xref: /linux/drivers/perf/alibaba_uncore_drw_pmu.c (revision 906fd46a65383cd639e5eec72a047efc33045d86)
1 // SPDX-License-Identifier: GPL-2.0
2 /*
3  * Alibaba DDR Sub-System Driveway PMU driver
4  *
5  * Copyright (C) 2022 Alibaba Inc
6  */
7 
8 #define ALI_DRW_PMUNAME		"ali_drw"
9 #define ALI_DRW_DRVNAME		ALI_DRW_PMUNAME "_pmu"
10 #define pr_fmt(fmt)		ALI_DRW_DRVNAME ": " fmt
11 
12 #include <linux/acpi.h>
13 #include <linux/bitfield.h>
14 #include <linux/bitmap.h>
15 #include <linux/bitops.h>
16 #include <linux/cpuhotplug.h>
17 #include <linux/cpumask.h>
18 #include <linux/device.h>
19 #include <linux/errno.h>
20 #include <linux/interrupt.h>
21 #include <linux/irq.h>
22 #include <linux/kernel.h>
23 #include <linux/list.h>
24 #include <linux/module.h>
25 #include <linux/mutex.h>
26 #include <linux/perf_event.h>
27 #include <linux/platform_device.h>
28 #include <linux/printk.h>
29 #include <linux/rculist.h>
30 #include <linux/refcount.h>
31 
32 
33 #define ALI_DRW_PMU_COMMON_MAX_COUNTERS			16
34 #define ALI_DRW_PMU_TEST_SEL_COMMON_COUNTER_BASE	19
35 
36 #define ALI_DRW_PMU_PA_SHIFT			12
37 #define ALI_DRW_PMU_CNT_INIT			0x00000000
38 #define ALI_DRW_CNT_MAX_PERIOD			0xffffffff
39 #define ALI_DRW_PMU_CYCLE_EVT_ID		0x80
40 
41 #define ALI_DRW_PMU_CNT_CTRL			0xC00
42 #define ALI_DRW_PMU_CNT_RST			BIT(2)
43 #define ALI_DRW_PMU_CNT_STOP			BIT(1)
44 #define ALI_DRW_PMU_CNT_START			BIT(0)
45 
46 #define ALI_DRW_PMU_CNT_STATE			0xC04
47 #define ALI_DRW_PMU_TEST_CTRL			0xC08
48 #define ALI_DRW_PMU_CNT_PRELOAD			0xC0C
49 
50 #define ALI_DRW_PMU_CYCLE_CNT_HIGH_MASK		GENMASK(23, 0)
51 #define ALI_DRW_PMU_CYCLE_CNT_LOW_MASK		GENMASK(31, 0)
52 #define ALI_DRW_PMU_CYCLE_CNT_HIGH		0xC10
53 #define ALI_DRW_PMU_CYCLE_CNT_LOW		0xC14
54 
55 /* PMU EVENT SEL 0-3 are paired in 32-bit registers on a 4-byte stride */
56 #define ALI_DRW_PMU_EVENT_SEL0			0xC68
57 /* counter 0-3 use sel0, counter 4-7 use sel1...*/
58 #define ALI_DRW_PMU_EVENT_SELn(n) \
59 	(ALI_DRW_PMU_EVENT_SEL0 + (n / 4) * 0x4)
60 #define ALI_DRW_PMCOM_CNT_EN			BIT(7)
61 #define ALI_DRW_PMCOM_CNT_EVENT_MASK		GENMASK(5, 0)
62 #define ALI_DRW_PMCOM_CNT_EVENT_OFFSET(n) \
63 	(8 * (n % 4))
64 
65 /* PMU COMMON COUNTER 0-15, are paired in 32-bit registers on a 4-byte stride */
66 #define ALI_DRW_PMU_COMMON_COUNTER0		0xC78
67 #define ALI_DRW_PMU_COMMON_COUNTERn(n) \
68 	(ALI_DRW_PMU_COMMON_COUNTER0 + 0x4 * (n))
69 
70 #define ALI_DRW_PMU_OV_INTR_ENABLE_CTL		0xCB8
71 #define ALI_DRW_PMU_OV_INTR_DISABLE_CTL		0xCBC
72 #define ALI_DRW_PMU_OV_INTR_ENABLE_STATUS	0xCC0
73 #define ALI_DRW_PMU_OV_INTR_CLR			0xCC4
74 #define ALI_DRW_PMU_OV_INTR_STATUS		0xCC8
75 #define ALI_DRW_PMCOM_CNT_OV_INTR_MASK		GENMASK(23, 8)
76 #define ALI_DRW_PMBW_CNT_OV_INTR_MASK		GENMASK(7, 0)
77 #define ALI_DRW_PMU_OV_INTR_MASK		GENMASK_ULL(63, 0)
78 
79 static int ali_drw_cpuhp_state_num;
80 
81 static LIST_HEAD(ali_drw_pmu_irqs);
82 static DEFINE_MUTEX(ali_drw_pmu_irqs_lock);
83 
84 struct ali_drw_pmu_irq {
85 	struct hlist_node node;
86 	struct list_head irqs_node;
87 	struct list_head pmus_node;
88 	int irq_num;
89 	int cpu;
90 	refcount_t refcount;
91 };
92 
93 struct ali_drw_pmu {
94 	void __iomem *cfg_base;
95 	struct device *dev;
96 
97 	struct list_head pmus_node;
98 	struct ali_drw_pmu_irq *irq;
99 	int irq_num;
100 	int cpu;
101 	DECLARE_BITMAP(used_mask, ALI_DRW_PMU_COMMON_MAX_COUNTERS);
102 	struct perf_event *events[ALI_DRW_PMU_COMMON_MAX_COUNTERS];
103 	int evtids[ALI_DRW_PMU_COMMON_MAX_COUNTERS];
104 
105 	struct pmu pmu;
106 };
107 
108 #define to_ali_drw_pmu(p) (container_of(p, struct ali_drw_pmu, pmu))
109 
110 #define DRW_CONFIG_EVENTID		GENMASK(7, 0)
111 #define GET_DRW_EVENTID(event)	FIELD_GET(DRW_CONFIG_EVENTID, (event)->attr.config)
112 
113 static ssize_t ali_drw_pmu_format_show(struct device *dev,
114 				struct device_attribute *attr, char *buf)
115 {
116 	struct dev_ext_attribute *eattr;
117 
118 	eattr = container_of(attr, struct dev_ext_attribute, attr);
119 
120 	return sprintf(buf, "%s\n", (char *)eattr->var);
121 }
122 
123 /*
124  * PMU event attributes
125  */
126 static ssize_t ali_drw_pmu_event_show(struct device *dev,
127 			       struct device_attribute *attr, char *page)
128 {
129 	struct dev_ext_attribute *eattr;
130 
131 	eattr = container_of(attr, struct dev_ext_attribute, attr);
132 
133 	return sprintf(page, "config=0x%lx\n", (unsigned long)eattr->var);
134 }
135 
136 #define ALI_DRW_PMU_ATTR(_name, _func, _config)                            \
137 		(&((struct dev_ext_attribute[]) {                               \
138 				{ __ATTR(_name, 0444, _func, NULL), (void *)_config }   \
139 		})[0].attr.attr)
140 
141 #define ALI_DRW_PMU_FORMAT_ATTR(_name, _config)            \
142 	ALI_DRW_PMU_ATTR(_name, ali_drw_pmu_format_show, (void *)_config)
143 #define ALI_DRW_PMU_EVENT_ATTR(_name, _config)             \
144 	ALI_DRW_PMU_ATTR(_name, ali_drw_pmu_event_show, (unsigned long)_config)
145 
146 static struct attribute *ali_drw_pmu_events_attrs[] = {
147 	ALI_DRW_PMU_EVENT_ATTR(hif_rd_or_wr,			0x0),
148 	ALI_DRW_PMU_EVENT_ATTR(hif_wr,				0x1),
149 	ALI_DRW_PMU_EVENT_ATTR(hif_rd,				0x2),
150 	ALI_DRW_PMU_EVENT_ATTR(hif_rmw,				0x3),
151 	ALI_DRW_PMU_EVENT_ATTR(hif_hi_pri_rd,			0x4),
152 	ALI_DRW_PMU_EVENT_ATTR(dfi_wr_data_cycles,		0x7),
153 	ALI_DRW_PMU_EVENT_ATTR(dfi_rd_data_cycles,		0x8),
154 	ALI_DRW_PMU_EVENT_ATTR(hpr_xact_when_critical,		0x9),
155 	ALI_DRW_PMU_EVENT_ATTR(lpr_xact_when_critical,		0xA),
156 	ALI_DRW_PMU_EVENT_ATTR(wr_xact_when_critical,		0xB),
157 	ALI_DRW_PMU_EVENT_ATTR(op_is_activate,			0xC),
158 	ALI_DRW_PMU_EVENT_ATTR(op_is_rd_or_wr,			0xD),
159 	ALI_DRW_PMU_EVENT_ATTR(op_is_rd_activate,		0xE),
160 	ALI_DRW_PMU_EVENT_ATTR(op_is_rd,			0xF),
161 	ALI_DRW_PMU_EVENT_ATTR(op_is_wr,			0x10),
162 	ALI_DRW_PMU_EVENT_ATTR(op_is_mwr,			0x11),
163 	ALI_DRW_PMU_EVENT_ATTR(op_is_precharge,			0x12),
164 	ALI_DRW_PMU_EVENT_ATTR(precharge_for_rdwr,		0x13),
165 	ALI_DRW_PMU_EVENT_ATTR(precharge_for_other,		0x14),
166 	ALI_DRW_PMU_EVENT_ATTR(rdwr_transitions,		0x15),
167 	ALI_DRW_PMU_EVENT_ATTR(write_combine,			0x16),
168 	ALI_DRW_PMU_EVENT_ATTR(war_hazard,			0x17),
169 	ALI_DRW_PMU_EVENT_ATTR(raw_hazard,			0x18),
170 	ALI_DRW_PMU_EVENT_ATTR(waw_hazard,			0x19),
171 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_selfref_rk0,		0x1A),
172 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_selfref_rk1,		0x1B),
173 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_selfref_rk2,		0x1C),
174 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_selfref_rk3,		0x1D),
175 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_powerdown_rk0,	0x1E),
176 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_powerdown_rk1,	0x1F),
177 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_powerdown_rk2,	0x20),
178 	ALI_DRW_PMU_EVENT_ATTR(op_is_enter_powerdown_rk3,	0x21),
179 	ALI_DRW_PMU_EVENT_ATTR(selfref_mode_rk0,		0x26),
180 	ALI_DRW_PMU_EVENT_ATTR(selfref_mode_rk1,		0x27),
181 	ALI_DRW_PMU_EVENT_ATTR(selfref_mode_rk2,		0x28),
182 	ALI_DRW_PMU_EVENT_ATTR(selfref_mode_rk3,		0x29),
183 	ALI_DRW_PMU_EVENT_ATTR(op_is_refresh,			0x2A),
184 	ALI_DRW_PMU_EVENT_ATTR(op_is_crit_ref,			0x2B),
185 	ALI_DRW_PMU_EVENT_ATTR(op_is_load_mode,			0x2D),
186 	ALI_DRW_PMU_EVENT_ATTR(op_is_zqcl,			0x2E),
187 	ALI_DRW_PMU_EVENT_ATTR(visible_window_limit_reached_rd, 0x30),
188 	ALI_DRW_PMU_EVENT_ATTR(visible_window_limit_reached_wr, 0x31),
189 	ALI_DRW_PMU_EVENT_ATTR(op_is_dqsosc_mpc,		0x34),
190 	ALI_DRW_PMU_EVENT_ATTR(op_is_dqsosc_mrr,		0x35),
191 	ALI_DRW_PMU_EVENT_ATTR(op_is_tcr_mrr,			0x36),
192 	ALI_DRW_PMU_EVENT_ATTR(op_is_zqstart,			0x37),
193 	ALI_DRW_PMU_EVENT_ATTR(op_is_zqlatch,			0x38),
194 	ALI_DRW_PMU_EVENT_ATTR(chi_txreq,			0x39),
195 	ALI_DRW_PMU_EVENT_ATTR(chi_txdat,			0x3A),
196 	ALI_DRW_PMU_EVENT_ATTR(chi_rxdat,			0x3B),
197 	ALI_DRW_PMU_EVENT_ATTR(chi_rxrsp,			0x3C),
198 	ALI_DRW_PMU_EVENT_ATTR(tsz_vio,				0x3D),
199 	ALI_DRW_PMU_EVENT_ATTR(cycle,				0x80),
200 	NULL,
201 };
202 
203 static struct attribute_group ali_drw_pmu_events_attr_group = {
204 	.name = "events",
205 	.attrs = ali_drw_pmu_events_attrs,
206 };
207 
208 static struct attribute *ali_drw_pmu_format_attr[] = {
209 	ALI_DRW_PMU_FORMAT_ATTR(event, "config:0-7"),
210 	NULL,
211 };
212 
213 static const struct attribute_group ali_drw_pmu_format_group = {
214 	.name = "format",
215 	.attrs = ali_drw_pmu_format_attr,
216 };
217 
218 static ssize_t ali_drw_pmu_cpumask_show(struct device *dev,
219 					struct device_attribute *attr,
220 					char *buf)
221 {
222 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(dev_get_drvdata(dev));
223 
224 	return cpumap_print_to_pagebuf(true, buf, cpumask_of(drw_pmu->cpu));
225 }
226 
227 static struct device_attribute ali_drw_pmu_cpumask_attr =
228 		__ATTR(cpumask, 0444, ali_drw_pmu_cpumask_show, NULL);
229 
230 static struct attribute *ali_drw_pmu_cpumask_attrs[] = {
231 	&ali_drw_pmu_cpumask_attr.attr,
232 	NULL,
233 };
234 
235 static const struct attribute_group ali_drw_pmu_cpumask_attr_group = {
236 	.attrs = ali_drw_pmu_cpumask_attrs,
237 };
238 
239 static umode_t ali_drw_pmu_identifier_attr_visible(struct kobject *kobj,
240 						struct attribute *attr, int n)
241 {
242 	return attr->mode;
243 }
244 
245 static DEVICE_STRING_ATTR_RO(ali_drw_pmu_identifier, 0444, "ali_drw_pmu");
246 
247 static struct attribute *ali_drw_pmu_identifier_attrs[] = {
248 	&dev_attr_ali_drw_pmu_identifier.attr.attr,
249 	NULL
250 };
251 
252 static const struct attribute_group ali_drw_pmu_identifier_attr_group = {
253 	.attrs = ali_drw_pmu_identifier_attrs,
254 	.is_visible = ali_drw_pmu_identifier_attr_visible
255 };
256 
257 static const struct attribute_group *ali_drw_pmu_attr_groups[] = {
258 	&ali_drw_pmu_events_attr_group,
259 	&ali_drw_pmu_cpumask_attr_group,
260 	&ali_drw_pmu_format_group,
261 	&ali_drw_pmu_identifier_attr_group,
262 	NULL,
263 };
264 
265 /* find a counter for event, then in add func, hw.idx will equal to counter */
266 static int ali_drw_get_counter_idx(struct perf_event *event)
267 {
268 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
269 	int idx;
270 
271 	for (idx = 0; idx < ALI_DRW_PMU_COMMON_MAX_COUNTERS; ++idx) {
272 		if (!test_and_set_bit(idx, drw_pmu->used_mask))
273 			return idx;
274 	}
275 
276 	/* The counters are all in use. */
277 	return -EBUSY;
278 }
279 
280 static u64 ali_drw_pmu_read_counter(struct perf_event *event)
281 {
282 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
283 	u64 cycle_high, cycle_low;
284 
285 	if (GET_DRW_EVENTID(event) == ALI_DRW_PMU_CYCLE_EVT_ID) {
286 		cycle_high = readl(drw_pmu->cfg_base + ALI_DRW_PMU_CYCLE_CNT_HIGH);
287 		cycle_high &= ALI_DRW_PMU_CYCLE_CNT_HIGH_MASK;
288 		cycle_low = readl(drw_pmu->cfg_base + ALI_DRW_PMU_CYCLE_CNT_LOW);
289 		cycle_low &= ALI_DRW_PMU_CYCLE_CNT_LOW_MASK;
290 		return (cycle_high << 32 | cycle_low);
291 	}
292 
293 	return readl(drw_pmu->cfg_base +
294 		     ALI_DRW_PMU_COMMON_COUNTERn(event->hw.idx));
295 }
296 
297 static void ali_drw_pmu_event_update(struct perf_event *event)
298 {
299 	struct hw_perf_event *hwc = &event->hw;
300 	u64 delta, prev, now;
301 
302 	do {
303 		prev = local64_read(&hwc->prev_count);
304 		now = ali_drw_pmu_read_counter(event);
305 	} while (local64_cmpxchg(&hwc->prev_count, prev, now) != prev);
306 
307 	/* handle overflow. */
308 	delta = now - prev;
309 	if (GET_DRW_EVENTID(event) == ALI_DRW_PMU_CYCLE_EVT_ID)
310 		delta &= ALI_DRW_PMU_OV_INTR_MASK;
311 	else
312 		delta &= ALI_DRW_CNT_MAX_PERIOD;
313 	local64_add(delta, &event->count);
314 }
315 
316 static void ali_drw_pmu_event_set_period(struct perf_event *event)
317 {
318 	u64 pre_val;
319 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
320 
321 	/* set a preload counter for test purpose */
322 	writel(ALI_DRW_PMU_TEST_SEL_COMMON_COUNTER_BASE + event->hw.idx,
323 	       drw_pmu->cfg_base + ALI_DRW_PMU_TEST_CTRL);
324 
325 	/* set conunter initial value */
326 	pre_val = ALI_DRW_PMU_CNT_INIT;
327 	writel(pre_val, drw_pmu->cfg_base + ALI_DRW_PMU_CNT_PRELOAD);
328 	local64_set(&event->hw.prev_count, pre_val);
329 
330 	/* set sel mode to zero to start test */
331 	writel(0x0, drw_pmu->cfg_base + ALI_DRW_PMU_TEST_CTRL);
332 }
333 
334 static void ali_drw_pmu_enable_counter(struct perf_event *event)
335 {
336 	u32 val, subval, reg, shift;
337 	int counter = event->hw.idx;
338 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
339 
340 	reg = ALI_DRW_PMU_EVENT_SELn(counter);
341 	val = readl(drw_pmu->cfg_base + reg);
342 	subval = FIELD_PREP(ALI_DRW_PMCOM_CNT_EN, 1) |
343 		 FIELD_PREP(ALI_DRW_PMCOM_CNT_EVENT_MASK, drw_pmu->evtids[counter]);
344 
345 	shift = ALI_DRW_PMCOM_CNT_EVENT_OFFSET(counter);
346 	val &= ~(GENMASK(7, 0) << shift);
347 	val |= subval << shift;
348 
349 	writel(val, drw_pmu->cfg_base + reg);
350 }
351 
352 static void ali_drw_pmu_disable_counter(struct perf_event *event)
353 {
354 	u32 val, reg, subval, shift;
355 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
356 	int counter = event->hw.idx;
357 
358 	reg = ALI_DRW_PMU_EVENT_SELn(counter);
359 	val = readl(drw_pmu->cfg_base + reg);
360 	subval = FIELD_PREP(ALI_DRW_PMCOM_CNT_EN, 0) |
361 		 FIELD_PREP(ALI_DRW_PMCOM_CNT_EVENT_MASK, 0);
362 
363 	shift = ALI_DRW_PMCOM_CNT_EVENT_OFFSET(counter);
364 	val &= ~(GENMASK(7, 0) << shift);
365 	val |= subval << shift;
366 
367 	writel(val, drw_pmu->cfg_base + reg);
368 }
369 
370 static irqreturn_t ali_drw_pmu_isr(int irq_num, void *data)
371 {
372 	struct ali_drw_pmu_irq *irq = data;
373 	struct ali_drw_pmu *drw_pmu;
374 	irqreturn_t ret = IRQ_NONE;
375 
376 	rcu_read_lock();
377 	list_for_each_entry_rcu(drw_pmu, &irq->pmus_node, pmus_node) {
378 		unsigned long status, clr_status;
379 		struct perf_event *event;
380 		unsigned int idx;
381 
382 		for (idx = 0; idx < ALI_DRW_PMU_COMMON_MAX_COUNTERS; idx++) {
383 			event = drw_pmu->events[idx];
384 			if (!event)
385 				continue;
386 			ali_drw_pmu_disable_counter(event);
387 		}
388 
389 		/* common counter intr status */
390 		status = readl(drw_pmu->cfg_base + ALI_DRW_PMU_OV_INTR_STATUS);
391 		status = FIELD_GET(ALI_DRW_PMCOM_CNT_OV_INTR_MASK, status);
392 		if (status) {
393 			for_each_set_bit(idx, &status,
394 					 ALI_DRW_PMU_COMMON_MAX_COUNTERS) {
395 				event = drw_pmu->events[idx];
396 				if (WARN_ON_ONCE(!event))
397 					continue;
398 				ali_drw_pmu_event_update(event);
399 				ali_drw_pmu_event_set_period(event);
400 			}
401 
402 			/* clear common counter intr status */
403 			clr_status = FIELD_PREP(ALI_DRW_PMCOM_CNT_OV_INTR_MASK, 1);
404 			writel(clr_status,
405 			       drw_pmu->cfg_base + ALI_DRW_PMU_OV_INTR_CLR);
406 		}
407 
408 		for (idx = 0; idx < ALI_DRW_PMU_COMMON_MAX_COUNTERS; idx++) {
409 			event = drw_pmu->events[idx];
410 			if (!event)
411 				continue;
412 			if (!(event->hw.state & PERF_HES_STOPPED))
413 				ali_drw_pmu_enable_counter(event);
414 		}
415 		if (status)
416 			ret = IRQ_HANDLED;
417 	}
418 	rcu_read_unlock();
419 	return ret;
420 }
421 
422 static struct ali_drw_pmu_irq *__ali_drw_pmu_init_irq(struct platform_device
423 						      *pdev, int irq_num)
424 {
425 	int ret;
426 	struct ali_drw_pmu_irq *irq;
427 
428 	list_for_each_entry(irq, &ali_drw_pmu_irqs, irqs_node) {
429 		if (irq->irq_num == irq_num
430 		    && refcount_inc_not_zero(&irq->refcount))
431 			return irq;
432 	}
433 
434 	irq = kzalloc(sizeof(*irq), GFP_KERNEL);
435 	if (!irq)
436 		return ERR_PTR(-ENOMEM);
437 
438 	INIT_LIST_HEAD(&irq->pmus_node);
439 
440 	/* Pick one CPU to be the preferred one to use */
441 	irq->cpu = smp_processor_id();
442 	refcount_set(&irq->refcount, 1);
443 
444 	/*
445 	 * FIXME: one of DDRSS Driveway PMU overflow interrupt shares the same
446 	 * irq number with MPAM ERR_IRQ. To register DDRSS PMU and MPAM drivers
447 	 * successfully, add IRQF_SHARED flag. Howerer, PMU interrupt should not
448 	 * share with other component.
449 	 */
450 	ret = devm_request_irq(&pdev->dev, irq_num, ali_drw_pmu_isr,
451 			       IRQF_SHARED, dev_name(&pdev->dev), irq);
452 	if (ret < 0) {
453 		dev_err(&pdev->dev,
454 			"Fail to request IRQ:%d ret:%d\n", irq_num, ret);
455 		goto out_free;
456 	}
457 
458 	ret = irq_set_affinity_hint(irq_num, cpumask_of(irq->cpu));
459 	if (ret)
460 		goto out_free;
461 
462 	ret = cpuhp_state_add_instance_nocalls(ali_drw_cpuhp_state_num,
463 					     &irq->node);
464 	if (ret)
465 		goto out_free;
466 
467 	irq->irq_num = irq_num;
468 	list_add(&irq->irqs_node, &ali_drw_pmu_irqs);
469 
470 	return irq;
471 
472 out_free:
473 	kfree(irq);
474 	return ERR_PTR(ret);
475 }
476 
477 static int ali_drw_pmu_init_irq(struct ali_drw_pmu *drw_pmu,
478 				struct platform_device *pdev)
479 {
480 	int irq_num;
481 	struct ali_drw_pmu_irq *irq;
482 
483 	/* Read and init IRQ */
484 	irq_num = platform_get_irq(pdev, 0);
485 	if (irq_num < 0)
486 		return irq_num;
487 
488 	mutex_lock(&ali_drw_pmu_irqs_lock);
489 	irq = __ali_drw_pmu_init_irq(pdev, irq_num);
490 	mutex_unlock(&ali_drw_pmu_irqs_lock);
491 
492 	if (IS_ERR(irq))
493 		return PTR_ERR(irq);
494 
495 	drw_pmu->irq = irq;
496 
497 	mutex_lock(&ali_drw_pmu_irqs_lock);
498 	list_add_rcu(&drw_pmu->pmus_node, &irq->pmus_node);
499 	mutex_unlock(&ali_drw_pmu_irqs_lock);
500 
501 	return 0;
502 }
503 
504 static void ali_drw_pmu_uninit_irq(struct ali_drw_pmu *drw_pmu)
505 {
506 	struct ali_drw_pmu_irq *irq = drw_pmu->irq;
507 
508 	mutex_lock(&ali_drw_pmu_irqs_lock);
509 	list_del_rcu(&drw_pmu->pmus_node);
510 
511 	if (!refcount_dec_and_test(&irq->refcount)) {
512 		mutex_unlock(&ali_drw_pmu_irqs_lock);
513 		return;
514 	}
515 
516 	list_del(&irq->irqs_node);
517 	mutex_unlock(&ali_drw_pmu_irqs_lock);
518 
519 	WARN_ON(irq_set_affinity_hint(irq->irq_num, NULL));
520 	cpuhp_state_remove_instance_nocalls(ali_drw_cpuhp_state_num,
521 					    &irq->node);
522 	kfree(irq);
523 }
524 
525 static int ali_drw_pmu_event_init(struct perf_event *event)
526 {
527 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
528 	struct hw_perf_event *hwc = &event->hw;
529 	struct perf_event *sibling;
530 	struct device *dev = drw_pmu->pmu.dev;
531 
532 	if (event->attr.type != event->pmu->type)
533 		return -ENOENT;
534 
535 	if (is_sampling_event(event)) {
536 		dev_err(dev, "Sampling not supported!\n");
537 		return -EOPNOTSUPP;
538 	}
539 
540 	if (event->attach_state & PERF_ATTACH_TASK) {
541 		dev_err(dev, "Per-task counter cannot allocate!\n");
542 		return -EOPNOTSUPP;
543 	}
544 
545 	event->cpu = drw_pmu->cpu;
546 	if (event->cpu < 0) {
547 		dev_err(dev, "Per-task mode not supported!\n");
548 		return -EOPNOTSUPP;
549 	}
550 
551 	if (event->group_leader != event &&
552 	    !is_software_event(event->group_leader)) {
553 		dev_err(dev, "driveway only allow one event!\n");
554 		return -EINVAL;
555 	}
556 
557 	for_each_sibling_event(sibling, event->group_leader) {
558 		if (sibling != event && !is_software_event(sibling)) {
559 			dev_err(dev, "driveway event not allowed!\n");
560 			return -EINVAL;
561 		}
562 	}
563 
564 	/* reset all the pmu counters */
565 	writel(ALI_DRW_PMU_CNT_RST, drw_pmu->cfg_base + ALI_DRW_PMU_CNT_CTRL);
566 
567 	hwc->idx = -1;
568 
569 	return 0;
570 }
571 
572 static void ali_drw_pmu_start(struct perf_event *event, int flags)
573 {
574 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
575 
576 	event->hw.state = 0;
577 
578 	if (GET_DRW_EVENTID(event) == ALI_DRW_PMU_CYCLE_EVT_ID) {
579 		writel(ALI_DRW_PMU_CNT_START,
580 		       drw_pmu->cfg_base + ALI_DRW_PMU_CNT_CTRL);
581 		return;
582 	}
583 
584 	ali_drw_pmu_event_set_period(event);
585 	if (flags & PERF_EF_RELOAD) {
586 		unsigned long prev_raw_count =
587 		    local64_read(&event->hw.prev_count);
588 		writel(prev_raw_count,
589 		       drw_pmu->cfg_base + ALI_DRW_PMU_CNT_PRELOAD);
590 	}
591 
592 	ali_drw_pmu_enable_counter(event);
593 
594 	writel(ALI_DRW_PMU_CNT_START, drw_pmu->cfg_base + ALI_DRW_PMU_CNT_CTRL);
595 }
596 
597 static void ali_drw_pmu_stop(struct perf_event *event, int flags)
598 {
599 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
600 
601 	if (event->hw.state & PERF_HES_STOPPED)
602 		return;
603 
604 	if (GET_DRW_EVENTID(event) != ALI_DRW_PMU_CYCLE_EVT_ID)
605 		ali_drw_pmu_disable_counter(event);
606 
607 	writel(ALI_DRW_PMU_CNT_STOP, drw_pmu->cfg_base + ALI_DRW_PMU_CNT_CTRL);
608 
609 	ali_drw_pmu_event_update(event);
610 	event->hw.state |= PERF_HES_STOPPED | PERF_HES_UPTODATE;
611 }
612 
613 static int ali_drw_pmu_add(struct perf_event *event, int flags)
614 {
615 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
616 	struct hw_perf_event *hwc = &event->hw;
617 	int idx = -1;
618 	int evtid;
619 
620 	evtid = GET_DRW_EVENTID(event);
621 
622 	if (evtid != ALI_DRW_PMU_CYCLE_EVT_ID) {
623 		idx = ali_drw_get_counter_idx(event);
624 		if (idx < 0)
625 			return idx;
626 		drw_pmu->events[idx] = event;
627 		drw_pmu->evtids[idx] = evtid;
628 	}
629 	hwc->idx = idx;
630 
631 	hwc->state = PERF_HES_STOPPED | PERF_HES_UPTODATE;
632 
633 	if (flags & PERF_EF_START)
634 		ali_drw_pmu_start(event, PERF_EF_RELOAD);
635 
636 	/* Propagate our changes to the userspace mapping. */
637 	perf_event_update_userpage(event);
638 
639 	return 0;
640 }
641 
642 static void ali_drw_pmu_del(struct perf_event *event, int flags)
643 {
644 	struct ali_drw_pmu *drw_pmu = to_ali_drw_pmu(event->pmu);
645 	struct hw_perf_event *hwc = &event->hw;
646 	int idx = hwc->idx;
647 
648 	ali_drw_pmu_stop(event, PERF_EF_UPDATE);
649 
650 	if (idx >= 0 && idx < ALI_DRW_PMU_COMMON_MAX_COUNTERS) {
651 		drw_pmu->events[idx] = NULL;
652 		drw_pmu->evtids[idx] = 0;
653 		clear_bit(idx, drw_pmu->used_mask);
654 	}
655 
656 	perf_event_update_userpage(event);
657 }
658 
659 static void ali_drw_pmu_read(struct perf_event *event)
660 {
661 	ali_drw_pmu_event_update(event);
662 }
663 
664 static int ali_drw_pmu_probe(struct platform_device *pdev)
665 {
666 	struct ali_drw_pmu *drw_pmu;
667 	struct resource *res;
668 	char *name;
669 	int ret;
670 
671 	drw_pmu = devm_kzalloc(&pdev->dev, sizeof(*drw_pmu), GFP_KERNEL);
672 	if (!drw_pmu)
673 		return -ENOMEM;
674 
675 	drw_pmu->dev = &pdev->dev;
676 	platform_set_drvdata(pdev, drw_pmu);
677 
678 	drw_pmu->cfg_base = devm_platform_get_and_ioremap_resource(pdev, 0, &res);
679 	if (IS_ERR(drw_pmu->cfg_base))
680 		return PTR_ERR(drw_pmu->cfg_base);
681 
682 	name = devm_kasprintf(drw_pmu->dev, GFP_KERNEL, "ali_drw_%llx",
683 			      (u64) (res->start >> ALI_DRW_PMU_PA_SHIFT));
684 	if (!name)
685 		return -ENOMEM;
686 
687 	writel(ALI_DRW_PMU_CNT_RST, drw_pmu->cfg_base + ALI_DRW_PMU_CNT_CTRL);
688 
689 	/* enable the generation of interrupt by all common counters */
690 	writel(ALI_DRW_PMCOM_CNT_OV_INTR_MASK,
691 	       drw_pmu->cfg_base + ALI_DRW_PMU_OV_INTR_ENABLE_CTL);
692 
693 	/* clearing interrupt status */
694 	writel(0xffffff, drw_pmu->cfg_base + ALI_DRW_PMU_OV_INTR_CLR);
695 
696 	drw_pmu->cpu = smp_processor_id();
697 
698 	ret = ali_drw_pmu_init_irq(drw_pmu, pdev);
699 	if (ret)
700 		return ret;
701 
702 	drw_pmu->pmu = (struct pmu) {
703 		.module		= THIS_MODULE,
704 		.parent		= &pdev->dev,
705 		.task_ctx_nr	= perf_invalid_context,
706 		.event_init	= ali_drw_pmu_event_init,
707 		.add		= ali_drw_pmu_add,
708 		.del		= ali_drw_pmu_del,
709 		.start		= ali_drw_pmu_start,
710 		.stop		= ali_drw_pmu_stop,
711 		.read		= ali_drw_pmu_read,
712 		.attr_groups	= ali_drw_pmu_attr_groups,
713 		.capabilities	= PERF_PMU_CAP_NO_EXCLUDE,
714 	};
715 
716 	ret = perf_pmu_register(&drw_pmu->pmu, name, -1);
717 	if (ret) {
718 		dev_err(drw_pmu->dev, "DRW Driveway PMU PMU register failed!\n");
719 		ali_drw_pmu_uninit_irq(drw_pmu);
720 	}
721 
722 	return ret;
723 }
724 
725 static void ali_drw_pmu_remove(struct platform_device *pdev)
726 {
727 	struct ali_drw_pmu *drw_pmu = platform_get_drvdata(pdev);
728 
729 	/* disable the generation of interrupt by all common counters */
730 	writel(ALI_DRW_PMCOM_CNT_OV_INTR_MASK,
731 	       drw_pmu->cfg_base + ALI_DRW_PMU_OV_INTR_DISABLE_CTL);
732 
733 	ali_drw_pmu_uninit_irq(drw_pmu);
734 	perf_pmu_unregister(&drw_pmu->pmu);
735 }
736 
737 static int ali_drw_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node)
738 {
739 	struct ali_drw_pmu_irq *irq;
740 	struct ali_drw_pmu *drw_pmu;
741 	unsigned int target;
742 
743 	irq = hlist_entry_safe(node, struct ali_drw_pmu_irq, node);
744 	if (cpu != irq->cpu)
745 		return 0;
746 
747 	target = cpumask_any_and_but(cpumask_of_node(cpu_to_node(cpu)),
748 				     cpu_online_mask, cpu);
749 	if (target >= nr_cpu_ids)
750 		target = cpumask_any_but(cpu_online_mask, cpu);
751 
752 	if (target >= nr_cpu_ids)
753 		return 0;
754 
755 	/* We're only reading, but this isn't the place to be involving RCU */
756 	mutex_lock(&ali_drw_pmu_irqs_lock);
757 	list_for_each_entry(drw_pmu, &irq->pmus_node, pmus_node)
758 		perf_pmu_migrate_context(&drw_pmu->pmu, irq->cpu, target);
759 	mutex_unlock(&ali_drw_pmu_irqs_lock);
760 
761 	WARN_ON(irq_set_affinity_hint(irq->irq_num, cpumask_of(target)));
762 	irq->cpu = target;
763 
764 	return 0;
765 }
766 
767 /*
768  * Due to historical reasons, the HID used in the production environment is
769  * ARMHD700, so we leave ARMHD700 as Compatible ID.
770  */
771 static const struct acpi_device_id ali_drw_acpi_match[] = {
772 	{"BABA5000", 0},
773 	{"ARMHD700", 0},
774 	{}
775 };
776 
777 MODULE_DEVICE_TABLE(acpi, ali_drw_acpi_match);
778 
779 static struct platform_driver ali_drw_pmu_driver = {
780 	.driver = {
781 		   .name = "ali_drw_pmu",
782 		   .acpi_match_table = ali_drw_acpi_match,
783 		   },
784 	.probe = ali_drw_pmu_probe,
785 	.remove_new = ali_drw_pmu_remove,
786 };
787 
788 static int __init ali_drw_pmu_init(void)
789 {
790 	int ret;
791 
792 	ret = cpuhp_setup_state_multi(CPUHP_AP_ONLINE_DYN,
793 				      "ali_drw_pmu:online",
794 				      NULL, ali_drw_pmu_offline_cpu);
795 
796 	if (ret < 0) {
797 		pr_err("DRW Driveway PMU: setup hotplug failed, ret = %d\n",
798 		       ret);
799 		return ret;
800 	}
801 	ali_drw_cpuhp_state_num = ret;
802 
803 	ret = platform_driver_register(&ali_drw_pmu_driver);
804 	if (ret)
805 		cpuhp_remove_multi_state(ali_drw_cpuhp_state_num);
806 
807 	return ret;
808 }
809 
810 static void __exit ali_drw_pmu_exit(void)
811 {
812 	platform_driver_unregister(&ali_drw_pmu_driver);
813 	cpuhp_remove_multi_state(ali_drw_cpuhp_state_num);
814 }
815 
816 module_init(ali_drw_pmu_init);
817 module_exit(ali_drw_pmu_exit);
818 
819 MODULE_AUTHOR("Hongbo Yao <yaohongbo@linux.alibaba.com>");
820 MODULE_AUTHOR("Neng Chen <nengchen@linux.alibaba.com>");
821 MODULE_AUTHOR("Shuai Xue <xueshuai@linux.alibaba.com>");
822 MODULE_DESCRIPTION("Alibaba DDR Sub-System Driveway PMU driver");
823 MODULE_LICENSE("GPL v2");
824