe("status", 0400, root,
2914 &mmc_dbg_card_status_fops);
2917 if (mmc_card_mmc(card)) {
2918 md->ext_csd_dentry =
2919 debugfs_create_file("ext_csd", S_IRUSR, root, card,
2920 &mmc_dbg_ext_csd_fops);
2924 static void mmc_blk_remove_debugfs(struct mmc_card *card,
2925 struct mmc_blk_data *md)
2927 if (!card->debugfs_root)
2930 debugfs_remove(md->status_dentry);
2931 md->status_dentry = NULL;
2933 debugfs_remove(md->ext_csd_dentry);
2934 md->ext_csd_dentry = NULL;
2939 static void mmc_blk_add_debugfs(struct mmc_card *card, struct mmc_blk_data *md)
2943 static void mmc_blk_remove_debugfs(struct mmc_card *card,
2944 struct mmc_blk_data *md)
2948 #endif /* CONFIG_DEBUG_FS */
2950 static int mmc_blk_probe(struct mmc_card *card)
2952 struct mmc_blk_data *md;
2956 * Check that the card supports the command class(es) we need.
2958 if (!(card->csd.cmdclass & CCC_BLOCK_READ))
2961 mmc_fixup_device(card, mmc_blk_fixups);
2963 card->complete_wq = alloc_workqueue("mmc_complete",
2964 WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2965 if (!card->complete_wq) {
2966 pr_err("Failed to create mmc completion workqueue");
2970 md = mmc_blk_alloc(card);
2976 ret = mmc_blk_alloc_parts(card, md);
2980 /* Add two debugfs entries */
2981 mmc_blk_add_debugfs(card, md);
2983 pm_runtime_set_autosuspend_delay(&card->dev, 3000);
2984 pm_runtime_use_autosuspend(&card->dev);
2987 * Don't enable runtime PM for SD-combo cards here. Leave that
2988 * decision to be taken during the SDIO init sequence instead.
2990 if (!mmc_card_sd_combo(card)) {
2991 pm_runtime_set_active(&card->dev);
2992 pm_runtime_enable(&card->dev);
2998 mmc_blk_remove_parts(card, md);
2999 mmc_blk_remove_req(md);
3001 destroy_workqueue(card->complete_wq);
3005 static void mmc_blk_remove(struct mmc_card *card)
3007 struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
3009 mmc_blk_remove_debugfs(card, md);
3010 mmc_blk_remove_parts(card, md);
3011 pm_runtime_get_sync(&card->dev);
3012 if (md->part_curr != md->part_type) {
3013 mmc_claim_host(card->host);
3014 mmc_blk_part_switch(card, md->part_type);
3015 mmc_release_host(card->host);
3017 if (!mmc_card_sd_combo(card))
3018 pm_runtime_disable(&card->dev);
3019 pm_runtime_put_noidle(&card->dev);
3020 mmc_blk_remove_req(md);
3021 destroy_workqueue(card->complete_wq);
3024 static int _mmc_blk_suspend(struct mmc_card *card)
3026 struct mmc_blk_data *part_md;
3027 struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
3030 mmc_queue_suspend(&md->queue);
3031 list_for_each_entry(part_md, &md->part, part) {
3032 mmc_queue_suspend(&part_md->queue);
3038 static void mmc_blk_shutdown(struct mmc_card *card)
3040 _mmc_blk_suspend(card);
3043 #ifdef CONFIG_PM_SLEEP
3044 static int mmc_blk_suspend(struct device *dev)
3046 struct mmc_card *card = mmc_dev_to_card(dev);
3048 return _mmc_blk_suspend(card);
3051 static int mmc_blk_resume(struct device *dev)
3053 struct mmc_blk_data *part_md;
3054 struct mmc_blk_data *md = dev_get_drvdata(dev);
3058 * Resume involves the card going into idle state,
3059 * so current partition is always the main one.
3061 md->part_curr = md->part_type;
3062 mmc_queue_resume(&md->queue);
3063 list_for_each_entry(part_md, &md->part, part) {
3064 mmc_queue_resume(&part_md->queue);
3071 static SIMPLE_DEV_PM_OPS(mmc_blk_pm_ops, mmc_blk_suspend, mmc_blk_resume);
3073 static struct mmc_driver mmc_driver = {
3076 .pm = &mmc_blk_pm_ops,
3078 .probe = mmc_blk_probe,
3079 .remove = mmc_blk_remove,
3080 .shutdown = mmc_blk_shutdown,
3083 static int __init mmc_blk_init(void)
3087 res = bus_register(&mmc_rpmb_bus_type);
3089 pr_err("mmcblk: could not register RPMB bus type\n");
3092 res = alloc_chrdev_region(&mmc_rpmb_devt, 0, MAX_DEVICES, "rpmb");
3094 pr_err("mmcblk: failed to allocate rpmb chrdev region\n");
3098 if (perdev_minors != CONFIG_MMC_BLOCK_MINORS)
3099 pr_info("mmcblk: using %d minors per device\n", perdev_minors);
3101 max_devices = min(MAX_DEVICES, (1 << MINORBITS) / perdev_minors);
3103 res = register_blkdev(MMC_BLOCK_MAJOR, "mmc");
3105 goto out_chrdev_unreg;
3107 res = mmc_register_driver(&mmc_driver);
3109 goto out_blkdev_unreg;
3114 unregister_blkdev(MMC_BLOCK_MAJOR, "mmc");
3116 unregister_chrdev_region(mmc_rpmb_devt, MAX_DEVICES);
3118 bus_unregister(&mmc_rpmb_bus_type);
3122 static void __exit mmc_blk_exit(void)
3124 mmc_unregister_driver(&mmc_driver);
3125 unregister_blkdev(MMC_BLOCK_MAJOR, "mmc");
3126 unregister_chrdev_region(mmc_rpmb_devt, MAX_DEVICES);
3127 bus_unregister(&mmc_rpmb_bus_type);
3130 module_init(mmc_blk_init);
3131 module_exit(mmc_blk_exit);
3133 MODULE_LICENSE("GPL");
3134 MODULE_DESCRIPTION("Multimedia Card (MMC) block device driver");