1 // SPDX-License-Identifier: GPL-2.0-only
2 /*
3  * Copyright (c) 2022 Qualcomm Innovation Center. All rights reserved.
4  *
5  * Authors:
6  *	Asutosh Das <quic_asutoshd@quicinc.com>
7  *	Can Guo <quic_cang@quicinc.com>
8  */
9 
10 #include <asm/unaligned.h>
11 #include <linux/dma-mapping.h>
12 #include <linux/module.h>
13 #include <linux/platform_device.h>
14 #include "ufshcd-priv.h"
15 #include <linux/delay.h>
16 #include <scsi/scsi_cmnd.h>
17 #include <linux/bitfield.h>
18 #include <linux/iopoll.h>
19 
20 #define MAX_QUEUE_SUP GENMASK(7, 0)
21 #define UFS_MCQ_MIN_RW_QUEUES 2
22 #define UFS_MCQ_MIN_READ_QUEUES 0
23 #define UFS_MCQ_MIN_POLL_QUEUES 0
24 #define QUEUE_EN_OFFSET 31
25 #define QUEUE_ID_OFFSET 16
26 
27 #define MCQ_CFG_MAC_MASK	GENMASK(16, 8)
28 #define MCQ_QCFG_SIZE		0x40
29 #define MCQ_ENTRY_SIZE_IN_DWORD	8
30 #define CQE_UCD_BA GENMASK_ULL(63, 7)
31 
32 /* Max mcq register polling time in microseconds */
33 #define MCQ_POLL_US 500000
34 
rw_queue_count_set(const char * val,const struct kernel_param * kp)35 static int rw_queue_count_set(const char *val, const struct kernel_param *kp)
36 {
37 	return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_RW_QUEUES,
38 				     num_possible_cpus());
39 }
40 
41 static const struct kernel_param_ops rw_queue_count_ops = {
42 	.set = rw_queue_count_set,
43 	.get = param_get_uint,
44 };
45 
46 static unsigned int rw_queues;
47 module_param_cb(rw_queues, &rw_queue_count_ops, &rw_queues, 0644);
48 MODULE_PARM_DESC(rw_queues,
49 		 "Number of interrupt driven I/O queues used for rw. Default value is nr_cpus");
50 
read_queue_count_set(const char * val,const struct kernel_param * kp)51 static int read_queue_count_set(const char *val, const struct kernel_param *kp)
52 {
53 	return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_READ_QUEUES,
54 				     num_possible_cpus());
55 }
56 
57 static const struct kernel_param_ops read_queue_count_ops = {
58 	.set = read_queue_count_set,
59 	.get = param_get_uint,
60 };
61 
62 static unsigned int read_queues;
63 module_param_cb(read_queues, &read_queue_count_ops, &read_queues, 0644);
64 MODULE_PARM_DESC(read_queues,
65 		 "Number of interrupt driven read queues used for read. Default value is 0");
66 
poll_queue_count_set(const char * val,const struct kernel_param * kp)67 static int poll_queue_count_set(const char *val, const struct kernel_param *kp)
68 {
69 	return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_POLL_QUEUES,
70 				     num_possible_cpus());
71 }
72 
73 static const struct kernel_param_ops poll_queue_count_ops = {
74 	.set = poll_queue_count_set,
75 	.get = param_get_uint,
76 };
77 
78 static unsigned int poll_queues = 1;
79 module_param_cb(poll_queues, &poll_queue_count_ops, &poll_queues, 0644);
80 MODULE_PARM_DESC(poll_queues,
81 		 "Number of poll queues used for r/w. Default value is 1");
82 
83 /**
84  * ufshcd_mcq_config_mac - Set the #Max Activ Cmds.
85  * @hba: per adapter instance
86  * @max_active_cmds: maximum # of active commands to the device at any time.
87  *
88  * The controller won't send more than the max_active_cmds to the device at
89  * any time.
90  */
ufshcd_mcq_config_mac(struct ufs_hba * hba,u32 max_active_cmds)91 void ufshcd_mcq_config_mac(struct ufs_hba *hba, u32 max_active_cmds)
92 {
93 	u32 val;
94 
95 	val = ufshcd_readl(hba, REG_UFS_MCQ_CFG);
96 	val &= ~MCQ_CFG_MAC_MASK;
97 	val |= FIELD_PREP(MCQ_CFG_MAC_MASK, max_active_cmds);
98 	ufshcd_writel(hba, val, REG_UFS_MCQ_CFG);
99 }
100 EXPORT_SYMBOL_GPL(ufshcd_mcq_config_mac);
101 
102 /**
103  * ufshcd_mcq_req_to_hwq - find the hardware queue on which the
104  * request would be issued.
105  * @hba: per adapter instance
106  * @req: pointer to the request to be issued
107  *
108  * Return: the hardware queue instance on which the request would
109  * be queued.
110  */
ufshcd_mcq_req_to_hwq(struct ufs_hba * hba,struct request * req)111 struct ufs_hw_queue *ufshcd_mcq_req_to_hwq(struct ufs_hba *hba,
112 					 struct request *req)
113 {
114 	u32 utag = blk_mq_unique_tag(req);
115 	u32 hwq = blk_mq_unique_tag_to_hwq(utag);
116 
117 	return &hba->uhq[hwq];
118 }
119 
120 /**
121  * ufshcd_mcq_decide_queue_depth - decide the queue depth
122  * @hba: per adapter instance
123  *
124  * Return: queue-depth on success, non-zero on error
125  *
126  * MAC - Max. Active Command of the Host Controller (HC)
127  * HC wouldn't send more than this commands to the device.
128  * It is mandatory to implement get_hba_mac() to enable MCQ mode.
129  * Calculates and adjusts the queue depth based on the depth
130  * supported by the HC and ufs device.
131  */
ufshcd_mcq_decide_queue_depth(struct ufs_hba * hba)132 int ufshcd_mcq_decide_queue_depth(struct ufs_hba *hba)
133 {
134 	int mac;
135 
136 	/* Mandatory to implement get_hba_mac() */
137 	mac = ufshcd_mcq_vops_get_hba_mac(hba);
138 	if (mac < 0) {
139 		dev_err(hba->dev, "Failed to get mac, err=%d\n", mac);
140 		return mac;
141 	}
142 
143 	WARN_ON_ONCE(!hba->dev_info.bqueuedepth);
144 	/*
145 	 * max. value of bqueuedepth = 256, mac is host dependent.
146 	 * It is mandatory for UFS device to define bQueueDepth if
147 	 * shared queuing architecture is enabled.
148 	 */
149 	return min_t(int, mac, hba->dev_info.bqueuedepth);
150 }
151 
ufshcd_mcq_config_nr_queues(struct ufs_hba * hba)152 static int ufshcd_mcq_config_nr_queues(struct ufs_hba *hba)
153 {
154 	int i;
155 	u32 hba_maxq, rem, tot_queues;
156 	struct Scsi_Host *host = hba->host;
157 
158 	/* maxq is 0 based value */
159 	hba_maxq = FIELD_GET(MAX_QUEUE_SUP, hba->mcq_capabilities) + 1;
160 
161 	tot_queues = read_queues + poll_queues + rw_queues;
162 
163 	if (hba_maxq < tot_queues) {
164 		dev_err(hba->dev, "Total queues (%d) exceeds HC capacity (%d)\n",
165 			tot_queues, hba_maxq);
166 		return -EOPNOTSUPP;
167 	}
168 
169 	rem = hba_maxq;
170 
171 	if (rw_queues) {
172 		hba->nr_queues[HCTX_TYPE_DEFAULT] = rw_queues;
173 		rem -= hba->nr_queues[HCTX_TYPE_DEFAULT];
174 	} else {
175 		rw_queues = num_possible_cpus();
176 	}
177 
178 	if (poll_queues) {
179 		hba->nr_queues[HCTX_TYPE_POLL] = poll_queues;
180 		rem -= hba->nr_queues[HCTX_TYPE_POLL];
181 	}
182 
183 	if (read_queues) {
184 		hba->nr_queues[HCTX_TYPE_READ] = read_queues;
185 		rem -= hba->nr_queues[HCTX_TYPE_READ];
186 	}
187 
188 	if (!hba->nr_queues[HCTX_TYPE_DEFAULT])
189 		hba->nr_queues[HCTX_TYPE_DEFAULT] = min3(rem, rw_queues,
190 							 num_possible_cpus());
191 
192 	for (i = 0; i < HCTX_MAX_TYPES; i++)
193 		host->nr_hw_queues += hba->nr_queues[i];
194 
195 	hba->nr_hw_queues = host->nr_hw_queues;
196 	return 0;
197 }
198 
ufshcd_mcq_memory_alloc(struct ufs_hba * hba)199 int ufshcd_mcq_memory_alloc(struct ufs_hba *hba)
200 {
201 	struct ufs_hw_queue *hwq;
202 	size_t utrdl_size, cqe_size;
203 	int i;
204 
205 	for (i = 0; i < hba->nr_hw_queues; i++) {
206 		hwq = &hba->uhq[i];
207 
208 		utrdl_size = sizeof(struct utp_transfer_req_desc) *
209 			     hwq->max_entries;
210 		hwq->sqe_base_addr = dmam_alloc_coherent(hba->dev, utrdl_size,
211 							 &hwq->sqe_dma_addr,
212 							 GFP_KERNEL);
213 		if (!hwq->sqe_dma_addr) {
214 			dev_err(hba->dev, "SQE allocation failed\n");
215 			return -ENOMEM;
216 		}
217 
218 		cqe_size = sizeof(struct cq_entry) * hwq->max_entries;
219 		hwq->cqe_base_addr = dmam_alloc_coherent(hba->dev, cqe_size,
220 							 &hwq->cqe_dma_addr,
221 							 GFP_KERNEL);
222 		if (!hwq->cqe_dma_addr) {
223 			dev_err(hba->dev, "CQE allocation failed\n");
224 			return -ENOMEM;
225 		}
226 	}
227 
228 	return 0;
229 }
230 
231 
232 /* Operation and runtime registers configuration */
233 #define MCQ_CFG_n(r, i)	((r) + MCQ_QCFG_SIZE * (i))
234 #define MCQ_OPR_OFFSET_n(p, i) \
235 	(hba->mcq_opr[(p)].offset + hba->mcq_opr[(p)].stride * (i))
236 
mcq_opr_base(struct ufs_hba * hba,enum ufshcd_mcq_opr n,int i)237 static void __iomem *mcq_opr_base(struct ufs_hba *hba,
238 					 enum ufshcd_mcq_opr n, int i)
239 {
240 	struct ufshcd_mcq_opr_info_t *opr = &hba->mcq_opr[n];
241 
242 	return opr->base + opr->stride * i;
243 }
244 
ufshcd_mcq_read_cqis(struct ufs_hba * hba,int i)245 u32 ufshcd_mcq_read_cqis(struct ufs_hba *hba, int i)
246 {
247 	return readl(mcq_opr_base(hba, OPR_CQIS, i) + REG_CQIS);
248 }
249 EXPORT_SYMBOL_GPL(ufshcd_mcq_read_cqis);
250 
ufshcd_mcq_write_cqis(struct ufs_hba * hba,u32 val,int i)251 void ufshcd_mcq_write_cqis(struct ufs_hba *hba, u32 val, int i)
252 {
253 	writel(val, mcq_opr_base(hba, OPR_CQIS, i) + REG_CQIS);
254 }
255 EXPORT_SYMBOL_GPL(ufshcd_mcq_write_cqis);
256 
257 /*
258  * Current MCQ specification doesn't provide a Task Tag or its equivalent in
259  * the Completion Queue Entry. Find the Task Tag using an indirect method.
260  */
ufshcd_mcq_get_tag(struct ufs_hba * hba,struct ufs_hw_queue * hwq,struct cq_entry * cqe)261 static int ufshcd_mcq_get_tag(struct ufs_hba *hba,
262 				     struct ufs_hw_queue *hwq,
263 				     struct cq_entry *cqe)
264 {
265 	u64 addr;
266 
267 	/* sizeof(struct utp_transfer_cmd_desc) must be a multiple of 128 */
268 	BUILD_BUG_ON(sizeof(struct utp_transfer_cmd_desc) & GENMASK(6, 0));
269 
270 	/* Bits 63:7 UCD base address, 6:5 are reserved, 4:0 is SQ ID */
271 	addr = (le64_to_cpu(cqe->command_desc_base_addr) & CQE_UCD_BA) -
272 		hba->ucdl_dma_addr;
273 
274 	return div_u64(addr, ufshcd_get_ucd_size(hba));
275 }
276 
ufshcd_mcq_process_cqe(struct ufs_hba * hba,struct ufs_hw_queue * hwq)277 static void ufshcd_mcq_process_cqe(struct ufs_hba *hba,
278 				   struct ufs_hw_queue *hwq)
279 {
280 	struct cq_entry *cqe = ufshcd_mcq_cur_cqe(hwq);
281 	int tag = ufshcd_mcq_get_tag(hba, hwq, cqe);
282 
283 	if (cqe->command_desc_base_addr) {
284 		ufshcd_compl_one_cqe(hba, tag, cqe);
285 		/* After processed the cqe, mark it empty (invalid) entry */
286 		cqe->command_desc_base_addr = 0;
287 	}
288 }
289 
ufshcd_mcq_compl_all_cqes_lock(struct ufs_hba * hba,struct ufs_hw_queue * hwq)290 void ufshcd_mcq_compl_all_cqes_lock(struct ufs_hba *hba,
291 				    struct ufs_hw_queue *hwq)
292 {
293 	unsigned long flags;
294 	u32 entries = hwq->max_entries;
295 
296 	spin_lock_irqsave(&hwq->cq_lock, flags);
297 	while (entries > 0) {
298 		ufshcd_mcq_process_cqe(hba, hwq);
299 		ufshcd_mcq_inc_cq_head_slot(hwq);
300 		entries--;
301 	}
302 
303 	ufshcd_mcq_update_cq_tail_slot(hwq);
304 	hwq->cq_head_slot = hwq->cq_tail_slot;
305 	spin_unlock_irqrestore(&hwq->cq_lock, flags);
306 }
307 
ufshcd_mcq_poll_cqe_lock(struct ufs_hba * hba,struct ufs_hw_queue * hwq)308 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
309 				       struct ufs_hw_queue *hwq)
310 {
311 	unsigned long completed_reqs = 0;
312 	unsigned long flags;
313 
314 	spin_lock_irqsave(&hwq->cq_lock, flags);
315 	ufshcd_mcq_update_cq_tail_slot(hwq);
316 	while (!ufshcd_mcq_is_cq_empty(hwq)) {
317 		ufshcd_mcq_process_cqe(hba, hwq);
318 		ufshcd_mcq_inc_cq_head_slot(hwq);
319 		completed_reqs++;
320 	}
321 
322 	if (completed_reqs)
323 		ufshcd_mcq_update_cq_head(hwq);
324 	spin_unlock_irqrestore(&hwq->cq_lock, flags);
325 
326 	return completed_reqs;
327 }
328 EXPORT_SYMBOL_GPL(ufshcd_mcq_poll_cqe_lock);
329 
ufshcd_mcq_make_queues_operational(struct ufs_hba * hba)330 void ufshcd_mcq_make_queues_operational(struct ufs_hba *hba)
331 {
332 	struct ufs_hw_queue *hwq;
333 	u16 qsize;
334 	int i;
335 
336 	for (i = 0; i < hba->nr_hw_queues; i++) {
337 		hwq = &hba->uhq[i];
338 		hwq->id = i;
339 		qsize = hwq->max_entries * MCQ_ENTRY_SIZE_IN_DWORD - 1;
340 
341 		/* Submission Queue Lower Base Address */
342 		ufsmcq_writelx(hba, lower_32_bits(hwq->sqe_dma_addr),
343 			      MCQ_CFG_n(REG_SQLBA, i));
344 		/* Submission Queue Upper Base Address */
345 		ufsmcq_writelx(hba, upper_32_bits(hwq->sqe_dma_addr),
346 			      MCQ_CFG_n(REG_SQUBA, i));
347 		/* Submission Queue Doorbell Address Offset */
348 		ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_SQD, i),
349 			      MCQ_CFG_n(REG_SQDAO, i));
350 		/* Submission Queue Interrupt Status Address Offset */
351 		ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_SQIS, i),
352 			      MCQ_CFG_n(REG_SQISAO, i));
353 
354 		/* Completion Queue Lower Base Address */
355 		ufsmcq_writelx(hba, lower_32_bits(hwq->cqe_dma_addr),
356 			      MCQ_CFG_n(REG_CQLBA, i));
357 		/* Completion Queue Upper Base Address */
358 		ufsmcq_writelx(hba, upper_32_bits(hwq->cqe_dma_addr),
359 			      MCQ_CFG_n(REG_CQUBA, i));
360 		/* Completion Queue Doorbell Address Offset */
361 		ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_CQD, i),
362 			      MCQ_CFG_n(REG_CQDAO, i));
363 		/* Completion Queue Interrupt Status Address Offset */
364 		ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_CQIS, i),
365 			      MCQ_CFG_n(REG_CQISAO, i));
366 
367 		/* Save the base addresses for quicker access */
368 		hwq->mcq_sq_head = mcq_opr_base(hba, OPR_SQD, i) + REG_SQHP;
369 		hwq->mcq_sq_tail = mcq_opr_base(hba, OPR_SQD, i) + REG_SQTP;
370 		hwq->mcq_cq_head = mcq_opr_base(hba, OPR_CQD, i) + REG_CQHP;
371 		hwq->mcq_cq_tail = mcq_opr_base(hba, OPR_CQD, i) + REG_CQTP;
372 
373 		/* Reinitializing is needed upon HC reset */
374 		hwq->sq_tail_slot = hwq->cq_tail_slot = hwq->cq_head_slot = 0;
375 
376 		/* Enable Tail Entry Push Status interrupt only for non-poll queues */
377 		if (i < hba->nr_hw_queues - hba->nr_queues[HCTX_TYPE_POLL])
378 			writel(1, mcq_opr_base(hba, OPR_CQIS, i) + REG_CQIE);
379 
380 		/* Completion Queue Enable|Size to Completion Queue Attribute */
381 		ufsmcq_writel(hba, (1 << QUEUE_EN_OFFSET) | qsize,
382 			      MCQ_CFG_n(REG_CQATTR, i));
383 
384 		/*
385 		 * Submission Qeueue Enable|Size|Completion Queue ID to
386 		 * Submission Queue Attribute
387 		 */
388 		ufsmcq_writel(hba, (1 << QUEUE_EN_OFFSET) | qsize |
389 			      (i << QUEUE_ID_OFFSET),
390 			      MCQ_CFG_n(REG_SQATTR, i));
391 	}
392 }
393 EXPORT_SYMBOL_GPL(ufshcd_mcq_make_queues_operational);
394 
ufshcd_mcq_enable_esi(struct ufs_hba * hba)395 void ufshcd_mcq_enable_esi(struct ufs_hba *hba)
396 {
397 	ufshcd_writel(hba, ufshcd_readl(hba, REG_UFS_MEM_CFG) | 0x2,
398 		      REG_UFS_MEM_CFG);
399 }
400 EXPORT_SYMBOL_GPL(ufshcd_mcq_enable_esi);
401 
ufshcd_mcq_config_esi(struct ufs_hba * hba,struct msi_msg * msg)402 void ufshcd_mcq_config_esi(struct ufs_hba *hba, struct msi_msg *msg)
403 {
404 	ufshcd_writel(hba, msg->address_lo, REG_UFS_ESILBA);
405 	ufshcd_writel(hba, msg->address_hi, REG_UFS_ESIUBA);
406 }
407 EXPORT_SYMBOL_GPL(ufshcd_mcq_config_esi);
408 
ufshcd_mcq_init(struct ufs_hba * hba)409 int ufshcd_mcq_init(struct ufs_hba *hba)
410 {
411 	struct Scsi_Host *host = hba->host;
412 	struct ufs_hw_queue *hwq;
413 	int ret, i;
414 
415 	ret = ufshcd_mcq_config_nr_queues(hba);
416 	if (ret)
417 		return ret;
418 
419 	ret = ufshcd_vops_mcq_config_resource(hba);
420 	if (ret)
421 		return ret;
422 
423 	ret = ufshcd_mcq_vops_op_runtime_config(hba);
424 	if (ret) {
425 		dev_err(hba->dev, "Operation runtime config failed, ret=%d\n",
426 			ret);
427 		return ret;
428 	}
429 	hba->uhq = devm_kzalloc(hba->dev,
430 				hba->nr_hw_queues * sizeof(struct ufs_hw_queue),
431 				GFP_KERNEL);
432 	if (!hba->uhq) {
433 		dev_err(hba->dev, "ufs hw queue memory allocation failed\n");
434 		return -ENOMEM;
435 	}
436 
437 	for (i = 0; i < hba->nr_hw_queues; i++) {
438 		hwq = &hba->uhq[i];
439 		hwq->max_entries = hba->nutrs + 1;
440 		spin_lock_init(&hwq->sq_lock);
441 		spin_lock_init(&hwq->cq_lock);
442 		mutex_init(&hwq->sq_mutex);
443 	}
444 
445 	/* The very first HW queue serves device commands */
446 	hba->dev_cmd_queue = &hba->uhq[0];
447 
448 	host->host_tagset = 1;
449 	return 0;
450 }
451 
ufshcd_mcq_sq_stop(struct ufs_hba * hba,struct ufs_hw_queue * hwq)452 static int ufshcd_mcq_sq_stop(struct ufs_hba *hba, struct ufs_hw_queue *hwq)
453 {
454 	void __iomem *reg;
455 	u32 id = hwq->id, val;
456 	int err;
457 
458 	if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
459 		return -ETIMEDOUT;
460 
461 	writel(SQ_STOP, mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTC);
462 	reg = mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTS;
463 	err = read_poll_timeout(readl, val, val & SQ_STS, 20,
464 				MCQ_POLL_US, false, reg);
465 	if (err)
466 		dev_err(hba->dev, "%s: failed. hwq-id=%d, err=%d\n",
467 			__func__, id, err);
468 	return err;
469 }
470 
ufshcd_mcq_sq_start(struct ufs_hba * hba,struct ufs_hw_queue * hwq)471 static int ufshcd_mcq_sq_start(struct ufs_hba *hba, struct ufs_hw_queue *hwq)
472 {
473 	void __iomem *reg;
474 	u32 id = hwq->id, val;
475 	int err;
476 
477 	if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
478 		return -ETIMEDOUT;
479 
480 	writel(SQ_START, mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTC);
481 	reg = mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTS;
482 	err = read_poll_timeout(readl, val, !(val & SQ_STS), 20,
483 				MCQ_POLL_US, false, reg);
484 	if (err)
485 		dev_err(hba->dev, "%s: failed. hwq-id=%d, err=%d\n",
486 			__func__, id, err);
487 	return err;
488 }
489 
490 /**
491  * ufshcd_mcq_sq_cleanup - Clean up submission queue resources
492  * associated with the pending command.
493  * @hba: per adapter instance.
494  * @task_tag: The command's task tag.
495  *
496  * Return: 0 for success; error code otherwise.
497  */
ufshcd_mcq_sq_cleanup(struct ufs_hba * hba,int task_tag)498 int ufshcd_mcq_sq_cleanup(struct ufs_hba *hba, int task_tag)
499 {
500 	struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
501 	struct scsi_cmnd *cmd = lrbp->cmd;
502 	struct ufs_hw_queue *hwq;
503 	void __iomem *reg, *opr_sqd_base;
504 	u32 nexus, id, val;
505 	int err;
506 
507 	if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
508 		return -ETIMEDOUT;
509 
510 	if (task_tag != hba->nutrs - UFSHCD_NUM_RESERVED) {
511 		if (!cmd)
512 			return -EINVAL;
513 		hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(cmd));
514 	} else {
515 		hwq = hba->dev_cmd_queue;
516 	}
517 
518 	id = hwq->id;
519 
520 	mutex_lock(&hwq->sq_mutex);
521 
522 	/* stop the SQ fetching before working on it */
523 	err = ufshcd_mcq_sq_stop(hba, hwq);
524 	if (err)
525 		goto unlock;
526 
527 	/* SQCTI = EXT_IID, IID, LUN, Task Tag */
528 	nexus = lrbp->lun << 8 | task_tag;
529 	opr_sqd_base = mcq_opr_base(hba, OPR_SQD, id);
530 	writel(nexus, opr_sqd_base + REG_SQCTI);
531 
532 	/* SQRTCy.ICU = 1 */
533 	writel(SQ_ICU, opr_sqd_base + REG_SQRTC);
534 
535 	/* Poll SQRTSy.CUS = 1. Return result from SQRTSy.RTC */
536 	reg = opr_sqd_base + REG_SQRTS;
537 	err = read_poll_timeout(readl, val, val & SQ_CUS, 20,
538 				MCQ_POLL_US, false, reg);
539 	if (err)
540 		dev_err(hba->dev, "%s: failed. hwq=%d, tag=%d err=%ld\n",
541 			__func__, id, task_tag,
542 			FIELD_GET(SQ_ICU_ERR_CODE_MASK, readl(reg)));
543 
544 	if (ufshcd_mcq_sq_start(hba, hwq))
545 		err = -ETIMEDOUT;
546 
547 unlock:
548 	mutex_unlock(&hwq->sq_mutex);
549 	return err;
550 }
551 
552 /**
553  * ufshcd_mcq_nullify_sqe - Nullify the submission queue entry.
554  * Write the sqe's Command Type to 0xF. The host controller will not
555  * fetch any sqe with Command Type = 0xF.
556  *
557  * @utrd: UTP Transfer Request Descriptor to be nullified.
558  */
ufshcd_mcq_nullify_sqe(struct utp_transfer_req_desc * utrd)559 static void ufshcd_mcq_nullify_sqe(struct utp_transfer_req_desc *utrd)
560 {
561 	utrd->header.command_type = 0xf;
562 }
563 
564 /**
565  * ufshcd_mcq_sqe_search - Search for the command in the submission queue
566  * If the command is in the submission queue and not issued to the device yet,
567  * nullify the sqe so the host controller will skip fetching the sqe.
568  *
569  * @hba: per adapter instance.
570  * @hwq: Hardware Queue to be searched.
571  * @task_tag: The command's task tag.
572  *
573  * Return: true if the SQE containing the command is present in the SQ
574  * (not fetched by the controller); returns false if the SQE is not in the SQ.
575  */
ufshcd_mcq_sqe_search(struct ufs_hba * hba,struct ufs_hw_queue * hwq,int task_tag)576 static bool ufshcd_mcq_sqe_search(struct ufs_hba *hba,
577 				  struct ufs_hw_queue *hwq, int task_tag)
578 {
579 	struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
580 	struct utp_transfer_req_desc *utrd;
581 	__le64  cmd_desc_base_addr;
582 	bool ret = false;
583 	u64 addr, match;
584 	u32 sq_head_slot;
585 
586 	if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
587 		return true;
588 
589 	mutex_lock(&hwq->sq_mutex);
590 
591 	ufshcd_mcq_sq_stop(hba, hwq);
592 	sq_head_slot = ufshcd_mcq_get_sq_head_slot(hwq);
593 	if (sq_head_slot == hwq->sq_tail_slot)
594 		goto out;
595 
596 	cmd_desc_base_addr = lrbp->utr_descriptor_ptr->command_desc_base_addr;
597 	addr = le64_to_cpu(cmd_desc_base_addr) & CQE_UCD_BA;
598 
599 	while (sq_head_slot != hwq->sq_tail_slot) {
600 		utrd = hwq->sqe_base_addr +
601 				sq_head_slot * sizeof(struct utp_transfer_req_desc);
602 		match = le64_to_cpu(utrd->command_desc_base_addr) & CQE_UCD_BA;
603 		if (addr == match) {
604 			ufshcd_mcq_nullify_sqe(utrd);
605 			ret = true;
606 			goto out;
607 		}
608 
609 		sq_head_slot++;
610 		if (sq_head_slot == hwq->max_entries)
611 			sq_head_slot = 0;
612 	}
613 
614 out:
615 	ufshcd_mcq_sq_start(hba, hwq);
616 	mutex_unlock(&hwq->sq_mutex);
617 	return ret;
618 }
619 
620 /**
621  * ufshcd_mcq_abort - Abort the command in MCQ.
622  * @cmd: The command to be aborted.
623  *
624  * Return: SUCCESS or FAILED error codes
625  */
ufshcd_mcq_abort(struct scsi_cmnd * cmd)626 int ufshcd_mcq_abort(struct scsi_cmnd *cmd)
627 {
628 	struct Scsi_Host *host = cmd->device->host;
629 	struct ufs_hba *hba = shost_priv(host);
630 	int tag = scsi_cmd_to_rq(cmd)->tag;
631 	struct ufshcd_lrb *lrbp = &hba->lrb[tag];
632 	struct ufs_hw_queue *hwq;
633 	unsigned long flags;
634 	int err = FAILED;
635 
636 	if (!ufshcd_cmd_inflight(lrbp->cmd)) {
637 		dev_err(hba->dev,
638 			"%s: skip abort. cmd at tag %d already completed.\n",
639 			__func__, tag);
640 		goto out;
641 	}
642 
643 	/* Skip task abort in case previous aborts failed and report failure */
644 	if (lrbp->req_abort_skip) {
645 		dev_err(hba->dev, "%s: skip abort. tag %d failed earlier\n",
646 			__func__, tag);
647 		goto out;
648 	}
649 
650 	hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(cmd));
651 
652 	if (ufshcd_mcq_sqe_search(hba, hwq, tag)) {
653 		/*
654 		 * Failure. The command should not be "stuck" in SQ for
655 		 * a long time which resulted in command being aborted.
656 		 */
657 		dev_err(hba->dev, "%s: cmd found in sq. hwq=%d, tag=%d\n",
658 			__func__, hwq->id, tag);
659 		goto out;
660 	}
661 
662 	/*
663 	 * The command is not in the submission queue, and it is not
664 	 * in the completion queue either. Query the device to see if
665 	 * the command is being processed in the device.
666 	 */
667 	if (ufshcd_try_to_abort_task(hba, tag)) {
668 		dev_err(hba->dev, "%s: device abort failed %d\n", __func__, err);
669 		lrbp->req_abort_skip = true;
670 		goto out;
671 	}
672 
673 	err = SUCCESS;
674 	spin_lock_irqsave(&hwq->cq_lock, flags);
675 	if (ufshcd_cmd_inflight(lrbp->cmd))
676 		ufshcd_release_scsi_cmd(hba, lrbp);
677 	spin_unlock_irqrestore(&hwq->cq_lock, flags);
678 
679 out:
680 	return err;
681 }
682