1	// SPDX-License-Identifier: GPL-2.0
2	/*
3	* Block driver for media (i.e., flash cards)
4	*
5	* Copyright 2002 Hewlett-Packard Company
6	* Copyright 2005-2008 Pierre Ossman
7	*
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.
11	*
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.
15	*
16	* Many thanks to Alessandro Rubini and Jonathan Corbet!
17	*
18	* Author: Andrew Christian
19	* 28 May 2002
20	*/
21	#include <linux/moduleparam.h>
22	#include <linux/module.h>
23	#include <linux/init.h>
24
25	#include <linux/kernel.h>
26	#include <linux/fs.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>
43
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>
49
50	#include <linux/uaccess.h>
51
52	#include "queue.h"
53	#include "block.h"
54	#include "core.h"
55	#include "card.h"
56	#include "crypto.h"
57	#include "host.h"
58	#include "bus.h"
59	#include "mmc_ops.h"
60	#include "quirks.h"
61	#include "sd_ops.h"
62
63	MODULE_ALIAS("mmc:block");
64	#ifdef MODULE_PARAM_PREFIX
65	#undef MODULE_PARAM_PREFIX
66	#endif
67	#define MODULE_PARAM_PREFIX "mmcblk."
68
69	/*
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
73	* ample.
74	*/
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)
78
79	static DEFINE_MUTEX(block_mutex);
80
81	/*
82	* The defaults come from config options but can be overriden by module
83	* or bootarg options.
84	*/
85	static int perdev_minors = CONFIG_MMC_BLOCK_MINORS;
86
87	/*
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.
91	*/
92	static int max_devices;
93
94	#define MAX_DEVICES 256
95
96	static DEFINE_IDA(mmc_blk_ida);
97	static DEFINE_IDA(mmc_rpmb_ida);
98
99	struct mmc_blk_busy_data {
100	struct mmc_card *card;
101	u32 status;
102	};
103
104	/*
105	* There is one mmc_blk_data per slot.
106	*/
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;
113
114	unsigned int flags;
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 */
117
118	struct kref kref;
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)
128
129	/*
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.
133	*/
134	unsigned int part_curr;
135	#define MMC_BLK_PART_INVALID UINT_MAX /* Unknown partition active */
136	int area_type;
137
138	/* debugfs files (only in main mmc_blk_data) */
139	struct dentry *status_dentry;
140	struct dentry *ext_csd_dentry;
141	};
142
143	/* Device type for RPMB character devices */
144	static dev_t mmc_rpmb_devt;
145
146	/* Bus type for RPMB character devices */
147	static const struct bus_type mmc_rpmb_bus_type = {
148	.name = "mmc_rpmb",
149	};
150
151	/**
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
159	*/
160	struct mmc_rpmb_data {
161	struct device dev;
162	struct cdev chrdev;
163	int id;
164	unsigned int part_index;
165	struct mmc_blk_data *md;
166	struct list_head node;
167	};
168
169	static DEFINE_MUTEX(open_lock);
170
171	module_param(perdev_minors, int, 0444);
172	MODULE_PARM_DESC(perdev_minors, "Minors numbers to allocate per device");
173
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,
178	int recovery_mode,
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);
182	static int mmc_blk_busy_cb(void *cb_data, bool *busy);
183
184	static struct mmc_blk_data *mmc_blk_get(struct gendisk *disk)
185	{
186	struct mmc_blk_data *md;
187
188	mutex_lock(&open_lock);
189	md = disk->private_data;
190	if (md && !kref_get_unless_zero(kref: &md->kref))
191	md = NULL;
192	mutex_unlock(lock: &open_lock);
193
194	return md;
195	}
196
197	static inline int mmc_get_devidx(struct gendisk *disk)
198	{
199	int devidx = disk->first_minor / perdev_minors;
200	return devidx;
201	}
202
203	static void mmc_blk_kref_release(struct kref *ref)
204	{
205	struct mmc_blk_data *md = container_of(ref, struct mmc_blk_data, kref);
206	int devidx;
207
208	devidx = mmc_get_devidx(disk: md->disk);
209	ida_free(&mmc_blk_ida, id: devidx);
210
211	mutex_lock(&open_lock);
212	md->disk->private_data = NULL;
213	mutex_unlock(lock: &open_lock);
214
215	put_disk(disk: md->disk);
216	kfree(objp: md);
217	}
218
219	static void mmc_blk_put(struct mmc_blk_data *md)
220	{
221	kref_put(kref: &md->kref, release: mmc_blk_kref_release);
222	}
223
224	static ssize_t power_ro_lock_show(struct device *dev,
225	struct device_attribute *attr, char *buf)
226	{
227	int ret;
228	struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
229	struct mmc_card *card = md->queue.card;
230	int locked = 0;
231
232	if (card->ext_csd.boot_ro_lock & EXT_CSD_BOOT_WP_B_PERM_WP_EN)
233	locked = 2;
234	else if (card->ext_csd.boot_ro_lock & EXT_CSD_BOOT_WP_B_PWR_WP_EN)
235	locked = 1;
236
237	ret = snprintf(buf, PAGE_SIZE, fmt: "%d\n", locked);
238
239	mmc_blk_put(md);
240
241	return ret;
242	}
243
244	static ssize_t power_ro_lock_store(struct device *dev,
245	struct device_attribute *attr, const char *buf, size_t count)
246	{
247	int ret;
248	struct mmc_blk_data *md, *part_md;
249	struct mmc_queue *mq;
250	struct request *req;
251	unsigned long set;
252
253	if (kstrtoul(s: buf, base: 0, res: &set))
254	return -EINVAL;
255
256	if (set != 1)
257	return count;
258
259	md = mmc_blk_get(dev_to_disk(dev));
260	mq = &md->queue;
261
262	/* Dispatch locking to the block layer */
263	req = blk_mq_alloc_request(q: mq->queue, opf: REQ_OP_DRV_OUT, flags: 0);
264	if (IS_ERR(ptr: req)) {
265	count = PTR_ERR(ptr: req);
266	goto out_put;
267	}
268	req_to_mmc_queue_req(rq: req)->drv_op = MMC_DRV_OP_BOOT_WP;
269	req_to_mmc_queue_req(rq: req)->drv_op_result = -EIO;
270	blk_execute_rq(rq: req, at_head: false);
271	ret = req_to_mmc_queue_req(rq: req)->drv_op_result;
272	blk_mq_free_request(rq: req);
273
274	if (!ret) {
275	pr_info("%s: Locking boot partition ro until next power on\n",
276	md->disk->disk_name);
277	set_disk_ro(disk: md->disk, read_only: 1);
278
279	list_for_each_entry(part_md, &md->part, part)
280	if (part_md->area_type == MMC_BLK_DATA_AREA_BOOT) {
281	pr_info("%s: Locking boot partition ro until next power on\n", part_md->disk->disk_name);
282	set_disk_ro(disk: part_md->disk, read_only: 1);
283	}
284	}
285	out_put:
286	mmc_blk_put(md);
287	return count;
288	}
289
290	static DEVICE_ATTR(ro_lock_until_next_power_on, 0,
291	power_ro_lock_show, power_ro_lock_store);
292
293	static ssize_t force_ro_show(struct device *dev, struct device_attribute *attr,
294	char *buf)
295	{
296	int ret;
297	struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
298
299	ret = snprintf(buf, PAGE_SIZE, fmt: "%d\n",
300	get_disk_ro(dev_to_disk(dev)) ^
301	md->read_only);
302	mmc_blk_put(md);
303	return ret;
304	}
305
306	static ssize_t force_ro_store(struct device *dev, struct device_attribute *attr,
307	const char *buf, size_t count)
308	{
309	int ret;
310	char *end;
311	struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
312	unsigned long set = simple_strtoul(buf, &end, 0);
313	if (end == buf) {
314	ret = -EINVAL;
315	goto out;
316	}
317
318	set_disk_ro(dev_to_disk(dev), read_only: set || md->read_only);
319	ret = count;
320	out:
321	mmc_blk_put(md);
322	return ret;
323	}
324
325	static DEVICE_ATTR(force_ro, 0644, force_ro_show, force_ro_store);
326
327	static struct attribute *mmc_disk_attrs[] = {
328	&dev_attr_force_ro.attr,
329	&dev_attr_ro_lock_until_next_power_on.attr,
330	NULL,
331	};
332
333	static umode_t mmc_disk_attrs_is_visible(struct kobject *kobj,
334	struct attribute *a, int n)
335	{
336	struct device *dev = kobj_to_dev(kobj);
337	struct mmc_blk_data *md = mmc_blk_get(dev_to_disk(dev));
338	umode_t mode = a->mode;
339
340	if (a == &dev_attr_ro_lock_until_next_power_on.attr &&
341	(md->area_type & MMC_BLK_DATA_AREA_BOOT) &&
342	md->queue.card->ext_csd.boot_ro_lockable) {
343	mode = S_IRUGO;
344	if (!(md->queue.card->ext_csd.boot_ro_lock &
345	EXT_CSD_BOOT_WP_B_PWR_WP_DIS))
346	mode |= S_IWUSR;
347	}
348
349	mmc_blk_put(md);
350	return mode;
351	}
352
353	static const struct attribute_group mmc_disk_attr_group = {
354	.is_visible = mmc_disk_attrs_is_visible,
355	.attrs = mmc_disk_attrs,
356	};
357
358	static const struct attribute_group *mmc_disk_attr_groups[] = {
359	&mmc_disk_attr_group,
360	NULL,
361	};
362
363	static int mmc_blk_open(struct gendisk *disk, blk_mode_t mode)
364	{
365	struct mmc_blk_data *md = mmc_blk_get(disk);
366	int ret = -ENXIO;
367
368	mutex_lock(&block_mutex);
369	if (md) {
370	ret = 0;
371	if ((mode & BLK_OPEN_WRITE) && md->read_only) {
372	mmc_blk_put(md);
373	ret = -EROFS;
374	}
375	}
376	mutex_unlock(lock: &block_mutex);
377
378	return ret;
379	}
380
381	static void mmc_blk_release(struct gendisk *disk)
382	{
383	struct mmc_blk_data *md = disk->private_data;
384
385	mutex_lock(&block_mutex);
386	mmc_blk_put(md);
387	mutex_unlock(lock: &block_mutex);
388	}
389
390	static int
391	mmc_blk_getgeo(struct block_device *bdev, struct hd_geometry *geo)
392	{
393	geo->cylinders = get_capacity(disk: bdev->bd_disk) / (4 * 16);
394	geo->heads = 4;
395	geo->sectors = 16;
396	return 0;
397	}
398
399	struct mmc_blk_ioc_data {
400	struct mmc_ioc_cmd ic;
401	unsigned char *buf;
402	u64 buf_bytes;
403	unsigned int flags;
404	#define MMC_BLK_IOC_DROP BIT(0) /* drop this mrq */
405	#define MMC_BLK_IOC_SBC BIT(1) /* use mrq.sbc */
406
407	struct mmc_rpmb_data *rpmb;
408	};
409
410	static struct mmc_blk_ioc_data *mmc_blk_ioctl_copy_from_user(
411	struct mmc_ioc_cmd __user *user)
412	{
413	struct mmc_blk_ioc_data *idata;
414	int err;
415
416	idata = kzalloc(size: sizeof(*idata), GFP_KERNEL);
417	if (!idata) {
418	err = -ENOMEM;
419	goto out;
420	}
421
422	if (copy_from_user(to: &idata->ic, from: user, n: sizeof(idata->ic))) {
423	err = -EFAULT;
424	goto idata_err;
425	}
426
427	idata->buf_bytes = (u64) idata->ic.blksz * idata->ic.blocks;
428	if (idata->buf_bytes > MMC_IOC_MAX_BYTES) {
429	err = -EOVERFLOW;
430	goto idata_err;
431	}
432
433	if (!idata->buf_bytes) {
434	idata->buf = NULL;
435	return idata;
436	}
437
438	idata->buf = memdup_user((void __user *)(unsigned long)
439	idata->ic.data_ptr, idata->buf_bytes);
440	if (IS_ERR(ptr: idata->buf)) {
441	err = PTR_ERR(ptr: idata->buf);
442	goto idata_err;
443	}
444
445	return idata;
446
447	idata_err:
448	kfree(objp: idata);
449	out:
450	return ERR_PTR(error: err);
451	}
452
453	static int mmc_blk_ioctl_copy_to_user(struct mmc_ioc_cmd __user *ic_ptr,
454	struct mmc_blk_ioc_data *idata)
455	{
456	struct mmc_ioc_cmd *ic = &idata->ic;
457
458	if (copy_to_user(to: &(ic_ptr->response), from: ic->response,
459	n: sizeof(ic->response)))
460	return -EFAULT;
461
462	if (!idata->ic.write_flag) {
463	if (copy_to_user(to: (void __user *)(unsigned long)ic->data_ptr,
464	from: idata->buf, n: idata->buf_bytes))
465	return -EFAULT;
466	}
467
468	return 0;
469	}
470
471	static int __mmc_blk_ioctl_cmd(struct mmc_card *card, struct mmc_blk_data *md,
472	struct mmc_blk_ioc_data **idatas, int i)
473	{
474	struct mmc_command cmd = {}, sbc = {};
475	struct mmc_data data = {};
476	struct mmc_request mrq = {};
477	struct scatterlist sg;
478	bool r1b_resp;
479	unsigned int busy_timeout_ms;
480	int err;
481	unsigned int target_part;
482	struct mmc_blk_ioc_data *idata = idatas[i];
483	struct mmc_blk_ioc_data *prev_idata = NULL;
484
485	if (!card || !md || !idata)
486	return -EINVAL;
487
488	if (idata->flags & MMC_BLK_IOC_DROP)
489	return 0;
490
491	if (idata->flags & MMC_BLK_IOC_SBC && i > 0)
492	prev_idata = idatas[i - 1];
493
494	/*
495	* The RPMB accesses comes in from the character device, so we
496	* need to target these explicitly. Else we just target the
497	* partition type for the block device the ioctl() was issued
498	* on.
499	*/
500	if (idata->rpmb) {
501	/* Support multiple RPMB partitions */
502	target_part = idata->rpmb->part_index;
503	target_part |= EXT_CSD_PART_CONFIG_ACC_RPMB;
504	} else {
505	target_part = md->part_type;
506	}
507
508	cmd.opcode = idata->ic.opcode;
509	cmd.arg = idata->ic.arg;
510	cmd.flags = idata->ic.flags;
511
512	if (idata->buf_bytes) {
513	data.sg = &sg;
514	data.sg_len = 1;
515	data.blksz = idata->ic.blksz;
516	data.blocks = idata->ic.blocks;
517
518	sg_init_one(data.sg, idata->buf, idata->buf_bytes);
519
520	if (idata->ic.write_flag)
521	data.flags = MMC_DATA_WRITE;
522	else
523	data.flags = MMC_DATA_READ;
524
525	/* data.flags must already be set before doing this. */
526	mmc_set_data_timeout(data: &data, card);
527
528	/* Allow overriding the timeout_ns for empirical tuning. */
529	if (idata->ic.data_timeout_ns)
530	data.timeout_ns = idata->ic.data_timeout_ns;
531
532	mrq.data = &data;
533	}
534
535	mrq.cmd = &cmd;
536
537	err = mmc_blk_part_switch(card, part_type: target_part);
538	if (err)
539	return err;
540
541	if (idata->ic.is_acmd) {
542	err = mmc_app_cmd(host: card->host, card);
543	if (err)
544	return err;
545	}
546
547	if (idata->rpmb || prev_idata) {
548	sbc.opcode = MMC_SET_BLOCK_COUNT;
549	/*
550	* We don't do any blockcount validation because the max size
551	* may be increased by a future standard. We just copy the
552	* 'Reliable Write' bit here.
553	*/
554	sbc.arg = data.blocks | (idata->ic.write_flag & BIT(31));
555	if (prev_idata)
556	sbc.arg = prev_idata->ic.arg;
557	sbc.flags = MMC_RSP_R1 | MMC_CMD_AC;
558	mrq.sbc = &sbc;
559	}
560
561	if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_SANITIZE_START) &&
562	(cmd.opcode == MMC_SWITCH))
563	return mmc_sanitize(card, timeout_ms: idata->ic.cmd_timeout_ms);
564
565	/* If it's an R1B response we need some more preparations. */
566	busy_timeout_ms = idata->ic.cmd_timeout_ms ? : MMC_BLK_TIMEOUT_MS;
567	r1b_resp = (cmd.flags & MMC_RSP_R1B) == MMC_RSP_R1B;
568	if (r1b_resp)
569	mmc_prepare_busy_cmd(host: card->host, cmd: &cmd, timeout_ms: busy_timeout_ms);
570
571	mmc_wait_for_req(host: card->host, mrq: &mrq);
572	memcpy(&idata->ic.response, cmd.resp, sizeof(cmd.resp));
573
574	if (prev_idata) {
575	memcpy(&prev_idata->ic.response, sbc.resp, sizeof(sbc.resp));
576	if (sbc.error) {
577	dev_err(mmc_dev(card->host), "%s: sbc error %d\n",
578	__func__, sbc.error);
579	return sbc.error;
580	}
581	}
582
583	if (cmd.error) {
584	dev_err(mmc_dev(card->host), "%s: cmd error %d\n",
585	__func__, cmd.error);
586	return cmd.error;
587	}
588	if (data.error) {
589	dev_err(mmc_dev(card->host), "%s: data error %d\n",
590	__func__, data.error);
591	return data.error;
592	}
593
594	/*
595	* Make sure the cache of the PARTITION_CONFIG register and
596	* PARTITION_ACCESS bits is updated in case the ioctl ext_csd write
597	* changed it successfully.
598	*/
599	if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_PART_CONFIG) &&
600	(cmd.opcode == MMC_SWITCH)) {
601	struct mmc_blk_data *main_md = dev_get_drvdata(dev: &card->dev);
602	u8 value = MMC_EXTRACT_VALUE_FROM_ARG(cmd.arg);
603
604	/*
605	* Update cache so the next mmc_blk_part_switch call operates
606	* on up-to-date data.
607	*/
608	card->ext_csd.part_config = value;
609	main_md->part_curr = value & EXT_CSD_PART_CONFIG_ACC_MASK;
610	}
611
612	/*
613	* Make sure to update CACHE_CTRL in case it was changed. The cache
614	* will get turned back on if the card is re-initialized, e.g.
615	* suspend/resume or hw reset in recovery.
616	*/
617	if ((MMC_EXTRACT_INDEX_FROM_ARG(cmd.arg) == EXT_CSD_CACHE_CTRL) &&
618	(cmd.opcode == MMC_SWITCH)) {
619	u8 value = MMC_EXTRACT_VALUE_FROM_ARG(cmd.arg) & 1;
620
621	card->ext_csd.cache_ctrl = value;
622	}
623
624	/*
625	* According to the SD specs, some commands require a delay after
626	* issuing the command.
627	*/
628	if (idata->ic.postsleep_min_us)
629	usleep_range(min: idata->ic.postsleep_min_us, max: idata->ic.postsleep_max_us);
630
631	if (mmc_host_is_spi(card->host)) {
632	if (idata->ic.write_flag || r1b_resp || cmd.flags & MMC_RSP_SPI_BUSY)
633	return mmc_spi_err_check(card);
634	return err;
635	}
636
637	/*
638	* Ensure RPMB, writes and R1B responses are completed by polling with
639	* CMD13. Note that, usually we don't need to poll when using HW busy
640	* detection, but here it's needed since some commands may indicate the
641	* error through the R1 status bits.
642	*/
643	if (idata->rpmb || idata->ic.write_flag || r1b_resp) {
644	struct mmc_blk_busy_data cb_data = {
645	.card = card,
646	};
647
648	err = __mmc_poll_for_busy(host: card->host, period_us: 0, timeout_ms: busy_timeout_ms,
649	busy_cb: &mmc_blk_busy_cb, cb_data: &cb_data);
650
651	idata->ic.response[0] = cb_data.status;
652	}
653
654	return err;
655	}
656
657	static int mmc_blk_ioctl_cmd(struct mmc_blk_data *md,
658	struct mmc_ioc_cmd __user *ic_ptr,
659	struct mmc_rpmb_data *rpmb)
660	{
661	struct mmc_blk_ioc_data *idata;
662	struct mmc_blk_ioc_data *idatas[1];
663	struct mmc_queue *mq;
664	struct mmc_card *card;
665	int err = 0, ioc_err = 0;
666	struct request *req;
667
668	idata = mmc_blk_ioctl_copy_from_user(user: ic_ptr);
669	if (IS_ERR(ptr: idata))
670	return PTR_ERR(ptr: idata);
671	/* This will be NULL on non-RPMB ioctl():s */
672	idata->rpmb = rpmb;
673
674	card = md->queue.card;
675	if (IS_ERR(ptr: card)) {
676	err = PTR_ERR(ptr: card);
677	goto cmd_done;
678	}
679
680	/*
681	* Dispatch the ioctl() into the block request queue.
682	*/
683	mq = &md->queue;
684	req = blk_mq_alloc_request(q: mq->queue,
685	opf: idata->ic.write_flag ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, flags: 0);
686	if (IS_ERR(ptr: req)) {
687	err = PTR_ERR(ptr: req);
688	goto cmd_done;
689	}
690	idatas[0] = idata;
691	req_to_mmc_queue_req(rq: req)->drv_op =
692	rpmb ? MMC_DRV_OP_IOCTL_RPMB : MMC_DRV_OP_IOCTL;
693	req_to_mmc_queue_req(rq: req)->drv_op_result = -EIO;
694	req_to_mmc_queue_req(rq: req)->drv_op_data = idatas;
695	req_to_mmc_queue_req(rq: req)->ioc_count = 1;
696	blk_execute_rq(rq: req, at_head: false);
697	ioc_err = req_to_mmc_queue_req(rq: req)->drv_op_result;
698	err = mmc_blk_ioctl_copy_to_user(ic_ptr, idata);
699	blk_mq_free_request(rq: req);
700
701	cmd_done:
702	kfree(objp: idata->buf);
703	kfree(objp: idata);
704	return ioc_err ? ioc_err : err;
705	}
706
707	static int mmc_blk_ioctl_multi_cmd(struct mmc_blk_data *md,
708	struct mmc_ioc_multi_cmd __user *user,
709	struct mmc_rpmb_data *rpmb)
710	{
711	struct mmc_blk_ioc_data **idata = NULL;
712	struct mmc_ioc_cmd __user *cmds = user->cmds;
713	struct mmc_card *card;
714	struct mmc_queue *mq;
715	int err = 0, ioc_err = 0;
716	__u64 num_of_cmds;
717	unsigned int i, n;
718	struct request *req;
719
720	if (copy_from_user(to: &num_of_cmds, from: &user->num_of_cmds,
721	n: sizeof(num_of_cmds)))
722	return -EFAULT;
723
724	if (!num_of_cmds)
725	return 0;
726
727	if (num_of_cmds > MMC_IOC_MAX_CMDS)
728	return -EINVAL;
729
730	n = num_of_cmds;
731	idata = kcalloc(n, size: sizeof(*idata), GFP_KERNEL);
732	if (!idata)
733	return -ENOMEM;
734
735	for (i = 0; i < n; i++) {
736	idata[i] = mmc_blk_ioctl_copy_from_user(user: &cmds[i]);
737	if (IS_ERR(ptr: idata[i])) {
738	err = PTR_ERR(ptr: idata[i]);
739	n = i;
740	goto cmd_err;
741	}
742	/* This will be NULL on non-RPMB ioctl():s */
743	idata[i]->rpmb = rpmb;
744	}
745
746	card = md->queue.card;
747	if (IS_ERR(ptr: card)) {
748	err = PTR_ERR(ptr: card);
749	goto cmd_err;
750	}
751
752
753	/*
754	* Dispatch the ioctl()s into the block request queue.
755	*/
756	mq = &md->queue;
757	req = blk_mq_alloc_request(q: mq->queue,
758	opf: idata[0]->ic.write_flag ? REQ_OP_DRV_OUT : REQ_OP_DRV_IN, flags: 0);
759	if (IS_ERR(ptr: req)) {
760	err = PTR_ERR(ptr: req);
761	goto cmd_err;
762	}
763	req_to_mmc_queue_req(rq: req)->drv_op =
764	rpmb ? MMC_DRV_OP_IOCTL_RPMB : MMC_DRV_OP_IOCTL;
765	req_to_mmc_queue_req(rq: req)->drv_op_result = -EIO;
766	req_to_mmc_queue_req(rq: req)->drv_op_data = idata;
767	req_to_mmc_queue_req(rq: req)->ioc_count = n;
768	blk_execute_rq(rq: req, at_head: false);
769	ioc_err = req_to_mmc_queue_req(rq: req)->drv_op_result;
770
771	/* copy to user if data and response */
772	for (i = 0; i < n && !err; i++)
773	err = mmc_blk_ioctl_copy_to_user(ic_ptr: &cmds[i], idata: idata[i]);
774
775	blk_mq_free_request(rq: req);
776
777	cmd_err:
778	for (i = 0; i < n; i++) {
779	kfree(objp: idata[i]->buf);
780	kfree(objp: idata[i]);
781	}
782	kfree(objp: idata);
783	return ioc_err ? ioc_err : err;
784	}
785
786	static int mmc_blk_check_blkdev(struct block_device *bdev)
787	{
788	/*
789	* The caller must have CAP_SYS_RAWIO, and must be calling this on the
790	* whole block device, not on a partition. This prevents overspray
791	* between sibling partitions.
792	*/
793	if (!capable(CAP_SYS_RAWIO) || bdev_is_partition(bdev))
794	return -EPERM;
795	return 0;
796	}
797
798	static int mmc_blk_ioctl(struct block_device *bdev, blk_mode_t mode,
799	unsigned int cmd, unsigned long arg)
800	{
801	struct mmc_blk_data *md;
802	int ret;
803
804	switch (cmd) {
805	case MMC_IOC_CMD:
806	ret = mmc_blk_check_blkdev(bdev);
807	if (ret)
808	return ret;
809	md = mmc_blk_get(disk: bdev->bd_disk);
810	if (!md)
811	return -EINVAL;
812	ret = mmc_blk_ioctl_cmd(md,
813	ic_ptr: (struct mmc_ioc_cmd __user *)arg,
814	NULL);
815	mmc_blk_put(md);
816	return ret;
817	case MMC_IOC_MULTI_CMD:
818	ret = mmc_blk_check_blkdev(bdev);
819	if (ret)
820	return ret;
821	md = mmc_blk_get(disk: bdev->bd_disk);
822	if (!md)
823	return -EINVAL;
824	ret = mmc_blk_ioctl_multi_cmd(md,
825	user: (struct mmc_ioc_multi_cmd __user *)arg,
826	NULL);
827	mmc_blk_put(md);
828	return ret;
829	default:
830	return -EINVAL;
831	}
832	}
833
834	#ifdef CONFIG_COMPAT
835	static int mmc_blk_compat_ioctl(struct block_device *bdev, blk_mode_t mode,
836	unsigned int cmd, unsigned long arg)
837	{
838	return mmc_blk_ioctl(bdev, mode, cmd, arg: (unsigned long) compat_ptr(uptr: arg));
839	}
840	#endif
841
842	static int mmc_blk_alternative_gpt_sector(struct gendisk *disk,
843	sector_t *sector)
844	{
845	struct mmc_blk_data *md;
846	int ret;
847
848	md = mmc_blk_get(disk);
849	if (!md)
850	return -EINVAL;
851
852	if (md->queue.card)
853	ret = mmc_card_alternative_gpt_sector(card: md->queue.card, sector);
854	else
855	ret = -ENODEV;
856
857	mmc_blk_put(md);
858
859	return ret;
860	}
861
862	static const struct block_device_operations mmc_bdops = {
863	.open = mmc_blk_open,
864	.release = mmc_blk_release,
865	.getgeo = mmc_blk_getgeo,
866	.owner = THIS_MODULE,
867	.ioctl = mmc_blk_ioctl,
868	#ifdef CONFIG_COMPAT
869	.compat_ioctl = mmc_blk_compat_ioctl,
870	#endif
871	.alternative_gpt_sector = mmc_blk_alternative_gpt_sector,
872	};
873
874	static int mmc_blk_part_switch_pre(struct mmc_card *card,
875	unsigned int part_type)
876	{
877	const unsigned int mask = EXT_CSD_PART_CONFIG_ACC_MASK;
878	const unsigned int rpmb = EXT_CSD_PART_CONFIG_ACC_RPMB;
879	int ret = 0;
880
881	if ((part_type & mask) == rpmb) {
882	if (card->ext_csd.cmdq_en) {
883	ret = mmc_cmdq_disable(card);
884	if (ret)
885	return ret;
886	}
887	mmc_retune_pause(host: card->host);
888	}
889
890	return ret;
891	}
892
893	static int mmc_blk_part_switch_post(struct mmc_card *card,
894	unsigned int part_type)
895	{
896	const unsigned int mask = EXT_CSD_PART_CONFIG_ACC_MASK;
897	const unsigned int rpmb = EXT_CSD_PART_CONFIG_ACC_RPMB;
898	int ret = 0;
899
900	if ((part_type & mask) == rpmb) {
901	mmc_retune_unpause(host: card->host);
902	if (card->reenable_cmdq && !card->ext_csd.cmdq_en)
903	ret = mmc_cmdq_enable(card);
904	}
905
906	return ret;
907	}
908
909	static inline int mmc_blk_part_switch(struct mmc_card *card,
910	unsigned int part_type)
911	{
912	int ret = 0;
913	struct mmc_blk_data *main_md = dev_get_drvdata(dev: &card->dev);
914
915	if (main_md->part_curr == part_type)
916	return 0;
917
918	if (mmc_card_mmc(card)) {
919	u8 part_config = card->ext_csd.part_config;
920
921	ret = mmc_blk_part_switch_pre(card, part_type);
922	if (ret)
923	return ret;
924
925	part_config &= ~EXT_CSD_PART_CONFIG_ACC_MASK;
926	part_config |= part_type;
927
928	ret = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
929	EXT_CSD_PART_CONFIG, value: part_config,
930	timeout_ms: card->ext_csd.part_time);
931	if (ret) {
932	mmc_blk_part_switch_post(card, part_type);
933	return ret;
934	}
935
936	card->ext_csd.part_config = part_config;
937
938	ret = mmc_blk_part_switch_post(card, part_type: main_md->part_curr);
939	}
940
941	main_md->part_curr = part_type;
942	return ret;
943	}
944
945	static int mmc_sd_num_wr_blocks(struct mmc_card *card, u32 *written_blocks)
946	{
947	int err;
948	u32 result;
949	__be32 *blocks;
950
951	struct mmc_request mrq = {};
952	struct mmc_command cmd = {};
953	struct mmc_data data = {};
954
955	struct scatterlist sg;
956
957	err = mmc_app_cmd(host: card->host, card);
958	if (err)
959	return err;
960
961	cmd.opcode = SD_APP_SEND_NUM_WR_BLKS;
962	cmd.arg = 0;
963	cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC;
964
965	data.blksz = 4;
966	data.blocks = 1;
967	data.flags = MMC_DATA_READ;
968	data.sg = &sg;
969	data.sg_len = 1;
970	mmc_set_data_timeout(data: &data, card);
971
972	mrq.cmd = &cmd;
973	mrq.data = &data;
974
975	blocks = kmalloc(size: 4, GFP_KERNEL);
976	if (!blocks)
977	return -ENOMEM;
978
979	sg_init_one(&sg, blocks, 4);
980
981	mmc_wait_for_req(host: card->host, mrq: &mrq);
982
983	result = ntohl(*blocks);
984	kfree(objp: blocks);
985
986	if (cmd.error || data.error)
987	return -EIO;
988
989	*written_blocks = result;
990
991	return 0;
992	}
993
994	static unsigned int mmc_blk_clock_khz(struct mmc_host *host)
995	{
996	if (host->actual_clock)
997	return host->actual_clock / 1000;
998
999	/* Clock may be subject to a divisor, fudge it by a factor of 2. */
1000	if (host->ios.clock)
1001	return host->ios.clock / 2000;
1002
1003	/* How can there be no clock */
1004	WARN_ON_ONCE(1);
1005	return 100; /* 100 kHz is minimum possible value */
1006	}
1007
1008	static unsigned int mmc_blk_data_timeout_ms(struct mmc_host *host,
1009	struct mmc_data *data)
1010	{
1011	unsigned int ms = DIV_ROUND_UP(data->timeout_ns, 1000000);
1012	unsigned int khz;
1013
1014	if (data->timeout_clks) {
1015	khz = mmc_blk_clock_khz(host);
1016	ms += DIV_ROUND_UP(data->timeout_clks, khz);
1017	}
1018
1019	return ms;
1020	}
1021
1022	/*
1023	* Attempts to reset the card and get back to the requested partition.
1024	* Therefore any error here must result in cancelling the block layer
1025	* request, it must not be reattempted without going through the mmc_blk
1026	* partition sanity checks.
1027	*/
1028	static int mmc_blk_reset(struct mmc_blk_data *md, struct mmc_host *host,
1029	int type)
1030	{
1031	int err;
1032	struct mmc_blk_data *main_md = dev_get_drvdata(dev: &host->card->dev);
1033
1034	if (md->reset_done & type)
1035	return -EEXIST;
1036
1037	md->reset_done |= type;
1038	err = mmc_hw_reset(card: host->card);
1039	/*
1040	* A successful reset will leave the card in the main partition, but
1041	* upon failure it might not be, so set it to MMC_BLK_PART_INVALID
1042	* in that case.
1043	*/
1044	main_md->part_curr = err ? MMC_BLK_PART_INVALID : main_md->part_type;
1045	if (err)
1046	return err;
1047	/* Ensure we switch back to the correct partition */
1048	if (mmc_blk_part_switch(card: host->card, part_type: md->part_type))
1049	/*
1050	* We have failed to get back into the correct
1051	* partition, so we need to abort the whole request.
1052	*/
1053	return -ENODEV;
1054	return 0;
1055	}
1056
1057	static inline void mmc_blk_reset_success(struct mmc_blk_data *md, int type)
1058	{
1059	md->reset_done &= ~type;
1060	}
1061
1062	static void mmc_blk_check_sbc(struct mmc_queue_req *mq_rq)
1063	{
1064	struct mmc_blk_ioc_data **idata = mq_rq->drv_op_data;
1065	int i;
1066
1067	for (i = 1; i < mq_rq->ioc_count; i++) {
1068	if (idata[i - 1]->ic.opcode == MMC_SET_BLOCK_COUNT &&
1069	mmc_op_multi(opcode: idata[i]->ic.opcode)) {
1070	idata[i - 1]->flags |= MMC_BLK_IOC_DROP;
1071	idata[i]->flags |= MMC_BLK_IOC_SBC;
1072	}
1073	}
1074	}
1075
1076	/*
1077	* The non-block commands come back from the block layer after it queued it and
1078	* processed it with all other requests and then they get issued in this
1079	* function.
1080	*/
1081	static void mmc_blk_issue_drv_op(struct mmc_queue *mq, struct request *req)
1082	{
1083	struct mmc_queue_req *mq_rq;
1084	struct mmc_card *card = mq->card;
1085	struct mmc_blk_data *md = mq->blkdata;
1086	struct mmc_blk_ioc_data **idata;
1087	bool rpmb_ioctl;
1088	u8 **ext_csd;
1089	u32 status;
1090	int ret;
1091	int i;
1092
1093	mq_rq = req_to_mmc_queue_req(rq: req);
1094	rpmb_ioctl = (mq_rq->drv_op == MMC_DRV_OP_IOCTL_RPMB);
1095
1096	switch (mq_rq->drv_op) {
1097	case MMC_DRV_OP_IOCTL:
1098	if (card->ext_csd.cmdq_en) {
1099	ret = mmc_cmdq_disable(card);
1100	if (ret)
1101	break;
1102	}
1103
1104	mmc_blk_check_sbc(mq_rq);
1105
1106	fallthrough;
1107	case MMC_DRV_OP_IOCTL_RPMB:
1108	idata = mq_rq->drv_op_data;
1109	for (i = 0, ret = 0; i < mq_rq->ioc_count; i++) {
1110	ret = __mmc_blk_ioctl_cmd(card, md, idatas: idata, i);
1111	if (ret)
1112	break;
1113	}
1114	/* Always switch back to main area after RPMB access */
1115	if (rpmb_ioctl)
1116	mmc_blk_part_switch(card, part_type: 0);
1117	else if (card->reenable_cmdq && !card->ext_csd.cmdq_en)
1118	mmc_cmdq_enable(card);
1119	break;
1120	case MMC_DRV_OP_BOOT_WP:
1121	ret = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL, EXT_CSD_BOOT_WP,
1122	value: card->ext_csd.boot_ro_lock |
1123	EXT_CSD_BOOT_WP_B_PWR_WP_EN,
1124	timeout_ms: card->ext_csd.part_time);
1125	if (ret)
1126	pr_err("%s: Locking boot partition ro until next power on failed: %d\n",
1127	md->disk->disk_name, ret);
1128	else
1129	card->ext_csd.boot_ro_lock |=
1130	EXT_CSD_BOOT_WP_B_PWR_WP_EN;
1131	break;
1132	case MMC_DRV_OP_GET_CARD_STATUS:
1133	ret = mmc_send_status(card, status: &status);
1134	if (!ret)
1135	ret = status;
1136	break;
1137	case MMC_DRV_OP_GET_EXT_CSD:
1138	ext_csd = mq_rq->drv_op_data;
1139	ret = mmc_get_ext_csd(card, new_ext_csd: ext_csd);
1140	break;
1141	default:
1142	pr_err("%s: unknown driver specific operation\n",
1143	md->disk->disk_name);
1144	ret = -EINVAL;
1145	break;
1146	}
1147	mq_rq->drv_op_result = ret;
1148	blk_mq_end_request(rq: req, error: ret ? BLK_STS_IOERR : BLK_STS_OK);
1149	}
1150
1151	static void mmc_blk_issue_erase_rq(struct mmc_queue *mq, struct request *req,
1152	int type, unsigned int erase_arg)
1153	{
1154	struct mmc_blk_data *md = mq->blkdata;
1155	struct mmc_card *card = md->queue.card;
1156	unsigned int from, nr;
1157	int err = 0;
1158	blk_status_t status = BLK_STS_OK;
1159
1160	if (!mmc_can_erase(card)) {
1161	status = BLK_STS_NOTSUPP;
1162	goto fail;
1163	}
1164
1165	from = blk_rq_pos(rq: req);
1166	nr = blk_rq_sectors(rq: req);
1167
1168	do {
1169	err = 0;
1170	if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1171	err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1172	INAND_CMD38_ARG_EXT_CSD,
1173	value: erase_arg == MMC_TRIM_ARG ?
1174	INAND_CMD38_ARG_TRIM :
1175	INAND_CMD38_ARG_ERASE,
1176	timeout_ms: card->ext_csd.generic_cmd6_time);
1177	}
1178	if (!err)
1179	err = mmc_erase(card, from, nr, arg: erase_arg);
1180	} while (err == -EIO && !mmc_blk_reset(md, host: card->host, type));
1181	if (err)
1182	status = BLK_STS_IOERR;
1183	else
1184	mmc_blk_reset_success(md, type);
1185	fail:
1186	blk_mq_end_request(rq: req, error: status);
1187	}
1188
1189	static void mmc_blk_issue_trim_rq(struct mmc_queue *mq, struct request *req)
1190	{
1191	mmc_blk_issue_erase_rq(mq, req, MMC_BLK_TRIM, MMC_TRIM_ARG);
1192	}
1193
1194	static void mmc_blk_issue_discard_rq(struct mmc_queue *mq, struct request *req)
1195	{
1196	struct mmc_blk_data *md = mq->blkdata;
1197	struct mmc_card *card = md->queue.card;
1198	unsigned int arg = card->erase_arg;
1199
1200	if (mmc_card_broken_sd_discard(c: card))
1201	arg = SD_ERASE_ARG;
1202
1203	mmc_blk_issue_erase_rq(mq, req, MMC_BLK_DISCARD, erase_arg: arg);
1204	}
1205
1206	static void mmc_blk_issue_secdiscard_rq(struct mmc_queue *mq,
1207	struct request *req)
1208	{
1209	struct mmc_blk_data *md = mq->blkdata;
1210	struct mmc_card *card = md->queue.card;
1211	unsigned int from, nr, arg;
1212	int err = 0, type = MMC_BLK_SECDISCARD;
1213	blk_status_t status = BLK_STS_OK;
1214
1215	if (!(mmc_can_secure_erase_trim(card))) {
1216	status = BLK_STS_NOTSUPP;
1217	goto out;
1218	}
1219
1220	from = blk_rq_pos(rq: req);
1221	nr = blk_rq_sectors(rq: req);
1222
1223	if (mmc_can_trim(card) && !mmc_erase_group_aligned(card, from, nr))
1224	arg = MMC_SECURE_TRIM1_ARG;
1225	else
1226	arg = MMC_SECURE_ERASE_ARG;
1227
1228	retry:
1229	if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1230	err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1231	INAND_CMD38_ARG_EXT_CSD,
1232	value: arg == MMC_SECURE_TRIM1_ARG ?
1233	INAND_CMD38_ARG_SECTRIM1 :
1234	INAND_CMD38_ARG_SECERASE,
1235	timeout_ms: card->ext_csd.generic_cmd6_time);
1236	if (err)
1237	goto out_retry;
1238	}
1239
1240	err = mmc_erase(card, from, nr, arg);
1241	if (err == -EIO)
1242	goto out_retry;
1243	if (err) {
1244	status = BLK_STS_IOERR;
1245	goto out;
1246	}
1247
1248	if (arg == MMC_SECURE_TRIM1_ARG) {
1249	if (card->quirks & MMC_QUIRK_INAND_CMD38) {
1250	err = mmc_switch(card, EXT_CSD_CMD_SET_NORMAL,
1251	INAND_CMD38_ARG_EXT_CSD,
1252	INAND_CMD38_ARG_SECTRIM2,
1253	timeout_ms: card->ext_csd.generic_cmd6_time);
1254	if (err)
1255	goto out_retry;
1256	}
1257
1258	err = mmc_erase(card, from, nr, MMC_SECURE_TRIM2_ARG);
1259	if (err == -EIO)
1260	goto out_retry;
1261	if (err) {
1262	status = BLK_STS_IOERR;
1263	goto out;
1264	}
1265	}
1266
1267	out_retry:
1268	if (err && !mmc_blk_reset(md, host: card->host, type))
1269	goto retry;
1270	if (!err)
1271	mmc_blk_reset_success(md, type);
1272	out:
1273	blk_mq_end_request(rq: req, error: status);
1274	}
1275
1276	static void mmc_blk_issue_flush(struct mmc_queue *mq, struct request *req)
1277	{
1278	struct mmc_blk_data *md = mq->blkdata;
1279	struct mmc_card *card = md->queue.card;
1280	int ret = 0;
1281
1282	ret = mmc_flush_cache(host: card->host);
1283	blk_mq_end_request(rq: req, error: ret ? BLK_STS_IOERR : BLK_STS_OK);
1284	}
1285
1286	/*
1287	* Reformat current write as a reliable write, supporting
1288	* both legacy and the enhanced reliable write MMC cards.
1289	* In each transfer we'll handle only as much as a single
1290	* reliable write can handle, thus finish the request in
1291	* partial completions.
1292	*/
1293	static inline void mmc_apply_rel_rw(struct mmc_blk_request *brq,
1294	struct mmc_card *card,
1295	struct request *req)
1296	{
1297	if (!(card->ext_csd.rel_param & EXT_CSD_WR_REL_PARAM_EN)) {
1298	/* Legacy mode imposes restrictions on transfers. */
1299	if (!IS_ALIGNED(blk_rq_pos(req), card->ext_csd.rel_sectors))
1300	brq->data.blocks = 1;
1301
1302	if (brq->data.blocks > card->ext_csd.rel_sectors)
1303	brq->data.blocks = card->ext_csd.rel_sectors;
1304	else if (brq->data.blocks < card->ext_csd.rel_sectors)
1305	brq->data.blocks = 1;
1306	}
1307	}
1308
1309	#define CMD_ERRORS_EXCL_OOR \
1310	(R1_ADDRESS_ERROR | /* Misaligned address */ \
1311	R1_BLOCK_LEN_ERROR | /* Transferred block length incorrect */\
1312	R1_WP_VIOLATION | /* Tried to write to protected block */ \
1313	R1_CARD_ECC_FAILED | /* Card ECC failed */ \
1314	R1_CC_ERROR | /* Card controller error */ \
1315	R1_ERROR) /* General/unknown error */
1316
1317	#define CMD_ERRORS \
1318	(CMD_ERRORS_EXCL_OOR | \
1319	R1_OUT_OF_RANGE) /* Command argument out of range */ \
1320
1321	static void mmc_blk_eval_resp_error(struct mmc_blk_request *brq)
1322	{
1323	u32 val;
1324
1325	/*
1326	* Per the SD specification(physical layer version 4.10)[1],
1327	* section 4.3.3, it explicitly states that "When the last
1328	* block of user area is read using CMD18, the host should
1329	* ignore OUT_OF_RANGE error that may occur even the sequence
1330	* is correct". And JESD84-B51 for eMMC also has a similar
1331	* statement on section 6.8.3.
1332	*
1333	* Multiple block read/write could be done by either predefined
1334	* method, namely CMD23, or open-ending mode. For open-ending mode,
1335	* we should ignore the OUT_OF_RANGE error as it's normal behaviour.
1336	*
1337	* However the spec[1] doesn't tell us whether we should also
1338	* ignore that for predefined method. But per the spec[1], section
1339	* 4.15 Set Block Count Command, it says"If illegal block count
1340	* is set, out of range error will be indicated during read/write
1341	* operation (For example, data transfer is stopped at user area
1342	* boundary)." In another word, we could expect a out of range error
1343	* in the response for the following CMD18/25. And if argument of
1344	* CMD23 + the argument of CMD18/25 exceed the max number of blocks,
1345	* we could also expect to get a -ETIMEDOUT or any error number from
1346	* the host drivers due to missing data response(for write)/data(for
1347	* read), as the cards will stop the data transfer by itself per the
1348	* spec. So we only need to check R1_OUT_OF_RANGE for open-ending mode.
1349	*/
1350
1351	if (!brq->stop.error) {
1352	bool oor_with_open_end;
1353	/* If there is no error yet, check R1 response */
1354
1355	val = brq->stop.resp[0] & CMD_ERRORS;
1356	oor_with_open_end = val & R1_OUT_OF_RANGE && !brq->mrq.sbc;
1357
1358	if (val && !oor_with_open_end)
1359	brq->stop.error = -EIO;
1360	}
1361	}
1362
1363	static void mmc_blk_data_prep(struct mmc_queue *mq, struct mmc_queue_req *mqrq,
1364	int recovery_mode, bool *do_rel_wr_p,
1365	bool *do_data_tag_p)
1366	{
1367	struct mmc_blk_data *md = mq->blkdata;
1368	struct mmc_card *card = md->queue.card;
1369	struct mmc_blk_request *brq = &mqrq->brq;
1370	struct request *req = mmc_queue_req_to_req(mqr: mqrq);
1371	bool do_rel_wr, do_data_tag;
1372
1373	/*
1374	* Reliable writes are used to implement Forced Unit Access and
1375	* are supported only on MMCs.
1376	*/
1377	do_rel_wr = (req->cmd_flags & REQ_FUA) &&
1378	rq_data_dir(req) == WRITE &&
1379	(md->flags & MMC_BLK_REL_WR);
1380
1381	memset(brq, 0, sizeof(struct mmc_blk_request));
1382
1383	mmc_crypto_prepare_req(mqrq);
1384
1385	brq->mrq.data = &brq->data;
1386	brq->mrq.tag = req->tag;
1387
1388	brq->stop.opcode = MMC_STOP_TRANSMISSION;
1389	brq->stop.arg = 0;
1390
1391	if (rq_data_dir(req) == READ) {
1392	brq->data.flags = MMC_DATA_READ;
1393	brq->stop.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC;
1394	} else {
1395	brq->data.flags = MMC_DATA_WRITE;
1396	brq->stop.flags = MMC_RSP_SPI_R1B | MMC_RSP_R1B | MMC_CMD_AC;
1397	}
1398
1399	brq->data.blksz = 512;
1400	brq->data.blocks = blk_rq_sectors(rq: req);
1401	brq->data.blk_addr = blk_rq_pos(rq: req);
1402
1403	/*
1404	* The command queue supports 2 priorities: "high" (1) and "simple" (0).
1405	* The eMMC will give "high" priority tasks priority over "simple"
1406	* priority tasks. Here we always set "simple" priority by not setting
1407	* MMC_DATA_PRIO.
1408	*/
1409
1410	/*
1411	* The block layer doesn't support all sector count
1412	* restrictions, so we need to be prepared for too big
1413	* requests.
1414	*/
1415	if (brq->data.blocks > card->host->max_blk_count)
1416	brq->data.blocks = card->host->max_blk_count;
1417
1418	if (brq->data.blocks > 1) {
1419	/*
1420	* Some SD cards in SPI mode return a CRC error or even lock up
1421	* completely when trying to read the last block using a
1422	* multiblock read command.
1423	*/
1424	if (mmc_host_is_spi(card->host) && (rq_data_dir(req) == READ) &&
1425	(blk_rq_pos(rq: req) + blk_rq_sectors(rq: req) ==
1426	get_capacity(disk: md->disk)))
1427	brq->data.blocks--;
1428
1429	/*
1430	* After a read error, we redo the request one (native) sector
1431	* at a time in order to accurately determine which
1432	* sectors can be read successfully.
1433	*/
1434	if (recovery_mode)
1435	brq->data.blocks = queue_physical_block_size(q: mq->queue) >> 9;
1436
1437	/*
1438	* Some controllers have HW issues while operating
1439	* in multiple I/O mode
1440	*/
1441	if (card->host->ops->multi_io_quirk)
1442	brq->data.blocks = card->host->ops->multi_io_quirk(card,
1443	(rq_data_dir(req) == READ) ?
1444	MMC_DATA_READ : MMC_DATA_WRITE,
1445	brq->data.blocks);
1446	}
1447
1448	if (do_rel_wr) {
1449	mmc_apply_rel_rw(brq, card, req);
1450	brq->data.flags |= MMC_DATA_REL_WR;
1451	}
1452
1453	/*
1454	* Data tag is used only during writing meta data to speed
1455	* up write and any subsequent read of this meta data
1456	*/
1457	do_data_tag = card->ext_csd.data_tag_unit_size &&
1458	(req->cmd_flags & REQ_META) &&
1459	(rq_data_dir(req) == WRITE) &&
1460	((brq->data.blocks * brq->data.blksz) >=
1461	card->ext_csd.data_tag_unit_size);
1462
1463	if (do_data_tag)
1464	brq->data.flags |= MMC_DATA_DAT_TAG;
1465
1466	mmc_set_data_timeout(data: &brq->data, card);
1467
1468	brq->data.sg = mqrq->sg;
1469	brq->data.sg_len = mmc_queue_map_sg(mq, mqrq);
1470
1471	/*
1472	* Adjust the sg list so it is the same size as the
1473	* request.
1474	*/
1475	if (brq->data.blocks != blk_rq_sectors(rq: req)) {
1476	int i, data_size = brq->data.blocks << 9;
1477	struct scatterlist *sg;
1478
1479	for_each_sg(brq->data.sg, sg, brq->data.sg_len, i) {
1480	data_size -= sg->length;
1481	if (data_size <= 0) {
1482	sg->length += data_size;
1483	i++;
1484	break;
1485	}
1486	}
1487	brq->data.sg_len = i;
1488	}
1489
1490	if (do_rel_wr_p)
1491	*do_rel_wr_p = do_rel_wr;
1492
1493	if (do_data_tag_p)
1494	*do_data_tag_p = do_data_tag;
1495	}
1496
1497	#define MMC_CQE_RETRIES 2
1498
1499	static void mmc_blk_cqe_complete_rq(struct mmc_queue *mq, struct request *req)
1500	{
1501	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1502	struct mmc_request *mrq = &mqrq->brq.mrq;
1503	struct request_queue *q = req->q;
1504	struct mmc_host *host = mq->card->host;
1505	enum mmc_issue_type issue_type = mmc_issue_type(mq, req);
1506	unsigned long flags;
1507	bool put_card;
1508	int err;
1509
1510	mmc_cqe_post_req(host, mrq);
1511
1512	if (mrq->cmd && mrq->cmd->error)
1513	err = mrq->cmd->error;
1514	else if (mrq->data && mrq->data->error)
1515	err = mrq->data->error;
1516	else
1517	err = 0;
1518
1519	if (err) {
1520	if (mqrq->retries++ < MMC_CQE_RETRIES)
1521	blk_mq_requeue_request(rq: req, kick_requeue_list: true);
1522	else
1523	blk_mq_end_request(rq: req, BLK_STS_IOERR);
1524	} else if (mrq->data) {
1525	if (blk_update_request(rq: req, BLK_STS_OK, nr_bytes: mrq->data->bytes_xfered))
1526	blk_mq_requeue_request(rq: req, kick_requeue_list: true);
1527	else
1528	__blk_mq_end_request(rq: req, BLK_STS_OK);
1529	} else if (mq->in_recovery) {
1530	blk_mq_requeue_request(rq: req, kick_requeue_list: true);
1531	} else {
1532	blk_mq_end_request(rq: req, BLK_STS_OK);
1533	}
1534
1535	spin_lock_irqsave(&mq->lock, flags);
1536
1537	mq->in_flight[issue_type] -= 1;
1538
1539	put_card = (mmc_tot_in_flight(mq) == 0);
1540
1541	mmc_cqe_check_busy(mq);
1542
1543	spin_unlock_irqrestore(lock: &mq->lock, flags);
1544
1545	if (!mq->cqe_busy)
1546	blk_mq_run_hw_queues(q, async: true);
1547
1548	if (put_card)
1549	mmc_put_card(card: mq->card, ctx: &mq->ctx);
1550	}
1551
1552	void mmc_blk_cqe_recovery(struct mmc_queue *mq)
1553	{
1554	struct mmc_card *card = mq->card;
1555	struct mmc_host *host = card->host;
1556	int err;
1557
1558	pr_debug("%s: CQE recovery start\n", mmc_hostname(host));
1559
1560	err = mmc_cqe_recovery(host);
1561	if (err)
1562	mmc_blk_reset(md: mq->blkdata, host, MMC_BLK_CQE_RECOVERY);
1563	mmc_blk_reset_success(md: mq->blkdata, MMC_BLK_CQE_RECOVERY);
1564
1565	pr_debug("%s: CQE recovery done\n", mmc_hostname(host));
1566	}
1567
1568	static void mmc_blk_cqe_req_done(struct mmc_request *mrq)
1569	{
1570	struct mmc_queue_req *mqrq = container_of(mrq, struct mmc_queue_req,
1571	brq.mrq);
1572	struct request *req = mmc_queue_req_to_req(mqr: mqrq);
1573	struct request_queue *q = req->q;
1574	struct mmc_queue *mq = q->queuedata;
1575
1576	/*
1577	* Block layer timeouts race with completions which means the normal
1578	* completion path cannot be used during recovery.
1579	*/
1580	if (mq->in_recovery)
1581	mmc_blk_cqe_complete_rq(mq, req);
1582	else if (likely(!blk_should_fake_timeout(req->q)))
1583	blk_mq_complete_request(rq: req);
1584	}
1585
1586	static int mmc_blk_cqe_start_req(struct mmc_host *host, struct mmc_request *mrq)
1587	{
1588	mrq->done = mmc_blk_cqe_req_done;
1589	mrq->recovery_notifier = mmc_cqe_recovery_notifier;
1590
1591	return mmc_cqe_start_req(host, mrq);
1592	}
1593
1594	static struct mmc_request *mmc_blk_cqe_prep_dcmd(struct mmc_queue_req *mqrq,
1595	struct request *req)
1596	{
1597	struct mmc_blk_request *brq = &mqrq->brq;
1598
1599	memset(brq, 0, sizeof(*brq));
1600
1601	brq->mrq.cmd = &brq->cmd;
1602	brq->mrq.tag = req->tag;
1603
1604	return &brq->mrq;
1605	}
1606
1607	static int mmc_blk_cqe_issue_flush(struct mmc_queue *mq, struct request *req)
1608	{
1609	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1610	struct mmc_request *mrq = mmc_blk_cqe_prep_dcmd(mqrq, req);
1611
1612	mrq->cmd->opcode = MMC_SWITCH;
1613	mrq->cmd->arg = (MMC_SWITCH_MODE_WRITE_BYTE << 24) |
1614	(EXT_CSD_FLUSH_CACHE << 16) |
1615	(1 << 8) |
1616	EXT_CSD_CMD_SET_NORMAL;
1617	mrq->cmd->flags = MMC_CMD_AC | MMC_RSP_R1B;
1618
1619	return mmc_blk_cqe_start_req(host: mq->card->host, mrq);
1620	}
1621
1622	static int mmc_blk_hsq_issue_rw_rq(struct mmc_queue *mq, struct request *req)
1623	{
1624	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1625	struct mmc_host *host = mq->card->host;
1626	int err;
1627
1628	mmc_blk_rw_rq_prep(mqrq, card: mq->card, recovery_mode: 0, mq);
1629	mqrq->brq.mrq.done = mmc_blk_hsq_req_done;
1630	mmc_pre_req(host, mrq: &mqrq->brq.mrq);
1631
1632	err = mmc_cqe_start_req(host, mrq: &mqrq->brq.mrq);
1633	if (err)
1634	mmc_post_req(host, mrq: &mqrq->brq.mrq, err);
1635
1636	return err;
1637	}
1638
1639	static int mmc_blk_cqe_issue_rw_rq(struct mmc_queue *mq, struct request *req)
1640	{
1641	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1642	struct mmc_host *host = mq->card->host;
1643
1644	if (host->hsq_enabled)
1645	return mmc_blk_hsq_issue_rw_rq(mq, req);
1646
1647	mmc_blk_data_prep(mq, mqrq, recovery_mode: 0, NULL, NULL);
1648
1649	return mmc_blk_cqe_start_req(host: mq->card->host, mrq: &mqrq->brq.mrq);
1650	}
1651
1652	static void mmc_blk_rw_rq_prep(struct mmc_queue_req *mqrq,
1653	struct mmc_card *card,
1654	int recovery_mode,
1655	struct mmc_queue *mq)
1656	{
1657	u32 readcmd, writecmd;
1658	struct mmc_blk_request *brq = &mqrq->brq;
1659	struct request *req = mmc_queue_req_to_req(mqr: mqrq);
1660	struct mmc_blk_data *md = mq->blkdata;
1661	bool do_rel_wr, do_data_tag;
1662
1663	mmc_blk_data_prep(mq, mqrq, recovery_mode, do_rel_wr_p: &do_rel_wr, do_data_tag_p: &do_data_tag);
1664
1665	brq->mrq.cmd = &brq->cmd;
1666
1667	brq->cmd.arg = blk_rq_pos(rq: req);
1668	if (!mmc_card_blockaddr(card))
1669	brq->cmd.arg <<= 9;
1670	brq->cmd.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_ADTC;
1671
1672	if (brq->data.blocks > 1 || do_rel_wr) {
1673	/* SPI multiblock writes terminate using a special
1674	* token, not a STOP_TRANSMISSION request.
1675	*/
1676	if (!mmc_host_is_spi(card->host) ||
1677	rq_data_dir(req) == READ)
1678	brq->mrq.stop = &brq->stop;
1679	readcmd = MMC_READ_MULTIPLE_BLOCK;
1680	writecmd = MMC_WRITE_MULTIPLE_BLOCK;
1681	} else {
1682	brq->mrq.stop = NULL;
1683	readcmd = MMC_READ_SINGLE_BLOCK;
1684	writecmd = MMC_WRITE_BLOCK;
1685	}
1686	brq->cmd.opcode = rq_data_dir(req) == READ ? readcmd : writecmd;
1687
1688	/*
1689	* Pre-defined multi-block transfers are preferable to
1690	* open ended-ones (and necessary for reliable writes).
1691	* However, it is not sufficient to just send CMD23,
1692	* and avoid the final CMD12, as on an error condition
1693	* CMD12 (stop) needs to be sent anyway. This, coupled
1694	* with Auto-CMD23 enhancements provided by some
1695	* hosts, means that the complexity of dealing
1696	* with this is best left to the host. If CMD23 is
1697	* supported by card and host, we'll fill sbc in and let
1698	* the host deal with handling it correctly. This means
1699	* that for hosts that don't expose MMC_CAP_CMD23, no
1700	* change of behavior will be observed.
1701	*
1702	* N.B: Some MMC cards experience perf degradation.
1703	* We'll avoid using CMD23-bounded multiblock writes for
1704	* these, while retaining features like reliable writes.
1705	*/
1706	if ((md->flags & MMC_BLK_CMD23) && mmc_op_multi(opcode: brq->cmd.opcode) &&
1707	(do_rel_wr || !(card->quirks & MMC_QUIRK_BLK_NO_CMD23) ||
1708	do_data_tag)) {
1709	brq->sbc.opcode = MMC_SET_BLOCK_COUNT;
1710	brq->sbc.arg = brq->data.blocks |
1711	(do_rel_wr ? (1 << 31) : 0) |
1712	(do_data_tag ? (1 << 29) : 0);
1713	brq->sbc.flags = MMC_RSP_R1 | MMC_CMD_AC;
1714	brq->mrq.sbc = &brq->sbc;
1715	}
1716	}
1717
1718	#define MMC_MAX_RETRIES 5
1719	#define MMC_DATA_RETRIES 2
1720	#define MMC_NO_RETRIES (MMC_MAX_RETRIES + 1)
1721
1722	static int mmc_blk_send_stop(struct mmc_card *card, unsigned int timeout)
1723	{
1724	struct mmc_command cmd = {
1725	.opcode = MMC_STOP_TRANSMISSION,
1726	.flags = MMC_RSP_SPI_R1 | MMC_RSP_R1 | MMC_CMD_AC,
1727	/* Some hosts wait for busy anyway, so provide a busy timeout */
1728	.busy_timeout = timeout,
1729	};
1730
1731	return mmc_wait_for_cmd(host: card->host, cmd: &cmd, retries: 5);
1732	}
1733
1734	static int mmc_blk_fix_state(struct mmc_card *card, struct request *req)
1735	{
1736	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1737	struct mmc_blk_request *brq = &mqrq->brq;
1738	unsigned int timeout = mmc_blk_data_timeout_ms(host: card->host, data: &brq->data);
1739	int err;
1740
1741	mmc_retune_hold_now(host: card->host);
1742
1743	mmc_blk_send_stop(card, timeout);
1744
1745	err = mmc_poll_for_busy(card, timeout_ms: timeout, retry_crc_err: false, busy_cmd: MMC_BUSY_IO);
1746
1747	mmc_retune_release(host: card->host);
1748
1749	return err;
1750	}
1751
1752	#define MMC_READ_SINGLE_RETRIES 2
1753
1754	/* Single (native) sector read during recovery */
1755	static void mmc_blk_read_single(struct mmc_queue *mq, struct request *req)
1756	{
1757	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1758	struct mmc_request *mrq = &mqrq->brq.mrq;
1759	struct mmc_card *card = mq->card;
1760	struct mmc_host *host = card->host;
1761	blk_status_t error = BLK_STS_OK;
1762	size_t bytes_per_read = queue_physical_block_size(q: mq->queue);
1763
1764	do {
1765	u32 status;
1766	int err;
1767	int retries = 0;
1768
1769	while (retries++ <= MMC_READ_SINGLE_RETRIES) {
1770	mmc_blk_rw_rq_prep(mqrq, card, recovery_mode: 1, mq);
1771
1772	mmc_wait_for_req(host, mrq);
1773
1774	err = mmc_send_status(card, status: &status);
1775	if (err)
1776	goto error_exit;
1777
1778	if (!mmc_host_is_spi(host) &&
1779	!mmc_ready_for_data(status)) {
1780	err = mmc_blk_fix_state(card, req);
1781	if (err)
1782	goto error_exit;
1783	}
1784
1785	if (!mrq->cmd->error)
1786	break;
1787	}
1788
1789	if (mrq->cmd->error ||
1790	mrq->data->error ||
1791	(!mmc_host_is_spi(host) &&
1792	(mrq->cmd->resp[0] & CMD_ERRORS || status & CMD_ERRORS)))
1793	error = BLK_STS_IOERR;
1794	else
1795	error = BLK_STS_OK;
1796
1797	} while (blk_update_request(rq: req, error, nr_bytes: bytes_per_read));
1798
1799	return;
1800
1801	error_exit:
1802	mrq->data->bytes_xfered = 0;
1803	blk_update_request(rq: req, BLK_STS_IOERR, nr_bytes: bytes_per_read);
1804	/* Let it try the remaining request again */
1805	if (mqrq->retries > MMC_MAX_RETRIES - 1)
1806	mqrq->retries = MMC_MAX_RETRIES - 1;
1807	}
1808
1809	static inline bool mmc_blk_oor_valid(struct mmc_blk_request *brq)
1810	{
1811	return !!brq->mrq.sbc;
1812	}
1813
1814	static inline u32 mmc_blk_stop_err_bits(struct mmc_blk_request *brq)
1815	{
1816	return mmc_blk_oor_valid(brq) ? CMD_ERRORS : CMD_ERRORS_EXCL_OOR;
1817	}
1818
1819	/*
1820	* Check for errors the host controller driver might not have seen such as
1821	* response mode errors or invalid card state.
1822	*/
1823	static bool mmc_blk_status_error(struct request *req, u32 status)
1824	{
1825	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1826	struct mmc_blk_request *brq = &mqrq->brq;
1827	struct mmc_queue *mq = req->q->queuedata;
1828	u32 stop_err_bits;
1829
1830	if (mmc_host_is_spi(mq->card->host))
1831	return false;
1832
1833	stop_err_bits = mmc_blk_stop_err_bits(brq);
1834
1835	return brq->cmd.resp[0] & CMD_ERRORS ||
1836	brq->stop.resp[0] & stop_err_bits ||
1837	status & stop_err_bits ||
1838	(rq_data_dir(req) == WRITE && !mmc_ready_for_data(status));
1839	}
1840
1841	static inline bool mmc_blk_cmd_started(struct mmc_blk_request *brq)
1842	{
1843	return !brq->sbc.error && !brq->cmd.error &&
1844	!(brq->cmd.resp[0] & CMD_ERRORS);
1845	}
1846
1847	/*
1848	* Requests are completed by mmc_blk_mq_complete_rq() which sets simple
1849	* policy:
1850	* 1. A request that has transferred at least some data is considered
1851	* successful and will be requeued if there is remaining data to
1852	* transfer.
1853	* 2. Otherwise the number of retries is incremented and the request
1854	* will be requeued if there are remaining retries.
1855	* 3. Otherwise the request will be errored out.
1856	* That means mmc_blk_mq_complete_rq() is controlled by bytes_xfered and
1857	* mqrq->retries. So there are only 4 possible actions here:
1858	* 1. do not accept the bytes_xfered value i.e. set it to zero
1859	* 2. change mqrq->retries to determine the number of retries
1860	* 3. try to reset the card
1861	* 4. read one sector at a time
1862	*/
1863	static void mmc_blk_mq_rw_recovery(struct mmc_queue *mq, struct request *req)
1864	{
1865	int type = rq_data_dir(req) == READ ? MMC_BLK_READ : MMC_BLK_WRITE;
1866	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
1867	struct mmc_blk_request *brq = &mqrq->brq;
1868	struct mmc_blk_data *md = mq->blkdata;
1869	struct mmc_card *card = mq->card;
1870	u32 status;
1871	u32 blocks;
1872	int err;
1873
1874	/*
1875	* Some errors the host driver might not have seen. Set the number of
1876	* bytes transferred to zero in that case.
1877	*/
1878	err = __mmc_send_status(card, status: &status, retries: 0);
1879	if (err || mmc_blk_status_error(req, status))
1880	brq->data.bytes_xfered = 0;
1881
1882	mmc_retune_release(host: card->host);
1883
1884	/*
1885	* Try again to get the status. This also provides an opportunity for
1886	* re-tuning.
1887	*/
1888	if (err)
1889	err = __mmc_send_status(card, status: &status, retries: 0);
1890
1891	/*
1892	* Nothing more to do after the number of bytes transferred has been
1893	* updated and there is no card.
1894	*/
1895	if (err && mmc_detect_card_removed(host: card->host))
1896	return;
1897
1898	/* Try to get back to "tran" state */
1899	if (!mmc_host_is_spi(mq->card->host) &&
1900	(err || !mmc_ready_for_data(status)))
1901	err = mmc_blk_fix_state(card: mq->card, req);
1902
1903	/*
1904	* Special case for SD cards where the card might record the number of
1905	* blocks written.
1906	*/
1907	if (!err && mmc_blk_cmd_started(brq) && mmc_card_sd(card) &&
1908	rq_data_dir(req) == WRITE) {
1909	if (mmc_sd_num_wr_blocks(card, written_blocks: &blocks))
1910	brq->data.bytes_xfered = 0;
1911	else
1912	brq->data.bytes_xfered = blocks << 9;
1913	}
1914
1915	/* Reset if the card is in a bad state */
1916	if (!mmc_host_is_spi(mq->card->host) &&
1917	err && mmc_blk_reset(md, host: card->host, type)) {
1918	pr_err("%s: recovery failed!\n", req->q->disk->disk_name);
1919	mqrq->retries = MMC_NO_RETRIES;
1920	return;
1921	}
1922
1923	/*
1924	* If anything was done, just return and if there is anything remaining
1925	* on the request it will get requeued.
1926	*/
1927	if (brq->data.bytes_xfered)
1928	return;
1929
1930	/* Reset before last retry */
1931	if (mqrq->retries + 1 == MMC_MAX_RETRIES &&
1932	mmc_blk_reset(md, host: card->host, type))
1933	return;
1934
1935	/* Command errors fail fast, so use all MMC_MAX_RETRIES */
1936	if (brq->sbc.error || brq->cmd.error)
1937	return;
1938
1939	/* Reduce the remaining retries for data errors */
1940	if (mqrq->retries < MMC_MAX_RETRIES - MMC_DATA_RETRIES) {
1941	mqrq->retries = MMC_MAX_RETRIES - MMC_DATA_RETRIES;
1942	return;
1943	}
1944
1945	if (rq_data_dir(req) == READ && brq->data.blocks >
1946	queue_physical_block_size(q: mq->queue) >> 9) {
1947	/* Read one (native) sector at a time */
1948	mmc_blk_read_single(mq, req);
1949	return;
1950	}
1951	}
1952
1953	static inline bool mmc_blk_rq_error(struct mmc_blk_request *brq)
1954	{
1955	mmc_blk_eval_resp_error(brq);
1956
1957	return brq->sbc.error || brq->cmd.error || brq->stop.error ||
1958	brq->data.error || brq->cmd.resp[0] & CMD_ERRORS;
1959	}
1960
1961	static int mmc_spi_err_check(struct mmc_card *card)
1962	{
1963	u32 status = 0;
1964	int err;
1965
1966	/*
1967	* SPI does not have a TRAN state we have to wait on, instead the
1968	* card is ready again when it no longer holds the line LOW.
1969	* We still have to ensure two things here before we know the write
1970	* was successful:
1971	* 1. The card has not disconnected during busy and we actually read our
1972	* own pull-up, thinking it was still connected, so ensure it
1973	* still responds.
1974	* 2. Check for any error bits, in particular R1_SPI_IDLE to catch a
1975	* just reconnected card after being disconnected during busy.
1976	*/
1977	err = __mmc_send_status(card, status: &status, retries: 0);
1978	if (err)
1979	return err;
1980	/* All R1 and R2 bits of SPI are errors in our case */
1981	if (status)
1982	return -EIO;
1983	return 0;
1984	}
1985
1986	static int mmc_blk_busy_cb(void *cb_data, bool *busy)
1987	{
1988	struct mmc_blk_busy_data *data = cb_data;
1989	u32 status = 0;
1990	int err;
1991
1992	err = mmc_send_status(card: data->card, status: &status);
1993	if (err)
1994	return err;
1995
1996	/* Accumulate response error bits. */
1997	data->status |= status;
1998
1999	*busy = !mmc_ready_for_data(status);
2000	return 0;
2001	}
2002
2003	static int mmc_blk_card_busy(struct mmc_card *card, struct request *req)
2004	{
2005	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
2006	struct mmc_blk_busy_data cb_data;
2007	int err;
2008
2009	if (rq_data_dir(req) == READ)
2010	return 0;
2011
2012	if (mmc_host_is_spi(card->host)) {
2013	err = mmc_spi_err_check(card);
2014	if (err)
2015	mqrq->brq.data.bytes_xfered = 0;
2016	return err;
2017	}
2018
2019	cb_data.card = card;
2020	cb_data.status = 0;
2021	err = __mmc_poll_for_busy(host: card->host, period_us: 0, MMC_BLK_TIMEOUT_MS,
2022	busy_cb: &mmc_blk_busy_cb, cb_data: &cb_data);
2023
2024	/*
2025	* Do not assume data transferred correctly if there are any error bits
2026	* set.
2027	*/
2028	if (cb_data.status & mmc_blk_stop_err_bits(brq: &mqrq->brq)) {
2029	mqrq->brq.data.bytes_xfered = 0;
2030	err = err ? err : -EIO;
2031	}
2032
2033	/* Copy the exception bit so it will be seen later on */
2034	if (mmc_card_mmc(card) && cb_data.status & R1_EXCEPTION_EVENT)
2035	mqrq->brq.cmd.resp[0] |= R1_EXCEPTION_EVENT;
2036
2037	return err;
2038	}
2039
2040	static inline void mmc_blk_rw_reset_success(struct mmc_queue *mq,
2041	struct request *req)
2042	{
2043	int type = rq_data_dir(req) == READ ? MMC_BLK_READ : MMC_BLK_WRITE;
2044
2045	mmc_blk_reset_success(md: mq->blkdata, type);
2046	}
2047
2048	static void mmc_blk_mq_complete_rq(struct mmc_queue *mq, struct request *req)
2049	{
2050	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
2051	unsigned int nr_bytes = mqrq->brq.data.bytes_xfered;
2052
2053	if (nr_bytes) {
2054	if (blk_update_request(rq: req, BLK_STS_OK, nr_bytes))
2055	blk_mq_requeue_request(rq: req, kick_requeue_list: true);
2056	else
2057	__blk_mq_end_request(rq: req, BLK_STS_OK);
2058	} else if (!blk_rq_bytes(rq: req)) {
2059	__blk_mq_end_request(rq: req, BLK_STS_IOERR);
2060	} else if (mqrq->retries++ < MMC_MAX_RETRIES) {
2061	blk_mq_requeue_request(rq: req, kick_requeue_list: true);
2062	} else {
2063	if (mmc_card_removed(mq->card))
2064	req->rq_flags |= RQF_QUIET;
2065	blk_mq_end_request(rq: req, BLK_STS_IOERR);
2066	}
2067	}
2068
2069	static bool mmc_blk_urgent_bkops_needed(struct mmc_queue *mq,
2070	struct mmc_queue_req *mqrq)
2071	{
2072	return mmc_card_mmc(mq->card) && !mmc_host_is_spi(mq->card->host) &&
2073	(mqrq->brq.cmd.resp[0] & R1_EXCEPTION_EVENT ||
2074	mqrq->brq.stop.resp[0] & R1_EXCEPTION_EVENT);
2075	}
2076
2077	static void mmc_blk_urgent_bkops(struct mmc_queue *mq,
2078	struct mmc_queue_req *mqrq)
2079	{
2080	if (mmc_blk_urgent_bkops_needed(mq, mqrq))
2081	mmc_run_bkops(card: mq->card);
2082	}
2083
2084	static void mmc_blk_hsq_req_done(struct mmc_request *mrq)
2085	{
2086	struct mmc_queue_req *mqrq =
2087	container_of(mrq, struct mmc_queue_req, brq.mrq);
2088	struct request *req = mmc_queue_req_to_req(mqr: mqrq);
2089	struct request_queue *q = req->q;
2090	struct mmc_queue *mq = q->queuedata;
2091	struct mmc_host *host = mq->card->host;
2092	unsigned long flags;
2093
2094	if (mmc_blk_rq_error(brq: &mqrq->brq) ||
2095	mmc_blk_urgent_bkops_needed(mq, mqrq)) {
2096	spin_lock_irqsave(&mq->lock, flags);
2097	mq->recovery_needed = true;
2098	mq->recovery_req = req;
2099	spin_unlock_irqrestore(lock: &mq->lock, flags);
2100
2101	host->cqe_ops->cqe_recovery_start(host);
2102
2103	schedule_work(work: &mq->recovery_work);
2104	return;
2105	}
2106
2107	mmc_blk_rw_reset_success(mq, req);
2108
2109	/*
2110	* Block layer timeouts race with completions which means the normal
2111	* completion path cannot be used during recovery.
2112	*/
2113	if (mq->in_recovery)
2114	mmc_blk_cqe_complete_rq(mq, req);
2115	else if (likely(!blk_should_fake_timeout(req->q)))
2116	blk_mq_complete_request(rq: req);
2117	}
2118
2119	void mmc_blk_mq_complete(struct request *req)
2120	{
2121	struct mmc_queue *mq = req->q->queuedata;
2122	struct mmc_host *host = mq->card->host;
2123
2124	if (host->cqe_enabled)
2125	mmc_blk_cqe_complete_rq(mq, req);
2126	else if (likely(!blk_should_fake_timeout(req->q)))
2127	mmc_blk_mq_complete_rq(mq, req);
2128	}
2129
2130	static void mmc_blk_mq_poll_completion(struct mmc_queue *mq,
2131	struct request *req)
2132	{
2133	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
2134	struct mmc_host *host = mq->card->host;
2135
2136	if (mmc_blk_rq_error(brq: &mqrq->brq) ||
2137	mmc_blk_card_busy(card: mq->card, req)) {
2138	mmc_blk_mq_rw_recovery(mq, req);
2139	} else {
2140	mmc_blk_rw_reset_success(mq, req);
2141	mmc_retune_release(host);
2142	}
2143
2144	mmc_blk_urgent_bkops(mq, mqrq);
2145	}
2146
2147	static void mmc_blk_mq_dec_in_flight(struct mmc_queue *mq, enum mmc_issue_type issue_type)
2148	{
2149	unsigned long flags;
2150	bool put_card;
2151
2152	spin_lock_irqsave(&mq->lock, flags);
2153
2154	mq->in_flight[issue_type] -= 1;
2155
2156	put_card = (mmc_tot_in_flight(mq) == 0);
2157
2158	spin_unlock_irqrestore(lock: &mq->lock, flags);
2159
2160	if (put_card)
2161	mmc_put_card(card: mq->card, ctx: &mq->ctx);
2162	}
2163
2164	static void mmc_blk_mq_post_req(struct mmc_queue *mq, struct request *req,
2165	bool can_sleep)
2166	{
2167	enum mmc_issue_type issue_type = mmc_issue_type(mq, req);
2168	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
2169	struct mmc_request *mrq = &mqrq->brq.mrq;
2170	struct mmc_host *host = mq->card->host;
2171
2172	mmc_post_req(host, mrq, err: 0);
2173
2174	/*
2175	* Block layer timeouts race with completions which means the normal
2176	* completion path cannot be used during recovery.
2177	*/
2178	if (mq->in_recovery) {
2179	mmc_blk_mq_complete_rq(mq, req);
2180	} else if (likely(!blk_should_fake_timeout(req->q))) {
2181	if (can_sleep)
2182	blk_mq_complete_request_direct(rq: req, complete: mmc_blk_mq_complete);
2183	else
2184	blk_mq_complete_request(rq: req);
2185	}
2186
2187	mmc_blk_mq_dec_in_flight(mq, issue_type);
2188	}
2189
2190	void mmc_blk_mq_recovery(struct mmc_queue *mq)
2191	{
2192	struct request *req = mq->recovery_req;
2193	struct mmc_host *host = mq->card->host;
2194	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
2195
2196	mq->recovery_req = NULL;
2197	mq->rw_wait = false;
2198
2199	if (mmc_blk_rq_error(brq: &mqrq->brq)) {
2200	mmc_retune_hold_now(host);
2201	mmc_blk_mq_rw_recovery(mq, req);
2202	}
2203
2204	mmc_blk_urgent_bkops(mq, mqrq);
2205
2206	mmc_blk_mq_post_req(mq, req, can_sleep: true);
2207	}
2208
2209	static void mmc_blk_mq_complete_prev_req(struct mmc_queue *mq,
2210	struct request **prev_req)
2211	{
2212	if (mmc_host_done_complete(host: mq->card->host))
2213	return;
2214
2215	mutex_lock(&mq->complete_lock);
2216
2217	if (!mq->complete_req)
2218	goto out_unlock;
2219
2220	mmc_blk_mq_poll_completion(mq, req: mq->complete_req);
2221
2222	if (prev_req)
2223	*prev_req = mq->complete_req;
2224	else
2225	mmc_blk_mq_post_req(mq, req: mq->complete_req, can_sleep: true);
2226
2227	mq->complete_req = NULL;
2228
2229	out_unlock:
2230	mutex_unlock(lock: &mq->complete_lock);
2231	}
2232
2233	void mmc_blk_mq_complete_work(struct work_struct *work)
2234	{
2235	struct mmc_queue *mq = container_of(work, struct mmc_queue,
2236	complete_work);
2237
2238	mmc_blk_mq_complete_prev_req(mq, NULL);
2239	}
2240
2241	static void mmc_blk_mq_req_done(struct mmc_request *mrq)
2242	{
2243	struct mmc_queue_req *mqrq = container_of(mrq, struct mmc_queue_req,
2244	brq.mrq);
2245	struct request *req = mmc_queue_req_to_req(mqr: mqrq);
2246	struct request_queue *q = req->q;
2247	struct mmc_queue *mq = q->queuedata;
2248	struct mmc_host *host = mq->card->host;
2249	unsigned long flags;
2250
2251	if (!mmc_host_done_complete(host)) {
2252	bool waiting;
2253
2254	/*
2255	* We cannot complete the request in this context, so record
2256	* that there is a request to complete, and that a following
2257	* request does not need to wait (although it does need to
2258	* complete complete_req first).
2259	*/
2260	spin_lock_irqsave(&mq->lock, flags);
2261	mq->complete_req = req;
2262	mq->rw_wait = false;
2263	waiting = mq->waiting;
2264	spin_unlock_irqrestore(lock: &mq->lock, flags);
2265
2266	/*
2267	* If 'waiting' then the waiting task will complete this
2268	* request, otherwise queue a work to do it. Note that
2269	* complete_work may still race with the dispatch of a following
2270	* request.
2271	*/
2272	if (waiting)
2273	wake_up(&mq->wait);
2274	else
2275	queue_work(wq: mq->card->complete_wq, work: &mq->complete_work);
2276
2277	return;
2278	}
2279
2280	/* Take the recovery path for errors or urgent background operations */
2281	if (mmc_blk_rq_error(brq: &mqrq->brq) ||
2282	mmc_blk_urgent_bkops_needed(mq, mqrq)) {
2283	spin_lock_irqsave(&mq->lock, flags);
2284	mq->recovery_needed = true;
2285	mq->recovery_req = req;
2286	spin_unlock_irqrestore(lock: &mq->lock, flags);
2287	wake_up(&mq->wait);
2288	schedule_work(work: &mq->recovery_work);
2289	return;
2290	}
2291
2292	mmc_blk_rw_reset_success(mq, req);
2293
2294	mq->rw_wait = false;
2295	wake_up(&mq->wait);
2296
2297	/* context unknown */
2298	mmc_blk_mq_post_req(mq, req, can_sleep: false);
2299	}
2300
2301	static bool mmc_blk_rw_wait_cond(struct mmc_queue *mq, int *err)
2302	{
2303	unsigned long flags;
2304	bool done;
2305
2306	/*
2307	* Wait while there is another request in progress, but not if recovery
2308	* is needed. Also indicate whether there is a request waiting to start.
2309	*/
2310	spin_lock_irqsave(&mq->lock, flags);
2311	if (mq->recovery_needed) {
2312	*err = -EBUSY;
2313	done = true;
2314	} else {
2315	done = !mq->rw_wait;
2316	}
2317	mq->waiting = !done;
2318	spin_unlock_irqrestore(lock: &mq->lock, flags);
2319
2320	return done;
2321	}
2322
2323	static int mmc_blk_rw_wait(struct mmc_queue *mq, struct request **prev_req)
2324	{
2325	int err = 0;
2326
2327	wait_event(mq->wait, mmc_blk_rw_wait_cond(mq, &err));
2328
2329	/* Always complete the previous request if there is one */
2330	mmc_blk_mq_complete_prev_req(mq, prev_req);
2331
2332	return err;
2333	}
2334
2335	static int mmc_blk_mq_issue_rw_rq(struct mmc_queue *mq,
2336	struct request *req)
2337	{
2338	struct mmc_queue_req *mqrq = req_to_mmc_queue_req(rq: req);
2339	struct mmc_host *host = mq->card->host;
2340	struct request *prev_req = NULL;
2341	int err = 0;
2342
2343	mmc_blk_rw_rq_prep(mqrq, card: mq->card, recovery_mode: 0, mq);
2344
2345	mqrq->brq.mrq.done = mmc_blk_mq_req_done;
2346
2347	mmc_pre_req(host, mrq: &mqrq->brq.mrq);
2348
2349	err = mmc_blk_rw_wait(mq, prev_req: &prev_req);
2350	if (err)
2351	goto out_post_req;
2352
2353	mq->rw_wait = true;
2354
2355	err = mmc_start_request(host, mrq: &mqrq->brq.mrq);
2356
2357	if (prev_req)
2358	mmc_blk_mq_post_req(mq, req: prev_req, can_sleep: true);
2359
2360	if (err)
2361	mq->rw_wait = false;
2362
2363	/* Release re-tuning here where there is no synchronization required */
2364	if (err || mmc_host_done_complete(host))
2365	mmc_retune_release(host);
2366
2367	out_post_req:
2368	if (err)
2369	mmc_post_req(host, mrq: &mqrq->brq.mrq, err);
2370
2371	return err;
2372	}
2373
2374	static int mmc_blk_wait_for_idle(struct mmc_queue *mq, struct mmc_host *host)
2375	{
2376	if (host->cqe_enabled)
2377	return host->cqe_ops->cqe_wait_for_idle(host);
2378
2379	return mmc_blk_rw_wait(mq, NULL);
2380	}
2381
2382	enum mmc_issued mmc_blk_mq_issue_rq(struct mmc_queue *mq, struct request *req)
2383	{
2384	struct mmc_blk_data *md = mq->blkdata;
2385	struct mmc_card *card = md->queue.card;
2386	struct mmc_host *host = card->host;
2387	int ret;
2388
2389	ret = mmc_blk_part_switch(card, part_type: md->part_type);
2390	if (ret)
2391	return MMC_REQ_FAILED_TO_START;
2392
2393	switch (mmc_issue_type(mq, req)) {
2394	case MMC_ISSUE_SYNC:
2395	ret = mmc_blk_wait_for_idle(mq, host);
2396	if (ret)
2397	return MMC_REQ_BUSY;
2398	switch (req_op(req)) {
2399	case REQ_OP_DRV_IN:
2400	case REQ_OP_DRV_OUT:
2401	mmc_blk_issue_drv_op(mq, req);
2402	break;
2403	case REQ_OP_DISCARD:
2404	mmc_blk_issue_discard_rq(mq, req);
2405	break;
2406	case REQ_OP_SECURE_ERASE:
2407	mmc_blk_issue_secdiscard_rq(mq, req);
2408	break;
2409	case REQ_OP_WRITE_ZEROES:
2410	mmc_blk_issue_trim_rq(mq, req);
2411	break;
2412	case REQ_OP_FLUSH:
2413	mmc_blk_issue_flush(mq, req);
2414	break;
2415	default:
2416	WARN_ON_ONCE(1);
2417	return MMC_REQ_FAILED_TO_START;
2418	}
2419	return MMC_REQ_FINISHED;
2420	case MMC_ISSUE_DCMD:
2421	case MMC_ISSUE_ASYNC:
2422	switch (req_op(req)) {
2423	case REQ_OP_FLUSH:
2424	if (!mmc_cache_enabled(host)) {
2425	blk_mq_end_request(rq: req, BLK_STS_OK);
2426	return MMC_REQ_FINISHED;
2427	}
2428	ret = mmc_blk_cqe_issue_flush(mq, req);
2429	break;
2430	case REQ_OP_WRITE:
2431	card->written_flag = true;
2432	fallthrough;
2433	case REQ_OP_READ:
2434	if (host->cqe_enabled)
2435	ret = mmc_blk_cqe_issue_rw_rq(mq, req);
2436	else
2437	ret = mmc_blk_mq_issue_rw_rq(mq, req);
2438	break;
2439	default:
2440	WARN_ON_ONCE(1);
2441	ret = -EINVAL;
2442	}
2443	if (!ret)
2444	return MMC_REQ_STARTED;
2445	return ret == -EBUSY ? MMC_REQ_BUSY : MMC_REQ_FAILED_TO_START;
2446	default:
2447	WARN_ON_ONCE(1);
2448	return MMC_REQ_FAILED_TO_START;
2449	}
2450	}
2451
2452	static inline int mmc_blk_readonly(struct mmc_card *card)
2453	{
2454	return mmc_card_readonly(card) ||
2455	!(card->csd.cmdclass & CCC_BLOCK_WRITE);
2456	}
2457
2458	static struct mmc_blk_data *mmc_blk_alloc_req(struct mmc_card *card,
2459	struct device *parent,
2460	sector_t size,
2461	bool default_ro,
2462	const char *subname,
2463	int area_type,
2464	unsigned int part_type)
2465	{
2466	struct mmc_blk_data *md;
2467	int devidx, ret;
2468	char cap_str[10];
2469	bool cache_enabled = false;
2470	bool fua_enabled = false;
2471
2472	devidx = ida_alloc_max(ida: &mmc_blk_ida, max: max_devices - 1, GFP_KERNEL);
2473	if (devidx < 0) {
2474	/*
2475	* We get -ENOSPC because there are no more any available
2476	* devidx. The reason may be that, either userspace haven't yet
2477	* unmounted the partitions, which postpones mmc_blk_release()
2478	* from being called, or the device has more partitions than
2479	* what we support.
2480	*/
2481	if (devidx == -ENOSPC)
2482	dev_err(mmc_dev(card->host),
2483	"no more device IDs available\n");
2484
2485	return ERR_PTR(error: devidx);
2486	}
2487
2488	md = kzalloc(size: sizeof(struct mmc_blk_data), GFP_KERNEL);
2489	if (!md) {
2490	ret = -ENOMEM;
2491	goto out;
2492	}
2493
2494	md->area_type = area_type;
2495
2496	/*
2497	* Set the read-only status based on the supported commands
2498	* and the write protect switch.
2499	*/
2500	md->read_only = mmc_blk_readonly(card);
2501
2502	md->disk = mmc_init_queue(mq: &md->queue, card);
2503	if (IS_ERR(ptr: md->disk)) {
2504	ret = PTR_ERR(ptr: md->disk);
2505	goto err_kfree;
2506	}
2507
2508	INIT_LIST_HEAD(list: &md->part);
2509	INIT_LIST_HEAD(list: &md->rpmbs);
2510	kref_init(kref: &md->kref);
2511
2512	md->queue.blkdata = md;
2513	md->part_type = part_type;
2514
2515	md->disk->major = MMC_BLOCK_MAJOR;
2516	md->disk->minors = perdev_minors;
2517	md->disk->first_minor = devidx * perdev_minors;
2518	md->disk->fops = &mmc_bdops;
2519	md->disk->private_data = md;
2520	md->parent = parent;
2521	set_disk_ro(disk: md->disk, read_only: md->read_only || default_ro);
2522	if (area_type & (MMC_BLK_DATA_AREA_RPMB | MMC_BLK_DATA_AREA_BOOT))
2523	md->disk->flags |= GENHD_FL_NO_PART;
2524
2525	/*
2526	* As discussed on lkml, GENHD_FL_REMOVABLE should:
2527	*
2528	* - be set for removable media with permanent block devices
2529	* - be unset for removable block devices with permanent media
2530	*
2531	* Since MMC block devices clearly fall under the second
2532	* case, we do not set GENHD_FL_REMOVABLE. Userspace
2533	* should use the block device creation/destruction hotplug
2534	* messages to tell when the card is present.
2535	*/
2536
2537	snprintf(buf: md->disk->disk_name, size: sizeof(md->disk->disk_name),
2538	fmt: "mmcblk%u%s", card->host->index, subname ? subname : "");
2539
2540	set_capacity(disk: md->disk, size);
2541
2542	if (mmc_host_cmd23(host: card->host)) {
2543	if ((mmc_card_mmc(card) &&
2544	card->csd.mmca_vsn >= CSD_SPEC_VER_3) ||
2545	(mmc_card_sd(card) &&
2546	card->scr.cmds & SD_SCR_CMD23_SUPPORT))
2547	md->flags |= MMC_BLK_CMD23;
2548	}
2549
2550	if (md->flags & MMC_BLK_CMD23 &&
2551	((card->ext_csd.rel_param & EXT_CSD_WR_REL_PARAM_EN) ||
2552	card->ext_csd.rel_sectors)) {
2553	md->flags |= MMC_BLK_REL_WR;
2554	fua_enabled = true;
2555	cache_enabled = true;
2556	}
2557	if (mmc_cache_enabled(host: card->host))
2558	cache_enabled = true;
2559
2560	blk_queue_write_cache(q: md->queue.queue, enabled: cache_enabled, fua: fua_enabled);
2561
2562	string_get_size(size: (u64)size, blk_size: 512, units: STRING_UNITS_2,
2563	buf: cap_str, len: sizeof(cap_str));
2564	pr_info("%s: %s %s %s%s\n",
2565	md->disk->disk_name, mmc_card_id(card), mmc_card_name(card),
2566	cap_str, md->read_only ? " (ro)" : "");
2567
2568	/* used in ->open, must be set before add_disk: */
2569	if (area_type == MMC_BLK_DATA_AREA_MAIN)
2570	dev_set_drvdata(dev: &card->dev, data: md);
2571	ret = device_add_disk(parent: md->parent, disk: md->disk, groups: mmc_disk_attr_groups);
2572	if (ret)
2573	goto err_put_disk;
2574	return md;
2575
2576	err_put_disk:
2577	put_disk(disk: md->disk);
2578	blk_mq_free_tag_set(set: &md->queue.tag_set);
2579	err_kfree:
2580	kfree(objp: md);
2581	out:
2582	ida_free(&mmc_blk_ida, id: devidx);
2583	return ERR_PTR(error: ret);
2584	}
2585
2586	static struct mmc_blk_data *mmc_blk_alloc(struct mmc_card *card)
2587	{
2588	sector_t size;
2589
2590	if (!mmc_card_sd(card) && mmc_card_blockaddr(card)) {
2591	/*
2592	* The EXT_CSD sector count is in number or 512 byte
2593	* sectors.
2594	*/
2595	size = card->ext_csd.sectors;
2596	} else {
2597	/*
2598	* The CSD capacity field is in units of read_blkbits.
2599	* set_capacity takes units of 512 bytes.
2600	*/
2601	size = (typeof(sector_t))card->csd.capacity
2602	<< (card->csd.read_blkbits - 9);
2603	}
2604
2605	return mmc_blk_alloc_req(card, parent: &card->dev, size, default_ro: false, NULL,
2606	MMC_BLK_DATA_AREA_MAIN, part_type: 0);
2607	}
2608
2609	static int mmc_blk_alloc_part(struct mmc_card *card,
2610	struct mmc_blk_data *md,
2611	unsigned int part_type,
2612	sector_t size,
2613	bool default_ro,
2614	const char *subname,
2615	int area_type)
2616	{
2617	struct mmc_blk_data *part_md;
2618
2619	part_md = mmc_blk_alloc_req(card, disk_to_dev(md->disk), size, default_ro,
2620	subname, area_type, part_type);
2621	if (IS_ERR(ptr: part_md))
2622	return PTR_ERR(ptr: part_md);
2623	list_add(new: &part_md->part, head: &md->part);
2624
2625	return 0;
2626	}
2627
2628	/**
2629	* mmc_rpmb_ioctl() - ioctl handler for the RPMB chardev
2630	* @filp: the character device file
2631	* @cmd: the ioctl() command
2632	* @arg: the argument from userspace
2633	*
2634	* This will essentially just redirect the ioctl()s coming in over to
2635	* the main block device spawning the RPMB character device.
2636	*/
2637	static long mmc_rpmb_ioctl(struct file *filp, unsigned int cmd,
2638	unsigned long arg)
2639	{
2640	struct mmc_rpmb_data *rpmb = filp->private_data;
2641	int ret;
2642
2643	switch (cmd) {
2644	case MMC_IOC_CMD:
2645	ret = mmc_blk_ioctl_cmd(md: rpmb->md,
2646	ic_ptr: (struct mmc_ioc_cmd __user *)arg,
2647	rpmb);
2648	break;
2649	case MMC_IOC_MULTI_CMD:
2650	ret = mmc_blk_ioctl_multi_cmd(md: rpmb->md,
2651	user: (struct mmc_ioc_multi_cmd __user *)arg,
2652	rpmb);
2653	break;
2654	default:
2655	ret = -EINVAL;
2656	break;
2657	}
2658
2659	return ret;
2660	}
2661
2662	#ifdef CONFIG_COMPAT
2663	static long mmc_rpmb_ioctl_compat(struct file *filp, unsigned int cmd,
2664	unsigned long arg)
2665	{
2666	return mmc_rpmb_ioctl(filp, cmd, arg: (unsigned long)compat_ptr(uptr: arg));
2667	}
2668	#endif
2669
2670	static int mmc_rpmb_chrdev_open(struct inode *inode, struct file *filp)
2671	{
2672	struct mmc_rpmb_data *rpmb = container_of(inode->i_cdev,
2673	struct mmc_rpmb_data, chrdev);
2674
2675	get_device(dev: &rpmb->dev);
2676	filp->private_data = rpmb;
2677	mmc_blk_get(disk: rpmb->md->disk);
2678
2679	return nonseekable_open(inode, filp);
2680	}
2681
2682	static int mmc_rpmb_chrdev_release(struct inode *inode, struct file *filp)
2683	{
2684	struct mmc_rpmb_data *rpmb = container_of(inode->i_cdev,
2685	struct mmc_rpmb_data, chrdev);
2686
2687	mmc_blk_put(md: rpmb->md);
2688	put_device(dev: &rpmb->dev);
2689
2690	return 0;
2691	}
2692
2693	static const struct file_operations mmc_rpmb_fileops = {
2694	.release = mmc_rpmb_chrdev_release,
2695	.open = mmc_rpmb_chrdev_open,
2696	.owner = THIS_MODULE,
2697	.llseek = no_llseek,
2698	.unlocked_ioctl = mmc_rpmb_ioctl,
2699	#ifdef CONFIG_COMPAT
2700	.compat_ioctl = mmc_rpmb_ioctl_compat,
2701	#endif
2702	};
2703
2704	static void mmc_blk_rpmb_device_release(struct device *dev)
2705	{
2706	struct mmc_rpmb_data *rpmb = dev_get_drvdata(dev);
2707
2708	ida_free(&mmc_rpmb_ida, id: rpmb->id);
2709	kfree(objp: rpmb);
2710	}
2711
2712	static int mmc_blk_alloc_rpmb_part(struct mmc_card *card,
2713	struct mmc_blk_data *md,
2714	unsigned int part_index,
2715	sector_t size,
2716	const char *subname)
2717	{
2718	int devidx, ret;
2719	char rpmb_name[DISK_NAME_LEN];
2720	char cap_str[10];
2721	struct mmc_rpmb_data *rpmb;
2722
2723	/* This creates the minor number for the RPMB char device */
2724	devidx = ida_alloc_max(ida: &mmc_rpmb_ida, max: max_devices - 1, GFP_KERNEL);
2725	if (devidx < 0)
2726	return devidx;
2727
2728	rpmb = kzalloc(size: sizeof(*rpmb), GFP_KERNEL);
2729	if (!rpmb) {
2730	ida_free(&mmc_rpmb_ida, id: devidx);
2731	return -ENOMEM;
2732	}
2733
2734	snprintf(buf: rpmb_name, size: sizeof(rpmb_name),
2735	fmt: "mmcblk%u%s", card->host->index, subname ? subname : "");
2736
2737	rpmb->id = devidx;
2738	rpmb->part_index = part_index;
2739	rpmb->dev.init_name = rpmb_name;
2740	rpmb->dev.bus = &mmc_rpmb_bus_type;
2741	rpmb->dev.devt = MKDEV(MAJOR(mmc_rpmb_devt), rpmb->id);
2742	rpmb->dev.parent = &card->dev;
2743	rpmb->dev.release = mmc_blk_rpmb_device_release;
2744	device_initialize(dev: &rpmb->dev);
2745	dev_set_drvdata(dev: &rpmb->dev, data: rpmb);
2746	rpmb->md = md;
2747
2748	cdev_init(&rpmb->chrdev, &mmc_rpmb_fileops);
2749	rpmb->chrdev.owner = THIS_MODULE;
2750	ret = cdev_device_add(cdev: &rpmb->chrdev, dev: &rpmb->dev);
2751	if (ret) {
2752	pr_err("%s: could not add character device\n", rpmb_name);
2753	goto out_put_device;
2754	}
2755
2756	list_add(new: &rpmb->node, head: &md->rpmbs);
2757
2758	string_get_size(size: (u64)size, blk_size: 512, units: STRING_UNITS_2,
2759	buf: cap_str, len: sizeof(cap_str));
2760
2761	pr_info("%s: %s %s %s, chardev (%d:%d)\n",
2762	rpmb_name, mmc_card_id(card), mmc_card_name(card), cap_str,
2763	MAJOR(mmc_rpmb_devt), rpmb->id);
2764
2765	return 0;
2766
2767	out_put_device:
2768	put_device(dev: &rpmb->dev);
2769	return ret;
2770	}
2771
2772	static void mmc_blk_remove_rpmb_part(struct mmc_rpmb_data *rpmb)
2773
2774	{
2775	cdev_device_del(cdev: &rpmb->chrdev, dev: &rpmb->dev);
2776	put_device(dev: &rpmb->dev);
2777	}
2778
2779	/* MMC Physical partitions consist of two boot partitions and
2780	* up to four general purpose partitions.
2781	* For each partition enabled in EXT_CSD a block device will be allocatedi
2782	* to provide access to the partition.
2783	*/
2784
2785	static int mmc_blk_alloc_parts(struct mmc_card *card, struct mmc_blk_data *md)
2786	{
2787	int idx, ret;
2788
2789	if (!mmc_card_mmc(card))
2790	return 0;
2791
2792	for (idx = 0; idx < card->nr_parts; idx++) {
2793	if (card->part[idx].area_type & MMC_BLK_DATA_AREA_RPMB) {
2794	/*
2795	* RPMB partitions does not provide block access, they
2796	* are only accessed using ioctl():s. Thus create
2797	* special RPMB block devices that do not have a
2798	* backing block queue for these.
2799	*/
2800	ret = mmc_blk_alloc_rpmb_part(card, md,
2801	part_index: card->part[idx].part_cfg,
2802	size: card->part[idx].size >> 9,
2803	subname: card->part[idx].name);
2804	if (ret)
2805	return ret;
2806	} else if (card->part[idx].size) {
2807	ret = mmc_blk_alloc_part(card, md,
2808	part_type: card->part[idx].part_cfg,
2809	size: card->part[idx].size >> 9,
2810	default_ro: card->part[idx].force_ro,
2811	subname: card->part[idx].name,
2812	area_type: card->part[idx].area_type);
2813	if (ret)
2814	return ret;
2815	}
2816	}
2817
2818	return 0;
2819	}
2820
2821	static void mmc_blk_remove_req(struct mmc_blk_data *md)
2822	{
2823	/*
2824	* Flush remaining requests and free queues. It is freeing the queue
2825	* that stops new requests from being accepted.
2826	*/
2827	del_gendisk(gp: md->disk);
2828	mmc_cleanup_queue(&md->queue);
2829	mmc_blk_put(md);
2830	}
2831
2832	static void mmc_blk_remove_parts(struct mmc_card *card,
2833	struct mmc_blk_data *md)
2834	{
2835	struct list_head *pos, *q;
2836	struct mmc_blk_data *part_md;
2837	struct mmc_rpmb_data *rpmb;
2838
2839	/* Remove RPMB partitions */
2840	list_for_each_safe(pos, q, &md->rpmbs) {
2841	rpmb = list_entry(pos, struct mmc_rpmb_data, node);
2842	list_del(entry: pos);
2843	mmc_blk_remove_rpmb_part(rpmb);
2844	}
2845	/* Remove block partitions */
2846	list_for_each_safe(pos, q, &md->part) {
2847	part_md = list_entry(pos, struct mmc_blk_data, part);
2848	list_del(entry: pos);
2849	mmc_blk_remove_req(md: part_md);
2850	}
2851	}
2852
2853	#ifdef CONFIG_DEBUG_FS
2854
2855	static int mmc_dbg_card_status_get(void *data, u64 *val)
2856	{
2857	struct mmc_card *card = data;
2858	struct mmc_blk_data *md = dev_get_drvdata(dev: &card->dev);
2859	struct mmc_queue *mq = &md->queue;
2860	struct request *req;
2861	int ret;
2862
2863	/* Ask the block layer about the card status */
2864	req = blk_mq_alloc_request(q: mq->queue, opf: REQ_OP_DRV_IN, flags: 0);
2865	if (IS_ERR(ptr: req))
2866	return PTR_ERR(ptr: req);
2867	req_to_mmc_queue_req(rq: req)->drv_op = MMC_DRV_OP_GET_CARD_STATUS;
2868	req_to_mmc_queue_req(rq: req)->drv_op_result = -EIO;
2869	blk_execute_rq(rq: req, at_head: false);
2870	ret = req_to_mmc_queue_req(rq: req)->drv_op_result;
2871	if (ret >= 0) {
2872	*val = ret;
2873	ret = 0;
2874	}
2875	blk_mq_free_request(rq: req);
2876
2877	return ret;
2878	}
2879	DEFINE_DEBUGFS_ATTRIBUTE(mmc_dbg_card_status_fops, mmc_dbg_card_status_get,
2880	NULL, "%08llx\n");
2881
2882	/* That is two digits * 512 + 1 for newline */
2883	#define EXT_CSD_STR_LEN 1025
2884
2885	static int mmc_ext_csd_open(struct inode *inode, struct file *filp)
2886	{
2887	struct mmc_card *card = inode->i_private;
2888	struct mmc_blk_data *md = dev_get_drvdata(dev: &card->dev);
2889	struct mmc_queue *mq = &md->queue;
2890	struct request *req;
2891	char *buf;
2892	ssize_t n = 0;
2893	u8 *ext_csd;
2894	int err, i;
2895
2896	buf = kmalloc(EXT_CSD_STR_LEN + 1, GFP_KERNEL);
2897	if (!buf)
2898	return -ENOMEM;
2899
2900	/* Ask the block layer for the EXT CSD */
2901	req = blk_mq_alloc_request(q: mq->queue, opf: REQ_OP_DRV_IN, flags: 0);
2902	if (IS_ERR(ptr: req)) {
2903	err = PTR_ERR(ptr: req);
2904	goto out_free;
2905	}
2906	req_to_mmc_queue_req(rq: req)->drv_op = MMC_DRV_OP_GET_EXT_CSD;
2907	req_to_mmc_queue_req(rq: req)->drv_op_result = -EIO;
2908	req_to_mmc_queue_req(rq: req)->drv_op_data = &ext_csd;
2909	blk_execute_rq(rq: req, at_head: false);
2910	err = req_to_mmc_queue_req(rq: req)->drv_op_result;
2911	blk_mq_free_request(rq: req);
2912	if (err) {
2913	pr_err("FAILED %d\n", err);
2914	goto out_free;
2915	}
2916
2917	for (i = 0; i < 512; i++)
2918	n += sprintf(buf: buf + n, fmt: "%02x", ext_csd[i]);
2919	n += sprintf(buf: buf + n, fmt: "\n");
2920
2921	if (n != EXT_CSD_STR_LEN) {
2922	err = -EINVAL;
2923	kfree(objp: ext_csd);
2924	goto out_free;
2925	}
2926
2927	filp->private_data = buf;
2928	kfree(objp: ext_csd);
2929	return 0;
2930
2931	out_free:
2932	kfree(objp: buf);
2933	return err;
2934	}
2935
2936	static ssize_t mmc_ext_csd_read(struct file *filp, char __user *ubuf,
2937	size_t cnt, loff_t *ppos)
2938	{
2939	char *buf = filp->private_data;
2940
2941	return simple_read_from_buffer(to: ubuf, count: cnt, ppos,
2942	from: buf, EXT_CSD_STR_LEN);
2943	}
2944
2945	static int mmc_ext_csd_release(struct inode *inode, struct file *file)
2946	{
2947	kfree(objp: file->private_data);
2948	return 0;
2949	}
2950
2951	static const struct file_operations mmc_dbg_ext_csd_fops = {
2952	.open = mmc_ext_csd_open,
2953	.read = mmc_ext_csd_read,
2954	.release = mmc_ext_csd_release,
2955	.llseek = default_llseek,
2956	};
2957
2958	static void mmc_blk_add_debugfs(struct mmc_card *card, struct mmc_blk_data *md)
2959	{
2960	struct dentry *root;
2961
2962	if (!card->debugfs_root)
2963	return;
2964
2965	root = card->debugfs_root;
2966
2967	if (mmc_card_mmc(card) || mmc_card_sd(card)) {
2968	md->status_dentry =
2969	debugfs_create_file_unsafe(name: "status", mode: 0400, parent: root,
2970	data: card,
2971	fops: &mmc_dbg_card_status_fops);
2972	}
2973
2974	if (mmc_card_mmc(card)) {
2975	md->ext_csd_dentry =
2976	debugfs_create_file(name: "ext_csd", S_IRUSR, parent: root, data: card,
2977	fops: &mmc_dbg_ext_csd_fops);
2978	}
2979	}
2980
2981	static void mmc_blk_remove_debugfs(struct mmc_card *card,
2982	struct mmc_blk_data *md)
2983	{
2984	if (!card->debugfs_root)
2985	return;
2986
2987	debugfs_remove(dentry: md->status_dentry);
2988	md->status_dentry = NULL;
2989
2990	debugfs_remove(dentry: md->ext_csd_dentry);
2991	md->ext_csd_dentry = NULL;
2992	}
2993
2994	#else
2995
2996	static void mmc_blk_add_debugfs(struct mmc_card *card, struct mmc_blk_data *md)
2997	{
2998	}
2999
3000	static void mmc_blk_remove_debugfs(struct mmc_card *card,
3001	struct mmc_blk_data *md)
3002	{
3003	}
3004
3005	#endif /* CONFIG_DEBUG_FS */
3006
3007	static int mmc_blk_probe(struct mmc_card *card)
3008	{
3009	struct mmc_blk_data *md;
3010	int ret = 0;
3011
3012	/*
3013	* Check that the card supports the command class(es) we need.
3014	*/
3015	if (!(card->csd.cmdclass & CCC_BLOCK_READ))
3016	return -ENODEV;
3017
3018	mmc_fixup_device(card, table: mmc_blk_fixups);
3019
3020	card->complete_wq = alloc_workqueue(fmt: "mmc_complete",
3021	flags: WQ_MEM_RECLAIM | WQ_HIGHPRI, max_active: 0);
3022	if (!card->complete_wq) {
3023	pr_err("Failed to create mmc completion workqueue");
3024	return -ENOMEM;
3025	}
3026
3027	md = mmc_blk_alloc(card);
3028	if (IS_ERR(ptr: md)) {
3029	ret = PTR_ERR(ptr: md);
3030	goto out_free;
3031	}
3032
3033	ret = mmc_blk_alloc_parts(card, md);
3034	if (ret)
3035	goto out;
3036
3037	/* Add two debugfs entries */
3038	mmc_blk_add_debugfs(card, md);
3039
3040	pm_runtime_set_autosuspend_delay(dev: &card->dev, delay: 3000);
3041	pm_runtime_use_autosuspend(dev: &card->dev);
3042
3043	/*
3044	* Don't enable runtime PM for SD-combo cards here. Leave that
3045	* decision to be taken during the SDIO init sequence instead.
3046	*/
3047	if (!mmc_card_sd_combo(card)) {
3048	pm_runtime_set_active(dev: &card->dev);
3049	pm_runtime_enable(dev: &card->dev);
3050	}
3051
3052	return 0;
3053
3054	out:
3055	mmc_blk_remove_parts(card, md);
3056	mmc_blk_remove_req(md);
3057	out_free:
3058	destroy_workqueue(wq: card->complete_wq);
3059	return ret;
3060	}
3061
3062	static void mmc_blk_remove(struct mmc_card *card)
3063	{
3064	struct mmc_blk_data *md = dev_get_drvdata(dev: &card->dev);
3065
3066	mmc_blk_remove_debugfs(card, md);
3067	mmc_blk_remove_parts(card, md);
3068	pm_runtime_get_sync(dev: &card->dev);
3069	if (md->part_curr != md->part_type) {
3070	mmc_claim_host(host: card->host);
3071	mmc_blk_part_switch(card, part_type: md->part_type);
3072	mmc_release_host(host: card->host);
3073	}
3074	if (!mmc_card_sd_combo(card))
3075	pm_runtime_disable(dev: &card->dev);
3076	pm_runtime_put_noidle(dev: &card->dev);
3077	mmc_blk_remove_req(md);
3078	destroy_workqueue(wq: card->complete_wq);
3079	}
3080
3081	static int _mmc_blk_suspend(struct mmc_card *card)
3082	{
3083	struct mmc_blk_data *part_md;
3084	struct mmc_blk_data *md = dev_get_drvdata(dev: &card->dev);
3085
3086	if (md) {
3087	mmc_queue_suspend(&md->queue);
3088	list_for_each_entry(part_md, &md->part, part) {
3089	mmc_queue_suspend(&part_md->queue);
3090	}
3091	}
3092	return 0;
3093	}
3094
3095	static void mmc_blk_shutdown(struct mmc_card *card)
3096	{
3097	_mmc_blk_suspend(card);
3098	}
3099
3100	#ifdef CONFIG_PM_SLEEP
3101	static int mmc_blk_suspend(struct device *dev)
3102	{
3103	struct mmc_card *card = mmc_dev_to_card(dev);
3104
3105	return _mmc_blk_suspend(card);
3106	}
3107
3108	static int mmc_blk_resume(struct device *dev)
3109	{
3110	struct mmc_blk_data *part_md;
3111	struct mmc_blk_data *md = dev_get_drvdata(dev);
3112
3113	if (md) {
3114	/*
3115	* Resume involves the card going into idle state,
3116	* so current partition is always the main one.
3117	*/
3118	md->part_curr = md->part_type;
3119	mmc_queue_resume(&md->queue);
3120	list_for_each_entry(part_md, &md->part, part) {
3121	mmc_queue_resume(&part_md->queue);
3122	}
3123	}
3124	return 0;
3125	}
3126	#endif
3127
3128	static SIMPLE_DEV_PM_OPS(mmc_blk_pm_ops, mmc_blk_suspend, mmc_blk_resume);
3129
3130	static struct mmc_driver mmc_driver = {
3131	.drv = {
3132	.name = "mmcblk",
3133	.pm = &mmc_blk_pm_ops,
3134	},
3135	.probe = mmc_blk_probe,
3136	.remove = mmc_blk_remove,
3137	.shutdown = mmc_blk_shutdown,
3138	};
3139
3140	static int __init mmc_blk_init(void)
3141	{
3142	int res;
3143
3144	res = bus_register(bus: &mmc_rpmb_bus_type);
3145	if (res < 0) {
3146	pr_err("mmcblk: could not register RPMB bus type\n");
3147	return res;
3148	}
3149	res = alloc_chrdev_region(&mmc_rpmb_devt, 0, MAX_DEVICES, "rpmb");
3150	if (res < 0) {
3151	pr_err("mmcblk: failed to allocate rpmb chrdev region\n");
3152	goto out_bus_unreg;
3153	}
3154
3155	if (perdev_minors != CONFIG_MMC_BLOCK_MINORS)
3156	pr_info("mmcblk: using %d minors per device\n", perdev_minors);
3157
3158	max_devices = min(MAX_DEVICES, (1 << MINORBITS) / perdev_minors);
3159
3160	res = register_blkdev(MMC_BLOCK_MAJOR, "mmc");
3161	if (res)
3162	goto out_chrdev_unreg;
3163
3164	res = mmc_register_driver(drv: &mmc_driver);
3165	if (res)
3166	goto out_blkdev_unreg;
3167
3168	return 0;
3169
3170	out_blkdev_unreg:
3171	unregister_blkdev(MMC_BLOCK_MAJOR, name: "mmc");
3172	out_chrdev_unreg:
3173	unregister_chrdev_region(mmc_rpmb_devt, MAX_DEVICES);
3174	out_bus_unreg:
3175	bus_unregister(bus: &mmc_rpmb_bus_type);
3176	return res;
3177	}
3178
3179	static void __exit mmc_blk_exit(void)
3180	{
3181	mmc_unregister_driver(drv: &mmc_driver);
3182	unregister_blkdev(MMC_BLOCK_MAJOR, name: "mmc");
3183	unregister_chrdev_region(mmc_rpmb_devt, MAX_DEVICES);
3184	bus_unregister(bus: &mmc_rpmb_bus_type);
3185	}
3186
3187	module_init(mmc_blk_init);
3188	module_exit(mmc_blk_exit);
3189
3190	MODULE_LICENSE("GPL");
3191	MODULE_DESCRIPTION("Multimedia Card (MMC) block device driver");
3192