1 // SPDX-License-Identifier: GPL-2.0-only
3 * Copyright (c) 2022 Qualcomm Innovation Center. All rights reserved.
6 * Asutosh Das <quic_asutoshd@quicinc.com>
7 * Can Guo <quic_cang@quicinc.com>
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>
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
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)
32 /* Max mcq register polling time in microseconds */
33 #define MCQ_POLL_US 500000
35 static int rw_queue_count_set(const char *val, const struct kernel_param *kp)
37 return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_RW_QUEUES,
41 static const struct kernel_param_ops rw_queue_count_ops = {
42 .set = rw_queue_count_set,
43 .get = param_get_uint,
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");
51 static int read_queue_count_set(const char *val, const struct kernel_param *kp)
53 return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_READ_QUEUES,
57 static const struct kernel_param_ops read_queue_count_ops = {
58 .set = read_queue_count_set,
59 .get = param_get_uint,
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");
67 static int poll_queue_count_set(const char *val, const struct kernel_param *kp)
69 return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_POLL_QUEUES,
73 static const struct kernel_param_ops poll_queue_count_ops = {
74 .set = poll_queue_count_set,
75 .get = param_get_uint,
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");
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.
88 * The controller won't send more than the max_active_cmds to the device at
91 void ufshcd_mcq_config_mac(struct ufs_hba *hba, u32 max_active_cmds)
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 - 1);
98 ufshcd_writel(hba, val, REG_UFS_MCQ_CFG);
100 EXPORT_SYMBOL_GPL(ufshcd_mcq_config_mac);
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
108 * Return: the hardware queue instance on which the request would
111 struct ufs_hw_queue *ufshcd_mcq_req_to_hwq(struct ufs_hba *hba,
114 u32 utag = blk_mq_unique_tag(req);
115 u32 hwq = blk_mq_unique_tag_to_hwq(utag);
117 return &hba->uhq[hwq];
121 * ufshcd_mcq_decide_queue_depth - decide the queue depth
122 * @hba: per adapter instance
124 * Return: queue-depth on success, non-zero on error
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.
132 int ufshcd_mcq_decide_queue_depth(struct ufs_hba *hba)
136 /* Mandatory to implement get_hba_mac() */
137 mac = ufshcd_mcq_vops_get_hba_mac(hba);
139 dev_err(hba->dev, "Failed to get mac, err=%d\n", mac);
143 WARN_ON_ONCE(!hba->dev_info.bqueuedepth);
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.
149 return min_t(int, mac, hba->dev_info.bqueuedepth);
152 static int ufshcd_mcq_config_nr_queues(struct ufs_hba *hba)
155 u32 hba_maxq, rem, tot_queues;
156 struct Scsi_Host *host = hba->host;
158 /* maxq is 0 based value */
159 hba_maxq = FIELD_GET(MAX_QUEUE_SUP, hba->mcq_capabilities) + 1;
161 tot_queues = read_queues + poll_queues + rw_queues;
163 if (hba_maxq < tot_queues) {
164 dev_err(hba->dev, "Total queues (%d) exceeds HC capacity (%d)\n",
165 tot_queues, hba_maxq);
172 hba->nr_queues[HCTX_TYPE_DEFAULT] = rw_queues;
173 rem -= hba->nr_queues[HCTX_TYPE_DEFAULT];
175 rw_queues = num_possible_cpus();
179 hba->nr_queues[HCTX_TYPE_POLL] = poll_queues;
180 rem -= hba->nr_queues[HCTX_TYPE_POLL];
184 hba->nr_queues[HCTX_TYPE_READ] = read_queues;
185 rem -= hba->nr_queues[HCTX_TYPE_READ];
188 if (!hba->nr_queues[HCTX_TYPE_DEFAULT])
189 hba->nr_queues[HCTX_TYPE_DEFAULT] = min3(rem, rw_queues,
190 num_possible_cpus());
192 for (i = 0; i < HCTX_MAX_TYPES; i++)
193 host->nr_hw_queues += hba->nr_queues[i];
195 hba->nr_hw_queues = host->nr_hw_queues;
199 int ufshcd_mcq_memory_alloc(struct ufs_hba *hba)
201 struct ufs_hw_queue *hwq;
202 size_t utrdl_size, cqe_size;
205 for (i = 0; i < hba->nr_hw_queues; i++) {
208 utrdl_size = sizeof(struct utp_transfer_req_desc) *
210 hwq->sqe_base_addr = dmam_alloc_coherent(hba->dev, utrdl_size,
213 if (!hwq->sqe_dma_addr) {
214 dev_err(hba->dev, "SQE allocation failed\n");
218 cqe_size = sizeof(struct cq_entry) * hwq->max_entries;
219 hwq->cqe_base_addr = dmam_alloc_coherent(hba->dev, cqe_size,
222 if (!hwq->cqe_dma_addr) {
223 dev_err(hba->dev, "CQE allocation failed\n");
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))
237 static void __iomem *mcq_opr_base(struct ufs_hba *hba,
238 enum ufshcd_mcq_opr n, int i)
240 struct ufshcd_mcq_opr_info_t *opr = &hba->mcq_opr[n];
242 return opr->base + opr->stride * i;
245 u32 ufshcd_mcq_read_cqis(struct ufs_hba *hba, int i)
247 return readl(mcq_opr_base(hba, OPR_CQIS, i) + REG_CQIS);
249 EXPORT_SYMBOL_GPL(ufshcd_mcq_read_cqis);
251 void ufshcd_mcq_write_cqis(struct ufs_hba *hba, u32 val, int i)
253 writel(val, mcq_opr_base(hba, OPR_CQIS, i) + REG_CQIS);
255 EXPORT_SYMBOL_GPL(ufshcd_mcq_write_cqis);
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.
261 static int ufshcd_mcq_get_tag(struct ufs_hba *hba, struct cq_entry *cqe)
265 /* sizeof(struct utp_transfer_cmd_desc) must be a multiple of 128 */
266 BUILD_BUG_ON(sizeof(struct utp_transfer_cmd_desc) & GENMASK(6, 0));
268 /* Bits 63:7 UCD base address, 6:5 are reserved, 4:0 is SQ ID */
269 addr = (le64_to_cpu(cqe->command_desc_base_addr) & CQE_UCD_BA) -
272 return div_u64(addr, ufshcd_get_ucd_size(hba));
275 static void ufshcd_mcq_process_cqe(struct ufs_hba *hba,
276 struct ufs_hw_queue *hwq)
278 struct cq_entry *cqe = ufshcd_mcq_cur_cqe(hwq);
279 int tag = ufshcd_mcq_get_tag(hba, cqe);
281 if (cqe->command_desc_base_addr) {
282 ufshcd_compl_one_cqe(hba, tag, cqe);
283 /* After processed the cqe, mark it empty (invalid) entry */
284 cqe->command_desc_base_addr = 0;
288 void ufshcd_mcq_compl_all_cqes_lock(struct ufs_hba *hba,
289 struct ufs_hw_queue *hwq)
292 u32 entries = hwq->max_entries;
294 spin_lock_irqsave(&hwq->cq_lock, flags);
295 while (entries > 0) {
296 ufshcd_mcq_process_cqe(hba, hwq);
297 ufshcd_mcq_inc_cq_head_slot(hwq);
301 ufshcd_mcq_update_cq_tail_slot(hwq);
302 hwq->cq_head_slot = hwq->cq_tail_slot;
303 spin_unlock_irqrestore(&hwq->cq_lock, flags);
306 unsigned long ufshcd_mcq_poll_cqe_lock(struct ufs_hba *hba,
307 struct ufs_hw_queue *hwq)
309 unsigned long completed_reqs = 0;
312 spin_lock_irqsave(&hwq->cq_lock, flags);
313 ufshcd_mcq_update_cq_tail_slot(hwq);
314 while (!ufshcd_mcq_is_cq_empty(hwq)) {
315 ufshcd_mcq_process_cqe(hba, hwq);
316 ufshcd_mcq_inc_cq_head_slot(hwq);
321 ufshcd_mcq_update_cq_head(hwq);
322 spin_unlock_irqrestore(&hwq->cq_lock, flags);
324 return completed_reqs;
326 EXPORT_SYMBOL_GPL(ufshcd_mcq_poll_cqe_lock);
328 void ufshcd_mcq_make_queues_operational(struct ufs_hba *hba)
330 struct ufs_hw_queue *hwq;
334 for (i = 0; i < hba->nr_hw_queues; i++) {
337 qsize = hwq->max_entries * MCQ_ENTRY_SIZE_IN_DWORD - 1;
339 /* Submission Queue Lower Base Address */
340 ufsmcq_writelx(hba, lower_32_bits(hwq->sqe_dma_addr),
341 MCQ_CFG_n(REG_SQLBA, i));
342 /* Submission Queue Upper Base Address */
343 ufsmcq_writelx(hba, upper_32_bits(hwq->sqe_dma_addr),
344 MCQ_CFG_n(REG_SQUBA, i));
345 /* Submission Queue Doorbell Address Offset */
346 ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_SQD, i),
347 MCQ_CFG_n(REG_SQDAO, i));
348 /* Submission Queue Interrupt Status Address Offset */
349 ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_SQIS, i),
350 MCQ_CFG_n(REG_SQISAO, i));
352 /* Completion Queue Lower Base Address */
353 ufsmcq_writelx(hba, lower_32_bits(hwq->cqe_dma_addr),
354 MCQ_CFG_n(REG_CQLBA, i));
355 /* Completion Queue Upper Base Address */
356 ufsmcq_writelx(hba, upper_32_bits(hwq->cqe_dma_addr),
357 MCQ_CFG_n(REG_CQUBA, i));
358 /* Completion Queue Doorbell Address Offset */
359 ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_CQD, i),
360 MCQ_CFG_n(REG_CQDAO, i));
361 /* Completion Queue Interrupt Status Address Offset */
362 ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_CQIS, i),
363 MCQ_CFG_n(REG_CQISAO, i));
365 /* Save the base addresses for quicker access */
366 hwq->mcq_sq_head = mcq_opr_base(hba, OPR_SQD, i) + REG_SQHP;
367 hwq->mcq_sq_tail = mcq_opr_base(hba, OPR_SQD, i) + REG_SQTP;
368 hwq->mcq_cq_head = mcq_opr_base(hba, OPR_CQD, i) + REG_CQHP;
369 hwq->mcq_cq_tail = mcq_opr_base(hba, OPR_CQD, i) + REG_CQTP;
371 /* Reinitializing is needed upon HC reset */
372 hwq->sq_tail_slot = hwq->cq_tail_slot = hwq->cq_head_slot = 0;
374 /* Enable Tail Entry Push Status interrupt only for non-poll queues */
375 if (i < hba->nr_hw_queues - hba->nr_queues[HCTX_TYPE_POLL])
376 writel(1, mcq_opr_base(hba, OPR_CQIS, i) + REG_CQIE);
378 /* Completion Queue Enable|Size to Completion Queue Attribute */
379 ufsmcq_writel(hba, (1 << QUEUE_EN_OFFSET) | qsize,
380 MCQ_CFG_n(REG_CQATTR, i));
383 * Submission Qeueue Enable|Size|Completion Queue ID to
384 * Submission Queue Attribute
386 ufsmcq_writel(hba, (1 << QUEUE_EN_OFFSET) | qsize |
387 (i << QUEUE_ID_OFFSET),
388 MCQ_CFG_n(REG_SQATTR, i));
391 EXPORT_SYMBOL_GPL(ufshcd_mcq_make_queues_operational);
393 void ufshcd_mcq_enable_esi(struct ufs_hba *hba)
395 ufshcd_writel(hba, ufshcd_readl(hba, REG_UFS_MEM_CFG) | 0x2,
398 EXPORT_SYMBOL_GPL(ufshcd_mcq_enable_esi);
400 void ufshcd_mcq_enable(struct ufs_hba *hba)
402 ufshcd_rmwl(hba, MCQ_MODE_SELECT, MCQ_MODE_SELECT, REG_UFS_MEM_CFG);
404 EXPORT_SYMBOL_GPL(ufshcd_mcq_enable);
406 void ufshcd_mcq_config_esi(struct ufs_hba *hba, struct msi_msg *msg)
408 ufshcd_writel(hba, msg->address_lo, REG_UFS_ESILBA);
409 ufshcd_writel(hba, msg->address_hi, REG_UFS_ESIUBA);
411 EXPORT_SYMBOL_GPL(ufshcd_mcq_config_esi);
413 int ufshcd_mcq_init(struct ufs_hba *hba)
415 struct Scsi_Host *host = hba->host;
416 struct ufs_hw_queue *hwq;
419 ret = ufshcd_mcq_config_nr_queues(hba);
423 ret = ufshcd_vops_mcq_config_resource(hba);
427 ret = ufshcd_mcq_vops_op_runtime_config(hba);
429 dev_err(hba->dev, "Operation runtime config failed, ret=%d\n",
433 hba->uhq = devm_kzalloc(hba->dev,
434 hba->nr_hw_queues * sizeof(struct ufs_hw_queue),
437 dev_err(hba->dev, "ufs hw queue memory allocation failed\n");
441 for (i = 0; i < hba->nr_hw_queues; i++) {
443 hwq->max_entries = hba->nutrs + 1;
444 spin_lock_init(&hwq->sq_lock);
445 spin_lock_init(&hwq->cq_lock);
446 mutex_init(&hwq->sq_mutex);
449 /* The very first HW queue serves device commands */
450 hba->dev_cmd_queue = &hba->uhq[0];
452 host->host_tagset = 1;
456 static int ufshcd_mcq_sq_stop(struct ufs_hba *hba, struct ufs_hw_queue *hwq)
459 u32 id = hwq->id, val;
462 if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
465 writel(SQ_STOP, mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTC);
466 reg = mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTS;
467 err = read_poll_timeout(readl, val, val & SQ_STS, 20,
468 MCQ_POLL_US, false, reg);
470 dev_err(hba->dev, "%s: failed. hwq-id=%d, err=%d\n",
475 static int ufshcd_mcq_sq_start(struct ufs_hba *hba, struct ufs_hw_queue *hwq)
478 u32 id = hwq->id, val;
481 if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
484 writel(SQ_START, mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTC);
485 reg = mcq_opr_base(hba, OPR_SQD, id) + REG_SQRTS;
486 err = read_poll_timeout(readl, val, !(val & SQ_STS), 20,
487 MCQ_POLL_US, false, reg);
489 dev_err(hba->dev, "%s: failed. hwq-id=%d, err=%d\n",
495 * ufshcd_mcq_sq_cleanup - Clean up submission queue resources
496 * associated with the pending command.
497 * @hba: per adapter instance.
498 * @task_tag: The command's task tag.
500 * Return: 0 for success; error code otherwise.
502 int ufshcd_mcq_sq_cleanup(struct ufs_hba *hba, int task_tag)
504 struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
505 struct scsi_cmnd *cmd = lrbp->cmd;
506 struct ufs_hw_queue *hwq;
507 void __iomem *reg, *opr_sqd_base;
511 if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
514 if (task_tag != hba->nutrs - UFSHCD_NUM_RESERVED) {
517 hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(cmd));
519 hwq = hba->dev_cmd_queue;
524 mutex_lock(&hwq->sq_mutex);
526 /* stop the SQ fetching before working on it */
527 err = ufshcd_mcq_sq_stop(hba, hwq);
531 /* SQCTI = EXT_IID, IID, LUN, Task Tag */
532 nexus = lrbp->lun << 8 | task_tag;
533 opr_sqd_base = mcq_opr_base(hba, OPR_SQD, id);
534 writel(nexus, opr_sqd_base + REG_SQCTI);
537 writel(SQ_ICU, opr_sqd_base + REG_SQRTC);
539 /* Poll SQRTSy.CUS = 1. Return result from SQRTSy.RTC */
540 reg = opr_sqd_base + REG_SQRTS;
541 err = read_poll_timeout(readl, val, val & SQ_CUS, 20,
542 MCQ_POLL_US, false, reg);
544 dev_err(hba->dev, "%s: failed. hwq=%d, tag=%d err=%ld\n",
545 __func__, id, task_tag,
546 FIELD_GET(SQ_ICU_ERR_CODE_MASK, readl(reg)));
548 if (ufshcd_mcq_sq_start(hba, hwq))
552 mutex_unlock(&hwq->sq_mutex);
557 * ufshcd_mcq_nullify_sqe - Nullify the submission queue entry.
558 * Write the sqe's Command Type to 0xF. The host controller will not
559 * fetch any sqe with Command Type = 0xF.
561 * @utrd: UTP Transfer Request Descriptor to be nullified.
563 static void ufshcd_mcq_nullify_sqe(struct utp_transfer_req_desc *utrd)
565 utrd->header.command_type = 0xf;
569 * ufshcd_mcq_sqe_search - Search for the command in the submission queue
570 * If the command is in the submission queue and not issued to the device yet,
571 * nullify the sqe so the host controller will skip fetching the sqe.
573 * @hba: per adapter instance.
574 * @hwq: Hardware Queue to be searched.
575 * @task_tag: The command's task tag.
577 * Return: true if the SQE containing the command is present in the SQ
578 * (not fetched by the controller); returns false if the SQE is not in the SQ.
580 static bool ufshcd_mcq_sqe_search(struct ufs_hba *hba,
581 struct ufs_hw_queue *hwq, int task_tag)
583 struct ufshcd_lrb *lrbp = &hba->lrb[task_tag];
584 struct utp_transfer_req_desc *utrd;
585 __le64 cmd_desc_base_addr;
590 if (hba->quirks & UFSHCD_QUIRK_MCQ_BROKEN_RTC)
593 mutex_lock(&hwq->sq_mutex);
595 ufshcd_mcq_sq_stop(hba, hwq);
596 sq_head_slot = ufshcd_mcq_get_sq_head_slot(hwq);
597 if (sq_head_slot == hwq->sq_tail_slot)
600 cmd_desc_base_addr = lrbp->utr_descriptor_ptr->command_desc_base_addr;
601 addr = le64_to_cpu(cmd_desc_base_addr) & CQE_UCD_BA;
603 while (sq_head_slot != hwq->sq_tail_slot) {
604 utrd = hwq->sqe_base_addr +
605 sq_head_slot * sizeof(struct utp_transfer_req_desc);
606 match = le64_to_cpu(utrd->command_desc_base_addr) & CQE_UCD_BA;
608 ufshcd_mcq_nullify_sqe(utrd);
614 if (sq_head_slot == hwq->max_entries)
619 ufshcd_mcq_sq_start(hba, hwq);
620 mutex_unlock(&hwq->sq_mutex);
625 * ufshcd_mcq_abort - Abort the command in MCQ.
626 * @cmd: The command to be aborted.
628 * Return: SUCCESS or FAILED error codes
630 int ufshcd_mcq_abort(struct scsi_cmnd *cmd)
632 struct Scsi_Host *host = cmd->device->host;
633 struct ufs_hba *hba = shost_priv(host);
634 int tag = scsi_cmd_to_rq(cmd)->tag;
635 struct ufshcd_lrb *lrbp = &hba->lrb[tag];
636 struct ufs_hw_queue *hwq;
640 if (!ufshcd_cmd_inflight(lrbp->cmd)) {
642 "%s: skip abort. cmd at tag %d already completed.\n",
647 /* Skip task abort in case previous aborts failed and report failure */
648 if (lrbp->req_abort_skip) {
649 dev_err(hba->dev, "%s: skip abort. tag %d failed earlier\n",
654 hwq = ufshcd_mcq_req_to_hwq(hba, scsi_cmd_to_rq(cmd));
656 if (ufshcd_mcq_sqe_search(hba, hwq, tag)) {
658 * Failure. The command should not be "stuck" in SQ for
659 * a long time which resulted in command being aborted.
661 dev_err(hba->dev, "%s: cmd found in sq. hwq=%d, tag=%d\n",
662 __func__, hwq->id, tag);
667 * The command is not in the submission queue, and it is not
668 * in the completion queue either. Query the device to see if
669 * the command is being processed in the device.
671 if (ufshcd_try_to_abort_task(hba, tag)) {
672 dev_err(hba->dev, "%s: device abort failed %d\n", __func__, err);
673 lrbp->req_abort_skip = true;
678 spin_lock_irqsave(&hwq->cq_lock, flags);
679 if (ufshcd_cmd_inflight(lrbp->cmd))
680 ufshcd_release_scsi_cmd(hba, lrbp);
681 spin_unlock_irqrestore(&hwq->cq_lock, flags);