xref: /kernel/linux/linux-5.10/drivers/mmc/core/block.c (revision 8c2ecf20)
1/*
2 * Block driver for media (i.e., flash cards)
3 *
4 * Copyright 2002 Hewlett-Packard Company
5 * Copyright 2005-2008 Pierre Ossman
6 *
7 * Use consistent with the GNU GPL is permitted,
8 * provided that this copyright notice is
9 * preserved in its entirety in all copies and derived works.
10 *
11 * HEWLETT-PACKARD COMPANY MAKES NO WARRANTIES, EXPRESSED OR IMPLIED,
12 * AS TO THE USEFULNESS OR CORRECTNESS OF THIS CODE OR ITS
13 * FITNESS FOR ANY PARTICULAR PURPOSE.
14 *
15 * Many thanks to Alessandro Rubini and Jonathan Corbet!
16 *
17 * Author:  Andrew Christian
18 *          28 May 2002
19 */
20#include <linux/moduleparam.h>
21#include <linux/module.h>
22#include <linux/init.h>
23
24#include <linux/kernel.h>
25#include <linux/fs.h>
26#include <linux/slab.h>
27#include <linux/errno.h>
28#include <linux/hdreg.h>
29#include <linux/kdev_t.h>
30#include <linux/blkdev.h>
31#include <linux/cdev.h>
32#include <linux/mutex.h>
33#include <linux/scatterlist.h>
34#include <linux/string_helpers.h>
35#include <linux/delay.h>
36#include <linux/capability.h>
37#include <linux/compat.h>
38#include <linux/pm_runtime.h>
39#include <linux/idr.h>
40#include <linux/debugfs.h>
41
42#include <linux/mmc/ioctl.h>
43#include <linux/mmc/card.h>
44#include <linux/mmc/host.h>
45#include <linux/mmc/mmc.h>
46#include <linux/mmc/sd.h>
47
48#include <linux/uaccess.h>
49
50#include "queue.h"
51#include "block.h"
52#include "core.h"
53#include "card.h"
54#include "host.h"
55#include "bus.h"
56#include "mmc_ops.h"
57#include "quirks.h"
58#include "sd_ops.h"
59
60MODULE_ALIAS("mmc:block");
61#ifdef MODULE_PARAM_PREFIX
62#undef MODULE_PARAM_PREFIX
63#endif
64#define MODULE_PARAM_PREFIX "mmcblk."
65
66/*
67 * Set a 10 second timeout for polling write request busy state. Note, mmc core
68 * is setting a 3 second timeout for SD cards, and SDHCI has long had a 10
69 * second software timer to timeout the whole request, so 10 seconds should be
70 * ample.
71 */
72#define MMC_BLK_TIMEOUT_MS  (10 * 1000)
73#define MMC_EXTRACT_INDEX_FROM_ARG(x) ((x & 0x00FF0000) >> 16)
74#define MMC_EXTRACT_VALUE_FROM_ARG(x) ((x & 0x0000FF00) >> 8)
75
76#define mmc_req_rel_wr(req)	((req->cmd_flags & REQ_FUA) && \
77				  (rq_data_dir(req) == WRITE))
78static DEFINE_MUTEX(block_mutex);
79
80/*
81 * The defaults come from config options but can be overriden by module
82 * or bootarg options.
83 */
84static int perdev_minors = CONFIG_MMC_BLOCK_MINORS;
85
86/*
87 * We've only got one major, so number of mmcblk devices is
88 * limited to (1 << 20) / number of minors per device.  It is also
89 * limited by the MAX_DEVICES below.
90 */
91static int max_devices;
92
93#define MAX_DEVICES 256
94
95static DEFINE_IDA(mmc_blk_ida);
96static DEFINE_IDA(mmc_rpmb_ida);
97
98/*
99 * There is one mmc_blk_data per slot.
100 */
101struct mmc_blk_data {
102	struct device	*parent;
103	struct gendisk	*disk;
104	struct mmc_queue queue;
105	struct list_head part;
106	struct list_head rpmbs;
107
108	unsigned int	flags;
109#define MMC_BLK_CMD23	(1 << 0)	/* Can do SET_BLOCK_COUNT for multiblock */
110#define MMC_BLK_REL_WR	(1 << 1)	/* MMC Reliable write support */
111
112	unsigned int	usage;
113	unsigned int	read_only;
114	unsigned int	part_type;
115	unsigned int	reset_done;
116#define MMC_BLK_READ		BIT(0)
117#define MMC_BLK_WRITE		BIT(1)
118#define MMC_BLK_DISCARD		BIT(2)
119#define MMC_BLK_SECDISCARD	BIT(3)
120#define MMC_BLK_CQE_RECOVERY	BIT(4)
121
122	/*
123	 * Only set in main mmc_blk_data associated
124	 * with mmc_card with dev_set_drvdata, and keeps
125	 * track of the current selected device partition.
126	 */
127	unsigned int	part_curr;
128	struct device_attribute force_ro;
129	struct device_attribute power_ro_lock;
130	int	area_type;
131
132	/* debugfs files (only in main mmc_blk_data) */
133	struct dentry *status_dentry;
134	struct dentry *ext_csd_dentry;
135};
136
137/* Device type for RPMB character devices */
138static dev_t mmc_rpmb_devt;
139
140/* Bus type for RPMB character devices */
141static struct bus_type mmc_rpmb_bus_type = {
142	.name = "mmc_rpmb",
143};
144
145/**
146 * struct mmc_rpmb_data - special RPMB device type for these areas
147 * @dev: the device for the RPMB area
148 * @chrdev: character device for the RPMB area
149 * @id: unique device ID number
150 * @part_index: partition index (0 on first)
151 * @md: parent MMC block device
152 * @node: list item, so we can put this device on a list
153 */
154struct mmc_rpmb_data {
155	struct device dev;
156	struct cdev chrdev;
157	int id;
158	unsigned int part_index;
159	struct mmc_blk_data *md;
160	struct list_head node;
161};
162
163static DEFINE_MUTEX(open_lock);
164
165module_param(perdev_minors, int, 0444);
166MODULE_PARM_DESC(perdev_minors, "Minors numbers to allocate per device");
167
168static inline int mmc_blk_part_switch(struct mmc_card *card,
169				      unsigned int part_type);
170static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
171			       struct mmc_card *card,
172			       int recovery_mode,
173			       struct mmc_queue *mq);
174static void mmc_blk_hsq_req_done(struct mmc_request *mrq);
175
176static struct mmc_blk_data *mmc_blk_get(struct gendisk *disk)
177{
178	struct mmc_blk_data *md;
179
180	mutex_lock(&open_lock);
181	md = disk->private_data;
182	if (md && md->usage == 0)
183		md = NULL;
184	if (md)
185		md->usage++;
186	mutex_unlock(&open_lock);
187
188	return md;
189}
190
191static inline int mmc_get_devidx(struct gendisk *disk)
192{
193	int devidx = disk->first_minor / perdev_minors;
194	return devidx;
195}
196
197static void mmc_blk_put(struct mmc_blk_data *md)
198{
199	mutex_lock(&open_lock);
200	md->usage--;
201	if (md->usage == 0) {
202		int devidx = mmc_get_devidx(md->disk);
203		blk_put_queue(md->queue.queue);
204		ida_simple_remove(&mmc_blk_ida, devidx);
205		put_disk(md->disk);
206		kfree(md);
207	}
208	mutex_unlock(&open_lock);
209}
210
211static ssize_t power_ro_lock_show(struct device *dev,
212		struct device_attribute *attr, char *buf)
213{
214	int ret;
215	struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
216	struct mmc_card *card = md->queue.card;
217	int locked = 0;
218
219	if (card->ext_csd.boot_ro_lock & EXT_CSD_BOOT_WP_B_PERM_WP_EN)
220		locked = 2;
221	else if (card->ext_csd.boot_ro_lock & EXT_CSD_BOOT_WP_B_PWR_WP_EN)
222		locked = 1;
223
224	ret = snprintf(buf, PAGE_SIZE, "%d\n", locked);
225
226	mmc_blk_put(md);
227
228	return ret;
229}
230
231static ssize_t power_ro_lock_store(struct device *dev,
232		struct device_attribute *attr, const char *buf, size_t count)
233{
234	int ret;
235	struct mmc_blk_data *md, *part_md;
236	struct mmc_queue *mq;
237	struct request *req;
238	unsigned long set;
239
240	if (kstrtoul(buf, 0, &set))
241		return -EINVAL;
242
243	if (set != 1)
244		return count;
245
246	md = mmc_blk_get(dev_to_disk(dev));
247	mq = &md->queue;
248
249	/* Dispatch locking to the block layer */
250	req = blk_get_request(mq->queue, REQ_OP_DRV_OUT, 0);
251	if (IS_ERR(req)) {
252		count = PTR_ERR(req);
253		goto out_put;
254	}
255	req_to_mmc_queue_req(req)->drv_op = MMC_DRV_OP_BOOT_WP;
256	req_to_mmc_queue_req(req)->drv_op_result = -EIO;
257	blk_execute_rq(mq->queue, NULL, req, 0);
258	ret = req_to_mmc_queue_req(req)->drv_op_result;
259	blk_put_request(req);
260
261	if (!ret) {
262		pr_info("%s: Locking boot partition ro until next power on\n",
263			md->disk->disk_name);
264		set_disk_ro(md->disk, 1);
265
266		list_for_each_entry(part_md, &md->part, part)
267			if (part_md->area_type == MMC_BLK_DATA_AREA_BOOT) {
268				pr_info("%s: Locking boot partition ro until next power on\n", part_md->disk->disk_name);
269				set_disk_ro(part_md->disk, 1);
270			}
271	}
272out_put:
273	mmc_blk_put(md);
274	return count;
275}
276
277static ssize_t force_ro_show(struct device *dev, struct device_attribute *attr,
278			     char *buf)
279{
280	int ret;
281	struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
282
283	ret = snprintf(buf, PAGE_SIZE, "%d\n",
284		       get_disk_ro(dev_to_disk(dev)) ^
285		       md->read_only);
286	mmc_blk_put(md);
287	return ret;
288}
289
290static ssize_t force_ro_store(struct device *dev, struct device_attribute *attr,
291			      const char *buf, size_t count)
292{
293	int ret;
294	char *end;
295	struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
296	unsigned long set = simple_strtoul(buf, &end, 0);
297	if (end == buf) {
298		ret = -EINVAL;
299		goto out;
300	}
301
302	set_disk_ro(dev_to_disk(dev), set || md->read_only);
303	ret = count;
304out:
305	mmc_blk_put(md);
306	return ret;
307}
308
309static int mmc_blk_open(struct block_device *bdev, fmode_t mode)
310{
311	struct mmc_blk_data *md = mmc_blk_get(bdev->bd_disk);
312	int ret = -ENXIO;
313
314	mutex_lock(&block_mutex);
315	if (md) {
316		ret = 0;
317		if ((mode & FMODE_WRITE) && md->read_only) {
318			mmc_blk_put(md);
319			ret = -EROFS;
320		}
321	}
322	mutex_unlock(&block_mutex);
323
324	return ret;
325}
326
327static void mmc_blk_release(struct gendisk *disk, fmode_t mode)
328{
329	struct mmc_blk_data *md = disk->private_data;
330
331	mutex_lock(&block_mutex);
332	mmc_blk_put(md);
333	mutex_unlock(&block_mutex);
334}
335
336static int
337mmc_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
338{
339	geo->cylinders = get_capacity(bdev->bd_disk) / (4 * 16);
340	geo->heads = 4;
341	geo->sectors = 16;
342	return 0;
343}
344
345struct mmc_blk_ioc_data {
346	struct mmc_ioc_cmd ic;
347	unsigned char *buf;
348	u64 buf_bytes;
349	unsigned int flags;
350#define MMC_BLK_IOC_DROP	BIT(0)	/* drop this mrq */
351#define MMC_BLK_IOC_SBC	BIT(1)	/* use mrq.sbc */
352
353	struct mmc_rpmb_data *rpmb;
354};
355
356static struct mmc_blk_ioc_data *mmc_blk_ioctl_copy_from_user(
357	struct mmc_ioc_cmd __user *user)
358{
359	struct mmc_blk_ioc_data *idata;
360	int err;
361
362	idata = kmalloc(sizeof(*idata), GFP_KERNEL);
363	if (!idata) {
364		err = -ENOMEM;
365		goto out;
366	}
367
368	if (copy_from_user(&idata->ic, user, sizeof(idata->ic))) {
369		err = -EFAULT;
370		goto idata_err;
371	}
372
373	idata->buf_bytes = (u64) idata->ic.blksz * idata->ic.blocks;
374	if (idata->buf_bytes > MMC_IOC_MAX_BYTES) {
375		err = -EOVERFLOW;
376		goto idata_err;
377	}
378
379	if (!idata->buf_bytes) {
380		idata->buf = NULL;
381		return idata;
382	}
383
384	idata->buf = memdup_user((void __user *)(unsigned long)
385				 idata->ic.data_ptr, idata->buf_bytes);
386	if (IS_ERR(idata->buf)) {
387		err = PTR_ERR(idata->buf);
388		goto idata_err;
389	}
390
391	return idata;
392
393idata_err:
394	kfree(idata);
395out:
396	return ERR_PTR(err);
397}
398
399static int mmc_blk_ioctl_copy_to_user(struct mmc_ioc_cmd __user *ic_ptr,
400				      struct mmc_blk_ioc_data *idata)
401{
402	struct mmc_ioc_cmd *ic = &idata->ic;
403
404	if (copy_to_user(&(ic_ptr->response), ic->response,
405			 sizeof(ic->response)))
406		return -EFAULT;
407
408	if (!idata->ic.write_flag) {
409		if (copy_to_user((void __user *)(unsigned long)ic->data_ptr,
410				 idata->buf, idata->buf_bytes))
411			return -EFAULT;
412	}
413
414	return 0;
415}
416
417static int card_busy_detect(struct mmc_card *card, unsigned int timeout_ms,
418			    u32 *resp_errs)
419{
420	unsigned long timeout = jiffies + msecs_to_jiffies(timeout_ms);
421	int err = 0;
422	u32 status;
423
424	do {
425		bool done = time_after(jiffies, timeout);
426
427		err = __mmc_send_status(card, &status, 5);
428		if (err) {
429			dev_err(mmc_dev(card->host),
430				"error %d requesting status\n", err);
431			return err;
432		}
433
434		/* Accumulate any response error bits seen */
435		if (resp_errs)
436			*resp_errs |= status;
437
438		/*
439		 * Timeout if the device never becomes ready for data and never
440		 * leaves the program state.
441		 */
442		if (done) {
443			dev_err(mmc_dev(card->host),
444				"Card stuck in wrong state! %s status: %#x\n",
445				 __func__, status);
446			return -ETIMEDOUT;
447		}
448	} while (!mmc_ready_for_data(status));
449
450	return err;
451}
452
453static int __mmc_blk_ioctl_cmd(struct mmc_card *card, struct mmc_blk_data *md,
454			       struct mmc_blk_ioc_data **idatas, int i)
455{
456	struct mmc_command cmd = {}, sbc = {};
457	struct mmc_data data = {};
458	struct mmc_request mrq = {};
459	struct scatterlist sg;
460	int err;
461	unsigned int target_part;
462	struct mmc_blk_ioc_data *idata = idatas[i];
463	struct mmc_blk_ioc_data *prev_idata = NULL;
464
465	if (!card || !md || !idata)
466		return -EINVAL;
467
468	if (idata->flags & MMC_BLK_IOC_DROP)
469		return 0;
470
471	if (idata->flags & MMC_BLK_IOC_SBC && i > 0)
472		prev_idata = idatas[i - 1];
473
474	/*
475	 * The RPMB accesses comes in from the character device, so we
476	 * need to target these explicitly. Else we just target the
477	 * partition type for the block device the ioctl() was issued
478	 * on.
479	 */
480	if (idata->rpmb) {
481		/* Support multiple RPMB partitions */
482		target_part = idata->rpmb->part_index;
483		target_part |= EXT_CSD_PART_CONFIG_ACC_RPMB;
484	} else {
485		target_part = md->part_type;
486	}
487
488	cmd.opcode = idata->ic.opcode;
489	cmd.arg = idata->ic.arg;
490	cmd.flags = idata->ic.flags;
491
492	if (idata->buf_bytes) {
493		data.sg = &sg;
494		data.sg_len = 1;
495		data.blksz = idata->ic.blksz;
496		data.blocks = idata->ic.blocks;
497
498		sg_init_one(data.sg, idata->buf, idata->buf_bytes);
499
500		if (idata->ic.write_flag)
501			data.flags = MMC_DATA_WRITE;
502		else
503			data.flags = MMC_DATA_READ;
504
505		/* data.flags must already be set before doing this. */
506		mmc_set_data_timeout(&data, card);
507
508		/* Allow overriding the timeout_ns for empirical tuning. */
509		if (idata->ic.data_timeout_ns)
510			data.timeout_ns = idata->ic.data_timeout_ns;
511
512		if ((cmd.flags & MMC_RSP_R1B) == MMC_RSP_R1B) {
513			/*
514			 * Pretend this is a data transfer and rely on the
515			 * host driver to compute timeout.  When all host
516			 * drivers support cmd.cmd_timeout for R1B, this
517			 * can be changed to:
518			 *
519			 *     mrq.data = NULL;
520			 *     cmd.cmd_timeout = idata->ic.cmd_timeout_ms;
521			 */
522			data.timeout_ns = idata->ic.cmd_timeout_ms * 1000000;
523		}
524
525		mrq.data = &data;
526	}
527
528	mrq.cmd = &cmd;
529
530	err = mmc_blk_part_switch(card, target_part);
531	if (err)
532		return err;
533
534	if (idata->ic.is_acmd) {
535		err = mmc_app_cmd(card->host, card);
536		if (err)
537			return err;
538	}
539
540	if (idata->rpmb || prev_idata) {
541		sbc.opcode = MMC_SET_BLOCK_COUNT;
542		/*
543		 * We don't do any blockcount validation because the max size
544		 * may be increased by a future standard. We just copy the
545		 * 'Reliable Write' bit here.
546		 */
547		sbc.arg = data.blocks | (idata->ic.write_flag & BIT(31));
548		if (prev_idata)
549			sbc.arg = prev_idata->ic.arg;
550		sbc.flags = MMC_RSP_R1 | MMC_CMD_AC;
551		mrq.sbc = &sbc;
552	}
553
554	if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_SANITIZE_START) &&
555	    (cmd.opcode == MMC_SWITCH))
556		return mmc_sanitize(card);
557
558	mmc_wait_for_req(card->host, &mrq);
559	memcpy(&idata->ic.response, cmd.resp, sizeof(cmd.resp));
560
561	if (prev_idata) {
562		memcpy(&prev_idata->ic.response, sbc.resp, sizeof(sbc.resp));
563		if (sbc.error) {
564			dev_err(mmc_dev(card->host), "%s: sbc error %d\n",
565							__func__, sbc.error);
566			return sbc.error;
567		}
568	}
569
570	if (cmd.error) {
571		dev_err(mmc_dev(card->host), "%s: cmd error %d\n",
572						__func__, cmd.error);
573		return cmd.error;
574	}
575	if (data.error) {
576		dev_err(mmc_dev(card->host), "%s: data error %d\n",
577						__func__, data.error);
578		return data.error;
579	}
580
581	/*
582	 * Make sure the cache of the PARTITION_CONFIG register and
583	 * PARTITION_ACCESS bits is updated in case the ioctl ext_csd write
584	 * changed it successfully.
585	 */
586	if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_PART_CONFIG) &&
587	    (cmd.opcode == MMC_SWITCH)) {
588		struct mmc_blk_data *main_md = dev_get_drvdata(&card->dev);
589		u8 value = MMC_EXTRACT_VALUE_FROM_ARG(cmd.arg);
590
591		/*
592		 * Update cache so the next mmc_blk_part_switch call operates
593		 * on up-to-date data.
594		 */
595		card->ext_csd.part_config = value;
596		main_md->part_curr = value & EXT_CSD_PART_CONFIG_ACC_MASK;
597	}
598
599	/*
600	 * Make sure to update CACHE_CTRL in case it was changed. The cache
601	 * will get turned back on if the card is re-initialized, e.g.
602	 * suspend/resume or hw reset in recovery.
603	 */
604	if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_CACHE_CTRL) &&
605	    (cmd.opcode == MMC_SWITCH)) {
606		u8 value = MMC_EXTRACT_VALUE_FROM_ARG(cmd.arg) & 1;
607
608		card->ext_csd.cache_ctrl = value;
609	}
610
611	/*
612	 * According to the SD specs, some commands require a delay after
613	 * issuing the command.
614	 */
615	if (idata->ic.postsleep_min_us)
616		usleep_range(idata->ic.postsleep_min_us, idata->ic.postsleep_max_us);
617
618	if (idata->rpmb || (cmd.flags & MMC_RSP_R1B) == MMC_RSP_R1B) {
619		/*
620		 * Ensure RPMB/R1B command has completed by polling CMD13
621		 * "Send Status".
622		 */
623		err = card_busy_detect(card, MMC_BLK_TIMEOUT_MS, NULL);
624	}
625
626	return err;
627}
628
629static int mmc_blk_ioctl_cmd(struct mmc_blk_data *md,
630			     struct mmc_ioc_cmd __user *ic_ptr,
631			     struct mmc_rpmb_data *rpmb)
632{
633	struct mmc_blk_ioc_data *idata;
634	struct mmc_blk_ioc_data *idatas[1];
635	struct mmc_queue *mq;
636	struct mmc_card *card;
637	int err = 0, ioc_err = 0;
638	struct request *req;
639
640	idata = mmc_blk_ioctl_copy_from_user(ic_ptr);
641	if (IS_ERR(idata))
642		return PTR_ERR(idata);
643	/* This will be NULL on non-RPMB ioctl():s */
644	idata->rpmb = rpmb;
645
646	card = md->queue.card;
647	if (IS_ERR(card)) {
648		err = PTR_ERR(card);
649		goto cmd_done;
650	}
651
652	/*
653	 * Dispatch the ioctl() into the block request queue.
654	 */
655	mq = &md->queue;
656	req = blk_get_request(mq->queue,
657		idata->ic.write_flag ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
658	if (IS_ERR(req)) {
659		err = PTR_ERR(req);
660		goto cmd_done;
661	}
662	idatas[0] = idata;
663	req_to_mmc_queue_req(req)->drv_op =
664		rpmb ? MMC_DRV_OP_IOCTL_RPMB : MMC_DRV_OP_IOCTL;
665	req_to_mmc_queue_req(req)->drv_op_result = -EIO;
666	req_to_mmc_queue_req(req)->drv_op_data = idatas;
667	req_to_mmc_queue_req(req)->ioc_count = 1;
668	blk_execute_rq(mq->queue, NULL, req, 0);
669	ioc_err = req_to_mmc_queue_req(req)->drv_op_result;
670	err = mmc_blk_ioctl_copy_to_user(ic_ptr, idata);
671	blk_put_request(req);
672
673cmd_done:
674	kfree(idata->buf);
675	kfree(idata);
676	return ioc_err ? ioc_err : err;
677}
678
679static int mmc_blk_ioctl_multi_cmd(struct mmc_blk_data *md,
680				   struct mmc_ioc_multi_cmd __user *user,
681				   struct mmc_rpmb_data *rpmb)
682{
683	struct mmc_blk_ioc_data **idata = NULL;
684	struct mmc_ioc_cmd __user *cmds = user->cmds;
685	struct mmc_card *card;
686	struct mmc_queue *mq;
687	int i, err = 0, ioc_err = 0;
688	__u64 num_of_cmds;
689	struct request *req;
690
691	if (copy_from_user(&num_of_cmds, &user->num_of_cmds,
692			   sizeof(num_of_cmds)))
693		return -EFAULT;
694
695	if (!num_of_cmds)
696		return 0;
697
698	if (num_of_cmds > MMC_IOC_MAX_CMDS)
699		return -EINVAL;
700
701	idata = kcalloc(num_of_cmds, sizeof(*idata), GFP_KERNEL);
702	if (!idata)
703		return -ENOMEM;
704
705	for (i = 0; i < num_of_cmds; i++) {
706		idata[i] = mmc_blk_ioctl_copy_from_user(&cmds[i]);
707		if (IS_ERR(idata[i])) {
708			err = PTR_ERR(idata[i]);
709			num_of_cmds = i;
710			goto cmd_err;
711		}
712		/* This will be NULL on non-RPMB ioctl():s */
713		idata[i]->rpmb = rpmb;
714	}
715
716	card = md->queue.card;
717	if (IS_ERR(card)) {
718		err = PTR_ERR(card);
719		goto cmd_err;
720	}
721
722
723	/*
724	 * Dispatch the ioctl()s into the block request queue.
725	 */
726	mq = &md->queue;
727	req = blk_get_request(mq->queue,
728		idata[0]->ic.write_flag ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, 0);
729	if (IS_ERR(req)) {
730		err = PTR_ERR(req);
731		goto cmd_err;
732	}
733	req_to_mmc_queue_req(req)->drv_op =
734		rpmb ? MMC_DRV_OP_IOCTL_RPMB : MMC_DRV_OP_IOCTL;
735	req_to_mmc_queue_req(req)->drv_op_result = -EIO;
736	req_to_mmc_queue_req(req)->drv_op_data = idata;
737	req_to_mmc_queue_req(req)->ioc_count = num_of_cmds;
738	blk_execute_rq(mq->queue, NULL, req, 0);
739	ioc_err = req_to_mmc_queue_req(req)->drv_op_result;
740
741	/* copy to user if data and response */
742	for (i = 0; i < num_of_cmds && !err; i++)
743		err = mmc_blk_ioctl_copy_to_user(&cmds[i], idata[i]);
744
745	blk_put_request(req);
746
747cmd_err:
748	for (i = 0; i < num_of_cmds; i++) {
749		kfree(idata[i]->buf);
750		kfree(idata[i]);
751	}
752	kfree(idata);
753	return ioc_err ? ioc_err : err;
754}
755
756static int mmc_blk_check_blkdev(struct block_device *bdev)
757{
758	/*
759	 * The caller must have CAP_SYS_RAWIO, and must be calling this on the
760	 * whole block device, not on a partition.  This prevents overspray
761	 * between sibling partitions.
762	 */
763	if (!capable(CAP_SYS_RAWIO) || bdev_is_partition(bdev))
764		return -EPERM;
765	return 0;
766}
767
768static int mmc_blk_ioctl(struct block_device *bdev, fmode_t mode,
769	unsigned int cmd, unsigned long arg)
770{
771	struct mmc_blk_data *md;
772	int ret;
773
774	switch (cmd) {
775	case MMC_IOC_CMD:
776		ret = mmc_blk_check_blkdev(bdev);
777		if (ret)
778			return ret;
779		md = mmc_blk_get(bdev->bd_disk);
780		if (!md)
781			return -EINVAL;
782		ret = mmc_blk_ioctl_cmd(md,
783					(struct mmc_ioc_cmd __user *)arg,
784					NULL);
785		mmc_blk_put(md);
786		return ret;
787	case MMC_IOC_MULTI_CMD:
788		ret = mmc_blk_check_blkdev(bdev);
789		if (ret)
790			return ret;
791		md = mmc_blk_get(bdev->bd_disk);
792		if (!md)
793			return -EINVAL;
794		ret = mmc_blk_ioctl_multi_cmd(md,
795					(struct mmc_ioc_multi_cmd __user *)arg,
796					NULL);
797		mmc_blk_put(md);
798		return ret;
799	default:
800		return -EINVAL;
801	}
802}
803
804#ifdef CONFIG_COMPAT
805static int mmc_blk_compat_ioctl(struct block_device *bdev, fmode_t mode,
806	unsigned int cmd, unsigned long arg)
807{
808	return mmc_blk_ioctl(bdev, mode, cmd, (unsigned long) compat_ptr(arg));
809}
810#endif
811
812static const struct block_device_operations mmc_bdops = {
813	.open			= mmc_blk_open,
814	.release		= mmc_blk_release,
815	.getgeo			= mmc_blk_getgeo,
816	.owner			= THIS_MODULE,
817	.ioctl			= mmc_blk_ioctl,
818#ifdef CONFIG_COMPAT
819	.compat_ioctl		= mmc_blk_compat_ioctl,
820#endif
821};
822
823static int mmc_blk_part_switch_pre(struct mmc_card *card,
824				   unsigned int part_type)
825{
826	const unsigned int mask = EXT_CSD_PART_CONFIG_ACC_RPMB;
827	int ret = 0;
828
829	if ((part_type & mask) == mask) {
830		if (card->ext_csd.cmdq_en) {
831			ret = mmc_cmdq_disable(card);
832			if (ret)
833				return ret;
834		}
835		mmc_retune_pause(card->host);
836	}
837
838	return ret;
839}
840
841static int mmc_blk_part_switch_post(struct mmc_card *card,
842				    unsigned int part_type)
843{
844	const unsigned int mask = EXT_CSD_PART_CONFIG_ACC_RPMB;
845	int ret = 0;
846
847	if ((part_type & mask) == mask) {
848		mmc_retune_unpause(card->host);
849		if (card->reenable_cmdq && !card->ext_csd.cmdq_en)
850			ret = mmc_cmdq_enable(card);
851	}
852
853	return ret;
854}
855
856static inline int mmc_blk_part_switch(struct mmc_card *card,
857				      unsigned int part_type)
858{
859	int ret = 0;
860	struct mmc_blk_data *main_md = dev_get_drvdata(&card->dev);
861
862	if (main_md->part_curr == part_type)
863		return 0;
864
865	if (mmc_card_mmc(card)) {
866		u8 part_config = card->ext_csd.part_config;
867
868		ret = mmc_blk_part_switch_pre(card, part_type);
869		if (ret)
870			return ret;
871
872		part_config &= ~EXT_CSD_PART_CONFIG_ACC_MASK;
873		part_config |= part_type;
874
875		ret = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
876				 EXT_CSD_PART_CONFIG, part_config,
877				 card->ext_csd.part_time);
878		if (ret) {
879			mmc_blk_part_switch_post(card, part_type);
880			return ret;
881		}
882
883		card->ext_csd.part_config = part_config;
884
885		ret = mmc_blk_part_switch_post(card, main_md->part_curr);
886	}
887
888	main_md->part_curr = part_type;
889	return ret;
890}
891
892static int mmc_sd_num_wr_blocks(struct mmc_card *card, u32 *written_blocks)
893{
894	int err;
895	u32 result;
896	__be32 *blocks;
897
898	struct mmc_request mrq = {};
899	struct mmc_command cmd = {};
900	struct mmc_data data = {};
901
902	struct scatterlist sg;
903
904	cmd.opcode = MMC_APP_CMD;
905	cmd.arg = card->rca << 16;
906	cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC;
907
908	err = mmc_wait_for_cmd(card->host, &cmd, 0);
909	if (err)
910		return err;
911	if (!mmc_host_is_spi(card->host) && !(cmd.resp[0] & R1_APP_CMD))
912		return -EIO;
913
914	memset(&cmd, 0, sizeof(struct mmc_command));
915
916	cmd.opcode = SD_APP_SEND_NUM_WR_BLKS;
917	cmd.arg = 0;
918	cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC;
919
920	data.blksz = 4;
921	data.blocks = 1;
922	data.flags = MMC_DATA_READ;
923	data.sg = &sg;
924	data.sg_len = 1;
925	mmc_set_data_timeout(&data, card);
926
927	mrq.cmd = &cmd;
928	mrq.data = &data;
929
930	blocks = kmalloc(4, GFP_KERNEL);
931	if (!blocks)
932		return -ENOMEM;
933
934	sg_init_one(&sg, blocks, 4);
935
936	mmc_wait_for_req(card->host, &mrq);
937
938	result = ntohl(*blocks);
939	kfree(blocks);
940
941	if (cmd.error || data.error)
942		return -EIO;
943
944	*written_blocks = result;
945
946	return 0;
947}
948
949static unsigned int mmc_blk_clock_khz(struct mmc_host *host)
950{
951	if (host->actual_clock)
952		return host->actual_clock / 1000;
953
954	/* Clock may be subject to a divisor, fudge it by a factor of 2. */
955	if (host->ios.clock)
956		return host->ios.clock / 2000;
957
958	/* How can there be no clock */
959	WARN_ON_ONCE(1);
960	return 100; /* 100 kHz is minimum possible value */
961}
962
963static unsigned int mmc_blk_data_timeout_ms(struct mmc_host *host,
964					    struct mmc_data *data)
965{
966	unsigned int ms = DIV_ROUND_UP(data->timeout_ns, 1000000);
967	unsigned int khz;
968
969	if (data->timeout_clks) {
970		khz = mmc_blk_clock_khz(host);
971		ms += DIV_ROUND_UP(data->timeout_clks, khz);
972	}
973
974	return ms;
975}
976
977static int mmc_blk_reset(struct mmc_blk_data *md, struct mmc_host *host,
978			 int type)
979{
980	int err;
981
982	if (md->reset_done & type)
983		return -EEXIST;
984
985	md->reset_done |= type;
986	err = mmc_hw_reset(host);
987	/* Ensure we switch back to the correct partition */
988	if (err != -EOPNOTSUPP) {
989		struct mmc_blk_data *main_md =
990			dev_get_drvdata(&host->card->dev);
991		int part_err;
992
993		main_md->part_curr = main_md->part_type;
994		part_err = mmc_blk_part_switch(host->card, md->part_type);
995		if (part_err) {
996			/*
997			 * We have failed to get back into the correct
998			 * partition, so we need to abort the whole request.
999			 */
1000			return -ENODEV;
1001		}
1002	}
1003	return err;
1004}
1005
1006static inline void mmc_blk_reset_success(struct mmc_blk_data *md, int type)
1007{
1008	md->reset_done &= ~type;
1009}
1010
1011static void mmc_blk_check_sbc(struct mmc_queue_req *mq_rq)
1012{
1013	struct mmc_blk_ioc_data **idata = mq_rq->drv_op_data;
1014	int i;
1015
1016	for (i = 1; i < mq_rq->ioc_count; i++) {
1017		if (idata[i - 1]->ic.opcode == MMC_SET_BLOCK_COUNT &&
1018		    mmc_op_multi(idata[i]->ic.opcode)) {
1019			idata[i - 1]->flags |= MMC_BLK_IOC_DROP;
1020			idata[i]->flags |= MMC_BLK_IOC_SBC;
1021		}
1022	}
1023}
1024
1025/*
1026 * The non-block commands come back from the block layer after it queued it and
1027 * processed it with all other requests and then they get issued in this
1028 * function.
1029 */
1030static void mmc_blk_issue_drv_op(struct mmc_queue *mq, struct request *req)
1031{
1032	struct mmc_queue_req *mq_rq;
1033	struct mmc_card *card = mq->card;
1034	struct mmc_blk_data *md = mq->blkdata;
1035	struct mmc_blk_ioc_data **idata;
1036	bool rpmb_ioctl;
1037	u8 **ext_csd;
1038	u32 status;
1039	int ret;
1040	int i;
1041
1042	mq_rq = req_to_mmc_queue_req(req);
1043	rpmb_ioctl = (mq_rq->drv_op == MMC_DRV_OP_IOCTL_RPMB);
1044
1045	switch (mq_rq->drv_op) {
1046	case MMC_DRV_OP_IOCTL:
1047		if (card->ext_csd.cmdq_en) {
1048			ret = mmc_cmdq_disable(card);
1049			if (ret)
1050				break;
1051		}
1052
1053		mmc_blk_check_sbc(mq_rq);
1054
1055		fallthrough;
1056	case MMC_DRV_OP_IOCTL_RPMB:
1057		idata = mq_rq->drv_op_data;
1058		for (i = 0, ret = 0; i < mq_rq->ioc_count; i++) {
1059			ret = __mmc_blk_ioctl_cmd(card, md, idata, i);
1060			if (ret)
1061				break;
1062		}
1063		/* Always switch back to main area after RPMB access */
1064		if (rpmb_ioctl)
1065			mmc_blk_part_switch(card, 0);
1066		else if (card->reenable_cmdq && !card->ext_csd.cmdq_en)
1067			mmc_cmdq_enable(card);
1068		break;
1069	case MMC_DRV_OP_BOOT_WP:
1070		ret = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL, EXT_CSD_BOOT_WP,
1071				 card->ext_csd.boot_ro_lock |
1072				 EXT_CSD_BOOT_WP_B_PWR_WP_EN,
1073				 card->ext_csd.part_time);
1074		if (ret)
1075			pr_err("%s: Locking boot partition ro until next power on failed: %d\n",
1076			       md->disk->disk_name, ret);
1077		else
1078			card->ext_csd.boot_ro_lock |=
1079				EXT_CSD_BOOT_WP_B_PWR_WP_EN;
1080		break;
1081	case MMC_DRV_OP_GET_CARD_STATUS:
1082		ret = mmc_send_status(card, &status);
1083		if (!ret)
1084			ret = status;
1085		break;
1086	case MMC_DRV_OP_GET_EXT_CSD:
1087		ext_csd = mq_rq->drv_op_data;
1088		ret = mmc_get_ext_csd(card, ext_csd);
1089		break;
1090	default:
1091		pr_err("%s: unknown driver specific operation\n",
1092		       md->disk->disk_name);
1093		ret = -EINVAL;
1094		break;
1095	}
1096	mq_rq->drv_op_result = ret;
1097	blk_mq_end_request(req, ret ? BLK_STS_IOERR : BLK_STS_OK);
1098}
1099
1100static void mmc_blk_issue_discard_rq(struct mmc_queue *mq, struct request *req)
1101{
1102	struct mmc_blk_data *md = mq->blkdata;
1103	struct mmc_card *card = md->queue.card;
1104	unsigned int from, nr;
1105	int err = 0, type = MMC_BLK_DISCARD;
1106	blk_status_t status = BLK_STS_OK;
1107
1108	if (!mmc_can_erase(card)) {
1109		status = BLK_STS_NOTSUPP;
1110		goto fail;
1111	}
1112
1113	from = blk_rq_pos(req);
1114	nr = blk_rq_sectors(req);
1115
1116	do {
1117		unsigned int erase_arg = card->erase_arg;
1118
1119		if (mmc_card_broken_sd_discard(card))
1120			erase_arg = SD_ERASE_ARG;
1121
1122		err = 0;
1123		if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1124			err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1125					 INAND_CMD38_ARG_EXT_CSD,
1126					 card->erase_arg == MMC_TRIM_ARG ?
1127					 INAND_CMD38_ARG_TRIM :
1128					 INAND_CMD38_ARG_ERASE,
1129					 card->ext_csd.generic_cmd6_time);
1130		}
1131		if (!err)
1132			err = mmc_erase(card, from, nr, erase_arg);
1133	} while (err == -EIO && !mmc_blk_reset(md, card->host, type));
1134	if (err)
1135		status = BLK_STS_IOERR;
1136	else
1137		mmc_blk_reset_success(md, type);
1138fail:
1139	blk_mq_end_request(req, status);
1140}
1141
1142static void mmc_blk_issue_secdiscard_rq(struct mmc_queue *mq,
1143				       struct request *req)
1144{
1145	struct mmc_blk_data *md = mq->blkdata;
1146	struct mmc_card *card = md->queue.card;
1147	unsigned int from, nr, arg;
1148	int err = 0, type = MMC_BLK_SECDISCARD;
1149	blk_status_t status = BLK_STS_OK;
1150
1151	if (!(mmc_can_secure_erase_trim(card))) {
1152		status = BLK_STS_NOTSUPP;
1153		goto out;
1154	}
1155
1156	from = blk_rq_pos(req);
1157	nr = blk_rq_sectors(req);
1158
1159	if (mmc_can_trim(card) && !mmc_erase_group_aligned(card, from, nr))
1160		arg = MMC_SECURE_TRIM1_ARG;
1161	else
1162		arg = MMC_SECURE_ERASE_ARG;
1163
1164retry:
1165	if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1166		err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1167				 INAND_CMD38_ARG_EXT_CSD,
1168				 arg == MMC_SECURE_TRIM1_ARG ?
1169				 INAND_CMD38_ARG_SECTRIM1 :
1170				 INAND_CMD38_ARG_SECERASE,
1171				 card->ext_csd.generic_cmd6_time);
1172		if (err)
1173			goto out_retry;
1174	}
1175
1176	err = mmc_erase(card, from, nr, arg);
1177	if (err == -EIO)
1178		goto out_retry;
1179	if (err) {
1180		status = BLK_STS_IOERR;
1181		goto out;
1182	}
1183
1184	if (arg == MMC_SECURE_TRIM1_ARG) {
1185		if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1186			err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1187					 INAND_CMD38_ARG_EXT_CSD,
1188					 INAND_CMD38_ARG_SECTRIM2,
1189					 card->ext_csd.generic_cmd6_time);
1190			if (err)
1191				goto out_retry;
1192		}
1193
1194		err = mmc_erase(card, from, nr, MMC_SECURE_TRIM2_ARG);
1195		if (err == -EIO)
1196			goto out_retry;
1197		if (err) {
1198			status = BLK_STS_IOERR;
1199			goto out;
1200		}
1201	}
1202
1203out_retry:
1204	if (err && !mmc_blk_reset(md, card->host, type))
1205		goto retry;
1206	if (!err)
1207		mmc_blk_reset_success(md, type);
1208out:
1209	blk_mq_end_request(req, status);
1210}
1211
1212static void mmc_blk_issue_flush(struct mmc_queue *mq, struct request *req)
1213{
1214	struct mmc_blk_data *md = mq->blkdata;
1215	struct mmc_card *card = md->queue.card;
1216	int ret = 0;
1217
1218	ret = mmc_flush_cache(card);
1219	blk_mq_end_request(req, ret ? BLK_STS_IOERR : BLK_STS_OK);
1220}
1221
1222/*
1223 * Reformat current write as a reliable write, supporting
1224 * both legacy and the enhanced reliable write MMC cards.
1225 * In each transfer we'll handle only as much as a single
1226 * reliable write can handle, thus finish the request in
1227 * partial completions.
1228 */
1229static inline void mmc_apply_rel_rw(struct mmc_blk_request *brq,
1230				    struct mmc_card *card,
1231				    struct request *req)
1232{
1233	if (!(card->ext_csd.rel_param & EXT_CSD_WR_REL_PARAM_EN)) {
1234		/* Legacy mode imposes restrictions on transfers. */
1235		if (!IS_ALIGNED(blk_rq_pos(req), card->ext_csd.rel_sectors))
1236			brq->data.blocks = 1;
1237
1238		if (brq->data.blocks > card->ext_csd.rel_sectors)
1239			brq->data.blocks = card->ext_csd.rel_sectors;
1240		else if (brq->data.blocks < card->ext_csd.rel_sectors)
1241			brq->data.blocks = 1;
1242	}
1243}
1244
1245#define CMD_ERRORS_EXCL_OOR						\
1246	(R1_ADDRESS_ERROR |	/* Misaligned address */		\
1247	 R1_BLOCK_LEN_ERROR |	/* Transferred block length incorrect */\
1248	 R1_WP_VIOLATION |	/* Tried to write to protected block */	\
1249	 R1_CARD_ECC_FAILED |	/* Card ECC failed */			\
1250	 R1_CC_ERROR |		/* Card controller error */		\
1251	 R1_ERROR)		/* General/unknown error */
1252
1253#define CMD_ERRORS							\
1254	(CMD_ERRORS_EXCL_OOR |						\
1255	 R1_OUT_OF_RANGE)	/* Command argument out of range */	\
1256
1257static void mmc_blk_eval_resp_error(struct mmc_blk_request *brq)
1258{
1259	u32 val;
1260
1261	/*
1262	 * Per the SD specification(physical layer version 4.10)[1],
1263	 * section 4.3.3, it explicitly states that "When the last
1264	 * block of user area is read using CMD18, the host should
1265	 * ignore OUT_OF_RANGE error that may occur even the sequence
1266	 * is correct". And JESD84-B51 for eMMC also has a similar
1267	 * statement on section 6.8.3.
1268	 *
1269	 * Multiple block read/write could be done by either predefined
1270	 * method, namely CMD23, or open-ending mode. For open-ending mode,
1271	 * we should ignore the OUT_OF_RANGE error as it's normal behaviour.
1272	 *
1273	 * However the spec[1] doesn't tell us whether we should also
1274	 * ignore that for predefined method. But per the spec[1], section
1275	 * 4.15 Set Block Count Command, it says"If illegal block count
1276	 * is set, out of range error will be indicated during read/write
1277	 * operation (For example, data transfer is stopped at user area
1278	 * boundary)." In another word, we could expect a out of range error
1279	 * in the response for the following CMD18/25. And if argument of
1280	 * CMD23 + the argument of CMD18/25 exceed the max number of blocks,
1281	 * we could also expect to get a -ETIMEDOUT or any error number from
1282	 * the host drivers due to missing data response(for write)/data(for
1283	 * read), as the cards will stop the data transfer by itself per the
1284	 * spec. So we only need to check R1_OUT_OF_RANGE for open-ending mode.
1285	 */
1286
1287	if (!brq->stop.error) {
1288		bool oor_with_open_end;
1289		/* If there is no error yet, check R1 response */
1290
1291		val = brq->stop.resp[0] & CMD_ERRORS;
1292		oor_with_open_end = val & R1_OUT_OF_RANGE && !brq->mrq.sbc;
1293
1294		if (val && !oor_with_open_end)
1295			brq->stop.error = -EIO;
1296	}
1297}
1298
1299static void mmc_blk_data_prep(struct mmc_queue *mq, struct mmc_queue_req *mqrq,
1300			      int recovery_mode, bool *do_rel_wr_p,
1301			      bool *do_data_tag_p)
1302{
1303	struct mmc_blk_data *md = mq->blkdata;
1304	struct mmc_card *card = md->queue.card;
1305	struct mmc_blk_request *brq = &mqrq->brq;
1306	struct request *req = mmc_queue_req_to_req(mqrq);
1307	bool do_rel_wr, do_data_tag;
1308
1309	/*
1310	 * Reliable writes are used to implement Forced Unit Access and
1311	 * are supported only on MMCs.
1312	 */
1313	do_rel_wr = (req->cmd_flags & REQ_FUA) &&
1314		    rq_data_dir(req) == WRITE &&
1315		    (md->flags & MMC_BLK_REL_WR);
1316
1317	memset(brq, 0, sizeof(struct mmc_blk_request));
1318
1319	brq->mrq.data = &brq->data;
1320	brq->mrq.tag = req->tag;
1321
1322	brq->stop.opcode = MMC_STOP_TRANSMISSION;
1323	brq->stop.arg = 0;
1324
1325	if (rq_data_dir(req) == READ) {
1326		brq->data.flags = MMC_DATA_READ;
1327		brq->stop.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC;
1328	} else {
1329		brq->data.flags = MMC_DATA_WRITE;
1330		brq->stop.flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC;
1331	}
1332
1333	brq->data.blksz = 512;
1334	brq->data.blocks = blk_rq_sectors(req);
1335	brq->data.blk_addr = blk_rq_pos(req);
1336
1337	/*
1338	 * The command queue supports 2 priorities: "high" (1) and "simple" (0).
1339	 * The eMMC will give "high" priority tasks priority over "simple"
1340	 * priority tasks. Here we always set "simple" priority by not setting
1341	 * MMC_DATA_PRIO.
1342	 */
1343
1344	/*
1345	 * The block layer doesn't support all sector count
1346	 * restrictions, so we need to be prepared for too big
1347	 * requests.
1348	 */
1349	if (brq->data.blocks > card->host->max_blk_count)
1350		brq->data.blocks = card->host->max_blk_count;
1351
1352	if (brq->data.blocks > 1) {
1353		/*
1354		 * Some SD cards in SPI mode return a CRC error or even lock up
1355		 * completely when trying to read the last block using a
1356		 * multiblock read command.
1357		 */
1358		if (mmc_host_is_spi(card->host) && (rq_data_dir(req) == READ) &&
1359		    (blk_rq_pos(req) + blk_rq_sectors(req) ==
1360		     get_capacity(md->disk)))
1361			brq->data.blocks--;
1362
1363		/*
1364		 * After a read error, we redo the request one (native) sector
1365		 * at a time in order to accurately determine which
1366		 * sectors can be read successfully.
1367		 */
1368		if (recovery_mode)
1369			brq->data.blocks = queue_physical_block_size(mq->queue) >> 9;
1370
1371		/*
1372		 * Some controllers have HW issues while operating
1373		 * in multiple I/O mode
1374		 */
1375		if (card->host->ops->multi_io_quirk)
1376			brq->data.blocks = card->host->ops->multi_io_quirk(card,
1377						(rq_data_dir(req) == READ) ?
1378						MMC_DATA_READ : MMC_DATA_WRITE,
1379						brq->data.blocks);
1380	}
1381
1382	if (do_rel_wr) {
1383		mmc_apply_rel_rw(brq, card, req);
1384		brq->data.flags |= MMC_DATA_REL_WR;
1385	}
1386
1387	/*
1388	 * Data tag is used only during writing meta data to speed
1389	 * up write and any subsequent read of this meta data
1390	 */
1391	do_data_tag = card->ext_csd.data_tag_unit_size &&
1392		      (req->cmd_flags & REQ_META) &&
1393		      (rq_data_dir(req) == WRITE) &&
1394		      ((brq->data.blocks * brq->data.blksz) >=
1395		       card->ext_csd.data_tag_unit_size);
1396
1397	if (do_data_tag)
1398		brq->data.flags |= MMC_DATA_DAT_TAG;
1399
1400	mmc_set_data_timeout(&brq->data, card);
1401
1402	brq->data.sg = mqrq->sg;
1403	brq->data.sg_len = mmc_queue_map_sg(mq, mqrq);
1404
1405	/*
1406	 * Adjust the sg list so it is the same size as the
1407	 * request.
1408	 */
1409	if (brq->data.blocks != blk_rq_sectors(req)) {
1410		int i, data_size = brq->data.blocks << 9;
1411		struct scatterlist *sg;
1412
1413		for_each_sg(brq->data.sg, sg, brq->data.sg_len, i) {
1414			data_size -= sg->length;
1415			if (data_size <= 0) {
1416				sg->length += data_size;
1417				i++;
1418				break;
1419			}
1420		}
1421		brq->data.sg_len = i;
1422	}
1423
1424	if (do_rel_wr_p)
1425		*do_rel_wr_p = do_rel_wr;
1426
1427	if (do_data_tag_p)
1428		*do_data_tag_p = do_data_tag;
1429}
1430
1431#define MMC_CQE_RETRIES 2
1432
1433static void mmc_blk_cqe_complete_rq(struct mmc_queue *mq, struct request *req)
1434{
1435	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1436	struct mmc_request *mrq = &mqrq->brq.mrq;
1437	struct request_queue *q = req->q;
1438	struct mmc_host *host = mq->card->host;
1439	enum mmc_issue_type issue_type = mmc_issue_type(mq, req);
1440	unsigned long flags;
1441	bool put_card;
1442	int err;
1443
1444	mmc_cqe_post_req(host, mrq);
1445
1446	if (mrq->cmd && mrq->cmd->error)
1447		err = mrq->cmd->error;
1448	else if (mrq->data && mrq->data->error)
1449		err = mrq->data->error;
1450	else
1451		err = 0;
1452
1453	if (err) {
1454		if (mqrq->retries++ < MMC_CQE_RETRIES)
1455			blk_mq_requeue_request(req, true);
1456		else
1457			blk_mq_end_request(req, BLK_STS_IOERR);
1458	} else if (mrq->data) {
1459		if (blk_update_request(req, BLK_STS_OK, mrq->data->bytes_xfered))
1460			blk_mq_requeue_request(req, true);
1461		else
1462			__blk_mq_end_request(req, BLK_STS_OK);
1463	} else if (mq->in_recovery) {
1464		blk_mq_requeue_request(req, true);
1465	} else {
1466		blk_mq_end_request(req, BLK_STS_OK);
1467	}
1468
1469	spin_lock_irqsave(&mq->lock, flags);
1470
1471	mq->in_flight[issue_type] -= 1;
1472
1473	put_card = (mmc_tot_in_flight(mq) == 0);
1474
1475	mmc_cqe_check_busy(mq);
1476
1477	spin_unlock_irqrestore(&mq->lock, flags);
1478
1479	if (!mq->cqe_busy)
1480		blk_mq_run_hw_queues(q, true);
1481
1482	if (put_card)
1483		mmc_put_card(mq->card, &mq->ctx);
1484}
1485
1486void mmc_blk_cqe_recovery(struct mmc_queue *mq)
1487{
1488	struct mmc_card *card = mq->card;
1489	struct mmc_host *host = card->host;
1490	int err;
1491
1492	pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
1493
1494	err = mmc_cqe_recovery(host);
1495	if (err)
1496		mmc_blk_reset(mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
1497	mmc_blk_reset_success(mq->blkdata, MMC_BLK_CQE_RECOVERY);
1498
1499	pr_debug("%s: CQE recovery done\n", mmc_hostname(host));
1500}
1501
1502static void mmc_blk_cqe_req_done(struct mmc_request *mrq)
1503{
1504	struct mmc_queue_req *mqrq = container_of(mrq, struct mmc_queue_req,
1505						  brq.mrq);
1506	struct request *req = mmc_queue_req_to_req(mqrq);
1507	struct request_queue *q = req->q;
1508	struct mmc_queue *mq = q->queuedata;
1509
1510	/*
1511	 * Block layer timeouts race with completions which means the normal
1512	 * completion path cannot be used during recovery.
1513	 */
1514	if (mq->in_recovery)
1515		mmc_blk_cqe_complete_rq(mq, req);
1516	else if (likely(!blk_should_fake_timeout(req->q)))
1517		blk_mq_complete_request(req);
1518}
1519
1520static int mmc_blk_cqe_start_req(struct mmc_host *host, struct mmc_request *mrq)
1521{
1522	mrq->done		= mmc_blk_cqe_req_done;
1523	mrq->recovery_notifier	= mmc_cqe_recovery_notifier;
1524
1525	return mmc_cqe_start_req(host, mrq);
1526}
1527
1528static struct mmc_request *mmc_blk_cqe_prep_dcmd(struct mmc_queue_req *mqrq,
1529						 struct request *req)
1530{
1531	struct mmc_blk_request *brq = &mqrq->brq;
1532
1533	memset(brq, 0, sizeof(*brq));
1534
1535	brq->mrq.cmd = &brq->cmd;
1536	brq->mrq.tag = req->tag;
1537
1538	return &brq->mrq;
1539}
1540
1541static int mmc_blk_cqe_issue_flush(struct mmc_queue *mq, struct request *req)
1542{
1543	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1544	struct mmc_request *mrq = mmc_blk_cqe_prep_dcmd(mqrq, req);
1545
1546	mrq->cmd->opcode = MMC_SWITCH;
1547	mrq->cmd->arg = (MMC_SWITCH_MODE_WRITE_BYTE << 24) |
1548			(EXT_CSD_FLUSH_CACHE << 16) |
1549			(1 << 8) |
1550			EXT_CSD_CMD_SET_NORMAL;
1551	mrq->cmd->flags = MMC_CMD_AC | MMC_RSP_R1B;
1552
1553	return mmc_blk_cqe_start_req(mq->card->host, mrq);
1554}
1555
1556static int mmc_blk_hsq_issue_rw_rq(struct mmc_queue *mq, struct request *req)
1557{
1558	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1559	struct mmc_host *host = mq->card->host;
1560	int err;
1561
1562	mmc_blk_rw_rq_prep(mqrq, mq->card, 0, mq);
1563	mqrq->brq.mrq.done = mmc_blk_hsq_req_done;
1564	mmc_pre_req(host, &mqrq->brq.mrq);
1565
1566	err = mmc_cqe_start_req(host, &mqrq->brq.mrq);
1567	if (err)
1568		mmc_post_req(host, &mqrq->brq.mrq, err);
1569
1570	return err;
1571}
1572
1573static int mmc_blk_cqe_issue_rw_rq(struct mmc_queue *mq, struct request *req)
1574{
1575	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1576	struct mmc_host *host = mq->card->host;
1577
1578	if (host->hsq_enabled)
1579		return mmc_blk_hsq_issue_rw_rq(mq, req);
1580
1581	mmc_blk_data_prep(mq, mqrq, 0, NULL, NULL);
1582
1583	return mmc_blk_cqe_start_req(mq->card->host, &mqrq->brq.mrq);
1584}
1585
1586static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
1587			       struct mmc_card *card,
1588			       int recovery_mode,
1589			       struct mmc_queue *mq)
1590{
1591	u32 readcmd, writecmd;
1592	struct mmc_blk_request *brq = &mqrq->brq;
1593	struct request *req = mmc_queue_req_to_req(mqrq);
1594	struct mmc_blk_data *md = mq->blkdata;
1595	bool do_rel_wr, do_data_tag;
1596
1597	mmc_blk_data_prep(mq, mqrq, recovery_mode, &do_rel_wr, &do_data_tag);
1598
1599	brq->mrq.cmd = &brq->cmd;
1600
1601	brq->cmd.arg = blk_rq_pos(req);
1602	if (!mmc_card_blockaddr(card))
1603		brq->cmd.arg <<= 9;
1604	brq->cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC;
1605
1606	if (brq->data.blocks > 1 || do_rel_wr) {
1607		/* SPI multiblock writes terminate using a special
1608		 * token, not a STOP_TRANSMISSION request.
1609		 */
1610		if (!mmc_host_is_spi(card->host) ||
1611		    rq_data_dir(req) == READ)
1612			brq->mrq.stop = &brq->stop;
1613		readcmd = MMC_READ_MULTIPLE_BLOCK;
1614		writecmd = MMC_WRITE_MULTIPLE_BLOCK;
1615	} else {
1616		brq->mrq.stop = NULL;
1617		readcmd = MMC_READ_SINGLE_BLOCK;
1618		writecmd = MMC_WRITE_BLOCK;
1619	}
1620	brq->cmd.opcode = rq_data_dir(req) == READ ? readcmd : writecmd;
1621
1622	/*
1623	 * Pre-defined multi-block transfers are preferable to
1624	 * open ended-ones (and necessary for reliable writes).
1625	 * However, it is not sufficient to just send CMD23,
1626	 * and avoid the final CMD12, as on an error condition
1627	 * CMD12 (stop) needs to be sent anyway. This, coupled
1628	 * with Auto-CMD23 enhancements provided by some
1629	 * hosts, means that the complexity of dealing
1630	 * with this is best left to the host. If CMD23 is
1631	 * supported by card and host, we'll fill sbc in and let
1632	 * the host deal with handling it correctly. This means
1633	 * that for hosts that don't expose MMC_CAP_CMD23, no
1634	 * change of behavior will be observed.
1635	 *
1636	 * N.B: Some MMC cards experience perf degradation.
1637	 * We'll avoid using CMD23-bounded multiblock writes for
1638	 * these, while retaining features like reliable writes.
1639	 */
1640	if ((md->flags & MMC_BLK_CMD23) && mmc_op_multi(brq->cmd.opcode) &&
1641	    (do_rel_wr || !(card->quirks & MMC_QUIRK_BLK_NO_CMD23) ||
1642	     do_data_tag)) {
1643		brq->sbc.opcode = MMC_SET_BLOCK_COUNT;
1644		brq->sbc.arg = brq->data.blocks |
1645			(do_rel_wr ? (1 << 31) : 0) |
1646			(do_data_tag ? (1 << 29) : 0);
1647		brq->sbc.flags = MMC_RSP_R1 | MMC_CMD_AC;
1648		brq->mrq.sbc = &brq->sbc;
1649	}
1650}
1651
1652#define MMC_MAX_RETRIES		5
1653#define MMC_DATA_RETRIES	2
1654#define MMC_NO_RETRIES		(MMC_MAX_RETRIES + 1)
1655
1656static int mmc_blk_send_stop(struct mmc_card *card, unsigned int timeout)
1657{
1658	struct mmc_command cmd = {
1659		.opcode = MMC_STOP_TRANSMISSION,
1660		.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC,
1661		/* Some hosts wait for busy anyway, so provide a busy timeout */
1662		.busy_timeout = timeout,
1663	};
1664
1665	return mmc_wait_for_cmd(card->host, &cmd, 5);
1666}
1667
1668static int mmc_blk_fix_state(struct mmc_card *card, struct request *req)
1669{
1670	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1671	struct mmc_blk_request *brq = &mqrq->brq;
1672	unsigned int timeout = mmc_blk_data_timeout_ms(card->host, &brq->data);
1673	int err;
1674
1675	mmc_retune_hold_now(card->host);
1676
1677	mmc_blk_send_stop(card, timeout);
1678
1679	err = card_busy_detect(card, timeout, NULL);
1680
1681	mmc_retune_release(card->host);
1682
1683	return err;
1684}
1685
1686#define MMC_READ_SINGLE_RETRIES	2
1687
1688/* Single (native) sector read during recovery */
1689static void mmc_blk_read_single(struct mmc_queue *mq, struct request *req)
1690{
1691	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1692	struct mmc_request *mrq = &mqrq->brq.mrq;
1693	struct mmc_card *card = mq->card;
1694	struct mmc_host *host = card->host;
1695	blk_status_t error = BLK_STS_OK;
1696	size_t bytes_per_read = queue_physical_block_size(mq->queue);
1697
1698	do {
1699		u32 status;
1700		int err;
1701		int retries = 0;
1702
1703		while (retries++ <= MMC_READ_SINGLE_RETRIES) {
1704			mmc_blk_rw_rq_prep(mqrq, card, 1, mq);
1705
1706			mmc_wait_for_req(host, mrq);
1707
1708			err = mmc_send_status(card, &status);
1709			if (err)
1710				goto error_exit;
1711
1712			if (!mmc_host_is_spi(host) &&
1713			    !mmc_ready_for_data(status)) {
1714				err = mmc_blk_fix_state(card, req);
1715				if (err)
1716					goto error_exit;
1717			}
1718
1719			if (!mrq->cmd->error)
1720				break;
1721		}
1722
1723		if (mrq->cmd->error ||
1724		    mrq->data->error ||
1725		    (!mmc_host_is_spi(host) &&
1726		     (mrq->cmd->resp[0] & CMD_ERRORS || status & CMD_ERRORS)))
1727			error = BLK_STS_IOERR;
1728		else
1729			error = BLK_STS_OK;
1730
1731	} while (blk_update_request(req, error, bytes_per_read));
1732
1733	return;
1734
1735error_exit:
1736	mrq->data->bytes_xfered = 0;
1737	blk_update_request(req, BLK_STS_IOERR, bytes_per_read);
1738	/* Let it try the remaining request again */
1739	if (mqrq->retries > MMC_MAX_RETRIES - 1)
1740		mqrq->retries = MMC_MAX_RETRIES - 1;
1741}
1742
1743static inline bool mmc_blk_oor_valid(struct mmc_blk_request *brq)
1744{
1745	return !!brq->mrq.sbc;
1746}
1747
1748static inline u32 mmc_blk_stop_err_bits(struct mmc_blk_request *brq)
1749{
1750	return mmc_blk_oor_valid(brq) ? CMD_ERRORS : CMD_ERRORS_EXCL_OOR;
1751}
1752
1753/*
1754 * Check for errors the host controller driver might not have seen such as
1755 * response mode errors or invalid card state.
1756 */
1757static bool mmc_blk_status_error(struct request *req, u32 status)
1758{
1759	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1760	struct mmc_blk_request *brq = &mqrq->brq;
1761	struct mmc_queue *mq = req->q->queuedata;
1762	u32 stop_err_bits;
1763
1764	if (mmc_host_is_spi(mq->card->host))
1765		return false;
1766
1767	stop_err_bits = mmc_blk_stop_err_bits(brq);
1768
1769	return brq->cmd.resp[0]  & CMD_ERRORS    ||
1770	       brq->stop.resp[0] & stop_err_bits ||
1771	       status            & stop_err_bits ||
1772	       (rq_data_dir(req) == WRITE && !mmc_ready_for_data(status));
1773}
1774
1775static inline bool mmc_blk_cmd_started(struct mmc_blk_request *brq)
1776{
1777	return !brq->sbc.error && !brq->cmd.error &&
1778	       !(brq->cmd.resp[0] & CMD_ERRORS);
1779}
1780
1781/*
1782 * Requests are completed by mmc_blk_mq_complete_rq() which sets simple
1783 * policy:
1784 * 1. A request that has transferred at least some data is considered
1785 * successful and will be requeued if there is remaining data to
1786 * transfer.
1787 * 2. Otherwise the number of retries is incremented and the request
1788 * will be requeued if there are remaining retries.
1789 * 3. Otherwise the request will be errored out.
1790 * That means mmc_blk_mq_complete_rq() is controlled by bytes_xfered and
1791 * mqrq->retries. So there are only 4 possible actions here:
1792 *	1. do not accept the bytes_xfered value i.e. set it to zero
1793 *	2. change mqrq->retries to determine the number of retries
1794 *	3. try to reset the card
1795 *	4. read one sector at a time
1796 */
1797static void mmc_blk_mq_rw_recovery(struct mmc_queue *mq, struct request *req)
1798{
1799	int type = rq_data_dir(req) == READ ? MMC_BLK_READ : MMC_BLK_WRITE;
1800	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1801	struct mmc_blk_request *brq = &mqrq->brq;
1802	struct mmc_blk_data *md = mq->blkdata;
1803	struct mmc_card *card = mq->card;
1804	u32 status;
1805	u32 blocks;
1806	int err;
1807
1808	/*
1809	 * Some errors the host driver might not have seen. Set the number of
1810	 * bytes transferred to zero in that case.
1811	 */
1812	err = __mmc_send_status(card, &status, 0);
1813	if (err || mmc_blk_status_error(req, status))
1814		brq->data.bytes_xfered = 0;
1815
1816	mmc_retune_release(card->host);
1817
1818	/*
1819	 * Try again to get the status. This also provides an opportunity for
1820	 * re-tuning.
1821	 */
1822	if (err)
1823		err = __mmc_send_status(card, &status, 0);
1824
1825	/*
1826	 * Nothing more to do after the number of bytes transferred has been
1827	 * updated and there is no card.
1828	 */
1829	if (err && mmc_detect_card_removed(card->host))
1830		return;
1831
1832	/* Try to get back to "tran" state */
1833	if (!mmc_host_is_spi(mq->card->host) &&
1834	    (err || !mmc_ready_for_data(status)))
1835		err = mmc_blk_fix_state(mq->card, req);
1836
1837	/*
1838	 * Special case for SD cards where the card might record the number of
1839	 * blocks written.
1840	 */
1841	if (!err && mmc_blk_cmd_started(brq) && mmc_card_sd(card) &&
1842	    rq_data_dir(req) == WRITE) {
1843		if (mmc_sd_num_wr_blocks(card, &blocks))
1844			brq->data.bytes_xfered = 0;
1845		else
1846			brq->data.bytes_xfered = blocks << 9;
1847	}
1848
1849	/* Reset if the card is in a bad state */
1850	if (!mmc_host_is_spi(mq->card->host) &&
1851	    err && mmc_blk_reset(md, card->host, type)) {
1852		pr_err("%s: recovery failed!\n", req->rq_disk->disk_name);
1853		mqrq->retries = MMC_NO_RETRIES;
1854		return;
1855	}
1856
1857	/*
1858	 * If anything was done, just return and if there is anything remaining
1859	 * on the request it will get requeued.
1860	 */
1861	if (brq->data.bytes_xfered)
1862		return;
1863
1864	/* Reset before last retry */
1865	if (mqrq->retries + 1 == MMC_MAX_RETRIES)
1866		mmc_blk_reset(md, card->host, type);
1867
1868	/* Command errors fail fast, so use all MMC_MAX_RETRIES */
1869	if (brq->sbc.error || brq->cmd.error)
1870		return;
1871
1872	/* Reduce the remaining retries for data errors */
1873	if (mqrq->retries < MMC_MAX_RETRIES - MMC_DATA_RETRIES) {
1874		mqrq->retries = MMC_MAX_RETRIES - MMC_DATA_RETRIES;
1875		return;
1876	}
1877
1878	if (rq_data_dir(req) == READ && brq->data.blocks >
1879			queue_physical_block_size(mq->queue) >> 9) {
1880		/* Read one (native) sector at a time */
1881		mmc_blk_read_single(mq, req);
1882		return;
1883	}
1884}
1885
1886static inline bool mmc_blk_rq_error(struct mmc_blk_request *brq)
1887{
1888	mmc_blk_eval_resp_error(brq);
1889
1890	return brq->sbc.error || brq->cmd.error || brq->stop.error ||
1891	       brq->data.error || brq->cmd.resp[0] & CMD_ERRORS;
1892}
1893
1894static int mmc_blk_card_busy(struct mmc_card *card, struct request *req)
1895{
1896	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1897	u32 status = 0;
1898	int err;
1899
1900	if (mmc_host_is_spi(card->host) || rq_data_dir(req) == READ)
1901		return 0;
1902
1903	err = card_busy_detect(card, MMC_BLK_TIMEOUT_MS, &status);
1904
1905	/*
1906	 * Do not assume data transferred correctly if there are any error bits
1907	 * set.
1908	 */
1909	if (status & mmc_blk_stop_err_bits(&mqrq->brq)) {
1910		mqrq->brq.data.bytes_xfered = 0;
1911		err = err ? err : -EIO;
1912	}
1913
1914	/* Copy the exception bit so it will be seen later on */
1915	if (mmc_card_mmc(card) && status & R1_EXCEPTION_EVENT)
1916		mqrq->brq.cmd.resp[0] |= R1_EXCEPTION_EVENT;
1917
1918	return err;
1919}
1920
1921static inline void mmc_blk_rw_reset_success(struct mmc_queue *mq,
1922					    struct request *req)
1923{
1924	int type = rq_data_dir(req) == READ ? MMC_BLK_READ : MMC_BLK_WRITE;
1925
1926	mmc_blk_reset_success(mq->blkdata, type);
1927}
1928
1929static void mmc_blk_mq_complete_rq(struct mmc_queue *mq, struct request *req)
1930{
1931	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
1932	unsigned int nr_bytes = mqrq->brq.data.bytes_xfered;
1933
1934	if (nr_bytes) {
1935		if (blk_update_request(req, BLK_STS_OK, nr_bytes))
1936			blk_mq_requeue_request(req, true);
1937		else
1938			__blk_mq_end_request(req, BLK_STS_OK);
1939	} else if (!blk_rq_bytes(req)) {
1940		__blk_mq_end_request(req, BLK_STS_IOERR);
1941	} else if (mqrq->retries++ < MMC_MAX_RETRIES) {
1942		blk_mq_requeue_request(req, true);
1943	} else {
1944		if (mmc_card_removed(mq->card))
1945			req->rq_flags |= RQF_QUIET;
1946		blk_mq_end_request(req, BLK_STS_IOERR);
1947	}
1948}
1949
1950static bool mmc_blk_urgent_bkops_needed(struct mmc_queue *mq,
1951					struct mmc_queue_req *mqrq)
1952{
1953	return mmc_card_mmc(mq->card) && !mmc_host_is_spi(mq->card->host) &&
1954	       (mqrq->brq.cmd.resp[0] & R1_EXCEPTION_EVENT ||
1955		mqrq->brq.stop.resp[0] & R1_EXCEPTION_EVENT);
1956}
1957
1958static void mmc_blk_urgent_bkops(struct mmc_queue *mq,
1959				 struct mmc_queue_req *mqrq)
1960{
1961	if (mmc_blk_urgent_bkops_needed(mq, mqrq))
1962		mmc_run_bkops(mq->card);
1963}
1964
1965static void mmc_blk_hsq_req_done(struct mmc_request *mrq)
1966{
1967	struct mmc_queue_req *mqrq =
1968		container_of(mrq, struct mmc_queue_req, brq.mrq);
1969	struct request *req = mmc_queue_req_to_req(mqrq);
1970	struct request_queue *q = req->q;
1971	struct mmc_queue *mq = q->queuedata;
1972	struct mmc_host *host = mq->card->host;
1973	unsigned long flags;
1974
1975	if (mmc_blk_rq_error(&mqrq->brq) ||
1976	    mmc_blk_urgent_bkops_needed(mq, mqrq)) {
1977		spin_lock_irqsave(&mq->lock, flags);
1978		mq->recovery_needed = true;
1979		mq->recovery_req = req;
1980		spin_unlock_irqrestore(&mq->lock, flags);
1981
1982		host->cqe_ops->cqe_recovery_start(host);
1983
1984		schedule_work(&mq->recovery_work);
1985		return;
1986	}
1987
1988	mmc_blk_rw_reset_success(mq, req);
1989
1990	/*
1991	 * Block layer timeouts race with completions which means the normal
1992	 * completion path cannot be used during recovery.
1993	 */
1994	if (mq->in_recovery)
1995		mmc_blk_cqe_complete_rq(mq, req);
1996	else if (likely(!blk_should_fake_timeout(req->q)))
1997		blk_mq_complete_request(req);
1998}
1999
2000void mmc_blk_mq_complete(struct request *req)
2001{
2002	struct mmc_queue *mq = req->q->queuedata;
2003
2004	if (mq->use_cqe)
2005		mmc_blk_cqe_complete_rq(mq, req);
2006	else if (likely(!blk_should_fake_timeout(req->q)))
2007		mmc_blk_mq_complete_rq(mq, req);
2008}
2009
2010static void mmc_blk_mq_poll_completion(struct mmc_queue *mq,
2011				       struct request *req)
2012{
2013	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2014	struct mmc_host *host = mq->card->host;
2015
2016	if (mmc_blk_rq_error(&mqrq->brq) ||
2017	    mmc_blk_card_busy(mq->card, req)) {
2018		mmc_blk_mq_rw_recovery(mq, req);
2019	} else {
2020		mmc_blk_rw_reset_success(mq, req);
2021		mmc_retune_release(host);
2022	}
2023
2024	mmc_blk_urgent_bkops(mq, mqrq);
2025}
2026
2027static void mmc_blk_mq_dec_in_flight(struct mmc_queue *mq, enum mmc_issue_type issue_type)
2028{
2029	unsigned long flags;
2030	bool put_card;
2031
2032	spin_lock_irqsave(&mq->lock, flags);
2033
2034	mq->in_flight[issue_type] -= 1;
2035
2036	put_card = (mmc_tot_in_flight(mq) == 0);
2037
2038	spin_unlock_irqrestore(&mq->lock, flags);
2039
2040	if (put_card)
2041		mmc_put_card(mq->card, &mq->ctx);
2042}
2043
2044static void mmc_blk_mq_post_req(struct mmc_queue *mq, struct request *req)
2045{
2046	enum mmc_issue_type issue_type = mmc_issue_type(mq, req);
2047	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2048	struct mmc_request *mrq = &mqrq->brq.mrq;
2049	struct mmc_host *host = mq->card->host;
2050
2051	mmc_post_req(host, mrq, 0);
2052
2053	/*
2054	 * Block layer timeouts race with completions which means the normal
2055	 * completion path cannot be used during recovery.
2056	 */
2057	if (mq->in_recovery)
2058		mmc_blk_mq_complete_rq(mq, req);
2059	else if (likely(!blk_should_fake_timeout(req->q)))
2060		blk_mq_complete_request(req);
2061
2062	mmc_blk_mq_dec_in_flight(mq, issue_type);
2063}
2064
2065void mmc_blk_mq_recovery(struct mmc_queue *mq)
2066{
2067	struct request *req = mq->recovery_req;
2068	struct mmc_host *host = mq->card->host;
2069	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2070
2071	mq->recovery_req = NULL;
2072	mq->rw_wait = false;
2073
2074	if (mmc_blk_rq_error(&mqrq->brq)) {
2075		mmc_retune_hold_now(host);
2076		mmc_blk_mq_rw_recovery(mq, req);
2077	}
2078
2079	mmc_blk_urgent_bkops(mq, mqrq);
2080
2081	mmc_blk_mq_post_req(mq, req);
2082}
2083
2084static void mmc_blk_mq_complete_prev_req(struct mmc_queue *mq,
2085					 struct request **prev_req)
2086{
2087	if (mmc_host_done_complete(mq->card->host))
2088		return;
2089
2090	mutex_lock(&mq->complete_lock);
2091
2092	if (!mq->complete_req)
2093		goto out_unlock;
2094
2095	mmc_blk_mq_poll_completion(mq, mq->complete_req);
2096
2097	if (prev_req)
2098		*prev_req = mq->complete_req;
2099	else
2100		mmc_blk_mq_post_req(mq, mq->complete_req);
2101
2102	mq->complete_req = NULL;
2103
2104out_unlock:
2105	mutex_unlock(&mq->complete_lock);
2106}
2107
2108void mmc_blk_mq_complete_work(struct work_struct *work)
2109{
2110	struct mmc_queue *mq = container_of(work, struct mmc_queue,
2111					    complete_work);
2112
2113	mmc_blk_mq_complete_prev_req(mq, NULL);
2114}
2115
2116static void mmc_blk_mq_req_done(struct mmc_request *mrq)
2117{
2118	struct mmc_queue_req *mqrq = container_of(mrq, struct mmc_queue_req,
2119						  brq.mrq);
2120	struct request *req = mmc_queue_req_to_req(mqrq);
2121	struct request_queue *q = req->q;
2122	struct mmc_queue *mq = q->queuedata;
2123	struct mmc_host *host = mq->card->host;
2124	unsigned long flags;
2125
2126	if (!mmc_host_done_complete(host)) {
2127		bool waiting;
2128
2129		/*
2130		 * We cannot complete the request in this context, so record
2131		 * that there is a request to complete, and that a following
2132		 * request does not need to wait (although it does need to
2133		 * complete complete_req first).
2134		 */
2135		spin_lock_irqsave(&mq->lock, flags);
2136		mq->complete_req = req;
2137		mq->rw_wait = false;
2138		waiting = mq->waiting;
2139		spin_unlock_irqrestore(&mq->lock, flags);
2140
2141		/*
2142		 * If 'waiting' then the waiting task will complete this
2143		 * request, otherwise queue a work to do it. Note that
2144		 * complete_work may still race with the dispatch of a following
2145		 * request.
2146		 */
2147		if (waiting)
2148			wake_up(&mq->wait);
2149		else
2150			queue_work(mq->card->complete_wq, &mq->complete_work);
2151
2152		return;
2153	}
2154
2155	/* Take the recovery path for errors or urgent background operations */
2156	if (mmc_blk_rq_error(&mqrq->brq) ||
2157	    mmc_blk_urgent_bkops_needed(mq, mqrq)) {
2158		spin_lock_irqsave(&mq->lock, flags);
2159		mq->recovery_needed = true;
2160		mq->recovery_req = req;
2161		spin_unlock_irqrestore(&mq->lock, flags);
2162		wake_up(&mq->wait);
2163		schedule_work(&mq->recovery_work);
2164		return;
2165	}
2166
2167	mmc_blk_rw_reset_success(mq, req);
2168
2169	mq->rw_wait = false;
2170	wake_up(&mq->wait);
2171
2172	mmc_blk_mq_post_req(mq, req);
2173}
2174
2175static bool mmc_blk_rw_wait_cond(struct mmc_queue *mq, int *err)
2176{
2177	unsigned long flags;
2178	bool done;
2179
2180	/*
2181	 * Wait while there is another request in progress, but not if recovery
2182	 * is needed. Also indicate whether there is a request waiting to start.
2183	 */
2184	spin_lock_irqsave(&mq->lock, flags);
2185	if (mq->recovery_needed) {
2186		*err = -EBUSY;
2187		done = true;
2188	} else {
2189		done = !mq->rw_wait;
2190	}
2191	mq->waiting = !done;
2192	spin_unlock_irqrestore(&mq->lock, flags);
2193
2194	return done;
2195}
2196
2197static int mmc_blk_rw_wait(struct mmc_queue *mq, struct request **prev_req)
2198{
2199	int err = 0;
2200
2201	wait_event(mq->wait, mmc_blk_rw_wait_cond(mq, &err));
2202
2203	/* Always complete the previous request if there is one */
2204	mmc_blk_mq_complete_prev_req(mq, prev_req);
2205
2206	return err;
2207}
2208
2209static int mmc_blk_mq_issue_rw_rq(struct mmc_queue *mq,
2210				  struct request *req)
2211{
2212	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(req);
2213	struct mmc_host *host = mq->card->host;
2214	struct request *prev_req = NULL;
2215	int err = 0;
2216
2217	mmc_blk_rw_rq_prep(mqrq, mq->card, 0, mq);
2218
2219	mqrq->brq.mrq.done = mmc_blk_mq_req_done;
2220
2221	mmc_pre_req(host, &mqrq->brq.mrq);
2222
2223	err = mmc_blk_rw_wait(mq, &prev_req);
2224	if (err)
2225		goto out_post_req;
2226
2227	mq->rw_wait = true;
2228
2229	err = mmc_start_request(host, &mqrq->brq.mrq);
2230
2231	if (prev_req)
2232		mmc_blk_mq_post_req(mq, prev_req);
2233
2234	if (err)
2235		mq->rw_wait = false;
2236
2237	/* Release re-tuning here where there is no synchronization required */
2238	if (err || mmc_host_done_complete(host))
2239		mmc_retune_release(host);
2240
2241out_post_req:
2242	if (err)
2243		mmc_post_req(host, &mqrq->brq.mrq, err);
2244
2245	return err;
2246}
2247
2248static int mmc_blk_wait_for_idle(struct mmc_queue *mq, struct mmc_host *host)
2249{
2250	if (mq->use_cqe)
2251		return host->cqe_ops->cqe_wait_for_idle(host);
2252
2253	return mmc_blk_rw_wait(mq, NULL);
2254}
2255
2256enum mmc_issued mmc_blk_mq_issue_rq(struct mmc_queue *mq, struct request *req)
2257{
2258	struct mmc_blk_data *md = mq->blkdata;
2259	struct mmc_card *card = md->queue.card;
2260	struct mmc_host *host = card->host;
2261	int ret;
2262
2263	ret = mmc_blk_part_switch(card, md->part_type);
2264	if (ret)
2265		return MMC_REQ_FAILED_TO_START;
2266
2267	switch (mmc_issue_type(mq, req)) {
2268	case MMC_ISSUE_SYNC:
2269		ret = mmc_blk_wait_for_idle(mq, host);
2270		if (ret)
2271			return MMC_REQ_BUSY;
2272		switch (req_op(req)) {
2273		case REQ_OP_DRV_IN:
2274		case REQ_OP_DRV_OUT:
2275			mmc_blk_issue_drv_op(mq, req);
2276			break;
2277		case REQ_OP_DISCARD:
2278			mmc_blk_issue_discard_rq(mq, req);
2279			break;
2280		case REQ_OP_SECURE_ERASE:
2281			mmc_blk_issue_secdiscard_rq(mq, req);
2282			break;
2283		case REQ_OP_FLUSH:
2284			mmc_blk_issue_flush(mq, req);
2285			break;
2286		default:
2287			WARN_ON_ONCE(1);
2288			return MMC_REQ_FAILED_TO_START;
2289		}
2290		return MMC_REQ_FINISHED;
2291	case MMC_ISSUE_DCMD:
2292	case MMC_ISSUE_ASYNC:
2293		switch (req_op(req)) {
2294		case REQ_OP_FLUSH:
2295			if (!mmc_cache_enabled(host)) {
2296				blk_mq_end_request(req, BLK_STS_OK);
2297				return MMC_REQ_FINISHED;
2298			}
2299			ret = mmc_blk_cqe_issue_flush(mq, req);
2300			break;
2301		case REQ_OP_READ:
2302		case REQ_OP_WRITE:
2303			if (mq->use_cqe)
2304				ret = mmc_blk_cqe_issue_rw_rq(mq, req);
2305			else
2306				ret = mmc_blk_mq_issue_rw_rq(mq, req);
2307			break;
2308		default:
2309			WARN_ON_ONCE(1);
2310			ret = -EINVAL;
2311		}
2312		if (!ret)
2313			return MMC_REQ_STARTED;
2314		return ret == -EBUSY ? MMC_REQ_BUSY : MMC_REQ_FAILED_TO_START;
2315	default:
2316		WARN_ON_ONCE(1);
2317		return MMC_REQ_FAILED_TO_START;
2318	}
2319}
2320
2321static inline int mmc_blk_readonly(struct mmc_card *card)
2322{
2323	return mmc_card_readonly(card) ||
2324	       !(card->csd.cmdclass & CCC_BLOCK_WRITE);
2325}
2326
2327static struct mmc_blk_data *mmc_blk_alloc_req(struct mmc_card *card,
2328					      struct device *parent,
2329					      sector_t size,
2330					      bool default_ro,
2331					      const char *subname,
2332					      int area_type)
2333{
2334	struct mmc_blk_data *md;
2335	int devidx, ret;
2336
2337	devidx = ida_simple_get(&mmc_blk_ida, 0, max_devices, GFP_KERNEL);
2338	if (devidx < 0) {
2339		/*
2340		 * We get -ENOSPC because there are no more any available
2341		 * devidx. The reason may be that, either userspace haven't yet
2342		 * unmounted the partitions, which postpones mmc_blk_release()
2343		 * from being called, or the device has more partitions than
2344		 * what we support.
2345		 */
2346		if (devidx == -ENOSPC)
2347			dev_err(mmc_dev(card->host),
2348				"no more device IDs available\n");
2349
2350		return ERR_PTR(devidx);
2351	}
2352
2353	md = kzalloc(sizeof(struct mmc_blk_data), GFP_KERNEL);
2354	if (!md) {
2355		ret = -ENOMEM;
2356		goto out;
2357	}
2358
2359	md->area_type = area_type;
2360
2361	/*
2362	 * Set the read-only status based on the supported commands
2363	 * and the write protect switch.
2364	 */
2365	md->read_only = mmc_blk_readonly(card);
2366
2367	md->disk = alloc_disk(perdev_minors);
2368	if (md->disk == NULL) {
2369		ret = -ENOMEM;
2370		goto err_kfree;
2371	}
2372
2373	INIT_LIST_HEAD(&md->part);
2374	INIT_LIST_HEAD(&md->rpmbs);
2375	md->usage = 1;
2376
2377	ret = mmc_init_queue(&md->queue, card);
2378	if (ret)
2379		goto err_putdisk;
2380
2381	md->queue.blkdata = md;
2382
2383	/*
2384	 * Keep an extra reference to the queue so that we can shutdown the
2385	 * queue (i.e. call blk_cleanup_queue()) while there are still
2386	 * references to the 'md'. The corresponding blk_put_queue() is in
2387	 * mmc_blk_put().
2388	 */
2389	if (!blk_get_queue(md->queue.queue)) {
2390		mmc_cleanup_queue(&md->queue);
2391		ret = -ENODEV;
2392		goto err_putdisk;
2393	}
2394
2395	md->disk->major	= MMC_BLOCK_MAJOR;
2396	md->disk->first_minor = devidx * perdev_minors;
2397	md->disk->fops = &mmc_bdops;
2398	md->disk->private_data = md;
2399	md->disk->queue = md->queue.queue;
2400	md->parent = parent;
2401	set_disk_ro(md->disk, md->read_only || default_ro);
2402	md->disk->flags = GENHD_FL_EXT_DEVT;
2403	if (area_type & (MMC_BLK_DATA_AREA_RPMB | MMC_BLK_DATA_AREA_BOOT))
2404		md->disk->flags |= GENHD_FL_NO_PART_SCAN
2405				   | GENHD_FL_SUPPRESS_PARTITION_INFO;
2406
2407	/*
2408	 * As discussed on lkml, GENHD_FL_REMOVABLE should:
2409	 *
2410	 * - be set for removable media with permanent block devices
2411	 * - be unset for removable block devices with permanent media
2412	 *
2413	 * Since MMC block devices clearly fall under the second
2414	 * case, we do not set GENHD_FL_REMOVABLE.  Userspace
2415	 * should use the block device creation/destruction hotplug
2416	 * messages to tell when the card is present.
2417	 */
2418
2419	snprintf(md->disk->disk_name, sizeof(md->disk->disk_name),
2420		 "mmcblk%u%s", card->host->index, subname ? subname : "");
2421
2422	set_capacity(md->disk, size);
2423
2424	if (mmc_host_cmd23(card->host)) {
2425		if ((mmc_card_mmc(card) &&
2426		     card->csd.mmca_vsn >= CSD_SPEC_VER_3) ||
2427		    (mmc_card_sd(card) &&
2428		     card->scr.cmds & SD_SCR_CMD23_SUPPORT))
2429			md->flags |= MMC_BLK_CMD23;
2430	}
2431
2432	if (mmc_card_mmc(card) &&
2433	    md->flags & MMC_BLK_CMD23 &&
2434	    ((card->ext_csd.rel_param & EXT_CSD_WR_REL_PARAM_EN) ||
2435	     card->ext_csd.rel_sectors)) {
2436		md->flags |= MMC_BLK_REL_WR;
2437		blk_queue_write_cache(md->queue.queue, true, true);
2438	}
2439
2440	return md;
2441
2442 err_putdisk:
2443	put_disk(md->disk);
2444 err_kfree:
2445	kfree(md);
2446 out:
2447	ida_simple_remove(&mmc_blk_ida, devidx);
2448	return ERR_PTR(ret);
2449}
2450
2451static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
2452{
2453	sector_t size;
2454
2455	if (!mmc_card_sd(card) && mmc_card_blockaddr(card)) {
2456		/*
2457		 * The EXT_CSD sector count is in number or 512 byte
2458		 * sectors.
2459		 */
2460		size = card->ext_csd.sectors;
2461	} else {
2462		/*
2463		 * The CSD capacity field is in units of read_blkbits.
2464		 * set_capacity takes units of 512 bytes.
2465		 */
2466		size = (typeof(sector_t))card->csd.capacity
2467			<< (card->csd.read_blkbits - 9);
2468	}
2469
2470	return mmc_blk_alloc_req(card, &card->dev, size, false, NULL,
2471					MMC_BLK_DATA_AREA_MAIN);
2472}
2473
2474static int mmc_blk_alloc_part(struct mmc_card *card,
2475			      struct mmc_blk_data *md,
2476			      unsigned int part_type,
2477			      sector_t size,
2478			      bool default_ro,
2479			      const char *subname,
2480			      int area_type)
2481{
2482	char cap_str[10];
2483	struct mmc_blk_data *part_md;
2484
2485	part_md = mmc_blk_alloc_req(card, disk_to_dev(md->disk), size, default_ro,
2486				    subname, area_type);
2487	if (IS_ERR(part_md))
2488		return PTR_ERR(part_md);
2489	part_md->part_type = part_type;
2490	list_add(&part_md->part, &md->part);
2491
2492	string_get_size((u64)get_capacity(part_md->disk), 512, STRING_UNITS_2,
2493			cap_str, sizeof(cap_str));
2494	pr_info("%s: %s %s partition %u %s\n",
2495	       part_md->disk->disk_name, mmc_card_id(card),
2496	       mmc_card_name(card), part_md->part_type, cap_str);
2497	return 0;
2498}
2499
2500/**
2501 * mmc_rpmb_ioctl() - ioctl handler for the RPMB chardev
2502 * @filp: the character device file
2503 * @cmd: the ioctl() command
2504 * @arg: the argument from userspace
2505 *
2506 * This will essentially just redirect the ioctl()s coming in over to
2507 * the main block device spawning the RPMB character device.
2508 */
2509static long mmc_rpmb_ioctl(struct file *filp, unsigned int cmd,
2510			   unsigned long arg)
2511{
2512	struct mmc_rpmb_data *rpmb = filp->private_data;
2513	int ret;
2514
2515	switch (cmd) {
2516	case MMC_IOC_CMD:
2517		ret = mmc_blk_ioctl_cmd(rpmb->md,
2518					(struct mmc_ioc_cmd __user *)arg,
2519					rpmb);
2520		break;
2521	case MMC_IOC_MULTI_CMD:
2522		ret = mmc_blk_ioctl_multi_cmd(rpmb->md,
2523					(struct mmc_ioc_multi_cmd __user *)arg,
2524					rpmb);
2525		break;
2526	default:
2527		ret = -EINVAL;
2528		break;
2529	}
2530
2531	return ret;
2532}
2533
2534#ifdef CONFIG_COMPAT
2535static long mmc_rpmb_ioctl_compat(struct file *filp, unsigned int cmd,
2536			      unsigned long arg)
2537{
2538	return mmc_rpmb_ioctl(filp, cmd, (unsigned long)compat_ptr(arg));
2539}
2540#endif
2541
2542static int mmc_rpmb_chrdev_open(struct inode *inode, struct file *filp)
2543{
2544	struct mmc_rpmb_data *rpmb = container_of(inode->i_cdev,
2545						  struct mmc_rpmb_data, chrdev);
2546
2547	get_device(&rpmb->dev);
2548	filp->private_data = rpmb;
2549	mmc_blk_get(rpmb->md->disk);
2550
2551	return nonseekable_open(inode, filp);
2552}
2553
2554static int mmc_rpmb_chrdev_release(struct inode *inode, struct file *filp)
2555{
2556	struct mmc_rpmb_data *rpmb = container_of(inode->i_cdev,
2557						  struct mmc_rpmb_data, chrdev);
2558
2559	mmc_blk_put(rpmb->md);
2560	put_device(&rpmb->dev);
2561
2562	return 0;
2563}
2564
2565static const struct file_operations mmc_rpmb_fileops = {
2566	.release = mmc_rpmb_chrdev_release,
2567	.open = mmc_rpmb_chrdev_open,
2568	.owner = THIS_MODULE,
2569	.llseek = no_llseek,
2570	.unlocked_ioctl = mmc_rpmb_ioctl,
2571#ifdef CONFIG_COMPAT
2572	.compat_ioctl = mmc_rpmb_ioctl_compat,
2573#endif
2574};
2575
2576static void mmc_blk_rpmb_device_release(struct device *dev)
2577{
2578	struct mmc_rpmb_data *rpmb = dev_get_drvdata(dev);
2579
2580	ida_simple_remove(&mmc_rpmb_ida, rpmb->id);
2581	kfree(rpmb);
2582}
2583
2584static int mmc_blk_alloc_rpmb_part(struct mmc_card *card,
2585				   struct mmc_blk_data *md,
2586				   unsigned int part_index,
2587				   sector_t size,
2588				   const char *subname)
2589{
2590	int devidx, ret;
2591	char rpmb_name[DISK_NAME_LEN];
2592	char cap_str[10];
2593	struct mmc_rpmb_data *rpmb;
2594
2595	/* This creates the minor number for the RPMB char device */
2596	devidx = ida_simple_get(&mmc_rpmb_ida, 0, max_devices, GFP_KERNEL);
2597	if (devidx < 0)
2598		return devidx;
2599
2600	rpmb = kzalloc(sizeof(*rpmb), GFP_KERNEL);
2601	if (!rpmb) {
2602		ida_simple_remove(&mmc_rpmb_ida, devidx);
2603		return -ENOMEM;
2604	}
2605
2606	snprintf(rpmb_name, sizeof(rpmb_name),
2607		 "mmcblk%u%s", card->host->index, subname ? subname : "");
2608
2609	rpmb->id = devidx;
2610	rpmb->part_index = part_index;
2611	rpmb->dev.init_name = rpmb_name;
2612	rpmb->dev.bus = &mmc_rpmb_bus_type;
2613	rpmb->dev.devt = MKDEV(MAJOR(mmc_rpmb_devt), rpmb->id);
2614	rpmb->dev.parent = &card->dev;
2615	rpmb->dev.release = mmc_blk_rpmb_device_release;
2616	device_initialize(&rpmb->dev);
2617	dev_set_drvdata(&rpmb->dev, rpmb);
2618	rpmb->md = md;
2619
2620	cdev_init(&rpmb->chrdev, &mmc_rpmb_fileops);
2621	rpmb->chrdev.owner = THIS_MODULE;
2622	ret = cdev_device_add(&rpmb->chrdev, &rpmb->dev);
2623	if (ret) {
2624		pr_err("%s: could not add character device\n", rpmb_name);
2625		goto out_put_device;
2626	}
2627
2628	list_add(&rpmb->node, &md->rpmbs);
2629
2630	string_get_size((u64)size, 512, STRING_UNITS_2,
2631			cap_str, sizeof(cap_str));
2632
2633	pr_info("%s: %s %s partition %u %s, chardev (%d:%d)\n",
2634		rpmb_name, mmc_card_id(card),
2635		mmc_card_name(card), EXT_CSD_PART_CONFIG_ACC_RPMB, cap_str,
2636		MAJOR(mmc_rpmb_devt), rpmb->id);
2637
2638	return 0;
2639
2640out_put_device:
2641	put_device(&rpmb->dev);
2642	return ret;
2643}
2644
2645static void mmc_blk_remove_rpmb_part(struct mmc_rpmb_data *rpmb)
2646
2647{
2648	cdev_device_del(&rpmb->chrdev, &rpmb->dev);
2649	put_device(&rpmb->dev);
2650}
2651
2652/* MMC Physical partitions consist of two boot partitions and
2653 * up to four general purpose partitions.
2654 * For each partition enabled in EXT_CSD a block device will be allocatedi
2655 * to provide access to the partition.
2656 */
2657
2658static int mmc_blk_alloc_parts(struct mmc_card *card, struct mmc_blk_data *md)
2659{
2660	int idx, ret;
2661
2662	if (!mmc_card_mmc(card))
2663		return 0;
2664
2665	for (idx = 0; idx < card->nr_parts; idx++) {
2666		if (card->part[idx].area_type & MMC_BLK_DATA_AREA_RPMB) {
2667			/*
2668			 * RPMB partitions does not provide block access, they
2669			 * are only accessed using ioctl():s. Thus create
2670			 * special RPMB block devices that do not have a
2671			 * backing block queue for these.
2672			 */
2673			ret = mmc_blk_alloc_rpmb_part(card, md,
2674				card->part[idx].part_cfg,
2675				card->part[idx].size >> 9,
2676				card->part[idx].name);
2677			if (ret)
2678				return ret;
2679		} else if (card->part[idx].size) {
2680			ret = mmc_blk_alloc_part(card, md,
2681				card->part[idx].part_cfg,
2682				card->part[idx].size >> 9,
2683				card->part[idx].force_ro,
2684				card->part[idx].name,
2685				card->part[idx].area_type);
2686			if (ret)
2687				return ret;
2688		}
2689	}
2690
2691	return 0;
2692}
2693
2694static void mmc_blk_remove_req(struct mmc_blk_data *md)
2695{
2696	struct mmc_card *card;
2697
2698	if (md) {
2699		/*
2700		 * Flush remaining requests and free queues. It
2701		 * is freeing the queue that stops new requests
2702		 * from being accepted.
2703		 */
2704		card = md->queue.card;
2705		if (md->disk->flags & GENHD_FL_UP) {
2706			device_remove_file(disk_to_dev(md->disk), &md->force_ro);
2707			if ((md->area_type & MMC_BLK_DATA_AREA_BOOT) &&
2708					card->ext_csd.boot_ro_lockable)
2709				device_remove_file(disk_to_dev(md->disk),
2710					&md->power_ro_lock);
2711
2712			del_gendisk(md->disk);
2713		}
2714		mmc_cleanup_queue(&md->queue);
2715		mmc_blk_put(md);
2716	}
2717}
2718
2719static void mmc_blk_remove_parts(struct mmc_card *card,
2720				 struct mmc_blk_data *md)
2721{
2722	struct list_head *pos, *q;
2723	struct mmc_blk_data *part_md;
2724	struct mmc_rpmb_data *rpmb;
2725
2726	/* Remove RPMB partitions */
2727	list_for_each_safe(pos, q, &md->rpmbs) {
2728		rpmb = list_entry(pos, struct mmc_rpmb_data, node);
2729		list_del(pos);
2730		mmc_blk_remove_rpmb_part(rpmb);
2731	}
2732	/* Remove block partitions */
2733	list_for_each_safe(pos, q, &md->part) {
2734		part_md = list_entry(pos, struct mmc_blk_data, part);
2735		list_del(pos);
2736		mmc_blk_remove_req(part_md);
2737	}
2738}
2739
2740static int mmc_add_disk(struct mmc_blk_data *md)
2741{
2742	int ret;
2743	struct mmc_card *card = md->queue.card;
2744
2745	device_add_disk(md->parent, md->disk, NULL);
2746	md->force_ro.show = force_ro_show;
2747	md->force_ro.store = force_ro_store;
2748	sysfs_attr_init(&md->force_ro.attr);
2749	md->force_ro.attr.name = "force_ro";
2750	md->force_ro.attr.mode = S_IRUGO | S_IWUSR;
2751	ret = device_create_file(disk_to_dev(md->disk), &md->force_ro);
2752	if (ret)
2753		goto force_ro_fail;
2754
2755	if ((md->area_type & MMC_BLK_DATA_AREA_BOOT) &&
2756	     card->ext_csd.boot_ro_lockable) {
2757		umode_t mode;
2758
2759		if (card->ext_csd.boot_ro_lock & EXT_CSD_BOOT_WP_B_PWR_WP_DIS)
2760			mode = S_IRUGO;
2761		else
2762			mode = S_IRUGO | S_IWUSR;
2763
2764		md->power_ro_lock.show = power_ro_lock_show;
2765		md->power_ro_lock.store = power_ro_lock_store;
2766		sysfs_attr_init(&md->power_ro_lock.attr);
2767		md->power_ro_lock.attr.mode = mode;
2768		md->power_ro_lock.attr.name =
2769					"ro_lock_until_next_power_on";
2770		ret = device_create_file(disk_to_dev(md->disk),
2771				&md->power_ro_lock);
2772		if (ret)
2773			goto power_ro_lock_fail;
2774	}
2775	return ret;
2776
2777power_ro_lock_fail:
2778	device_remove_file(disk_to_dev(md->disk), &md->force_ro);
2779force_ro_fail:
2780	del_gendisk(md->disk);
2781
2782	return ret;
2783}
2784
2785#ifdef CONFIG_DEBUG_FS
2786
2787static int mmc_dbg_card_status_get(void *data, u64 *val)
2788{
2789	struct mmc_card *card = data;
2790	struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
2791	struct mmc_queue *mq = &md->queue;
2792	struct request *req;
2793	int ret;
2794
2795	/* Ask the block layer about the card status */
2796	req = blk_get_request(mq->queue, REQ_OP_DRV_IN, 0);
2797	if (IS_ERR(req))
2798		return PTR_ERR(req);
2799	req_to_mmc_queue_req(req)->drv_op = MMC_DRV_OP_GET_CARD_STATUS;
2800	req_to_mmc_queue_req(req)->drv_op_result = -EIO;
2801	blk_execute_rq(mq->queue, NULL, req, 0);
2802	ret = req_to_mmc_queue_req(req)->drv_op_result;
2803	if (ret >= 0) {
2804		*val = ret;
2805		ret = 0;
2806	}
2807	blk_put_request(req);
2808
2809	return ret;
2810}
2811DEFINE_DEBUGFS_ATTRIBUTE(mmc_dbg_card_status_fops, mmc_dbg_card_status_get,
2812			 NULL, "%08llx\n");
2813
2814/* That is two digits * 512 + 1 for newline */
2815#define EXT_CSD_STR_LEN 1025
2816
2817static int mmc_ext_csd_open(struct inode *inode, struct file *filp)
2818{
2819	struct mmc_card *card = inode->i_private;
2820	struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
2821	struct mmc_queue *mq = &md->queue;
2822	struct request *req;
2823	char *buf;
2824	ssize_t n = 0;
2825	u8 *ext_csd;
2826	int err, i;
2827
2828	buf = kmalloc(EXT_CSD_STR_LEN + 1, GFP_KERNEL);
2829	if (!buf)
2830		return -ENOMEM;
2831
2832	/* Ask the block layer for the EXT CSD */
2833	req = blk_get_request(mq->queue, REQ_OP_DRV_IN, 0);
2834	if (IS_ERR(req)) {
2835		err = PTR_ERR(req);
2836		goto out_free;
2837	}
2838	req_to_mmc_queue_req(req)->drv_op = MMC_DRV_OP_GET_EXT_CSD;
2839	req_to_mmc_queue_req(req)->drv_op_result = -EIO;
2840	req_to_mmc_queue_req(req)->drv_op_data = &ext_csd;
2841	blk_execute_rq(mq->queue, NULL, req, 0);
2842	err = req_to_mmc_queue_req(req)->drv_op_result;
2843	blk_put_request(req);
2844	if (err) {
2845		pr_err("FAILED %d\n", err);
2846		goto out_free;
2847	}
2848
2849	for (i = 0; i < 512; i++)
2850		n += sprintf(buf + n, "%02x", ext_csd[i]);
2851	n += sprintf(buf + n, "\n");
2852
2853	if (n != EXT_CSD_STR_LEN) {
2854		err = -EINVAL;
2855		kfree(ext_csd);
2856		goto out_free;
2857	}
2858
2859	filp->private_data = buf;
2860	kfree(ext_csd);
2861	return 0;
2862
2863out_free:
2864	kfree(buf);
2865	return err;
2866}
2867
2868static ssize_t mmc_ext_csd_read(struct file *filp, char __user *ubuf,
2869				size_t cnt, loff_t *ppos)
2870{
2871	char *buf = filp->private_data;
2872
2873	return simple_read_from_buffer(ubuf, cnt, ppos,
2874				       buf, EXT_CSD_STR_LEN);
2875}
2876
2877static int mmc_ext_csd_release(struct inode *inode, struct file *file)
2878{
2879	kfree(file->private_data);
2880	return 0;
2881}
2882
2883static const struct file_operations mmc_dbg_ext_csd_fops = {
2884	.open		= mmc_ext_csd_open,
2885	.read		= mmc_ext_csd_read,
2886	.release	= mmc_ext_csd_release,
2887	.llseek		= default_llseek,
2888};
2889
2890static int mmc_blk_add_debugfs(struct mmc_card *card, struct mmc_blk_data *md)
2891{
2892	struct dentry *root;
2893
2894	if (!card->debugfs_root)
2895		return 0;
2896
2897	root = card->debugfs_root;
2898
2899	if (mmc_card_mmc(card) || mmc_card_sd(card)) {
2900		md->status_dentry =
2901			debugfs_create_file_unsafe("status", 0400, root,
2902						   card,
2903						   &mmc_dbg_card_status_fops);
2904		if (!md->status_dentry)
2905			return -EIO;
2906	}
2907
2908	if (mmc_card_mmc(card)) {
2909		md->ext_csd_dentry =
2910			debugfs_create_file("ext_csd", S_IRUSR, root, card,
2911					    &mmc_dbg_ext_csd_fops);
2912		if (!md->ext_csd_dentry)
2913			return -EIO;
2914	}
2915
2916	return 0;
2917}
2918
2919static void mmc_blk_remove_debugfs(struct mmc_card *card,
2920				   struct mmc_blk_data *md)
2921{
2922	if (!card->debugfs_root)
2923		return;
2924
2925	if (!IS_ERR_OR_NULL(md->status_dentry)) {
2926		debugfs_remove(md->status_dentry);
2927		md->status_dentry = NULL;
2928	}
2929
2930	if (!IS_ERR_OR_NULL(md->ext_csd_dentry)) {
2931		debugfs_remove(md->ext_csd_dentry);
2932		md->ext_csd_dentry = NULL;
2933	}
2934}
2935
2936#else
2937
2938static int mmc_blk_add_debugfs(struct mmc_card *card, struct mmc_blk_data *md)
2939{
2940	return 0;
2941}
2942
2943static void mmc_blk_remove_debugfs(struct mmc_card *card,
2944				   struct mmc_blk_data *md)
2945{
2946}
2947
2948#endif /* CONFIG_DEBUG_FS */
2949
2950static int mmc_blk_probe(struct mmc_card *card)
2951{
2952	struct mmc_blk_data *md, *part_md;
2953	char cap_str[10];
2954
2955	/*
2956	 * Check that the card supports the command class(es) we need.
2957	 */
2958	if (!(card->csd.cmdclass & CCC_BLOCK_READ))
2959		return -ENODEV;
2960
2961	mmc_fixup_device(card, mmc_blk_fixups);
2962
2963	card->complete_wq = alloc_workqueue("mmc_complete",
2964					WQ_MEM_RECLAIM | WQ_HIGHPRI, 0);
2965	if (unlikely(!card->complete_wq)) {
2966		pr_err("Failed to create mmc completion workqueue");
2967		return -ENOMEM;
2968	}
2969
2970	md = mmc_blk_alloc(card);
2971	if (IS_ERR(md))
2972		return PTR_ERR(md);
2973
2974	string_get_size((u64)get_capacity(md->disk), 512, STRING_UNITS_2,
2975			cap_str, sizeof(cap_str));
2976	pr_info("%s: %s %s %s %s\n",
2977		md->disk->disk_name, mmc_card_id(card), mmc_card_name(card),
2978		cap_str, md->read_only ? "(ro)" : "");
2979
2980	if (mmc_blk_alloc_parts(card, md))
2981		goto out;
2982
2983	dev_set_drvdata(&card->dev, md);
2984
2985	if (mmc_add_disk(md))
2986		goto out;
2987
2988	list_for_each_entry(part_md, &md->part, part) {
2989		if (mmc_add_disk(part_md))
2990			goto out;
2991	}
2992
2993	/* Add two debugfs entries */
2994	mmc_blk_add_debugfs(card, md);
2995
2996	pm_runtime_set_autosuspend_delay(&card->dev, 3000);
2997	pm_runtime_use_autosuspend(&card->dev);
2998
2999	/*
3000	 * Don't enable runtime PM for SD-combo cards here. Leave that
3001	 * decision to be taken during the SDIO init sequence instead.
3002	 */
3003	if (card->type != MMC_TYPE_SD_COMBO) {
3004		pm_runtime_set_active(&card->dev);
3005		pm_runtime_enable(&card->dev);
3006	}
3007
3008	return 0;
3009
3010 out:
3011	mmc_blk_remove_parts(card, md);
3012	mmc_blk_remove_req(md);
3013	return 0;
3014}
3015
3016static void mmc_blk_remove(struct mmc_card *card)
3017{
3018	struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
3019
3020	mmc_blk_remove_debugfs(card, md);
3021	mmc_blk_remove_parts(card, md);
3022	pm_runtime_get_sync(&card->dev);
3023	if (md->part_curr != md->part_type) {
3024		mmc_claim_host(card->host);
3025		mmc_blk_part_switch(card, md->part_type);
3026		mmc_release_host(card->host);
3027	}
3028	if (card->type != MMC_TYPE_SD_COMBO)
3029		pm_runtime_disable(&card->dev);
3030	pm_runtime_put_noidle(&card->dev);
3031	mmc_blk_remove_req(md);
3032	dev_set_drvdata(&card->dev, NULL);
3033	destroy_workqueue(card->complete_wq);
3034}
3035
3036static int _mmc_blk_suspend(struct mmc_card *card)
3037{
3038	struct mmc_blk_data *part_md;
3039	struct mmc_blk_data *md = dev_get_drvdata(&card->dev);
3040
3041	if (md) {
3042		mmc_queue_suspend(&md->queue);
3043		list_for_each_entry(part_md, &md->part, part) {
3044			mmc_queue_suspend(&part_md->queue);
3045		}
3046	}
3047	return 0;
3048}
3049
3050static void mmc_blk_shutdown(struct mmc_card *card)
3051{
3052	_mmc_blk_suspend(card);
3053}
3054
3055#ifdef CONFIG_PM_SLEEP
3056static int mmc_blk_suspend(struct device *dev)
3057{
3058	struct mmc_card *card = mmc_dev_to_card(dev);
3059
3060	return _mmc_blk_suspend(card);
3061}
3062
3063static int mmc_blk_resume(struct device *dev)
3064{
3065	struct mmc_blk_data *part_md;
3066	struct mmc_blk_data *md = dev_get_drvdata(dev);
3067
3068	if (md) {
3069		/*
3070		 * Resume involves the card going into idle state,
3071		 * so current partition is always the main one.
3072		 */
3073		md->part_curr = md->part_type;
3074		mmc_queue_resume(&md->queue);
3075		list_for_each_entry(part_md, &md->part, part) {
3076			mmc_queue_resume(&part_md->queue);
3077		}
3078	}
3079	return 0;
3080}
3081#endif
3082
3083static SIMPLE_DEV_PM_OPS(mmc_blk_pm_ops, mmc_blk_suspend, mmc_blk_resume);
3084
3085static struct mmc_driver mmc_driver = {
3086	.drv		= {
3087		.name	= "mmcblk",
3088		.pm	= &mmc_blk_pm_ops,
3089	},
3090	.probe		= mmc_blk_probe,
3091	.remove		= mmc_blk_remove,
3092	.shutdown	= mmc_blk_shutdown,
3093};
3094
3095static int __init mmc_blk_init(void)
3096{
3097	int res;
3098
3099	res  = bus_register(&mmc_rpmb_bus_type);
3100	if (res < 0) {
3101		pr_err("mmcblk: could not register RPMB bus type\n");
3102		return res;
3103	}
3104	res = alloc_chrdev_region(&mmc_rpmb_devt, 0, MAX_DEVICES, "rpmb");
3105	if (res < 0) {
3106		pr_err("mmcblk: failed to allocate rpmb chrdev region\n");
3107		goto out_bus_unreg;
3108	}
3109
3110	if (perdev_minors != CONFIG_MMC_BLOCK_MINORS)
3111		pr_info("mmcblk: using %d minors per device\n", perdev_minors);
3112
3113	max_devices = min(MAX_DEVICES, (1 << MINORBITS) / perdev_minors);
3114
3115	res = register_blkdev(MMC_BLOCK_MAJOR, "mmc");
3116	if (res)
3117		goto out_chrdev_unreg;
3118
3119	res = mmc_register_driver(&mmc_driver);
3120	if (res)
3121		goto out_blkdev_unreg;
3122
3123	return 0;
3124
3125out_blkdev_unreg:
3126	unregister_blkdev(MMC_BLOCK_MAJOR, "mmc");
3127out_chrdev_unreg:
3128	unregister_chrdev_region(mmc_rpmb_devt, MAX_DEVICES);
3129out_bus_unreg:
3130	bus_unregister(&mmc_rpmb_bus_type);
3131	return res;
3132}
3133
3134static void __exit mmc_blk_exit(void)
3135{
3136	mmc_unregister_driver(&mmc_driver);
3137	unregister_blkdev(MMC_BLOCK_MAJOR, "mmc");
3138	unregister_chrdev_region(mmc_rpmb_devt, MAX_DEVICES);
3139	bus_unregister(&mmc_rpmb_bus_type);
3140}
3141
3142module_init(mmc_blk_init);
3143module_exit(mmc_blk_exit);
3144
3145MODULE_LICENSE("GPL");
3146MODULE_DESCRIPTION("Multimedia Card (MMC) block device driver");
3147