Commit | Line | Data |
---|---|---|
57b1c0ef AD |
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 | ||
16 | #define MAX_QUEUE_SUP GENMASK(7, 0) | |
17 | #define UFS_MCQ_MIN_RW_QUEUES 2 | |
18 | #define UFS_MCQ_MIN_READ_QUEUES 0 | |
19 | #define UFS_MCQ_NUM_DEV_CMD_QUEUES 1 | |
20 | #define UFS_MCQ_MIN_POLL_QUEUES 0 | |
2468da61 AD |
21 | #define QUEUE_EN_OFFSET 31 |
22 | #define QUEUE_ID_OFFSET 16 | |
57b1c0ef | 23 | |
7224c806 AD |
24 | #define MAX_DEV_CMD_ENTRIES 2 |
25 | #define MCQ_CFG_MAC_MASK GENMASK(16, 8) | |
2468da61 AD |
26 | #define MCQ_QCFG_SIZE 0x40 |
27 | #define MCQ_ENTRY_SIZE_IN_DWORD 8 | |
7224c806 | 28 | |
57b1c0ef AD |
29 | static int rw_queue_count_set(const char *val, const struct kernel_param *kp) |
30 | { | |
31 | return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_RW_QUEUES, | |
32 | num_possible_cpus()); | |
33 | } | |
34 | ||
35 | static const struct kernel_param_ops rw_queue_count_ops = { | |
36 | .set = rw_queue_count_set, | |
37 | .get = param_get_uint, | |
38 | }; | |
39 | ||
40 | static unsigned int rw_queues; | |
41 | module_param_cb(rw_queues, &rw_queue_count_ops, &rw_queues, 0644); | |
42 | MODULE_PARM_DESC(rw_queues, | |
43 | "Number of interrupt driven I/O queues used for rw. Default value is nr_cpus"); | |
44 | ||
45 | static int read_queue_count_set(const char *val, const struct kernel_param *kp) | |
46 | { | |
47 | return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_READ_QUEUES, | |
48 | num_possible_cpus()); | |
49 | } | |
50 | ||
51 | static const struct kernel_param_ops read_queue_count_ops = { | |
52 | .set = read_queue_count_set, | |
53 | .get = param_get_uint, | |
54 | }; | |
55 | ||
56 | static unsigned int read_queues; | |
57 | module_param_cb(read_queues, &read_queue_count_ops, &read_queues, 0644); | |
58 | MODULE_PARM_DESC(read_queues, | |
59 | "Number of interrupt driven read queues used for read. Default value is 0"); | |
60 | ||
61 | static int poll_queue_count_set(const char *val, const struct kernel_param *kp) | |
62 | { | |
63 | return param_set_uint_minmax(val, kp, UFS_MCQ_MIN_POLL_QUEUES, | |
64 | num_possible_cpus()); | |
65 | } | |
66 | ||
67 | static const struct kernel_param_ops poll_queue_count_ops = { | |
68 | .set = poll_queue_count_set, | |
69 | .get = param_get_uint, | |
70 | }; | |
71 | ||
72 | static unsigned int poll_queues = 1; | |
73 | module_param_cb(poll_queues, &poll_queue_count_ops, &poll_queues, 0644); | |
74 | MODULE_PARM_DESC(poll_queues, | |
75 | "Number of poll queues used for r/w. Default value is 1"); | |
76 | ||
2468da61 AD |
77 | /** |
78 | * ufshcd_mcq_config_mac - Set the #Max Activ Cmds. | |
79 | * @hba - per adapter instance | |
80 | * @max_active_cmds - maximum # of active commands to the device at any time. | |
81 | * | |
82 | * The controller won't send more than the max_active_cmds to the device at | |
83 | * any time. | |
84 | */ | |
85 | void ufshcd_mcq_config_mac(struct ufs_hba *hba, u32 max_active_cmds) | |
86 | { | |
87 | u32 val; | |
88 | ||
89 | val = ufshcd_readl(hba, REG_UFS_MCQ_CFG); | |
90 | val &= ~MCQ_CFG_MAC_MASK; | |
91 | val |= FIELD_PREP(MCQ_CFG_MAC_MASK, max_active_cmds); | |
92 | ufshcd_writel(hba, val, REG_UFS_MCQ_CFG); | |
93 | } | |
94 | ||
7224c806 AD |
95 | /** |
96 | * ufshcd_mcq_decide_queue_depth - decide the queue depth | |
97 | * @hba - per adapter instance | |
98 | * | |
99 | * Returns queue-depth on success, non-zero on error | |
100 | * | |
101 | * MAC - Max. Active Command of the Host Controller (HC) | |
102 | * HC wouldn't send more than this commands to the device. | |
103 | * It is mandatory to implement get_hba_mac() to enable MCQ mode. | |
104 | * Calculates and adjusts the queue depth based on the depth | |
105 | * supported by the HC and ufs device. | |
106 | */ | |
107 | int ufshcd_mcq_decide_queue_depth(struct ufs_hba *hba) | |
108 | { | |
109 | int mac; | |
110 | ||
111 | /* Mandatory to implement get_hba_mac() */ | |
112 | mac = ufshcd_mcq_vops_get_hba_mac(hba); | |
113 | if (mac < 0) { | |
114 | dev_err(hba->dev, "Failed to get mac, err=%d\n", mac); | |
115 | return mac; | |
116 | } | |
117 | ||
118 | WARN_ON_ONCE(!hba->dev_info.bqueuedepth); | |
119 | /* | |
120 | * max. value of bqueuedepth = 256, mac is host dependent. | |
121 | * It is mandatory for UFS device to define bQueueDepth if | |
122 | * shared queuing architecture is enabled. | |
123 | */ | |
124 | return min_t(int, mac, hba->dev_info.bqueuedepth); | |
125 | } | |
126 | ||
57b1c0ef AD |
127 | static int ufshcd_mcq_config_nr_queues(struct ufs_hba *hba) |
128 | { | |
129 | int i; | |
130 | u32 hba_maxq, rem, tot_queues; | |
131 | struct Scsi_Host *host = hba->host; | |
132 | ||
133 | hba_maxq = FIELD_GET(MAX_QUEUE_SUP, hba->mcq_capabilities); | |
134 | ||
135 | tot_queues = UFS_MCQ_NUM_DEV_CMD_QUEUES + read_queues + poll_queues + | |
136 | rw_queues; | |
137 | ||
138 | if (hba_maxq < tot_queues) { | |
139 | dev_err(hba->dev, "Total queues (%d) exceeds HC capacity (%d)\n", | |
140 | tot_queues, hba_maxq); | |
141 | return -EOPNOTSUPP; | |
142 | } | |
143 | ||
144 | rem = hba_maxq - UFS_MCQ_NUM_DEV_CMD_QUEUES; | |
145 | ||
146 | if (rw_queues) { | |
147 | hba->nr_queues[HCTX_TYPE_DEFAULT] = rw_queues; | |
148 | rem -= hba->nr_queues[HCTX_TYPE_DEFAULT]; | |
149 | } else { | |
150 | rw_queues = num_possible_cpus(); | |
151 | } | |
152 | ||
153 | if (poll_queues) { | |
154 | hba->nr_queues[HCTX_TYPE_POLL] = poll_queues; | |
155 | rem -= hba->nr_queues[HCTX_TYPE_POLL]; | |
156 | } | |
157 | ||
158 | if (read_queues) { | |
159 | hba->nr_queues[HCTX_TYPE_READ] = read_queues; | |
160 | rem -= hba->nr_queues[HCTX_TYPE_READ]; | |
161 | } | |
162 | ||
163 | if (!hba->nr_queues[HCTX_TYPE_DEFAULT]) | |
164 | hba->nr_queues[HCTX_TYPE_DEFAULT] = min3(rem, rw_queues, | |
165 | num_possible_cpus()); | |
166 | ||
167 | for (i = 0; i < HCTX_MAX_TYPES; i++) | |
168 | host->nr_hw_queues += hba->nr_queues[i]; | |
169 | ||
170 | hba->nr_hw_queues = host->nr_hw_queues + UFS_MCQ_NUM_DEV_CMD_QUEUES; | |
171 | return 0; | |
172 | } | |
173 | ||
4682abfa AD |
174 | int ufshcd_mcq_memory_alloc(struct ufs_hba *hba) |
175 | { | |
176 | struct ufs_hw_queue *hwq; | |
177 | size_t utrdl_size, cqe_size; | |
178 | int i; | |
179 | ||
180 | for (i = 0; i < hba->nr_hw_queues; i++) { | |
181 | hwq = &hba->uhq[i]; | |
182 | ||
183 | utrdl_size = sizeof(struct utp_transfer_req_desc) * | |
184 | hwq->max_entries; | |
185 | hwq->sqe_base_addr = dmam_alloc_coherent(hba->dev, utrdl_size, | |
186 | &hwq->sqe_dma_addr, | |
187 | GFP_KERNEL); | |
188 | if (!hwq->sqe_dma_addr) { | |
189 | dev_err(hba->dev, "SQE allocation failed\n"); | |
190 | return -ENOMEM; | |
191 | } | |
192 | ||
193 | cqe_size = sizeof(struct cq_entry) * hwq->max_entries; | |
194 | hwq->cqe_base_addr = dmam_alloc_coherent(hba->dev, cqe_size, | |
195 | &hwq->cqe_dma_addr, | |
196 | GFP_KERNEL); | |
197 | if (!hwq->cqe_dma_addr) { | |
198 | dev_err(hba->dev, "CQE allocation failed\n"); | |
199 | return -ENOMEM; | |
200 | } | |
201 | } | |
202 | ||
203 | return 0; | |
204 | } | |
205 | ||
206 | ||
2468da61 AD |
207 | /* Operation and runtime registers configuration */ |
208 | #define MCQ_CFG_n(r, i) ((r) + MCQ_QCFG_SIZE * (i)) | |
209 | #define MCQ_OPR_OFFSET_n(p, i) \ | |
210 | (hba->mcq_opr[(p)].offset + hba->mcq_opr[(p)].stride * (i)) | |
211 | ||
212 | static void __iomem *mcq_opr_base(struct ufs_hba *hba, | |
213 | enum ufshcd_mcq_opr n, int i) | |
214 | { | |
215 | struct ufshcd_mcq_opr_info_t *opr = &hba->mcq_opr[n]; | |
216 | ||
217 | return opr->base + opr->stride * i; | |
218 | } | |
219 | ||
220 | void ufshcd_mcq_make_queues_operational(struct ufs_hba *hba) | |
221 | { | |
222 | struct ufs_hw_queue *hwq; | |
223 | u16 qsize; | |
224 | int i; | |
225 | ||
226 | for (i = 0; i < hba->nr_hw_queues; i++) { | |
227 | hwq = &hba->uhq[i]; | |
228 | hwq->id = i; | |
229 | qsize = hwq->max_entries * MCQ_ENTRY_SIZE_IN_DWORD - 1; | |
230 | ||
231 | /* Submission Queue Lower Base Address */ | |
232 | ufsmcq_writelx(hba, lower_32_bits(hwq->sqe_dma_addr), | |
233 | MCQ_CFG_n(REG_SQLBA, i)); | |
234 | /* Submission Queue Upper Base Address */ | |
235 | ufsmcq_writelx(hba, upper_32_bits(hwq->sqe_dma_addr), | |
236 | MCQ_CFG_n(REG_SQUBA, i)); | |
237 | /* Submission Queue Doorbell Address Offset */ | |
238 | ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_SQD, i), | |
239 | MCQ_CFG_n(REG_SQDAO, i)); | |
240 | /* Submission Queue Interrupt Status Address Offset */ | |
241 | ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_SQIS, i), | |
242 | MCQ_CFG_n(REG_SQISAO, i)); | |
243 | ||
244 | /* Completion Queue Lower Base Address */ | |
245 | ufsmcq_writelx(hba, lower_32_bits(hwq->cqe_dma_addr), | |
246 | MCQ_CFG_n(REG_CQLBA, i)); | |
247 | /* Completion Queue Upper Base Address */ | |
248 | ufsmcq_writelx(hba, upper_32_bits(hwq->cqe_dma_addr), | |
249 | MCQ_CFG_n(REG_CQUBA, i)); | |
250 | /* Completion Queue Doorbell Address Offset */ | |
251 | ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_CQD, i), | |
252 | MCQ_CFG_n(REG_CQDAO, i)); | |
253 | /* Completion Queue Interrupt Status Address Offset */ | |
254 | ufsmcq_writelx(hba, MCQ_OPR_OFFSET_n(OPR_CQIS, i), | |
255 | MCQ_CFG_n(REG_CQISAO, i)); | |
256 | ||
257 | /* Save the base addresses for quicker access */ | |
258 | hwq->mcq_sq_head = mcq_opr_base(hba, OPR_SQD, i) + REG_SQHP; | |
259 | hwq->mcq_sq_tail = mcq_opr_base(hba, OPR_SQD, i) + REG_SQTP; | |
260 | hwq->mcq_cq_head = mcq_opr_base(hba, OPR_CQD, i) + REG_CQHP; | |
261 | hwq->mcq_cq_tail = mcq_opr_base(hba, OPR_CQD, i) + REG_CQTP; | |
262 | ||
263 | /* Enable Tail Entry Push Status interrupt only for non-poll queues */ | |
264 | if (i < hba->nr_hw_queues - hba->nr_queues[HCTX_TYPE_POLL]) | |
265 | writel(1, mcq_opr_base(hba, OPR_CQIS, i) + REG_CQIE); | |
266 | ||
267 | /* Completion Queue Enable|Size to Completion Queue Attribute */ | |
268 | ufsmcq_writel(hba, (1 << QUEUE_EN_OFFSET) | qsize, | |
269 | MCQ_CFG_n(REG_CQATTR, i)); | |
270 | ||
271 | /* | |
272 | * Submission Qeueue Enable|Size|Completion Queue ID to | |
273 | * Submission Queue Attribute | |
274 | */ | |
275 | ufsmcq_writel(hba, (1 << QUEUE_EN_OFFSET) | qsize | | |
276 | (i << QUEUE_ID_OFFSET), | |
277 | MCQ_CFG_n(REG_SQATTR, i)); | |
278 | } | |
279 | } | |
280 | ||
57b1c0ef AD |
281 | int ufshcd_mcq_init(struct ufs_hba *hba) |
282 | { | |
0d33728f | 283 | struct Scsi_Host *host = hba->host; |
4682abfa AD |
284 | struct ufs_hw_queue *hwq; |
285 | int ret, i; | |
57b1c0ef AD |
286 | |
287 | ret = ufshcd_mcq_config_nr_queues(hba); | |
c263b4ef AD |
288 | if (ret) |
289 | return ret; | |
57b1c0ef | 290 | |
c263b4ef | 291 | ret = ufshcd_vops_mcq_config_resource(hba); |
4682abfa AD |
292 | if (ret) |
293 | return ret; | |
294 | ||
2468da61 AD |
295 | ret = ufshcd_mcq_vops_op_runtime_config(hba); |
296 | if (ret) { | |
297 | dev_err(hba->dev, "Operation runtime config failed, ret=%d\n", | |
298 | ret); | |
299 | return ret; | |
300 | } | |
4682abfa AD |
301 | hba->uhq = devm_kzalloc(hba->dev, |
302 | hba->nr_hw_queues * sizeof(struct ufs_hw_queue), | |
303 | GFP_KERNEL); | |
304 | if (!hba->uhq) { | |
305 | dev_err(hba->dev, "ufs hw queue memory allocation failed\n"); | |
306 | return -ENOMEM; | |
307 | } | |
308 | ||
309 | for (i = 0; i < hba->nr_hw_queues; i++) { | |
310 | hwq = &hba->uhq[i]; | |
311 | hwq->max_entries = hba->nutrs; | |
22a2d563 | 312 | spin_lock_init(&hwq->sq_lock); |
4682abfa AD |
313 | } |
314 | ||
315 | /* The very first HW queue serves device commands */ | |
316 | hba->dev_cmd_queue = &hba->uhq[0]; | |
317 | /* Give dev_cmd_queue the minimal number of entries */ | |
318 | hba->dev_cmd_queue->max_entries = MAX_DEV_CMD_ENTRIES; | |
319 | ||
0d33728f | 320 | host->host_tagset = 1; |
4682abfa | 321 | return 0; |
57b1c0ef | 322 | } |