xref: /linux/drivers/interconnect/qcom/icc-rpm.c (revision 4b660dbd9ee2059850fd30e0df420ca7a38a1856)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Copyright (C) 2020 Linaro Ltd
4  */
5 
6 #include <linux/device.h>
7 #include <linux/interconnect-provider.h>
8 #include <linux/io.h>
9 #include <linux/module.h>
10 #include <linux/of.h>
11 #include <linux/of_platform.h>
12 #include <linux/platform_device.h>
13 #include <linux/regmap.h>
14 #include <linux/slab.h>
15 
16 #include "icc-common.h"
17 #include "icc-rpm.h"
18 
19 /* QNOC QoS */
20 #define QNOC_QOS_MCTL_LOWn_ADDR(n)	(0x8 + (n * 0x1000))
21 #define QNOC_QOS_MCTL_DFLT_PRIO_MASK	0x70
22 #define QNOC_QOS_MCTL_DFLT_PRIO_SHIFT	4
23 #define QNOC_QOS_MCTL_URGFWD_EN_MASK	0x8
24 #define QNOC_QOS_MCTL_URGFWD_EN_SHIFT	3
25 
26 /* BIMC QoS */
27 #define M_BKE_REG_BASE(n)		(0x300 + (0x4000 * n))
28 #define M_BKE_EN_ADDR(n)		(M_BKE_REG_BASE(n))
29 #define M_BKE_HEALTH_CFG_ADDR(i, n)	(M_BKE_REG_BASE(n) + 0x40 + (0x4 * i))
30 
31 #define M_BKE_HEALTH_CFG_LIMITCMDS_MASK	0x80000000
32 #define M_BKE_HEALTH_CFG_AREQPRIO_MASK	0x300
33 #define M_BKE_HEALTH_CFG_PRIOLVL_MASK	0x3
34 #define M_BKE_HEALTH_CFG_AREQPRIO_SHIFT	0x8
35 #define M_BKE_HEALTH_CFG_LIMITCMDS_SHIFT 0x1f
36 
37 #define M_BKE_EN_EN_BMASK		0x1
38 
39 /* NoC QoS */
40 #define NOC_QOS_PRIORITYn_ADDR(n)	(0x8 + (n * 0x1000))
41 #define NOC_QOS_PRIORITY_P1_MASK	0xc
42 #define NOC_QOS_PRIORITY_P0_MASK	0x3
43 #define NOC_QOS_PRIORITY_P1_SHIFT	0x2
44 
45 #define NOC_QOS_MODEn_ADDR(n)		(0xc + (n * 0x1000))
46 #define NOC_QOS_MODEn_MASK		0x3
47 
48 #define NOC_QOS_MODE_FIXED_VAL		0x0
49 #define NOC_QOS_MODE_BYPASS_VAL		0x2
50 
51 #define ICC_BUS_CLK_MIN_RATE		19200ULL /* kHz */
52 
53 static int qcom_icc_set_qnoc_qos(struct icc_node *src)
54 {
55 	struct icc_provider *provider = src->provider;
56 	struct qcom_icc_provider *qp = to_qcom_provider(provider);
57 	struct qcom_icc_node *qn = src->data;
58 	struct qcom_icc_qos *qos = &qn->qos;
59 	int rc;
60 
61 	rc = regmap_update_bits(qp->regmap,
62 			qp->qos_offset + QNOC_QOS_MCTL_LOWn_ADDR(qos->qos_port),
63 			QNOC_QOS_MCTL_DFLT_PRIO_MASK,
64 			qos->areq_prio << QNOC_QOS_MCTL_DFLT_PRIO_SHIFT);
65 	if (rc)
66 		return rc;
67 
68 	return regmap_update_bits(qp->regmap,
69 			qp->qos_offset + QNOC_QOS_MCTL_LOWn_ADDR(qos->qos_port),
70 			QNOC_QOS_MCTL_URGFWD_EN_MASK,
71 			!!qos->urg_fwd_en << QNOC_QOS_MCTL_URGFWD_EN_SHIFT);
72 }
73 
74 static int qcom_icc_bimc_set_qos_health(struct qcom_icc_provider *qp,
75 					struct qcom_icc_qos *qos,
76 					int regnum)
77 {
78 	u32 val;
79 	u32 mask;
80 
81 	val = qos->prio_level;
82 	mask = M_BKE_HEALTH_CFG_PRIOLVL_MASK;
83 
84 	val |= qos->areq_prio << M_BKE_HEALTH_CFG_AREQPRIO_SHIFT;
85 	mask |= M_BKE_HEALTH_CFG_AREQPRIO_MASK;
86 
87 	/* LIMITCMDS is not present on M_BKE_HEALTH_3 */
88 	if (regnum != 3) {
89 		val |= qos->limit_commands << M_BKE_HEALTH_CFG_LIMITCMDS_SHIFT;
90 		mask |= M_BKE_HEALTH_CFG_LIMITCMDS_MASK;
91 	}
92 
93 	return regmap_update_bits(qp->regmap,
94 				  qp->qos_offset + M_BKE_HEALTH_CFG_ADDR(regnum, qos->qos_port),
95 				  mask, val);
96 }
97 
98 static int qcom_icc_set_bimc_qos(struct icc_node *src)
99 {
100 	struct qcom_icc_provider *qp;
101 	struct qcom_icc_node *qn;
102 	struct icc_provider *provider;
103 	u32 mode = NOC_QOS_MODE_BYPASS;
104 	u32 val = 0;
105 	int i, rc = 0;
106 
107 	qn = src->data;
108 	provider = src->provider;
109 	qp = to_qcom_provider(provider);
110 
111 	if (qn->qos.qos_mode != NOC_QOS_MODE_INVALID)
112 		mode = qn->qos.qos_mode;
113 
114 	/* QoS Priority: The QoS Health parameters are getting considered
115 	 * only if we are NOT in Bypass Mode.
116 	 */
117 	if (mode != NOC_QOS_MODE_BYPASS) {
118 		for (i = 3; i >= 0; i--) {
119 			rc = qcom_icc_bimc_set_qos_health(qp,
120 							  &qn->qos, i);
121 			if (rc)
122 				return rc;
123 		}
124 
125 		/* Set BKE_EN to 1 when Fixed, Regulator or Limiter Mode */
126 		val = 1;
127 	}
128 
129 	return regmap_update_bits(qp->regmap,
130 				  qp->qos_offset + M_BKE_EN_ADDR(qn->qos.qos_port),
131 				  M_BKE_EN_EN_BMASK, val);
132 }
133 
134 static int qcom_icc_noc_set_qos_priority(struct qcom_icc_provider *qp,
135 					 struct qcom_icc_qos *qos)
136 {
137 	u32 val;
138 	int rc;
139 
140 	/* Must be updated one at a time, P1 first, P0 last */
141 	val = qos->areq_prio << NOC_QOS_PRIORITY_P1_SHIFT;
142 	rc = regmap_update_bits(qp->regmap,
143 				qp->qos_offset + NOC_QOS_PRIORITYn_ADDR(qos->qos_port),
144 				NOC_QOS_PRIORITY_P1_MASK, val);
145 	if (rc)
146 		return rc;
147 
148 	return regmap_update_bits(qp->regmap,
149 				  qp->qos_offset + NOC_QOS_PRIORITYn_ADDR(qos->qos_port),
150 				  NOC_QOS_PRIORITY_P0_MASK, qos->prio_level);
151 }
152 
153 static int qcom_icc_set_noc_qos(struct icc_node *src)
154 {
155 	struct qcom_icc_provider *qp;
156 	struct qcom_icc_node *qn;
157 	struct icc_provider *provider;
158 	u32 mode = NOC_QOS_MODE_BYPASS_VAL;
159 	int rc = 0;
160 
161 	qn = src->data;
162 	provider = src->provider;
163 	qp = to_qcom_provider(provider);
164 
165 	if (qn->qos.qos_port < 0) {
166 		dev_dbg(src->provider->dev,
167 			"NoC QoS: Skipping %s: vote aggregated on parent.\n",
168 			qn->name);
169 		return 0;
170 	}
171 
172 	if (qn->qos.qos_mode == NOC_QOS_MODE_FIXED) {
173 		dev_dbg(src->provider->dev, "NoC QoS: %s: Set Fixed mode\n", qn->name);
174 		mode = NOC_QOS_MODE_FIXED_VAL;
175 		rc = qcom_icc_noc_set_qos_priority(qp, &qn->qos);
176 		if (rc)
177 			return rc;
178 	} else if (qn->qos.qos_mode == NOC_QOS_MODE_BYPASS) {
179 		dev_dbg(src->provider->dev, "NoC QoS: %s: Set Bypass mode\n", qn->name);
180 		mode = NOC_QOS_MODE_BYPASS_VAL;
181 	} else {
182 		/* How did we get here? */
183 	}
184 
185 	return regmap_update_bits(qp->regmap,
186 				  qp->qos_offset + NOC_QOS_MODEn_ADDR(qn->qos.qos_port),
187 				  NOC_QOS_MODEn_MASK, mode);
188 }
189 
190 static int qcom_icc_qos_set(struct icc_node *node)
191 {
192 	struct qcom_icc_provider *qp = to_qcom_provider(node->provider);
193 	struct qcom_icc_node *qn = node->data;
194 
195 	dev_dbg(node->provider->dev, "Setting QoS for %s\n", qn->name);
196 
197 	switch (qp->type) {
198 	case QCOM_ICC_BIMC:
199 		return qcom_icc_set_bimc_qos(node);
200 	case QCOM_ICC_QNOC:
201 		return qcom_icc_set_qnoc_qos(node);
202 	default:
203 		return qcom_icc_set_noc_qos(node);
204 	}
205 }
206 
207 static int qcom_icc_rpm_set(struct qcom_icc_node *qn, u64 *bw)
208 {
209 	int ret, rpm_ctx = 0;
210 	u64 bw_bps;
211 
212 	if (qn->qos.ap_owned)
213 		return 0;
214 
215 	for (rpm_ctx = 0; rpm_ctx < QCOM_SMD_RPM_STATE_NUM; rpm_ctx++) {
216 		bw_bps = icc_units_to_bps(bw[rpm_ctx]);
217 
218 		if (qn->mas_rpm_id != -1) {
219 			ret = qcom_icc_rpm_smd_send(rpm_ctx,
220 						    RPM_BUS_MASTER_REQ,
221 						    qn->mas_rpm_id,
222 						    bw_bps);
223 			if (ret) {
224 				pr_err("qcom_icc_rpm_smd_send mas %d error %d\n",
225 				qn->mas_rpm_id, ret);
226 				return ret;
227 			}
228 		}
229 
230 		if (qn->slv_rpm_id != -1) {
231 			ret = qcom_icc_rpm_smd_send(rpm_ctx,
232 						    RPM_BUS_SLAVE_REQ,
233 						    qn->slv_rpm_id,
234 						    bw_bps);
235 			if (ret) {
236 				pr_err("qcom_icc_rpm_smd_send slv %d error %d\n",
237 				qn->slv_rpm_id, ret);
238 				return ret;
239 			}
240 		}
241 	}
242 
243 	return 0;
244 }
245 
246 /**
247  * qcom_icc_pre_bw_aggregate - cleans up values before re-aggregate requests
248  * @node: icc node to operate on
249  */
250 static void qcom_icc_pre_bw_aggregate(struct icc_node *node)
251 {
252 	struct qcom_icc_node *qn;
253 	size_t i;
254 
255 	qn = node->data;
256 	for (i = 0; i < QCOM_SMD_RPM_STATE_NUM; i++) {
257 		qn->sum_avg[i] = 0;
258 		qn->max_peak[i] = 0;
259 	}
260 }
261 
262 /**
263  * qcom_icc_bw_aggregate - aggregate bw for buckets indicated by tag
264  * @node: node to aggregate
265  * @tag: tag to indicate which buckets to aggregate
266  * @avg_bw: new bw to sum aggregate
267  * @peak_bw: new bw to max aggregate
268  * @agg_avg: existing aggregate avg bw val
269  * @agg_peak: existing aggregate peak bw val
270  */
271 static int qcom_icc_bw_aggregate(struct icc_node *node, u32 tag, u32 avg_bw,
272 				 u32 peak_bw, u32 *agg_avg, u32 *agg_peak)
273 {
274 	size_t i;
275 	struct qcom_icc_node *qn;
276 
277 	qn = node->data;
278 
279 	if (!tag)
280 		tag = RPM_ALWAYS_TAG;
281 
282 	for (i = 0; i < QCOM_SMD_RPM_STATE_NUM; i++) {
283 		if (tag & BIT(i)) {
284 			qn->sum_avg[i] += avg_bw;
285 			qn->max_peak[i] = max_t(u32, qn->max_peak[i], peak_bw);
286 		}
287 	}
288 
289 	*agg_avg += avg_bw;
290 	*agg_peak = max_t(u32, *agg_peak, peak_bw);
291 	return 0;
292 }
293 
294 static u64 qcom_icc_calc_rate(struct qcom_icc_provider *qp, struct qcom_icc_node *qn, int ctx)
295 {
296 	u64 agg_avg_rate, agg_peak_rate, agg_rate;
297 
298 	if (qn->channels)
299 		agg_avg_rate = div_u64(qn->sum_avg[ctx], qn->channels);
300 	else
301 		agg_avg_rate = qn->sum_avg[ctx];
302 
303 	if (qn->ab_coeff) {
304 		agg_avg_rate = agg_avg_rate * qn->ab_coeff;
305 		agg_avg_rate = div_u64(agg_avg_rate, 100);
306 	}
307 
308 	if (qn->ib_coeff) {
309 		agg_peak_rate = qn->max_peak[ctx] * 100;
310 		agg_peak_rate = div_u64(agg_peak_rate, qn->ib_coeff);
311 	} else {
312 		agg_peak_rate = qn->max_peak[ctx];
313 	}
314 
315 	agg_rate = max_t(u64, agg_avg_rate, agg_peak_rate);
316 
317 	return div_u64(agg_rate, qn->buswidth);
318 }
319 
320 /**
321  * qcom_icc_bus_aggregate - calculate bus clock rates by traversing all nodes
322  * @provider: generic interconnect provider
323  * @agg_clk_rate: array containing the aggregated clock rates in kHz
324  */
325 static void qcom_icc_bus_aggregate(struct icc_provider *provider, u64 *agg_clk_rate)
326 {
327 	struct qcom_icc_provider *qp = to_qcom_provider(provider);
328 	struct qcom_icc_node *qn;
329 	struct icc_node *node;
330 	int ctx;
331 
332 	/*
333 	 * Iterate nodes on the provider, aggregate bandwidth requests for
334 	 * every bucket and convert them into bus clock rates.
335 	 */
336 	list_for_each_entry(node, &provider->nodes, node_list) {
337 		qn = node->data;
338 		for (ctx = 0; ctx < QCOM_SMD_RPM_STATE_NUM; ctx++) {
339 			agg_clk_rate[ctx] = max_t(u64, agg_clk_rate[ctx],
340 						  qcom_icc_calc_rate(qp, qn, ctx));
341 		}
342 	}
343 }
344 
345 static int qcom_icc_set(struct icc_node *src, struct icc_node *dst)
346 {
347 	struct qcom_icc_node *src_qn = NULL, *dst_qn = NULL;
348 	u64 agg_clk_rate[QCOM_SMD_RPM_STATE_NUM] = { 0 };
349 	struct icc_provider *provider;
350 	struct qcom_icc_provider *qp;
351 	u64 active_rate, sleep_rate;
352 	int ret;
353 
354 	src_qn = src->data;
355 	if (dst)
356 		dst_qn = dst->data;
357 	provider = src->provider;
358 	qp = to_qcom_provider(provider);
359 
360 	qcom_icc_bus_aggregate(provider, agg_clk_rate);
361 	active_rate = agg_clk_rate[QCOM_SMD_RPM_ACTIVE_STATE];
362 	sleep_rate = agg_clk_rate[QCOM_SMD_RPM_SLEEP_STATE];
363 
364 	ret = qcom_icc_rpm_set(src_qn, src_qn->sum_avg);
365 	if (ret)
366 		return ret;
367 
368 	if (dst_qn) {
369 		ret = qcom_icc_rpm_set(dst_qn, dst_qn->sum_avg);
370 		if (ret)
371 			return ret;
372 	}
373 
374 	/* Some providers don't have a bus clock to scale */
375 	if (!qp->bus_clk_desc && !qp->bus_clk)
376 		return 0;
377 
378 	/*
379 	 * Downstream checks whether the requested rate is zero, but it makes little sense
380 	 * to vote for a value that's below the lower threshold, so let's not do so.
381 	 */
382 	if (qp->keep_alive)
383 		active_rate = max(ICC_BUS_CLK_MIN_RATE, active_rate);
384 
385 	/* Some providers have a non-RPM-owned bus clock - convert kHz->Hz for the CCF */
386 	if (qp->bus_clk) {
387 		active_rate = max_t(u64, active_rate, sleep_rate);
388 		/* ARM32 caps clk_set_rate arg to u32.. Nothing we can do about that! */
389 		active_rate = min_t(u64, 1000ULL * active_rate, ULONG_MAX);
390 		return clk_set_rate(qp->bus_clk, active_rate);
391 	}
392 
393 	/* RPM only accepts <=INT_MAX rates */
394 	active_rate = min_t(u64, active_rate, INT_MAX);
395 	sleep_rate = min_t(u64, sleep_rate, INT_MAX);
396 
397 	if (active_rate != qp->bus_clk_rate[QCOM_SMD_RPM_ACTIVE_STATE]) {
398 		ret = qcom_icc_rpm_set_bus_rate(qp->bus_clk_desc, QCOM_SMD_RPM_ACTIVE_STATE,
399 						active_rate);
400 		if (ret)
401 			return ret;
402 
403 		/* Cache the rate after we've successfully commited it to RPM */
404 		qp->bus_clk_rate[QCOM_SMD_RPM_ACTIVE_STATE] = active_rate;
405 	}
406 
407 	if (sleep_rate != qp->bus_clk_rate[QCOM_SMD_RPM_SLEEP_STATE]) {
408 		ret = qcom_icc_rpm_set_bus_rate(qp->bus_clk_desc, QCOM_SMD_RPM_SLEEP_STATE,
409 						sleep_rate);
410 		if (ret)
411 			return ret;
412 
413 		/* Cache the rate after we've successfully commited it to RPM */
414 		qp->bus_clk_rate[QCOM_SMD_RPM_SLEEP_STATE] = sleep_rate;
415 	}
416 
417 	/* Handle the node-specific clock */
418 	if (!src_qn->bus_clk_desc)
419 		return 0;
420 
421 	active_rate = qcom_icc_calc_rate(qp, src_qn, QCOM_SMD_RPM_ACTIVE_STATE);
422 	sleep_rate = qcom_icc_calc_rate(qp, src_qn, QCOM_SMD_RPM_SLEEP_STATE);
423 
424 	if (active_rate != src_qn->bus_clk_rate[QCOM_SMD_RPM_ACTIVE_STATE]) {
425 		ret = qcom_icc_rpm_set_bus_rate(src_qn->bus_clk_desc, QCOM_SMD_RPM_ACTIVE_STATE,
426 						active_rate);
427 		if (ret)
428 			return ret;
429 
430 		/* Cache the rate after we've successfully committed it to RPM */
431 		src_qn->bus_clk_rate[QCOM_SMD_RPM_ACTIVE_STATE] = active_rate;
432 	}
433 
434 	if (sleep_rate != src_qn->bus_clk_rate[QCOM_SMD_RPM_SLEEP_STATE]) {
435 		ret = qcom_icc_rpm_set_bus_rate(src_qn->bus_clk_desc, QCOM_SMD_RPM_SLEEP_STATE,
436 						sleep_rate);
437 		if (ret)
438 			return ret;
439 
440 		/* Cache the rate after we've successfully committed it to RPM */
441 		src_qn->bus_clk_rate[QCOM_SMD_RPM_SLEEP_STATE] = sleep_rate;
442 	}
443 
444 	return 0;
445 }
446 
447 int qnoc_probe(struct platform_device *pdev)
448 {
449 	struct device *dev = &pdev->dev;
450 	const struct qcom_icc_desc *desc;
451 	struct icc_onecell_data *data;
452 	struct icc_provider *provider;
453 	struct qcom_icc_node * const *qnodes;
454 	struct qcom_icc_provider *qp;
455 	struct icc_node *node;
456 	size_t num_nodes, i;
457 	const char * const *cds = NULL;
458 	int cd_num;
459 	int ret;
460 
461 	/* wait for the RPM proxy */
462 	if (!qcom_icc_rpm_smd_available())
463 		return -EPROBE_DEFER;
464 
465 	desc = of_device_get_match_data(dev);
466 	if (!desc)
467 		return -EINVAL;
468 
469 	qnodes = desc->nodes;
470 	num_nodes = desc->num_nodes;
471 
472 	if (desc->num_intf_clocks) {
473 		cds = desc->intf_clocks;
474 		cd_num = desc->num_intf_clocks;
475 	} else {
476 		/* 0 intf clocks is perfectly fine */
477 		cd_num = 0;
478 	}
479 
480 	qp = devm_kzalloc(dev, sizeof(*qp), GFP_KERNEL);
481 	if (!qp)
482 		return -ENOMEM;
483 
484 	qp->intf_clks = devm_kcalloc(dev, cd_num, sizeof(*qp->intf_clks), GFP_KERNEL);
485 	if (!qp->intf_clks)
486 		return -ENOMEM;
487 
488 	if (desc->bus_clk_desc) {
489 		qp->bus_clk_desc = devm_kzalloc(dev, sizeof(*qp->bus_clk_desc),
490 						GFP_KERNEL);
491 		if (!qp->bus_clk_desc)
492 			return -ENOMEM;
493 
494 		qp->bus_clk_desc = desc->bus_clk_desc;
495 	} else {
496 		/* Some older SoCs may have a single non-RPM-owned bus clock. */
497 		qp->bus_clk = devm_clk_get_optional(dev, "bus");
498 		if (IS_ERR(qp->bus_clk))
499 			return PTR_ERR(qp->bus_clk);
500 	}
501 
502 	data = devm_kzalloc(dev, struct_size(data, nodes, num_nodes),
503 			    GFP_KERNEL);
504 	if (!data)
505 		return -ENOMEM;
506 
507 	qp->num_intf_clks = cd_num;
508 	for (i = 0; i < cd_num; i++)
509 		qp->intf_clks[i].id = cds[i];
510 
511 	qp->keep_alive = desc->keep_alive;
512 	qp->type = desc->type;
513 	qp->qos_offset = desc->qos_offset;
514 
515 	if (desc->regmap_cfg) {
516 		struct resource *res;
517 		void __iomem *mmio;
518 
519 		res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
520 		if (!res) {
521 			/* Try parent's regmap */
522 			qp->regmap = dev_get_regmap(dev->parent, NULL);
523 			if (qp->regmap)
524 				goto regmap_done;
525 			return -ENODEV;
526 		}
527 
528 		mmio = devm_ioremap_resource(dev, res);
529 		if (IS_ERR(mmio))
530 			return PTR_ERR(mmio);
531 
532 		qp->regmap = devm_regmap_init_mmio(dev, mmio, desc->regmap_cfg);
533 		if (IS_ERR(qp->regmap)) {
534 			dev_err(dev, "Cannot regmap interconnect bus resource\n");
535 			return PTR_ERR(qp->regmap);
536 		}
537 	}
538 
539 regmap_done:
540 	ret = clk_prepare_enable(qp->bus_clk);
541 	if (ret)
542 		return ret;
543 
544 	ret = devm_clk_bulk_get(dev, qp->num_intf_clks, qp->intf_clks);
545 	if (ret)
546 		goto err_disable_unprepare_clk;
547 
548 	provider = &qp->provider;
549 	provider->dev = dev;
550 	provider->set = qcom_icc_set;
551 	provider->pre_aggregate = qcom_icc_pre_bw_aggregate;
552 	provider->aggregate = qcom_icc_bw_aggregate;
553 	provider->xlate_extended = qcom_icc_xlate_extended;
554 	provider->data = data;
555 
556 	icc_provider_init(provider);
557 
558 	/* If this fails, bus accesses will crash the platform! */
559 	ret = clk_bulk_prepare_enable(qp->num_intf_clks, qp->intf_clks);
560 	if (ret)
561 		goto err_disable_unprepare_clk;
562 
563 	for (i = 0; i < num_nodes; i++) {
564 		size_t j;
565 
566 		if (!qnodes[i]->ab_coeff)
567 			qnodes[i]->ab_coeff = qp->ab_coeff;
568 
569 		if (!qnodes[i]->ib_coeff)
570 			qnodes[i]->ib_coeff = qp->ib_coeff;
571 
572 		node = icc_node_create(qnodes[i]->id);
573 		if (IS_ERR(node)) {
574 			clk_bulk_disable_unprepare(qp->num_intf_clks,
575 						   qp->intf_clks);
576 			ret = PTR_ERR(node);
577 			goto err_remove_nodes;
578 		}
579 
580 		node->name = qnodes[i]->name;
581 		node->data = qnodes[i];
582 		icc_node_add(node, provider);
583 
584 		for (j = 0; j < qnodes[i]->num_links; j++)
585 			icc_link_create(node, qnodes[i]->links[j]);
586 
587 		/* Set QoS registers (we only need to do it once, generally) */
588 		if (qnodes[i]->qos.ap_owned &&
589 		    qnodes[i]->qos.qos_mode != NOC_QOS_MODE_INVALID) {
590 			ret = qcom_icc_qos_set(node);
591 			if (ret) {
592 				clk_bulk_disable_unprepare(qp->num_intf_clks,
593 							   qp->intf_clks);
594 				goto err_remove_nodes;
595 			}
596 		}
597 
598 		data->nodes[i] = node;
599 	}
600 	data->num_nodes = num_nodes;
601 
602 	clk_bulk_disable_unprepare(qp->num_intf_clks, qp->intf_clks);
603 
604 	ret = icc_provider_register(provider);
605 	if (ret)
606 		goto err_remove_nodes;
607 
608 	platform_set_drvdata(pdev, qp);
609 
610 	/* Populate child NoC devices if any */
611 	if (of_get_child_count(dev->of_node) > 0) {
612 		ret = of_platform_populate(dev->of_node, NULL, NULL, dev);
613 		if (ret)
614 			goto err_deregister_provider;
615 	}
616 
617 	return 0;
618 
619 err_deregister_provider:
620 	icc_provider_deregister(provider);
621 err_remove_nodes:
622 	icc_nodes_remove(provider);
623 err_disable_unprepare_clk:
624 	clk_disable_unprepare(qp->bus_clk);
625 
626 	return ret;
627 }
628 EXPORT_SYMBOL(qnoc_probe);
629 
630 void qnoc_remove(struct platform_device *pdev)
631 {
632 	struct qcom_icc_provider *qp = platform_get_drvdata(pdev);
633 
634 	icc_provider_deregister(&qp->provider);
635 	icc_nodes_remove(&qp->provider);
636 	clk_disable_unprepare(qp->bus_clk);
637 }
638 EXPORT_SYMBOL(qnoc_remove);
639