xref: /linux/drivers/perf/arm_cspmu/arm_cspmu.c (revision 6a4aee277740d04ac0fd54cfa17cc28261932ddc)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * ARM CoreSight Architecture PMU driver.
4  *
5  * This driver adds support for uncore PMU based on ARM CoreSight Performance
6  * Monitoring Unit Architecture. The PMU is accessible via MMIO registers and
7  * like other uncore PMUs, it does not support process specific events and
8  * cannot be used in sampling mode.
9  *
10  * This code is based on other uncore PMUs like ARM DSU PMU. It provides a
11  * generic implementation to operate the PMU according to CoreSight PMU
12  * architecture and ACPI ARM PMU table (APMT) documents below:
13  *   - ARM CoreSight PMU architecture document number: ARM IHI 0091 A.a-00bet0.
14  *   - APMT document number: ARM DEN0117.
15  *
16  * The user should refer to the vendor technical documentation to get details
17  * about the supported events.
18  *
19  * Copyright (c) 2022-2023, NVIDIA CORPORATION & AFFILIATES. All rights reserved.
20  *
21  */
22 
23 #include <linux/acpi.h>
24 #include <linux/cacheinfo.h>
25 #include <linux/ctype.h>
26 #include <linux/interrupt.h>
27 #include <linux/io-64-nonatomic-lo-hi.h>
28 #include <linux/module.h>
29 #include <linux/mutex.h>
30 #include <linux/of.h>
31 #include <linux/perf_event.h>
32 #include <linux/platform_device.h>
33 
34 #include "arm_cspmu.h"
35 
36 #define PMUNAME "arm_cspmu"
37 #define DRVNAME "arm-cs-arch-pmu"
38 
39 #define ARM_CSPMU_CPUMASK_ATTR(_name, _config)			\
40 	ARM_CSPMU_EXT_ATTR(_name, arm_cspmu_cpumask_show,	\
41 				(unsigned long)_config)
42 
43 /*
44  * CoreSight PMU Arch register offsets.
45  */
46 #define PMEVCNTR_LO					0x0
47 #define PMEVCNTR_HI					0x4
48 #define PMEVTYPER					0x400
49 #define PMCCFILTR					0x47C
50 #define PMEVFILTR					0xA00
51 #define PMCNTENSET					0xC00
52 #define PMCNTENCLR					0xC20
53 #define PMINTENSET					0xC40
54 #define PMINTENCLR					0xC60
55 #define PMOVSCLR					0xC80
56 #define PMOVSSET					0xCC0
57 #define PMCFGR						0xE00
58 #define PMCR						0xE04
59 #define PMIIDR						0xE08
60 
61 /* PMCFGR register field */
62 #define PMCFGR_NCG					GENMASK(31, 28)
63 #define PMCFGR_HDBG					BIT(24)
64 #define PMCFGR_TRO					BIT(23)
65 #define PMCFGR_SS					BIT(22)
66 #define PMCFGR_FZO					BIT(21)
67 #define PMCFGR_MSI					BIT(20)
68 #define PMCFGR_UEN					BIT(19)
69 #define PMCFGR_NA					BIT(17)
70 #define PMCFGR_EX					BIT(16)
71 #define PMCFGR_CCD					BIT(15)
72 #define PMCFGR_CC					BIT(14)
73 #define PMCFGR_SIZE					GENMASK(13, 8)
74 #define PMCFGR_N					GENMASK(7, 0)
75 
76 /* PMCR register field */
77 #define PMCR_TRO					BIT(11)
78 #define PMCR_HDBG					BIT(10)
79 #define PMCR_FZO					BIT(9)
80 #define PMCR_NA						BIT(8)
81 #define PMCR_DP						BIT(5)
82 #define PMCR_X						BIT(4)
83 #define PMCR_D						BIT(3)
84 #define PMCR_C						BIT(2)
85 #define PMCR_P						BIT(1)
86 #define PMCR_E						BIT(0)
87 
88 /* Each SET/CLR register supports up to 32 counters. */
89 #define ARM_CSPMU_SET_CLR_COUNTER_SHIFT		5
90 #define ARM_CSPMU_SET_CLR_COUNTER_NUM		\
91 	(1 << ARM_CSPMU_SET_CLR_COUNTER_SHIFT)
92 
93 /* Convert counter idx into SET/CLR register number. */
94 #define COUNTER_TO_SET_CLR_ID(idx)			\
95 	(idx >> ARM_CSPMU_SET_CLR_COUNTER_SHIFT)
96 
97 /* Convert counter idx into SET/CLR register bit. */
98 #define COUNTER_TO_SET_CLR_BIT(idx)			\
99 	(idx & (ARM_CSPMU_SET_CLR_COUNTER_NUM - 1))
100 
101 #define ARM_CSPMU_ACTIVE_CPU_MASK		0x0
102 #define ARM_CSPMU_ASSOCIATED_CPU_MASK		0x1
103 
104 /*
105  * Maximum poll count for reading counter value using high-low-high sequence.
106  */
107 #define HILOHI_MAX_POLL	1000
108 
109 static unsigned long arm_cspmu_cpuhp_state;
110 
111 static DEFINE_MUTEX(arm_cspmu_lock);
112 
113 static void arm_cspmu_set_ev_filter(struct arm_cspmu *cspmu,
114 				    struct hw_perf_event *hwc, u32 filter);
115 
116 static struct acpi_apmt_node *arm_cspmu_apmt_node(struct device *dev)
117 {
118 	struct acpi_apmt_node **ptr = dev_get_platdata(dev);
119 
120 	return ptr ? *ptr : NULL;
121 }
122 
123 /*
124  * In CoreSight PMU architecture, all of the MMIO registers are 32-bit except
125  * counter register. The counter register can be implemented as 32-bit or 64-bit
126  * register depending on the value of PMCFGR.SIZE field. For 64-bit access,
127  * single-copy 64-bit atomic support is implementation defined. APMT node flag
128  * is used to identify if the PMU supports 64-bit single copy atomic. If 64-bit
129  * single copy atomic is not supported, the driver treats the register as a pair
130  * of 32-bit register.
131  */
132 
133 /*
134  * Read 64-bit register as a pair of 32-bit registers using hi-lo-hi sequence.
135  */
136 static u64 read_reg64_hilohi(const void __iomem *addr, u32 max_poll_count)
137 {
138 	u32 val_lo, val_hi;
139 	u64 val;
140 
141 	/* Use high-low-high sequence to avoid tearing */
142 	do {
143 		if (max_poll_count-- == 0) {
144 			pr_err("ARM CSPMU: timeout hi-low-high sequence\n");
145 			return 0;
146 		}
147 
148 		val_hi = readl(addr + 4);
149 		val_lo = readl(addr);
150 	} while (val_hi != readl(addr + 4));
151 
152 	val = (((u64)val_hi << 32) | val_lo);
153 
154 	return val;
155 }
156 
157 /* Check if cycle counter is supported. */
158 static inline bool supports_cycle_counter(const struct arm_cspmu *cspmu)
159 {
160 	return (cspmu->pmcfgr & PMCFGR_CC);
161 }
162 
163 /* Get counter size, which is (PMCFGR_SIZE + 1). */
164 static inline u32 counter_size(const struct arm_cspmu *cspmu)
165 {
166 	return FIELD_GET(PMCFGR_SIZE, cspmu->pmcfgr) + 1;
167 }
168 
169 /* Get counter mask. */
170 static inline u64 counter_mask(const struct arm_cspmu *cspmu)
171 {
172 	return GENMASK_ULL(counter_size(cspmu) - 1, 0);
173 }
174 
175 /* Check if counter is implemented as 64-bit register. */
176 static inline bool use_64b_counter_reg(const struct arm_cspmu *cspmu)
177 {
178 	return (counter_size(cspmu) > 32);
179 }
180 
181 ssize_t arm_cspmu_sysfs_event_show(struct device *dev,
182 				struct device_attribute *attr, char *buf)
183 {
184 	struct perf_pmu_events_attr *pmu_attr;
185 
186 	pmu_attr = container_of(attr, typeof(*pmu_attr), attr);
187 	return sysfs_emit(buf, "event=0x%llx\n", pmu_attr->id);
188 }
189 EXPORT_SYMBOL_GPL(arm_cspmu_sysfs_event_show);
190 
191 /* Default event list. */
192 static struct attribute *arm_cspmu_event_attrs[] = {
193 	ARM_CSPMU_EVENT_ATTR(cycles, ARM_CSPMU_EVT_CYCLES_DEFAULT),
194 	NULL,
195 };
196 
197 static struct attribute **
198 arm_cspmu_get_event_attrs(const struct arm_cspmu *cspmu)
199 {
200 	struct attribute **attrs;
201 
202 	attrs = devm_kmemdup(cspmu->dev, arm_cspmu_event_attrs,
203 		sizeof(arm_cspmu_event_attrs), GFP_KERNEL);
204 
205 	return attrs;
206 }
207 
208 static umode_t
209 arm_cspmu_event_attr_is_visible(struct kobject *kobj,
210 				struct attribute *attr, int unused)
211 {
212 	struct device *dev = kobj_to_dev(kobj);
213 	struct arm_cspmu *cspmu = to_arm_cspmu(dev_get_drvdata(dev));
214 	struct perf_pmu_events_attr *eattr;
215 
216 	eattr = container_of(attr, typeof(*eattr), attr.attr);
217 
218 	/* Hide cycle event if not supported */
219 	if (!supports_cycle_counter(cspmu) &&
220 	    eattr->id == ARM_CSPMU_EVT_CYCLES_DEFAULT)
221 		return 0;
222 
223 	return attr->mode;
224 }
225 
226 ssize_t arm_cspmu_sysfs_format_show(struct device *dev,
227 				struct device_attribute *attr,
228 				char *buf)
229 {
230 	struct dev_ext_attribute *eattr =
231 		container_of(attr, struct dev_ext_attribute, attr);
232 	return sysfs_emit(buf, "%s\n", (char *)eattr->var);
233 }
234 EXPORT_SYMBOL_GPL(arm_cspmu_sysfs_format_show);
235 
236 static struct attribute *arm_cspmu_format_attrs[] = {
237 	ARM_CSPMU_FORMAT_EVENT_ATTR,
238 	ARM_CSPMU_FORMAT_FILTER_ATTR,
239 	NULL,
240 };
241 
242 static struct attribute **
243 arm_cspmu_get_format_attrs(const struct arm_cspmu *cspmu)
244 {
245 	struct attribute **attrs;
246 
247 	attrs = devm_kmemdup(cspmu->dev, arm_cspmu_format_attrs,
248 		sizeof(arm_cspmu_format_attrs), GFP_KERNEL);
249 
250 	return attrs;
251 }
252 
253 static u32 arm_cspmu_event_type(const struct perf_event *event)
254 {
255 	return event->attr.config & ARM_CSPMU_EVENT_MASK;
256 }
257 
258 static bool arm_cspmu_is_cycle_counter_event(const struct perf_event *event)
259 {
260 	return (event->attr.config == ARM_CSPMU_EVT_CYCLES_DEFAULT);
261 }
262 
263 static u32 arm_cspmu_event_filter(const struct perf_event *event)
264 {
265 	return event->attr.config1 & ARM_CSPMU_FILTER_MASK;
266 }
267 
268 static ssize_t arm_cspmu_identifier_show(struct device *dev,
269 					 struct device_attribute *attr,
270 					 char *page)
271 {
272 	struct arm_cspmu *cspmu = to_arm_cspmu(dev_get_drvdata(dev));
273 
274 	return sysfs_emit(page, "%s\n", cspmu->identifier);
275 }
276 
277 static struct device_attribute arm_cspmu_identifier_attr =
278 	__ATTR(identifier, 0444, arm_cspmu_identifier_show, NULL);
279 
280 static struct attribute *arm_cspmu_identifier_attrs[] = {
281 	&arm_cspmu_identifier_attr.attr,
282 	NULL,
283 };
284 
285 static struct attribute_group arm_cspmu_identifier_attr_group = {
286 	.attrs = arm_cspmu_identifier_attrs,
287 };
288 
289 static const char *arm_cspmu_get_identifier(const struct arm_cspmu *cspmu)
290 {
291 	const char *identifier =
292 		devm_kasprintf(cspmu->dev, GFP_KERNEL, "%x",
293 			       cspmu->impl.pmiidr);
294 	return identifier;
295 }
296 
297 static const char *arm_cspmu_type_str[ACPI_APMT_NODE_TYPE_COUNT] = {
298 	"mc",
299 	"smmu",
300 	"pcie",
301 	"acpi",
302 	"cache",
303 };
304 
305 static const char *arm_cspmu_get_name(const struct arm_cspmu *cspmu)
306 {
307 	struct device *dev;
308 	struct acpi_apmt_node *apmt_node;
309 	u8 pmu_type;
310 	char *name;
311 	char acpi_hid_string[ACPI_ID_LEN] = { 0 };
312 	static atomic_t pmu_idx[ACPI_APMT_NODE_TYPE_COUNT] = { 0 };
313 
314 	dev = cspmu->dev;
315 	apmt_node = arm_cspmu_apmt_node(dev);
316 	if (!apmt_node)
317 		return devm_kasprintf(dev, GFP_KERNEL, PMUNAME "_%u",
318 				      atomic_fetch_inc(&pmu_idx[0]));
319 
320 	pmu_type = apmt_node->type;
321 
322 	if (pmu_type >= ACPI_APMT_NODE_TYPE_COUNT) {
323 		dev_err(dev, "unsupported PMU type-%u\n", pmu_type);
324 		return NULL;
325 	}
326 
327 	if (pmu_type == ACPI_APMT_NODE_TYPE_ACPI) {
328 		memcpy(acpi_hid_string,
329 			&apmt_node->inst_primary,
330 			sizeof(apmt_node->inst_primary));
331 		name = devm_kasprintf(dev, GFP_KERNEL, "%s_%s_%s_%u", PMUNAME,
332 				      arm_cspmu_type_str[pmu_type],
333 				      acpi_hid_string,
334 				      apmt_node->inst_secondary);
335 	} else {
336 		name = devm_kasprintf(dev, GFP_KERNEL, "%s_%s_%d", PMUNAME,
337 				      arm_cspmu_type_str[pmu_type],
338 				      atomic_fetch_inc(&pmu_idx[pmu_type]));
339 	}
340 
341 	return name;
342 }
343 
344 static ssize_t arm_cspmu_cpumask_show(struct device *dev,
345 				      struct device_attribute *attr,
346 				      char *buf)
347 {
348 	struct pmu *pmu = dev_get_drvdata(dev);
349 	struct arm_cspmu *cspmu = to_arm_cspmu(pmu);
350 	struct dev_ext_attribute *eattr =
351 		container_of(attr, struct dev_ext_attribute, attr);
352 	unsigned long mask_id = (unsigned long)eattr->var;
353 	const cpumask_t *cpumask;
354 
355 	switch (mask_id) {
356 	case ARM_CSPMU_ACTIVE_CPU_MASK:
357 		cpumask = &cspmu->active_cpu;
358 		break;
359 	case ARM_CSPMU_ASSOCIATED_CPU_MASK:
360 		cpumask = &cspmu->associated_cpus;
361 		break;
362 	default:
363 		return 0;
364 	}
365 	return cpumap_print_to_pagebuf(true, buf, cpumask);
366 }
367 
368 static struct attribute *arm_cspmu_cpumask_attrs[] = {
369 	ARM_CSPMU_CPUMASK_ATTR(cpumask, ARM_CSPMU_ACTIVE_CPU_MASK),
370 	ARM_CSPMU_CPUMASK_ATTR(associated_cpus, ARM_CSPMU_ASSOCIATED_CPU_MASK),
371 	NULL,
372 };
373 
374 static struct attribute_group arm_cspmu_cpumask_attr_group = {
375 	.attrs = arm_cspmu_cpumask_attrs,
376 };
377 
378 static struct arm_cspmu_impl_match impl_match[] = {
379 	{
380 		.module_name	= "nvidia_cspmu",
381 		.pmiidr_val	= ARM_CSPMU_IMPL_ID_NVIDIA,
382 		.pmiidr_mask	= ARM_CSPMU_PMIIDR_IMPLEMENTER,
383 		.module		= NULL,
384 		.impl_init_ops	= NULL,
385 	},
386 	{
387 		.module_name	= "ampere_cspmu",
388 		.pmiidr_val	= ARM_CSPMU_IMPL_ID_AMPERE,
389 		.pmiidr_mask	= ARM_CSPMU_PMIIDR_IMPLEMENTER,
390 		.module		= NULL,
391 		.impl_init_ops	= NULL,
392 	},
393 
394 	{0}
395 };
396 
397 static struct arm_cspmu_impl_match *arm_cspmu_impl_match_get(u32 pmiidr)
398 {
399 	struct arm_cspmu_impl_match *match = impl_match;
400 
401 	for (; match->pmiidr_val; match++) {
402 		u32 mask = match->pmiidr_mask;
403 
404 		if ((match->pmiidr_val & mask) == (pmiidr & mask))
405 			return match;
406 	}
407 
408 	return NULL;
409 }
410 
411 #define DEFAULT_IMPL_OP(name)	.name = arm_cspmu_##name
412 
413 static int arm_cspmu_init_impl_ops(struct arm_cspmu *cspmu)
414 {
415 	int ret = 0;
416 	struct acpi_apmt_node *apmt_node = arm_cspmu_apmt_node(cspmu->dev);
417 	struct arm_cspmu_impl_match *match;
418 
419 	/* Start with a default PMU implementation */
420 	cspmu->impl.module = THIS_MODULE;
421 	cspmu->impl.pmiidr = readl(cspmu->base0 + PMIIDR);
422 	cspmu->impl.ops = (struct arm_cspmu_impl_ops) {
423 		DEFAULT_IMPL_OP(get_event_attrs),
424 		DEFAULT_IMPL_OP(get_format_attrs),
425 		DEFAULT_IMPL_OP(get_identifier),
426 		DEFAULT_IMPL_OP(get_name),
427 		DEFAULT_IMPL_OP(is_cycle_counter_event),
428 		DEFAULT_IMPL_OP(event_type),
429 		DEFAULT_IMPL_OP(event_filter),
430 		DEFAULT_IMPL_OP(set_ev_filter),
431 		DEFAULT_IMPL_OP(event_attr_is_visible),
432 	};
433 
434 	/* Firmware may override implementer/product ID from PMIIDR */
435 	if (apmt_node && apmt_node->impl_id)
436 		cspmu->impl.pmiidr = apmt_node->impl_id;
437 
438 	/* Find implementer specific attribute ops. */
439 	match = arm_cspmu_impl_match_get(cspmu->impl.pmiidr);
440 
441 	/* Load implementer module and initialize the callbacks. */
442 	if (match) {
443 		mutex_lock(&arm_cspmu_lock);
444 
445 		if (match->impl_init_ops) {
446 			/* Prevent unload until PMU registration is done. */
447 			if (try_module_get(match->module)) {
448 				cspmu->impl.module = match->module;
449 				cspmu->impl.match = match;
450 				ret = match->impl_init_ops(cspmu);
451 				if (ret)
452 					module_put(match->module);
453 			} else {
454 				WARN(1, "arm_cspmu failed to get module: %s\n",
455 					match->module_name);
456 				ret = -EINVAL;
457 			}
458 		} else {
459 			request_module_nowait(match->module_name);
460 			ret = -EPROBE_DEFER;
461 		}
462 
463 		mutex_unlock(&arm_cspmu_lock);
464 	}
465 
466 	return ret;
467 }
468 
469 static struct attribute_group *
470 arm_cspmu_alloc_event_attr_group(struct arm_cspmu *cspmu)
471 {
472 	struct attribute_group *event_group;
473 	struct device *dev = cspmu->dev;
474 	const struct arm_cspmu_impl_ops *impl_ops = &cspmu->impl.ops;
475 
476 	event_group =
477 		devm_kzalloc(dev, sizeof(struct attribute_group), GFP_KERNEL);
478 	if (!event_group)
479 		return NULL;
480 
481 	event_group->name = "events";
482 	event_group->is_visible = impl_ops->event_attr_is_visible;
483 	event_group->attrs = impl_ops->get_event_attrs(cspmu);
484 
485 	if (!event_group->attrs)
486 		return NULL;
487 
488 	return event_group;
489 }
490 
491 static struct attribute_group *
492 arm_cspmu_alloc_format_attr_group(struct arm_cspmu *cspmu)
493 {
494 	struct attribute_group *format_group;
495 	struct device *dev = cspmu->dev;
496 
497 	format_group =
498 		devm_kzalloc(dev, sizeof(struct attribute_group), GFP_KERNEL);
499 	if (!format_group)
500 		return NULL;
501 
502 	format_group->name = "format";
503 	format_group->attrs = cspmu->impl.ops.get_format_attrs(cspmu);
504 
505 	if (!format_group->attrs)
506 		return NULL;
507 
508 	return format_group;
509 }
510 
511 static int arm_cspmu_alloc_attr_groups(struct arm_cspmu *cspmu)
512 {
513 	const struct attribute_group **attr_groups = cspmu->attr_groups;
514 	const struct arm_cspmu_impl_ops *impl_ops = &cspmu->impl.ops;
515 
516 	cspmu->identifier = impl_ops->get_identifier(cspmu);
517 	cspmu->name = impl_ops->get_name(cspmu);
518 
519 	if (!cspmu->identifier || !cspmu->name)
520 		return -ENOMEM;
521 
522 	attr_groups[0] = arm_cspmu_alloc_event_attr_group(cspmu);
523 	attr_groups[1] = arm_cspmu_alloc_format_attr_group(cspmu);
524 	attr_groups[2] = &arm_cspmu_identifier_attr_group;
525 	attr_groups[3] = &arm_cspmu_cpumask_attr_group;
526 
527 	if (!attr_groups[0] || !attr_groups[1])
528 		return -ENOMEM;
529 
530 	return 0;
531 }
532 
533 static inline void arm_cspmu_reset_counters(struct arm_cspmu *cspmu)
534 {
535 	writel(PMCR_C | PMCR_P, cspmu->base0 + PMCR);
536 }
537 
538 static inline void arm_cspmu_start_counters(struct arm_cspmu *cspmu)
539 {
540 	writel(PMCR_E, cspmu->base0 + PMCR);
541 }
542 
543 static inline void arm_cspmu_stop_counters(struct arm_cspmu *cspmu)
544 {
545 	writel(0, cspmu->base0 + PMCR);
546 }
547 
548 static void arm_cspmu_enable(struct pmu *pmu)
549 {
550 	bool disabled;
551 	struct arm_cspmu *cspmu = to_arm_cspmu(pmu);
552 
553 	disabled = bitmap_empty(cspmu->hw_events.used_ctrs,
554 				cspmu->num_logical_ctrs);
555 
556 	if (disabled)
557 		return;
558 
559 	arm_cspmu_start_counters(cspmu);
560 }
561 
562 static void arm_cspmu_disable(struct pmu *pmu)
563 {
564 	struct arm_cspmu *cspmu = to_arm_cspmu(pmu);
565 
566 	arm_cspmu_stop_counters(cspmu);
567 }
568 
569 static int arm_cspmu_get_event_idx(struct arm_cspmu_hw_events *hw_events,
570 				struct perf_event *event)
571 {
572 	int idx, ret;
573 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
574 
575 	if (supports_cycle_counter(cspmu)) {
576 		if (cspmu->impl.ops.is_cycle_counter_event(event)) {
577 			/* Search for available cycle counter. */
578 			if (test_and_set_bit(cspmu->cycle_counter_logical_idx,
579 					     hw_events->used_ctrs))
580 				return -EAGAIN;
581 
582 			return cspmu->cycle_counter_logical_idx;
583 		}
584 
585 		/*
586 		 * Search a regular counter from the used counter bitmap.
587 		 * The cycle counter divides the bitmap into two parts. Search
588 		 * the first then second half to exclude the cycle counter bit.
589 		 */
590 		idx = find_first_zero_bit(hw_events->used_ctrs,
591 					  cspmu->cycle_counter_logical_idx);
592 		if (idx >= cspmu->cycle_counter_logical_idx) {
593 			idx = find_next_zero_bit(
594 				hw_events->used_ctrs,
595 				cspmu->num_logical_ctrs,
596 				cspmu->cycle_counter_logical_idx + 1);
597 		}
598 	} else {
599 		idx = find_first_zero_bit(hw_events->used_ctrs,
600 					  cspmu->num_logical_ctrs);
601 	}
602 
603 	if (idx >= cspmu->num_logical_ctrs)
604 		return -EAGAIN;
605 
606 	if (cspmu->impl.ops.validate_event) {
607 		ret = cspmu->impl.ops.validate_event(cspmu, event);
608 		if (ret)
609 			return ret;
610 	}
611 
612 	set_bit(idx, hw_events->used_ctrs);
613 
614 	return idx;
615 }
616 
617 static bool arm_cspmu_validate_event(struct pmu *pmu,
618 				 struct arm_cspmu_hw_events *hw_events,
619 				 struct perf_event *event)
620 {
621 	if (is_software_event(event))
622 		return true;
623 
624 	/* Reject groups spanning multiple HW PMUs. */
625 	if (event->pmu != pmu)
626 		return false;
627 
628 	return (arm_cspmu_get_event_idx(hw_events, event) >= 0);
629 }
630 
631 /*
632  * Make sure the group of events can be scheduled at once
633  * on the PMU.
634  */
635 static bool arm_cspmu_validate_group(struct perf_event *event)
636 {
637 	struct perf_event *sibling, *leader = event->group_leader;
638 	struct arm_cspmu_hw_events fake_hw_events;
639 
640 	if (event->group_leader == event)
641 		return true;
642 
643 	memset(&fake_hw_events, 0, sizeof(fake_hw_events));
644 
645 	if (!arm_cspmu_validate_event(event->pmu, &fake_hw_events, leader))
646 		return false;
647 
648 	for_each_sibling_event(sibling, leader) {
649 		if (!arm_cspmu_validate_event(event->pmu, &fake_hw_events,
650 						  sibling))
651 			return false;
652 	}
653 
654 	return arm_cspmu_validate_event(event->pmu, &fake_hw_events, event);
655 }
656 
657 static int arm_cspmu_event_init(struct perf_event *event)
658 {
659 	struct arm_cspmu *cspmu;
660 	struct hw_perf_event *hwc = &event->hw;
661 
662 	cspmu = to_arm_cspmu(event->pmu);
663 
664 	if (event->attr.type != event->pmu->type)
665 		return -ENOENT;
666 
667 	/*
668 	 * Following other "uncore" PMUs, we do not support sampling mode or
669 	 * attach to a task (per-process mode).
670 	 */
671 	if (is_sampling_event(event)) {
672 		dev_dbg(cspmu->pmu.dev,
673 			"Can't support sampling events\n");
674 		return -EOPNOTSUPP;
675 	}
676 
677 	if (event->cpu < 0 || event->attach_state & PERF_ATTACH_TASK) {
678 		dev_dbg(cspmu->pmu.dev,
679 			"Can't support per-task counters\n");
680 		return -EINVAL;
681 	}
682 
683 	/*
684 	 * Make sure the CPU assignment is on one of the CPUs associated with
685 	 * this PMU.
686 	 */
687 	if (!cpumask_test_cpu(event->cpu, &cspmu->associated_cpus)) {
688 		dev_dbg(cspmu->pmu.dev,
689 			"Requested cpu is not associated with the PMU\n");
690 		return -EINVAL;
691 	}
692 
693 	/* Enforce the current active CPU to handle the events in this PMU. */
694 	event->cpu = cpumask_first(&cspmu->active_cpu);
695 	if (event->cpu >= nr_cpu_ids)
696 		return -EINVAL;
697 
698 	if (!arm_cspmu_validate_group(event))
699 		return -EINVAL;
700 
701 	/*
702 	 * The logical counter id is tracked with hw_perf_event.extra_reg.idx.
703 	 * The physical counter id is tracked with hw_perf_event.idx.
704 	 * We don't assign an index until we actually place the event onto
705 	 * hardware. Use -1 to signify that we haven't decided where to put it
706 	 * yet.
707 	 */
708 	hwc->idx = -1;
709 	hwc->extra_reg.idx = -1;
710 	hwc->config = cspmu->impl.ops.event_type(event);
711 
712 	return 0;
713 }
714 
715 static inline u32 counter_offset(u32 reg_sz, u32 ctr_idx)
716 {
717 	return (PMEVCNTR_LO + (reg_sz * ctr_idx));
718 }
719 
720 static void arm_cspmu_write_counter(struct perf_event *event, u64 val)
721 {
722 	u32 offset;
723 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
724 
725 	if (use_64b_counter_reg(cspmu)) {
726 		offset = counter_offset(sizeof(u64), event->hw.idx);
727 
728 		if (cspmu->has_atomic_dword)
729 			writeq(val, cspmu->base1 + offset);
730 		else
731 			lo_hi_writeq(val, cspmu->base1 + offset);
732 	} else {
733 		offset = counter_offset(sizeof(u32), event->hw.idx);
734 
735 		writel(lower_32_bits(val), cspmu->base1 + offset);
736 	}
737 }
738 
739 static u64 arm_cspmu_read_counter(struct perf_event *event)
740 {
741 	u32 offset;
742 	const void __iomem *counter_addr;
743 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
744 
745 	if (use_64b_counter_reg(cspmu)) {
746 		offset = counter_offset(sizeof(u64), event->hw.idx);
747 		counter_addr = cspmu->base1 + offset;
748 
749 		return cspmu->has_atomic_dword ?
750 			       readq(counter_addr) :
751 			       read_reg64_hilohi(counter_addr, HILOHI_MAX_POLL);
752 	}
753 
754 	offset = counter_offset(sizeof(u32), event->hw.idx);
755 	return readl(cspmu->base1 + offset);
756 }
757 
758 /*
759  * arm_cspmu_set_event_period: Set the period for the counter.
760  *
761  * To handle cases of extreme interrupt latency, we program
762  * the counter with half of the max count for the counters.
763  */
764 static void arm_cspmu_set_event_period(struct perf_event *event)
765 {
766 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
767 	u64 val = counter_mask(cspmu) >> 1ULL;
768 
769 	local64_set(&event->hw.prev_count, val);
770 	arm_cspmu_write_counter(event, val);
771 }
772 
773 static void arm_cspmu_enable_counter(struct arm_cspmu *cspmu, int idx)
774 {
775 	u32 reg_id, reg_bit, inten_off, cnten_off;
776 
777 	reg_id = COUNTER_TO_SET_CLR_ID(idx);
778 	reg_bit = COUNTER_TO_SET_CLR_BIT(idx);
779 
780 	inten_off = PMINTENSET + (4 * reg_id);
781 	cnten_off = PMCNTENSET + (4 * reg_id);
782 
783 	writel(BIT(reg_bit), cspmu->base0 + inten_off);
784 	writel(BIT(reg_bit), cspmu->base0 + cnten_off);
785 }
786 
787 static void arm_cspmu_disable_counter(struct arm_cspmu *cspmu, int idx)
788 {
789 	u32 reg_id, reg_bit, inten_off, cnten_off;
790 
791 	reg_id = COUNTER_TO_SET_CLR_ID(idx);
792 	reg_bit = COUNTER_TO_SET_CLR_BIT(idx);
793 
794 	inten_off = PMINTENCLR + (4 * reg_id);
795 	cnten_off = PMCNTENCLR + (4 * reg_id);
796 
797 	writel(BIT(reg_bit), cspmu->base0 + cnten_off);
798 	writel(BIT(reg_bit), cspmu->base0 + inten_off);
799 }
800 
801 static void arm_cspmu_event_update(struct perf_event *event)
802 {
803 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
804 	struct hw_perf_event *hwc = &event->hw;
805 	u64 delta, prev, now;
806 
807 	do {
808 		prev = local64_read(&hwc->prev_count);
809 		now = arm_cspmu_read_counter(event);
810 	} while (local64_cmpxchg(&hwc->prev_count, prev, now) != prev);
811 
812 	delta = (now - prev) & counter_mask(cspmu);
813 	local64_add(delta, &event->count);
814 }
815 
816 static inline void arm_cspmu_set_event(struct arm_cspmu *cspmu,
817 					struct hw_perf_event *hwc)
818 {
819 	u32 offset = PMEVTYPER + (4 * hwc->idx);
820 
821 	writel(hwc->config, cspmu->base0 + offset);
822 }
823 
824 static void arm_cspmu_set_ev_filter(struct arm_cspmu *cspmu,
825 					struct hw_perf_event *hwc,
826 					u32 filter)
827 {
828 	u32 offset = PMEVFILTR + (4 * hwc->idx);
829 
830 	writel(filter, cspmu->base0 + offset);
831 }
832 
833 static inline void arm_cspmu_set_cc_filter(struct arm_cspmu *cspmu, u32 filter)
834 {
835 	u32 offset = PMCCFILTR;
836 
837 	writel(filter, cspmu->base0 + offset);
838 }
839 
840 static void arm_cspmu_start(struct perf_event *event, int pmu_flags)
841 {
842 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
843 	struct hw_perf_event *hwc = &event->hw;
844 	u32 filter;
845 
846 	/* We always reprogram the counter */
847 	if (pmu_flags & PERF_EF_RELOAD)
848 		WARN_ON(!(hwc->state & PERF_HES_UPTODATE));
849 
850 	arm_cspmu_set_event_period(event);
851 
852 	filter = cspmu->impl.ops.event_filter(event);
853 
854 	if (event->hw.extra_reg.idx == cspmu->cycle_counter_logical_idx) {
855 		arm_cspmu_set_cc_filter(cspmu, filter);
856 	} else {
857 		arm_cspmu_set_event(cspmu, hwc);
858 		cspmu->impl.ops.set_ev_filter(cspmu, hwc, filter);
859 	}
860 
861 	hwc->state = 0;
862 
863 	arm_cspmu_enable_counter(cspmu, hwc->idx);
864 }
865 
866 static void arm_cspmu_stop(struct perf_event *event, int pmu_flags)
867 {
868 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
869 	struct hw_perf_event *hwc = &event->hw;
870 
871 	if (hwc->state & PERF_HES_STOPPED)
872 		return;
873 
874 	arm_cspmu_disable_counter(cspmu, hwc->idx);
875 	arm_cspmu_event_update(event);
876 
877 	hwc->state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
878 }
879 
880 static inline u32 to_phys_idx(struct arm_cspmu *cspmu, u32 idx)
881 {
882 	return (idx == cspmu->cycle_counter_logical_idx) ?
883 		ARM_CSPMU_CYCLE_CNTR_IDX : idx;
884 }
885 
886 static int arm_cspmu_add(struct perf_event *event, int flags)
887 {
888 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
889 	struct arm_cspmu_hw_events *hw_events = &cspmu->hw_events;
890 	struct hw_perf_event *hwc = &event->hw;
891 	int idx;
892 
893 	if (WARN_ON_ONCE(!cpumask_test_cpu(smp_processor_id(),
894 					   &cspmu->associated_cpus)))
895 		return -ENOENT;
896 
897 	idx = arm_cspmu_get_event_idx(hw_events, event);
898 	if (idx < 0)
899 		return idx;
900 
901 	hw_events->events[idx] = event;
902 	hwc->idx = to_phys_idx(cspmu, idx);
903 	hwc->extra_reg.idx = idx;
904 	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
905 
906 	if (flags & PERF_EF_START)
907 		arm_cspmu_start(event, PERF_EF_RELOAD);
908 
909 	/* Propagate changes to the userspace mapping. */
910 	perf_event_update_userpage(event);
911 
912 	return 0;
913 }
914 
915 static void arm_cspmu_del(struct perf_event *event, int flags)
916 {
917 	struct arm_cspmu *cspmu = to_arm_cspmu(event->pmu);
918 	struct arm_cspmu_hw_events *hw_events = &cspmu->hw_events;
919 	struct hw_perf_event *hwc = &event->hw;
920 	int idx = hwc->extra_reg.idx;
921 
922 	arm_cspmu_stop(event, PERF_EF_UPDATE);
923 
924 	hw_events->events[idx] = NULL;
925 
926 	clear_bit(idx, hw_events->used_ctrs);
927 
928 	perf_event_update_userpage(event);
929 }
930 
931 static void arm_cspmu_read(struct perf_event *event)
932 {
933 	arm_cspmu_event_update(event);
934 }
935 
936 static struct arm_cspmu *arm_cspmu_alloc(struct platform_device *pdev)
937 {
938 	struct acpi_apmt_node *apmt_node;
939 	struct arm_cspmu *cspmu;
940 	struct device *dev = &pdev->dev;
941 
942 	cspmu = devm_kzalloc(dev, sizeof(*cspmu), GFP_KERNEL);
943 	if (!cspmu)
944 		return NULL;
945 
946 	cspmu->dev = dev;
947 	platform_set_drvdata(pdev, cspmu);
948 
949 	apmt_node = arm_cspmu_apmt_node(dev);
950 	if (apmt_node) {
951 		cspmu->has_atomic_dword = apmt_node->flags & ACPI_APMT_FLAGS_ATOMIC;
952 	} else {
953 		u32 width = 0;
954 
955 		device_property_read_u32(dev, "reg-io-width", &width);
956 		cspmu->has_atomic_dword = (width == 8);
957 	}
958 
959 	return cspmu;
960 }
961 
962 static int arm_cspmu_init_mmio(struct arm_cspmu *cspmu)
963 {
964 	struct device *dev;
965 	struct platform_device *pdev;
966 
967 	dev = cspmu->dev;
968 	pdev = to_platform_device(dev);
969 
970 	/* Base address for page 0. */
971 	cspmu->base0 = devm_platform_ioremap_resource(pdev, 0);
972 	if (IS_ERR(cspmu->base0)) {
973 		dev_err(dev, "ioremap failed for page-0 resource\n");
974 		return PTR_ERR(cspmu->base0);
975 	}
976 
977 	/* Base address for page 1 if supported. Otherwise point to page 0. */
978 	cspmu->base1 = cspmu->base0;
979 	if (platform_get_resource(pdev, IORESOURCE_MEM, 1)) {
980 		cspmu->base1 = devm_platform_ioremap_resource(pdev, 1);
981 		if (IS_ERR(cspmu->base1)) {
982 			dev_err(dev, "ioremap failed for page-1 resource\n");
983 			return PTR_ERR(cspmu->base1);
984 		}
985 	}
986 
987 	cspmu->pmcfgr = readl(cspmu->base0 + PMCFGR);
988 
989 	cspmu->num_logical_ctrs = FIELD_GET(PMCFGR_N, cspmu->pmcfgr) + 1;
990 
991 	cspmu->cycle_counter_logical_idx = ARM_CSPMU_MAX_HW_CNTRS;
992 
993 	if (supports_cycle_counter(cspmu)) {
994 		/*
995 		 * The last logical counter is mapped to cycle counter if
996 		 * there is a gap between regular and cycle counter. Otherwise,
997 		 * logical and physical have 1-to-1 mapping.
998 		 */
999 		cspmu->cycle_counter_logical_idx =
1000 			(cspmu->num_logical_ctrs <= ARM_CSPMU_CYCLE_CNTR_IDX) ?
1001 				cspmu->num_logical_ctrs - 1 :
1002 				ARM_CSPMU_CYCLE_CNTR_IDX;
1003 	}
1004 
1005 	cspmu->num_set_clr_reg =
1006 		DIV_ROUND_UP(cspmu->num_logical_ctrs,
1007 				ARM_CSPMU_SET_CLR_COUNTER_NUM);
1008 
1009 	cspmu->hw_events.events =
1010 		devm_kcalloc(dev, cspmu->num_logical_ctrs,
1011 			     sizeof(*cspmu->hw_events.events), GFP_KERNEL);
1012 
1013 	if (!cspmu->hw_events.events)
1014 		return -ENOMEM;
1015 
1016 	return 0;
1017 }
1018 
1019 static inline int arm_cspmu_get_reset_overflow(struct arm_cspmu *cspmu,
1020 					       u32 *pmovs)
1021 {
1022 	int i;
1023 	u32 pmovclr_offset = PMOVSCLR;
1024 	u32 has_overflowed = 0;
1025 
1026 	for (i = 0; i < cspmu->num_set_clr_reg; ++i) {
1027 		pmovs[i] = readl(cspmu->base1 + pmovclr_offset);
1028 		has_overflowed |= pmovs[i];
1029 		writel(pmovs[i], cspmu->base1 + pmovclr_offset);
1030 		pmovclr_offset += sizeof(u32);
1031 	}
1032 
1033 	return has_overflowed != 0;
1034 }
1035 
1036 static irqreturn_t arm_cspmu_handle_irq(int irq_num, void *dev)
1037 {
1038 	int idx, has_overflowed;
1039 	struct perf_event *event;
1040 	struct arm_cspmu *cspmu = dev;
1041 	DECLARE_BITMAP(pmovs, ARM_CSPMU_MAX_HW_CNTRS);
1042 	bool handled = false;
1043 
1044 	arm_cspmu_stop_counters(cspmu);
1045 
1046 	has_overflowed = arm_cspmu_get_reset_overflow(cspmu, (u32 *)pmovs);
1047 	if (!has_overflowed)
1048 		goto done;
1049 
1050 	for_each_set_bit(idx, cspmu->hw_events.used_ctrs,
1051 			cspmu->num_logical_ctrs) {
1052 		event = cspmu->hw_events.events[idx];
1053 
1054 		if (!event)
1055 			continue;
1056 
1057 		if (!test_bit(event->hw.idx, pmovs))
1058 			continue;
1059 
1060 		arm_cspmu_event_update(event);
1061 		arm_cspmu_set_event_period(event);
1062 
1063 		handled = true;
1064 	}
1065 
1066 done:
1067 	arm_cspmu_start_counters(cspmu);
1068 	return IRQ_RETVAL(handled);
1069 }
1070 
1071 static int arm_cspmu_request_irq(struct arm_cspmu *cspmu)
1072 {
1073 	int irq, ret;
1074 	struct device *dev;
1075 	struct platform_device *pdev;
1076 
1077 	dev = cspmu->dev;
1078 	pdev = to_platform_device(dev);
1079 
1080 	/* Skip IRQ request if the PMU does not support overflow interrupt. */
1081 	irq = platform_get_irq_optional(pdev, 0);
1082 	if (irq < 0)
1083 		return irq == -ENXIO ? 0 : irq;
1084 
1085 	ret = devm_request_irq(dev, irq, arm_cspmu_handle_irq,
1086 			       IRQF_NOBALANCING | IRQF_NO_THREAD, dev_name(dev),
1087 			       cspmu);
1088 	if (ret) {
1089 		dev_err(dev, "Could not request IRQ %d\n", irq);
1090 		return ret;
1091 	}
1092 
1093 	cspmu->irq = irq;
1094 
1095 	return 0;
1096 }
1097 
1098 #if defined(CONFIG_ACPI) && defined(CONFIG_ARM64)
1099 #include <acpi/processor.h>
1100 
1101 static inline int arm_cspmu_find_cpu_container(int cpu, u32 container_uid)
1102 {
1103 	struct device *cpu_dev;
1104 	struct acpi_device *acpi_dev;
1105 
1106 	cpu_dev = get_cpu_device(cpu);
1107 	if (!cpu_dev)
1108 		return -ENODEV;
1109 
1110 	acpi_dev = ACPI_COMPANION(cpu_dev);
1111 	while (acpi_dev) {
1112 		if (acpi_dev_hid_uid_match(acpi_dev, ACPI_PROCESSOR_CONTAINER_HID, container_uid))
1113 			return 0;
1114 
1115 		acpi_dev = acpi_dev_parent(acpi_dev);
1116 	}
1117 
1118 	return -ENODEV;
1119 }
1120 
1121 static int arm_cspmu_acpi_get_cpus(struct arm_cspmu *cspmu)
1122 {
1123 	struct acpi_apmt_node *apmt_node;
1124 	int affinity_flag;
1125 	int cpu;
1126 
1127 	apmt_node = arm_cspmu_apmt_node(cspmu->dev);
1128 	affinity_flag = apmt_node->flags & ACPI_APMT_FLAGS_AFFINITY;
1129 
1130 	if (affinity_flag == ACPI_APMT_FLAGS_AFFINITY_PROC) {
1131 		for_each_possible_cpu(cpu) {
1132 			if (apmt_node->proc_affinity ==
1133 			    get_acpi_id_for_cpu(cpu)) {
1134 				cpumask_set_cpu(cpu, &cspmu->associated_cpus);
1135 				break;
1136 			}
1137 		}
1138 	} else {
1139 		for_each_possible_cpu(cpu) {
1140 			if (arm_cspmu_find_cpu_container(
1141 				    cpu, apmt_node->proc_affinity))
1142 				continue;
1143 
1144 			cpumask_set_cpu(cpu, &cspmu->associated_cpus);
1145 		}
1146 	}
1147 
1148 	return 0;
1149 }
1150 #else
1151 static int arm_cspmu_acpi_get_cpus(struct arm_cspmu *cspmu)
1152 {
1153 	return -ENODEV;
1154 }
1155 #endif
1156 
1157 static int arm_cspmu_of_get_cpus(struct arm_cspmu *cspmu)
1158 {
1159 	struct of_phandle_iterator it;
1160 	int ret, cpu;
1161 
1162 	of_for_each_phandle(&it, ret, dev_of_node(cspmu->dev), "cpus", NULL, 0) {
1163 		cpu = of_cpu_node_to_id(it.node);
1164 		if (cpu < 0)
1165 			continue;
1166 		cpumask_set_cpu(cpu, &cspmu->associated_cpus);
1167 	}
1168 	return ret == -ENOENT ? 0 : ret;
1169 }
1170 
1171 static int arm_cspmu_get_cpus(struct arm_cspmu *cspmu)
1172 {
1173 	int ret = 0;
1174 
1175 	if (arm_cspmu_apmt_node(cspmu->dev))
1176 		ret = arm_cspmu_acpi_get_cpus(cspmu);
1177 	else if (device_property_present(cspmu->dev, "cpus"))
1178 		ret = arm_cspmu_of_get_cpus(cspmu);
1179 	else
1180 		cpumask_copy(&cspmu->associated_cpus, cpu_possible_mask);
1181 
1182 	if (!ret && cpumask_empty(&cspmu->associated_cpus)) {
1183 		dev_dbg(cspmu->dev, "No cpu associated with the PMU\n");
1184 		ret = -ENODEV;
1185 	}
1186 	return ret;
1187 }
1188 
1189 static int arm_cspmu_register_pmu(struct arm_cspmu *cspmu)
1190 {
1191 	int ret, capabilities;
1192 
1193 	ret = arm_cspmu_alloc_attr_groups(cspmu);
1194 	if (ret)
1195 		return ret;
1196 
1197 	ret = cpuhp_state_add_instance(arm_cspmu_cpuhp_state,
1198 				       &cspmu->cpuhp_node);
1199 	if (ret)
1200 		return ret;
1201 
1202 	capabilities = PERF_PMU_CAP_NO_EXCLUDE;
1203 	if (cspmu->irq == 0)
1204 		capabilities |= PERF_PMU_CAP_NO_INTERRUPT;
1205 
1206 	cspmu->pmu = (struct pmu){
1207 		.task_ctx_nr	= perf_invalid_context,
1208 		.module		= cspmu->impl.module,
1209 		.pmu_enable	= arm_cspmu_enable,
1210 		.pmu_disable	= arm_cspmu_disable,
1211 		.event_init	= arm_cspmu_event_init,
1212 		.add		= arm_cspmu_add,
1213 		.del		= arm_cspmu_del,
1214 		.start		= arm_cspmu_start,
1215 		.stop		= arm_cspmu_stop,
1216 		.read		= arm_cspmu_read,
1217 		.attr_groups	= cspmu->attr_groups,
1218 		.capabilities	= capabilities,
1219 	};
1220 
1221 	/* Hardware counter init */
1222 	arm_cspmu_reset_counters(cspmu);
1223 
1224 	ret = perf_pmu_register(&cspmu->pmu, cspmu->name, -1);
1225 	if (ret) {
1226 		cpuhp_state_remove_instance(arm_cspmu_cpuhp_state,
1227 					    &cspmu->cpuhp_node);
1228 	}
1229 
1230 	return ret;
1231 }
1232 
1233 static int arm_cspmu_device_probe(struct platform_device *pdev)
1234 {
1235 	int ret;
1236 	struct arm_cspmu *cspmu;
1237 
1238 	cspmu = arm_cspmu_alloc(pdev);
1239 	if (!cspmu)
1240 		return -ENOMEM;
1241 
1242 	ret = arm_cspmu_init_mmio(cspmu);
1243 	if (ret)
1244 		return ret;
1245 
1246 	ret = arm_cspmu_request_irq(cspmu);
1247 	if (ret)
1248 		return ret;
1249 
1250 	ret = arm_cspmu_get_cpus(cspmu);
1251 	if (ret)
1252 		return ret;
1253 
1254 	ret = arm_cspmu_init_impl_ops(cspmu);
1255 	if (ret)
1256 		return ret;
1257 
1258 	ret = arm_cspmu_register_pmu(cspmu);
1259 
1260 	/* Matches arm_cspmu_init_impl_ops() above. */
1261 	if (cspmu->impl.module != THIS_MODULE)
1262 		module_put(cspmu->impl.module);
1263 
1264 	return ret;
1265 }
1266 
1267 static void arm_cspmu_device_remove(struct platform_device *pdev)
1268 {
1269 	struct arm_cspmu *cspmu = platform_get_drvdata(pdev);
1270 
1271 	perf_pmu_unregister(&cspmu->pmu);
1272 	cpuhp_state_remove_instance(arm_cspmu_cpuhp_state, &cspmu->cpuhp_node);
1273 }
1274 
1275 static const struct platform_device_id arm_cspmu_id[] = {
1276 	{DRVNAME, 0},
1277 	{ },
1278 };
1279 MODULE_DEVICE_TABLE(platform, arm_cspmu_id);
1280 
1281 static const struct of_device_id arm_cspmu_of_match[] = {
1282 	{ .compatible = "arm,coresight-pmu" },
1283 	{}
1284 };
1285 MODULE_DEVICE_TABLE(of, arm_cspmu_of_match);
1286 
1287 static struct platform_driver arm_cspmu_driver = {
1288 	.driver = {
1289 		.name = DRVNAME,
1290 		.of_match_table = arm_cspmu_of_match,
1291 		.suppress_bind_attrs = true,
1292 	},
1293 	.probe = arm_cspmu_device_probe,
1294 	.remove_new = arm_cspmu_device_remove,
1295 	.id_table = arm_cspmu_id,
1296 };
1297 
1298 static void arm_cspmu_set_active_cpu(int cpu, struct arm_cspmu *cspmu)
1299 {
1300 	cpumask_set_cpu(cpu, &cspmu->active_cpu);
1301 	if (cspmu->irq)
1302 		WARN_ON(irq_set_affinity(cspmu->irq, &cspmu->active_cpu));
1303 }
1304 
1305 static int arm_cspmu_cpu_online(unsigned int cpu, struct hlist_node *node)
1306 {
1307 	struct arm_cspmu *cspmu =
1308 		hlist_entry_safe(node, struct arm_cspmu, cpuhp_node);
1309 
1310 	if (!cpumask_test_cpu(cpu, &cspmu->associated_cpus))
1311 		return 0;
1312 
1313 	/* If the PMU is already managed, there is nothing to do */
1314 	if (!cpumask_empty(&cspmu->active_cpu))
1315 		return 0;
1316 
1317 	/* Use this CPU for event counting */
1318 	arm_cspmu_set_active_cpu(cpu, cspmu);
1319 
1320 	return 0;
1321 }
1322 
1323 static int arm_cspmu_cpu_teardown(unsigned int cpu, struct hlist_node *node)
1324 {
1325 	int dst;
1326 	struct cpumask online_supported;
1327 
1328 	struct arm_cspmu *cspmu =
1329 		hlist_entry_safe(node, struct arm_cspmu, cpuhp_node);
1330 
1331 	/* Nothing to do if this CPU doesn't own the PMU */
1332 	if (!cpumask_test_and_clear_cpu(cpu, &cspmu->active_cpu))
1333 		return 0;
1334 
1335 	/* Choose a new CPU to migrate ownership of the PMU to */
1336 	cpumask_and(&online_supported, &cspmu->associated_cpus,
1337 		    cpu_online_mask);
1338 	dst = cpumask_any_but(&online_supported, cpu);
1339 	if (dst >= nr_cpu_ids)
1340 		return 0;
1341 
1342 	/* Use this CPU for event counting */
1343 	perf_pmu_migrate_context(&cspmu->pmu, cpu, dst);
1344 	arm_cspmu_set_active_cpu(dst, cspmu);
1345 
1346 	return 0;
1347 }
1348 
1349 static int __init arm_cspmu_init(void)
1350 {
1351 	int ret;
1352 
1353 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
1354 					"perf/arm/cspmu:online",
1355 					arm_cspmu_cpu_online,
1356 					arm_cspmu_cpu_teardown);
1357 	if (ret < 0)
1358 		return ret;
1359 	arm_cspmu_cpuhp_state = ret;
1360 	return platform_driver_register(&arm_cspmu_driver);
1361 }
1362 
1363 static void __exit arm_cspmu_exit(void)
1364 {
1365 	platform_driver_unregister(&arm_cspmu_driver);
1366 	cpuhp_remove_multi_state(arm_cspmu_cpuhp_state);
1367 }
1368 
1369 int arm_cspmu_impl_register(const struct arm_cspmu_impl_match *impl_match)
1370 {
1371 	struct arm_cspmu_impl_match *match;
1372 	int ret = 0;
1373 
1374 	match = arm_cspmu_impl_match_get(impl_match->pmiidr_val);
1375 
1376 	if (match) {
1377 		mutex_lock(&arm_cspmu_lock);
1378 
1379 		if (!match->impl_init_ops) {
1380 			match->module = impl_match->module;
1381 			match->impl_init_ops = impl_match->impl_init_ops;
1382 		} else {
1383 			/* Broken match table may contain non-unique entries */
1384 			WARN(1, "arm_cspmu backend already registered for module: %s, pmiidr: 0x%x, mask: 0x%x\n",
1385 				match->module_name,
1386 				match->pmiidr_val,
1387 				match->pmiidr_mask);
1388 
1389 			ret = -EINVAL;
1390 		}
1391 
1392 		mutex_unlock(&arm_cspmu_lock);
1393 
1394 		if (!ret)
1395 			ret = driver_attach(&arm_cspmu_driver.driver);
1396 	} else {
1397 		pr_err("arm_cspmu reg failed, unable to find a match for pmiidr: 0x%x\n",
1398 			impl_match->pmiidr_val);
1399 
1400 		ret = -EINVAL;
1401 	}
1402 
1403 	return ret;
1404 }
1405 EXPORT_SYMBOL_GPL(arm_cspmu_impl_register);
1406 
1407 static int arm_cspmu_match_device(struct device *dev, const void *match)
1408 {
1409 	struct arm_cspmu *cspmu = platform_get_drvdata(to_platform_device(dev));
1410 
1411 	return (cspmu && cspmu->impl.match == match) ? 1 : 0;
1412 }
1413 
1414 void arm_cspmu_impl_unregister(const struct arm_cspmu_impl_match *impl_match)
1415 {
1416 	struct device *dev;
1417 	struct arm_cspmu_impl_match *match;
1418 
1419 	match = arm_cspmu_impl_match_get(impl_match->pmiidr_val);
1420 
1421 	if (WARN_ON(!match))
1422 		return;
1423 
1424 	/* Unbind the driver from all matching backend devices. */
1425 	while ((dev = driver_find_device(&arm_cspmu_driver.driver, NULL,
1426 			match, arm_cspmu_match_device)))
1427 		device_release_driver(dev);
1428 
1429 	mutex_lock(&arm_cspmu_lock);
1430 
1431 	match->module = NULL;
1432 	match->impl_init_ops = NULL;
1433 
1434 	mutex_unlock(&arm_cspmu_lock);
1435 }
1436 EXPORT_SYMBOL_GPL(arm_cspmu_impl_unregister);
1437 
1438 module_init(arm_cspmu_init);
1439 module_exit(arm_cspmu_exit);
1440 
1441 MODULE_LICENSE("GPL v2");
1442