xref: /openbmc/linux/drivers/perf/xgene_pmu.c (revision 96de2506)
1 /*
2  * APM X-Gene SoC PMU (Performance Monitor Unit)
3  *
4  * Copyright (c) 2016, Applied Micro Circuits Corporation
5  * Author: Hoan Tran <hotran@apm.com>
6  *         Tai Nguyen <ttnguyen@apm.com>
7  *
8  * This program is free software; you can redistribute  it and/or modify it
9  * under  the terms of  the GNU General  Public License as published by the
10  * Free Software Foundation;  either version 2 of the  License, or (at your
11  * option) any later version.
12  *
13  * This program is distributed in the hope that it will be useful,
14  * but WITHOUT ANY WARRANTY; without even the implied warranty of
15  * MERCHANTABILITY or FITNESS FOR A PARTICULAR PURPOSE.  See the
16  * GNU General Public License for more details.
17  *
18  * You should have received a copy of the GNU General Public License
19  * along with this program.  If not, see <http://www.gnu.org/licenses/>.
20  */
21 
22 #include <linux/acpi.h>
23 #include <linux/clk.h>
24 #include <linux/cpumask.h>
25 #include <linux/interrupt.h>
26 #include <linux/io.h>
27 #include <linux/mfd/syscon.h>
28 #include <linux/module.h>
29 #include <linux/of_address.h>
30 #include <linux/of_fdt.h>
31 #include <linux/of_irq.h>
32 #include <linux/of_platform.h>
33 #include <linux/perf_event.h>
34 #include <linux/platform_device.h>
35 #include <linux/regmap.h>
36 #include <linux/slab.h>
37 
38 #define CSW_CSWCR                       0x0000
39 #define  CSW_CSWCR_DUALMCB_MASK         BIT(0)
40 #define  CSW_CSWCR_MCB0_ROUTING(x)	(((x) & 0x0C) >> 2)
41 #define  CSW_CSWCR_MCB1_ROUTING(x)	(((x) & 0x30) >> 4)
42 #define MCBADDRMR                       0x0000
43 #define  MCBADDRMR_DUALMCU_MODE_MASK    BIT(2)
44 
45 #define PCPPMU_INTSTATUS_REG	0x000
46 #define PCPPMU_INTMASK_REG	0x004
47 #define  PCPPMU_INTMASK		0x0000000F
48 #define  PCPPMU_INTENMASK	0xFFFFFFFF
49 #define  PCPPMU_INTCLRMASK	0xFFFFFFF0
50 #define  PCPPMU_INT_MCU		BIT(0)
51 #define  PCPPMU_INT_MCB		BIT(1)
52 #define  PCPPMU_INT_L3C		BIT(2)
53 #define  PCPPMU_INT_IOB		BIT(3)
54 
55 #define  PCPPMU_V3_INTMASK	0x00FF33FF
56 #define  PCPPMU_V3_INTENMASK	0xFFFFFFFF
57 #define  PCPPMU_V3_INTCLRMASK	0xFF00CC00
58 #define  PCPPMU_V3_INT_MCU	0x000000FF
59 #define  PCPPMU_V3_INT_MCB	0x00000300
60 #define  PCPPMU_V3_INT_L3C	0x00FF0000
61 #define  PCPPMU_V3_INT_IOB	0x00003000
62 
63 #define PMU_MAX_COUNTERS	4
64 #define PMU_CNT_MAX_PERIOD	0xFFFFFFFFULL
65 #define PMU_V3_CNT_MAX_PERIOD	0xFFFFFFFFFFFFFFFFULL
66 #define PMU_OVERFLOW_MASK	0xF
67 #define PMU_PMCR_E		BIT(0)
68 #define PMU_PMCR_P		BIT(1)
69 
70 #define PMU_PMEVCNTR0		0x000
71 #define PMU_PMEVCNTR1		0x004
72 #define PMU_PMEVCNTR2		0x008
73 #define PMU_PMEVCNTR3		0x00C
74 #define PMU_PMEVTYPER0		0x400
75 #define PMU_PMEVTYPER1		0x404
76 #define PMU_PMEVTYPER2		0x408
77 #define PMU_PMEVTYPER3		0x40C
78 #define PMU_PMAMR0		0xA00
79 #define PMU_PMAMR1		0xA04
80 #define PMU_PMCNTENSET		0xC00
81 #define PMU_PMCNTENCLR		0xC20
82 #define PMU_PMINTENSET		0xC40
83 #define PMU_PMINTENCLR		0xC60
84 #define PMU_PMOVSR		0xC80
85 #define PMU_PMCR		0xE04
86 
87 /* PMU registers for V3 */
88 #define PMU_PMOVSCLR		0xC80
89 #define PMU_PMOVSSET		0xCC0
90 
91 #define to_pmu_dev(p)     container_of(p, struct xgene_pmu_dev, pmu)
92 #define GET_CNTR(ev)      (ev->hw.idx)
93 #define GET_EVENTID(ev)   (ev->hw.config & 0xFFULL)
94 #define GET_AGENTID(ev)   (ev->hw.config_base & 0xFFFFFFFFUL)
95 #define GET_AGENT1ID(ev)  ((ev->hw.config_base >> 32) & 0xFFFFFFFFUL)
96 
97 struct hw_pmu_info {
98 	u32 type;
99 	u32 enable_mask;
100 	void __iomem *csr;
101 };
102 
103 struct xgene_pmu_dev {
104 	struct hw_pmu_info *inf;
105 	struct xgene_pmu *parent;
106 	struct pmu pmu;
107 	u8 max_counters;
108 	DECLARE_BITMAP(cntr_assign_mask, PMU_MAX_COUNTERS);
109 	u64 max_period;
110 	const struct attribute_group **attr_groups;
111 	struct perf_event *pmu_counter_event[PMU_MAX_COUNTERS];
112 };
113 
114 struct xgene_pmu_ops {
115 	void (*mask_int)(struct xgene_pmu *pmu);
116 	void (*unmask_int)(struct xgene_pmu *pmu);
117 	u64 (*read_counter)(struct xgene_pmu_dev *pmu, int idx);
118 	void (*write_counter)(struct xgene_pmu_dev *pmu, int idx, u64 val);
119 	void (*write_evttype)(struct xgene_pmu_dev *pmu_dev, int idx, u32 val);
120 	void (*write_agentmsk)(struct xgene_pmu_dev *pmu_dev, u32 val);
121 	void (*write_agent1msk)(struct xgene_pmu_dev *pmu_dev, u32 val);
122 	void (*enable_counter)(struct xgene_pmu_dev *pmu_dev, int idx);
123 	void (*disable_counter)(struct xgene_pmu_dev *pmu_dev, int idx);
124 	void (*enable_counter_int)(struct xgene_pmu_dev *pmu_dev, int idx);
125 	void (*disable_counter_int)(struct xgene_pmu_dev *pmu_dev, int idx);
126 	void (*reset_counters)(struct xgene_pmu_dev *pmu_dev);
127 	void (*start_counters)(struct xgene_pmu_dev *pmu_dev);
128 	void (*stop_counters)(struct xgene_pmu_dev *pmu_dev);
129 };
130 
131 struct xgene_pmu {
132 	struct device *dev;
133 	int version;
134 	void __iomem *pcppmu_csr;
135 	u32 mcb_active_mask;
136 	u32 mc_active_mask;
137 	u32 l3c_active_mask;
138 	cpumask_t cpu;
139 	raw_spinlock_t lock;
140 	const struct xgene_pmu_ops *ops;
141 	struct list_head l3cpmus;
142 	struct list_head iobpmus;
143 	struct list_head mcbpmus;
144 	struct list_head mcpmus;
145 };
146 
147 struct xgene_pmu_dev_ctx {
148 	char *name;
149 	struct list_head next;
150 	struct xgene_pmu_dev *pmu_dev;
151 	struct hw_pmu_info inf;
152 };
153 
154 struct xgene_pmu_data {
155 	int id;
156 	u32 data;
157 };
158 
159 enum xgene_pmu_version {
160 	PCP_PMU_V1 = 1,
161 	PCP_PMU_V2,
162 	PCP_PMU_V3,
163 };
164 
165 enum xgene_pmu_dev_type {
166 	PMU_TYPE_L3C = 0,
167 	PMU_TYPE_IOB,
168 	PMU_TYPE_IOB_SLOW,
169 	PMU_TYPE_MCB,
170 	PMU_TYPE_MC,
171 };
172 
173 /*
174  * sysfs format attributes
175  */
176 static ssize_t xgene_pmu_format_show(struct device *dev,
177 				     struct device_attribute *attr, char *buf)
178 {
179 	struct dev_ext_attribute *eattr;
180 
181 	eattr = container_of(attr, struct dev_ext_attribute, attr);
182 	return sprintf(buf, "%s\n", (char *) eattr->var);
183 }
184 
185 #define XGENE_PMU_FORMAT_ATTR(_name, _config)		\
186 	(&((struct dev_ext_attribute[]) {		\
187 		{ .attr = __ATTR(_name, S_IRUGO, xgene_pmu_format_show, NULL), \
188 		  .var = (void *) _config, }		\
189 	})[0].attr.attr)
190 
191 static struct attribute *l3c_pmu_format_attrs[] = {
192 	XGENE_PMU_FORMAT_ATTR(l3c_eventid, "config:0-7"),
193 	XGENE_PMU_FORMAT_ATTR(l3c_agentid, "config1:0-9"),
194 	NULL,
195 };
196 
197 static struct attribute *iob_pmu_format_attrs[] = {
198 	XGENE_PMU_FORMAT_ATTR(iob_eventid, "config:0-7"),
199 	XGENE_PMU_FORMAT_ATTR(iob_agentid, "config1:0-63"),
200 	NULL,
201 };
202 
203 static struct attribute *mcb_pmu_format_attrs[] = {
204 	XGENE_PMU_FORMAT_ATTR(mcb_eventid, "config:0-5"),
205 	XGENE_PMU_FORMAT_ATTR(mcb_agentid, "config1:0-9"),
206 	NULL,
207 };
208 
209 static struct attribute *mc_pmu_format_attrs[] = {
210 	XGENE_PMU_FORMAT_ATTR(mc_eventid, "config:0-28"),
211 	NULL,
212 };
213 
214 static const struct attribute_group l3c_pmu_format_attr_group = {
215 	.name = "format",
216 	.attrs = l3c_pmu_format_attrs,
217 };
218 
219 static const struct attribute_group iob_pmu_format_attr_group = {
220 	.name = "format",
221 	.attrs = iob_pmu_format_attrs,
222 };
223 
224 static const struct attribute_group mcb_pmu_format_attr_group = {
225 	.name = "format",
226 	.attrs = mcb_pmu_format_attrs,
227 };
228 
229 static const struct attribute_group mc_pmu_format_attr_group = {
230 	.name = "format",
231 	.attrs = mc_pmu_format_attrs,
232 };
233 
234 static struct attribute *l3c_pmu_v3_format_attrs[] = {
235 	XGENE_PMU_FORMAT_ATTR(l3c_eventid, "config:0-39"),
236 	NULL,
237 };
238 
239 static struct attribute *iob_pmu_v3_format_attrs[] = {
240 	XGENE_PMU_FORMAT_ATTR(iob_eventid, "config:0-47"),
241 	NULL,
242 };
243 
244 static struct attribute *iob_slow_pmu_v3_format_attrs[] = {
245 	XGENE_PMU_FORMAT_ATTR(iob_slow_eventid, "config:0-16"),
246 	NULL,
247 };
248 
249 static struct attribute *mcb_pmu_v3_format_attrs[] = {
250 	XGENE_PMU_FORMAT_ATTR(mcb_eventid, "config:0-35"),
251 	NULL,
252 };
253 
254 static struct attribute *mc_pmu_v3_format_attrs[] = {
255 	XGENE_PMU_FORMAT_ATTR(mc_eventid, "config:0-44"),
256 	NULL,
257 };
258 
259 static const struct attribute_group l3c_pmu_v3_format_attr_group = {
260 	.name = "format",
261 	.attrs = l3c_pmu_v3_format_attrs,
262 };
263 
264 static const struct attribute_group iob_pmu_v3_format_attr_group = {
265 	.name = "format",
266 	.attrs = iob_pmu_v3_format_attrs,
267 };
268 
269 static const struct attribute_group iob_slow_pmu_v3_format_attr_group = {
270 	.name = "format",
271 	.attrs = iob_slow_pmu_v3_format_attrs,
272 };
273 
274 static const struct attribute_group mcb_pmu_v3_format_attr_group = {
275 	.name = "format",
276 	.attrs = mcb_pmu_v3_format_attrs,
277 };
278 
279 static const struct attribute_group mc_pmu_v3_format_attr_group = {
280 	.name = "format",
281 	.attrs = mc_pmu_v3_format_attrs,
282 };
283 
284 /*
285  * sysfs event attributes
286  */
287 static ssize_t xgene_pmu_event_show(struct device *dev,
288 				    struct device_attribute *attr, char *buf)
289 {
290 	struct dev_ext_attribute *eattr;
291 
292 	eattr = container_of(attr, struct dev_ext_attribute, attr);
293 	return sprintf(buf, "config=0x%lx\n", (unsigned long) eattr->var);
294 }
295 
296 #define XGENE_PMU_EVENT_ATTR(_name, _config)		\
297 	(&((struct dev_ext_attribute[]) {		\
298 		{ .attr = __ATTR(_name, S_IRUGO, xgene_pmu_event_show, NULL), \
299 		  .var = (void *) _config, }		\
300 	 })[0].attr.attr)
301 
302 static struct attribute *l3c_pmu_events_attrs[] = {
303 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
304 	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
305 	XGENE_PMU_EVENT_ATTR(read-hit,				0x02),
306 	XGENE_PMU_EVENT_ATTR(read-miss,				0x03),
307 	XGENE_PMU_EVENT_ATTR(write-need-replacement,		0x06),
308 	XGENE_PMU_EVENT_ATTR(write-not-need-replacement,	0x07),
309 	XGENE_PMU_EVENT_ATTR(tq-full,				0x08),
310 	XGENE_PMU_EVENT_ATTR(ackq-full,				0x09),
311 	XGENE_PMU_EVENT_ATTR(wdb-full,				0x0a),
312 	XGENE_PMU_EVENT_ATTR(bank-fifo-full,			0x0b),
313 	XGENE_PMU_EVENT_ATTR(odb-full,				0x0c),
314 	XGENE_PMU_EVENT_ATTR(wbq-full,				0x0d),
315 	XGENE_PMU_EVENT_ATTR(bank-conflict-fifo-issue,		0x0e),
316 	XGENE_PMU_EVENT_ATTR(bank-fifo-issue,			0x0f),
317 	NULL,
318 };
319 
320 static struct attribute *iob_pmu_events_attrs[] = {
321 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
322 	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
323 	XGENE_PMU_EVENT_ATTR(axi0-read,				0x02),
324 	XGENE_PMU_EVENT_ATTR(axi0-read-partial,			0x03),
325 	XGENE_PMU_EVENT_ATTR(axi1-read,				0x04),
326 	XGENE_PMU_EVENT_ATTR(axi1-read-partial,			0x05),
327 	XGENE_PMU_EVENT_ATTR(csw-read-block,			0x06),
328 	XGENE_PMU_EVENT_ATTR(csw-read-partial,			0x07),
329 	XGENE_PMU_EVENT_ATTR(axi0-write,			0x10),
330 	XGENE_PMU_EVENT_ATTR(axi0-write-partial,		0x11),
331 	XGENE_PMU_EVENT_ATTR(axi1-write,			0x13),
332 	XGENE_PMU_EVENT_ATTR(axi1-write-partial,		0x14),
333 	XGENE_PMU_EVENT_ATTR(csw-inbound-dirty,			0x16),
334 	NULL,
335 };
336 
337 static struct attribute *mcb_pmu_events_attrs[] = {
338 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
339 	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
340 	XGENE_PMU_EVENT_ATTR(csw-read,				0x02),
341 	XGENE_PMU_EVENT_ATTR(csw-write-request,			0x03),
342 	XGENE_PMU_EVENT_ATTR(mcb-csw-stall,			0x04),
343 	XGENE_PMU_EVENT_ATTR(cancel-read-gack,			0x05),
344 	NULL,
345 };
346 
347 static struct attribute *mc_pmu_events_attrs[] = {
348 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
349 	XGENE_PMU_EVENT_ATTR(cycle-count-div-64,		0x01),
350 	XGENE_PMU_EVENT_ATTR(act-cmd-sent,			0x02),
351 	XGENE_PMU_EVENT_ATTR(pre-cmd-sent,			0x03),
352 	XGENE_PMU_EVENT_ATTR(rd-cmd-sent,			0x04),
353 	XGENE_PMU_EVENT_ATTR(rda-cmd-sent,			0x05),
354 	XGENE_PMU_EVENT_ATTR(wr-cmd-sent,			0x06),
355 	XGENE_PMU_EVENT_ATTR(wra-cmd-sent,			0x07),
356 	XGENE_PMU_EVENT_ATTR(pde-cmd-sent,			0x08),
357 	XGENE_PMU_EVENT_ATTR(sre-cmd-sent,			0x09),
358 	XGENE_PMU_EVENT_ATTR(prea-cmd-sent,			0x0a),
359 	XGENE_PMU_EVENT_ATTR(ref-cmd-sent,			0x0b),
360 	XGENE_PMU_EVENT_ATTR(rd-rda-cmd-sent,			0x0c),
361 	XGENE_PMU_EVENT_ATTR(wr-wra-cmd-sent,			0x0d),
362 	XGENE_PMU_EVENT_ATTR(in-rd-collision,			0x0e),
363 	XGENE_PMU_EVENT_ATTR(in-wr-collision,			0x0f),
364 	XGENE_PMU_EVENT_ATTR(collision-queue-not-empty,		0x10),
365 	XGENE_PMU_EVENT_ATTR(collision-queue-full,		0x11),
366 	XGENE_PMU_EVENT_ATTR(mcu-request,			0x12),
367 	XGENE_PMU_EVENT_ATTR(mcu-rd-request,			0x13),
368 	XGENE_PMU_EVENT_ATTR(mcu-hp-rd-request,			0x14),
369 	XGENE_PMU_EVENT_ATTR(mcu-wr-request,			0x15),
370 	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-all,		0x16),
371 	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-cancel,		0x17),
372 	XGENE_PMU_EVENT_ATTR(mcu-rd-response,			0x18),
373 	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-speculative-all,	0x19),
374 	XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-speculative-cancel,	0x1a),
375 	XGENE_PMU_EVENT_ATTR(mcu-wr-proceed-all,		0x1b),
376 	XGENE_PMU_EVENT_ATTR(mcu-wr-proceed-cancel,		0x1c),
377 	NULL,
378 };
379 
380 static const struct attribute_group l3c_pmu_events_attr_group = {
381 	.name = "events",
382 	.attrs = l3c_pmu_events_attrs,
383 };
384 
385 static const struct attribute_group iob_pmu_events_attr_group = {
386 	.name = "events",
387 	.attrs = iob_pmu_events_attrs,
388 };
389 
390 static const struct attribute_group mcb_pmu_events_attr_group = {
391 	.name = "events",
392 	.attrs = mcb_pmu_events_attrs,
393 };
394 
395 static const struct attribute_group mc_pmu_events_attr_group = {
396 	.name = "events",
397 	.attrs = mc_pmu_events_attrs,
398 };
399 
400 static struct attribute *l3c_pmu_v3_events_attrs[] = {
401 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
402 	XGENE_PMU_EVENT_ATTR(read-hit,				0x01),
403 	XGENE_PMU_EVENT_ATTR(read-miss,				0x02),
404 	XGENE_PMU_EVENT_ATTR(index-flush-eviction,		0x03),
405 	XGENE_PMU_EVENT_ATTR(write-caused-replacement,		0x04),
406 	XGENE_PMU_EVENT_ATTR(write-not-caused-replacement,	0x05),
407 	XGENE_PMU_EVENT_ATTR(clean-eviction,			0x06),
408 	XGENE_PMU_EVENT_ATTR(dirty-eviction,			0x07),
409 	XGENE_PMU_EVENT_ATTR(read,				0x08),
410 	XGENE_PMU_EVENT_ATTR(write,				0x09),
411 	XGENE_PMU_EVENT_ATTR(request,				0x0a),
412 	XGENE_PMU_EVENT_ATTR(tq-bank-conflict-issue-stall,	0x0b),
413 	XGENE_PMU_EVENT_ATTR(tq-full,				0x0c),
414 	XGENE_PMU_EVENT_ATTR(ackq-full,				0x0d),
415 	XGENE_PMU_EVENT_ATTR(wdb-full,				0x0e),
416 	XGENE_PMU_EVENT_ATTR(odb-full,				0x10),
417 	XGENE_PMU_EVENT_ATTR(wbq-full,				0x11),
418 	XGENE_PMU_EVENT_ATTR(input-req-async-fifo-stall,	0x12),
419 	XGENE_PMU_EVENT_ATTR(output-req-async-fifo-stall,	0x13),
420 	XGENE_PMU_EVENT_ATTR(output-data-async-fifo-stall,	0x14),
421 	XGENE_PMU_EVENT_ATTR(total-insertion,			0x15),
422 	XGENE_PMU_EVENT_ATTR(sip-insertions-r-set,		0x16),
423 	XGENE_PMU_EVENT_ATTR(sip-insertions-r-clear,		0x17),
424 	XGENE_PMU_EVENT_ATTR(dip-insertions-r-set,		0x18),
425 	XGENE_PMU_EVENT_ATTR(dip-insertions-r-clear,		0x19),
426 	XGENE_PMU_EVENT_ATTR(dip-insertions-force-r-set,	0x1a),
427 	XGENE_PMU_EVENT_ATTR(egression,				0x1b),
428 	XGENE_PMU_EVENT_ATTR(replacement,			0x1c),
429 	XGENE_PMU_EVENT_ATTR(old-replacement,			0x1d),
430 	XGENE_PMU_EVENT_ATTR(young-replacement,			0x1e),
431 	XGENE_PMU_EVENT_ATTR(r-set-replacement,			0x1f),
432 	XGENE_PMU_EVENT_ATTR(r-clear-replacement,		0x20),
433 	XGENE_PMU_EVENT_ATTR(old-r-replacement,			0x21),
434 	XGENE_PMU_EVENT_ATTR(old-nr-replacement,		0x22),
435 	XGENE_PMU_EVENT_ATTR(young-r-replacement,		0x23),
436 	XGENE_PMU_EVENT_ATTR(young-nr-replacement,		0x24),
437 	XGENE_PMU_EVENT_ATTR(bloomfilter-clearing,		0x25),
438 	XGENE_PMU_EVENT_ATTR(generation-flip,			0x26),
439 	XGENE_PMU_EVENT_ATTR(vcc-droop-detected,		0x27),
440 	NULL,
441 };
442 
443 static struct attribute *iob_fast_pmu_v3_events_attrs[] = {
444 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
445 	XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-all,		0x01),
446 	XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-rd,		0x02),
447 	XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-wr,		0x03),
448 	XGENE_PMU_EVENT_ATTR(pa-all-cp-req,			0x04),
449 	XGENE_PMU_EVENT_ATTR(pa-cp-blk-req,			0x05),
450 	XGENE_PMU_EVENT_ATTR(pa-cp-ptl-req,			0x06),
451 	XGENE_PMU_EVENT_ATTR(pa-cp-rd-req,			0x07),
452 	XGENE_PMU_EVENT_ATTR(pa-cp-wr-req,			0x08),
453 	XGENE_PMU_EVENT_ATTR(ba-all-req,			0x09),
454 	XGENE_PMU_EVENT_ATTR(ba-rd-req,				0x0a),
455 	XGENE_PMU_EVENT_ATTR(ba-wr-req,				0x0b),
456 	XGENE_PMU_EVENT_ATTR(pa-rd-shared-req-issued,		0x10),
457 	XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-req-issued,	0x11),
458 	XGENE_PMU_EVENT_ATTR(pa-wr-invalidate-req-issued-stashable, 0x12),
459 	XGENE_PMU_EVENT_ATTR(pa-wr-invalidate-req-issued-nonstashable, 0x13),
460 	XGENE_PMU_EVENT_ATTR(pa-wr-back-req-issued-stashable,	0x14),
461 	XGENE_PMU_EVENT_ATTR(pa-wr-back-req-issued-nonstashable, 0x15),
462 	XGENE_PMU_EVENT_ATTR(pa-ptl-wr-req,			0x16),
463 	XGENE_PMU_EVENT_ATTR(pa-ptl-rd-req,			0x17),
464 	XGENE_PMU_EVENT_ATTR(pa-wr-back-clean-data,		0x18),
465 	XGENE_PMU_EVENT_ATTR(pa-wr-back-cancelled-on-SS,	0x1b),
466 	XGENE_PMU_EVENT_ATTR(pa-barrier-occurrence,		0x1c),
467 	XGENE_PMU_EVENT_ATTR(pa-barrier-cycles,			0x1d),
468 	XGENE_PMU_EVENT_ATTR(pa-total-cp-snoops,		0x20),
469 	XGENE_PMU_EVENT_ATTR(pa-rd-shared-snoop,		0x21),
470 	XGENE_PMU_EVENT_ATTR(pa-rd-shared-snoop-hit,		0x22),
471 	XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-snoop,		0x23),
472 	XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-snoop-hit,		0x24),
473 	XGENE_PMU_EVENT_ATTR(pa-rd-wr-invalid-snoop,		0x25),
474 	XGENE_PMU_EVENT_ATTR(pa-rd-wr-invalid-snoop-hit,	0x26),
475 	XGENE_PMU_EVENT_ATTR(pa-req-buffer-full,		0x28),
476 	XGENE_PMU_EVENT_ATTR(cswlf-outbound-req-fifo-full,	0x29),
477 	XGENE_PMU_EVENT_ATTR(cswlf-inbound-snoop-fifo-backpressure, 0x2a),
478 	XGENE_PMU_EVENT_ATTR(cswlf-outbound-lack-fifo-full,	0x2b),
479 	XGENE_PMU_EVENT_ATTR(cswlf-inbound-gack-fifo-backpressure, 0x2c),
480 	XGENE_PMU_EVENT_ATTR(cswlf-outbound-data-fifo-full,	0x2d),
481 	XGENE_PMU_EVENT_ATTR(cswlf-inbound-data-fifo-backpressure, 0x2e),
482 	XGENE_PMU_EVENT_ATTR(cswlf-inbound-req-backpressure,	0x2f),
483 	NULL,
484 };
485 
486 static struct attribute *iob_slow_pmu_v3_events_attrs[] = {
487 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
488 	XGENE_PMU_EVENT_ATTR(pa-axi0-rd-req,			0x01),
489 	XGENE_PMU_EVENT_ATTR(pa-axi0-wr-req,			0x02),
490 	XGENE_PMU_EVENT_ATTR(pa-axi1-rd-req,			0x03),
491 	XGENE_PMU_EVENT_ATTR(pa-axi1-wr-req,			0x04),
492 	XGENE_PMU_EVENT_ATTR(ba-all-axi-req,			0x07),
493 	XGENE_PMU_EVENT_ATTR(ba-axi-rd-req,			0x08),
494 	XGENE_PMU_EVENT_ATTR(ba-axi-wr-req,			0x09),
495 	XGENE_PMU_EVENT_ATTR(ba-free-list-empty,		0x10),
496 	NULL,
497 };
498 
499 static struct attribute *mcb_pmu_v3_events_attrs[] = {
500 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
501 	XGENE_PMU_EVENT_ATTR(req-receive,			0x01),
502 	XGENE_PMU_EVENT_ATTR(rd-req-recv,			0x02),
503 	XGENE_PMU_EVENT_ATTR(rd-req-recv-2,			0x03),
504 	XGENE_PMU_EVENT_ATTR(wr-req-recv,			0x04),
505 	XGENE_PMU_EVENT_ATTR(wr-req-recv-2,			0x05),
506 	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-mcu,		0x06),
507 	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-mcu-2,		0x07),
508 	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-spec-mcu,		0x08),
509 	XGENE_PMU_EVENT_ATTR(rd-req-sent-to-spec-mcu-2,		0x09),
510 	XGENE_PMU_EVENT_ATTR(glbl-ack-recv-for-rd-sent-to-spec-mcu, 0x0a),
511 	XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-for-rd-sent-to-spec-mcu, 0x0b),
512 	XGENE_PMU_EVENT_ATTR(glbl-ack-nogo-recv-for-rd-sent-to-spec-mcu, 0x0c),
513 	XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-any-rd-req,	0x0d),
514 	XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-any-rd-req-2,	0x0e),
515 	XGENE_PMU_EVENT_ATTR(wr-req-sent-to-mcu,		0x0f),
516 	XGENE_PMU_EVENT_ATTR(gack-recv,				0x10),
517 	XGENE_PMU_EVENT_ATTR(rd-gack-recv,			0x11),
518 	XGENE_PMU_EVENT_ATTR(wr-gack-recv,			0x12),
519 	XGENE_PMU_EVENT_ATTR(cancel-rd-gack,			0x13),
520 	XGENE_PMU_EVENT_ATTR(cancel-wr-gack,			0x14),
521 	XGENE_PMU_EVENT_ATTR(mcb-csw-req-stall,			0x15),
522 	XGENE_PMU_EVENT_ATTR(mcu-req-intf-blocked,		0x16),
523 	XGENE_PMU_EVENT_ATTR(mcb-mcu-rd-intf-stall,		0x17),
524 	XGENE_PMU_EVENT_ATTR(csw-rd-intf-blocked,		0x18),
525 	XGENE_PMU_EVENT_ATTR(csw-local-ack-intf-blocked,	0x19),
526 	XGENE_PMU_EVENT_ATTR(mcu-req-table-full,		0x1a),
527 	XGENE_PMU_EVENT_ATTR(mcu-stat-table-full,		0x1b),
528 	XGENE_PMU_EVENT_ATTR(mcu-wr-table-full,			0x1c),
529 	XGENE_PMU_EVENT_ATTR(mcu-rdreceipt-resp,		0x1d),
530 	XGENE_PMU_EVENT_ATTR(mcu-wrcomplete-resp,		0x1e),
531 	XGENE_PMU_EVENT_ATTR(mcu-retryack-resp,			0x1f),
532 	XGENE_PMU_EVENT_ATTR(mcu-pcrdgrant-resp,		0x20),
533 	XGENE_PMU_EVENT_ATTR(mcu-req-from-lastload,		0x21),
534 	XGENE_PMU_EVENT_ATTR(mcu-req-from-bypass,		0x22),
535 	XGENE_PMU_EVENT_ATTR(volt-droop-detect,			0x23),
536 	NULL,
537 };
538 
539 static struct attribute *mc_pmu_v3_events_attrs[] = {
540 	XGENE_PMU_EVENT_ATTR(cycle-count,			0x00),
541 	XGENE_PMU_EVENT_ATTR(act-sent,				0x01),
542 	XGENE_PMU_EVENT_ATTR(pre-sent,				0x02),
543 	XGENE_PMU_EVENT_ATTR(rd-sent,				0x03),
544 	XGENE_PMU_EVENT_ATTR(rda-sent,				0x04),
545 	XGENE_PMU_EVENT_ATTR(wr-sent,				0x05),
546 	XGENE_PMU_EVENT_ATTR(wra-sent,				0x06),
547 	XGENE_PMU_EVENT_ATTR(pd-entry-vld,			0x07),
548 	XGENE_PMU_EVENT_ATTR(sref-entry-vld,			0x08),
549 	XGENE_PMU_EVENT_ATTR(prea-sent,				0x09),
550 	XGENE_PMU_EVENT_ATTR(ref-sent,				0x0a),
551 	XGENE_PMU_EVENT_ATTR(rd-rda-sent,			0x0b),
552 	XGENE_PMU_EVENT_ATTR(wr-wra-sent,			0x0c),
553 	XGENE_PMU_EVENT_ATTR(raw-hazard,			0x0d),
554 	XGENE_PMU_EVENT_ATTR(war-hazard,			0x0e),
555 	XGENE_PMU_EVENT_ATTR(waw-hazard,			0x0f),
556 	XGENE_PMU_EVENT_ATTR(rar-hazard,			0x10),
557 	XGENE_PMU_EVENT_ATTR(raw-war-waw-hazard,		0x11),
558 	XGENE_PMU_EVENT_ATTR(hprd-lprd-wr-req-vld,		0x12),
559 	XGENE_PMU_EVENT_ATTR(lprd-req-vld,			0x13),
560 	XGENE_PMU_EVENT_ATTR(hprd-req-vld,			0x14),
561 	XGENE_PMU_EVENT_ATTR(hprd-lprd-req-vld,			0x15),
562 	XGENE_PMU_EVENT_ATTR(wr-req-vld,			0x16),
563 	XGENE_PMU_EVENT_ATTR(partial-wr-req-vld,		0x17),
564 	XGENE_PMU_EVENT_ATTR(rd-retry,				0x18),
565 	XGENE_PMU_EVENT_ATTR(wr-retry,				0x19),
566 	XGENE_PMU_EVENT_ATTR(retry-gnt,				0x1a),
567 	XGENE_PMU_EVENT_ATTR(rank-change,			0x1b),
568 	XGENE_PMU_EVENT_ATTR(dir-change,			0x1c),
569 	XGENE_PMU_EVENT_ATTR(rank-dir-change,			0x1d),
570 	XGENE_PMU_EVENT_ATTR(rank-active,			0x1e),
571 	XGENE_PMU_EVENT_ATTR(rank-idle,				0x1f),
572 	XGENE_PMU_EVENT_ATTR(rank-pd,				0x20),
573 	XGENE_PMU_EVENT_ATTR(rank-sref,				0x21),
574 	XGENE_PMU_EVENT_ATTR(queue-fill-gt-thresh,		0x22),
575 	XGENE_PMU_EVENT_ATTR(queue-rds-gt-thresh,		0x23),
576 	XGENE_PMU_EVENT_ATTR(queue-wrs-gt-thresh,		0x24),
577 	XGENE_PMU_EVENT_ATTR(phy-updt-complt,			0x25),
578 	XGENE_PMU_EVENT_ATTR(tz-fail,				0x26),
579 	XGENE_PMU_EVENT_ATTR(dram-errc,				0x27),
580 	XGENE_PMU_EVENT_ATTR(dram-errd,				0x28),
581 	XGENE_PMU_EVENT_ATTR(rd-enq,				0x29),
582 	XGENE_PMU_EVENT_ATTR(wr-enq,				0x2a),
583 	XGENE_PMU_EVENT_ATTR(tmac-limit-reached,		0x2b),
584 	XGENE_PMU_EVENT_ATTR(tmaw-tracker-full,			0x2c),
585 	NULL,
586 };
587 
588 static const struct attribute_group l3c_pmu_v3_events_attr_group = {
589 	.name = "events",
590 	.attrs = l3c_pmu_v3_events_attrs,
591 };
592 
593 static const struct attribute_group iob_fast_pmu_v3_events_attr_group = {
594 	.name = "events",
595 	.attrs = iob_fast_pmu_v3_events_attrs,
596 };
597 
598 static const struct attribute_group iob_slow_pmu_v3_events_attr_group = {
599 	.name = "events",
600 	.attrs = iob_slow_pmu_v3_events_attrs,
601 };
602 
603 static const struct attribute_group mcb_pmu_v3_events_attr_group = {
604 	.name = "events",
605 	.attrs = mcb_pmu_v3_events_attrs,
606 };
607 
608 static const struct attribute_group mc_pmu_v3_events_attr_group = {
609 	.name = "events",
610 	.attrs = mc_pmu_v3_events_attrs,
611 };
612 
613 /*
614  * sysfs cpumask attributes
615  */
616 static ssize_t xgene_pmu_cpumask_show(struct device *dev,
617 				      struct device_attribute *attr, char *buf)
618 {
619 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(dev_get_drvdata(dev));
620 
621 	return cpumap_print_to_pagebuf(true, buf, &pmu_dev->parent->cpu);
622 }
623 
624 static DEVICE_ATTR(cpumask, S_IRUGO, xgene_pmu_cpumask_show, NULL);
625 
626 static struct attribute *xgene_pmu_cpumask_attrs[] = {
627 	&dev_attr_cpumask.attr,
628 	NULL,
629 };
630 
631 static const struct attribute_group pmu_cpumask_attr_group = {
632 	.attrs = xgene_pmu_cpumask_attrs,
633 };
634 
635 /*
636  * Per PMU device attribute groups of PMU v1 and v2
637  */
638 static const struct attribute_group *l3c_pmu_attr_groups[] = {
639 	&l3c_pmu_format_attr_group,
640 	&pmu_cpumask_attr_group,
641 	&l3c_pmu_events_attr_group,
642 	NULL
643 };
644 
645 static const struct attribute_group *iob_pmu_attr_groups[] = {
646 	&iob_pmu_format_attr_group,
647 	&pmu_cpumask_attr_group,
648 	&iob_pmu_events_attr_group,
649 	NULL
650 };
651 
652 static const struct attribute_group *mcb_pmu_attr_groups[] = {
653 	&mcb_pmu_format_attr_group,
654 	&pmu_cpumask_attr_group,
655 	&mcb_pmu_events_attr_group,
656 	NULL
657 };
658 
659 static const struct attribute_group *mc_pmu_attr_groups[] = {
660 	&mc_pmu_format_attr_group,
661 	&pmu_cpumask_attr_group,
662 	&mc_pmu_events_attr_group,
663 	NULL
664 };
665 
666 /*
667  * Per PMU device attribute groups of PMU v3
668  */
669 static const struct attribute_group *l3c_pmu_v3_attr_groups[] = {
670 	&l3c_pmu_v3_format_attr_group,
671 	&pmu_cpumask_attr_group,
672 	&l3c_pmu_v3_events_attr_group,
673 	NULL
674 };
675 
676 static const struct attribute_group *iob_fast_pmu_v3_attr_groups[] = {
677 	&iob_pmu_v3_format_attr_group,
678 	&pmu_cpumask_attr_group,
679 	&iob_fast_pmu_v3_events_attr_group,
680 	NULL
681 };
682 
683 static const struct attribute_group *iob_slow_pmu_v3_attr_groups[] = {
684 	&iob_slow_pmu_v3_format_attr_group,
685 	&pmu_cpumask_attr_group,
686 	&iob_slow_pmu_v3_events_attr_group,
687 	NULL
688 };
689 
690 static const struct attribute_group *mcb_pmu_v3_attr_groups[] = {
691 	&mcb_pmu_v3_format_attr_group,
692 	&pmu_cpumask_attr_group,
693 	&mcb_pmu_v3_events_attr_group,
694 	NULL
695 };
696 
697 static const struct attribute_group *mc_pmu_v3_attr_groups[] = {
698 	&mc_pmu_v3_format_attr_group,
699 	&pmu_cpumask_attr_group,
700 	&mc_pmu_v3_events_attr_group,
701 	NULL
702 };
703 
704 static int get_next_avail_cntr(struct xgene_pmu_dev *pmu_dev)
705 {
706 	int cntr;
707 
708 	cntr = find_first_zero_bit(pmu_dev->cntr_assign_mask,
709 				pmu_dev->max_counters);
710 	if (cntr == pmu_dev->max_counters)
711 		return -ENOSPC;
712 	set_bit(cntr, pmu_dev->cntr_assign_mask);
713 
714 	return cntr;
715 }
716 
717 static void clear_avail_cntr(struct xgene_pmu_dev *pmu_dev, int cntr)
718 {
719 	clear_bit(cntr, pmu_dev->cntr_assign_mask);
720 }
721 
722 static inline void xgene_pmu_mask_int(struct xgene_pmu *xgene_pmu)
723 {
724 	writel(PCPPMU_INTENMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
725 }
726 
727 static inline void xgene_pmu_v3_mask_int(struct xgene_pmu *xgene_pmu)
728 {
729 	writel(PCPPMU_V3_INTENMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
730 }
731 
732 static inline void xgene_pmu_unmask_int(struct xgene_pmu *xgene_pmu)
733 {
734 	writel(PCPPMU_INTCLRMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
735 }
736 
737 static inline void xgene_pmu_v3_unmask_int(struct xgene_pmu *xgene_pmu)
738 {
739 	writel(PCPPMU_V3_INTCLRMASK,
740 	       xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG);
741 }
742 
743 static inline u64 xgene_pmu_read_counter32(struct xgene_pmu_dev *pmu_dev,
744 					   int idx)
745 {
746 	return readl(pmu_dev->inf->csr + PMU_PMEVCNTR0 + (4 * idx));
747 }
748 
749 static inline u64 xgene_pmu_read_counter64(struct xgene_pmu_dev *pmu_dev,
750 					   int idx)
751 {
752 	u32 lo, hi;
753 
754 	/*
755 	 * v3 has 64-bit counter registers composed by 2 32-bit registers
756 	 * This can be a problem if the counter increases and carries
757 	 * out of bit [31] between 2 reads. The extra reads would help
758 	 * to prevent this issue.
759 	 */
760 	do {
761 		hi = xgene_pmu_read_counter32(pmu_dev, 2 * idx + 1);
762 		lo = xgene_pmu_read_counter32(pmu_dev, 2 * idx);
763 	} while (hi != xgene_pmu_read_counter32(pmu_dev, 2 * idx + 1));
764 
765 	return (((u64)hi << 32) | lo);
766 }
767 
768 static inline void
769 xgene_pmu_write_counter32(struct xgene_pmu_dev *pmu_dev, int idx, u64 val)
770 {
771 	writel(val, pmu_dev->inf->csr + PMU_PMEVCNTR0 + (4 * idx));
772 }
773 
774 static inline void
775 xgene_pmu_write_counter64(struct xgene_pmu_dev *pmu_dev, int idx, u64 val)
776 {
777 	u32 cnt_lo, cnt_hi;
778 
779 	cnt_hi = upper_32_bits(val);
780 	cnt_lo = lower_32_bits(val);
781 
782 	/* v3 has 64-bit counter registers composed by 2 32-bit registers */
783 	xgene_pmu_write_counter32(pmu_dev, 2 * idx, cnt_lo);
784 	xgene_pmu_write_counter32(pmu_dev, 2 * idx + 1, cnt_hi);
785 }
786 
787 static inline void
788 xgene_pmu_write_evttype(struct xgene_pmu_dev *pmu_dev, int idx, u32 val)
789 {
790 	writel(val, pmu_dev->inf->csr + PMU_PMEVTYPER0 + (4 * idx));
791 }
792 
793 static inline void
794 xgene_pmu_write_agentmsk(struct xgene_pmu_dev *pmu_dev, u32 val)
795 {
796 	writel(val, pmu_dev->inf->csr + PMU_PMAMR0);
797 }
798 
799 static inline void
800 xgene_pmu_v3_write_agentmsk(struct xgene_pmu_dev *pmu_dev, u32 val) { }
801 
802 static inline void
803 xgene_pmu_write_agent1msk(struct xgene_pmu_dev *pmu_dev, u32 val)
804 {
805 	writel(val, pmu_dev->inf->csr + PMU_PMAMR1);
806 }
807 
808 static inline void
809 xgene_pmu_v3_write_agent1msk(struct xgene_pmu_dev *pmu_dev, u32 val) { }
810 
811 static inline void
812 xgene_pmu_enable_counter(struct xgene_pmu_dev *pmu_dev, int idx)
813 {
814 	u32 val;
815 
816 	val = readl(pmu_dev->inf->csr + PMU_PMCNTENSET);
817 	val |= 1 << idx;
818 	writel(val, pmu_dev->inf->csr + PMU_PMCNTENSET);
819 }
820 
821 static inline void
822 xgene_pmu_disable_counter(struct xgene_pmu_dev *pmu_dev, int idx)
823 {
824 	u32 val;
825 
826 	val = readl(pmu_dev->inf->csr + PMU_PMCNTENCLR);
827 	val |= 1 << idx;
828 	writel(val, pmu_dev->inf->csr + PMU_PMCNTENCLR);
829 }
830 
831 static inline void
832 xgene_pmu_enable_counter_int(struct xgene_pmu_dev *pmu_dev, int idx)
833 {
834 	u32 val;
835 
836 	val = readl(pmu_dev->inf->csr + PMU_PMINTENSET);
837 	val |= 1 << idx;
838 	writel(val, pmu_dev->inf->csr + PMU_PMINTENSET);
839 }
840 
841 static inline void
842 xgene_pmu_disable_counter_int(struct xgene_pmu_dev *pmu_dev, int idx)
843 {
844 	u32 val;
845 
846 	val = readl(pmu_dev->inf->csr + PMU_PMINTENCLR);
847 	val |= 1 << idx;
848 	writel(val, pmu_dev->inf->csr + PMU_PMINTENCLR);
849 }
850 
851 static inline void xgene_pmu_reset_counters(struct xgene_pmu_dev *pmu_dev)
852 {
853 	u32 val;
854 
855 	val = readl(pmu_dev->inf->csr + PMU_PMCR);
856 	val |= PMU_PMCR_P;
857 	writel(val, pmu_dev->inf->csr + PMU_PMCR);
858 }
859 
860 static inline void xgene_pmu_start_counters(struct xgene_pmu_dev *pmu_dev)
861 {
862 	u32 val;
863 
864 	val = readl(pmu_dev->inf->csr + PMU_PMCR);
865 	val |= PMU_PMCR_E;
866 	writel(val, pmu_dev->inf->csr + PMU_PMCR);
867 }
868 
869 static inline void xgene_pmu_stop_counters(struct xgene_pmu_dev *pmu_dev)
870 {
871 	u32 val;
872 
873 	val = readl(pmu_dev->inf->csr + PMU_PMCR);
874 	val &= ~PMU_PMCR_E;
875 	writel(val, pmu_dev->inf->csr + PMU_PMCR);
876 }
877 
878 static void xgene_perf_pmu_enable(struct pmu *pmu)
879 {
880 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(pmu);
881 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
882 	int enabled = bitmap_weight(pmu_dev->cntr_assign_mask,
883 			pmu_dev->max_counters);
884 
885 	if (!enabled)
886 		return;
887 
888 	xgene_pmu->ops->start_counters(pmu_dev);
889 }
890 
891 static void xgene_perf_pmu_disable(struct pmu *pmu)
892 {
893 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(pmu);
894 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
895 
896 	xgene_pmu->ops->stop_counters(pmu_dev);
897 }
898 
899 static int xgene_perf_event_init(struct perf_event *event)
900 {
901 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
902 	struct hw_perf_event *hw = &event->hw;
903 	struct perf_event *sibling;
904 
905 	/* Test the event attr type check for PMU enumeration */
906 	if (event->attr.type != event->pmu->type)
907 		return -ENOENT;
908 
909 	/*
910 	 * SOC PMU counters are shared across all cores.
911 	 * Therefore, it does not support per-process mode.
912 	 * Also, it does not support event sampling mode.
913 	 */
914 	if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK)
915 		return -EINVAL;
916 
917 	/* SOC counters do not have usr/os/guest/host bits */
918 	if (event->attr.exclude_user || event->attr.exclude_kernel ||
919 	    event->attr.exclude_host || event->attr.exclude_guest)
920 		return -EINVAL;
921 
922 	if (event->cpu < 0)
923 		return -EINVAL;
924 	/*
925 	 * Many perf core operations (eg. events rotation) operate on a
926 	 * single CPU context. This is obvious for CPU PMUs, where one
927 	 * expects the same sets of events being observed on all CPUs,
928 	 * but can lead to issues for off-core PMUs, where each
929 	 * event could be theoretically assigned to a different CPU. To
930 	 * mitigate this, we enforce CPU assignment to one, selected
931 	 * processor (the one described in the "cpumask" attribute).
932 	 */
933 	event->cpu = cpumask_first(&pmu_dev->parent->cpu);
934 
935 	hw->config = event->attr.config;
936 	/*
937 	 * Each bit of the config1 field represents an agent from which the
938 	 * request of the event come. The event is counted only if it's caused
939 	 * by a request of an agent has the bit cleared.
940 	 * By default, the event is counted for all agents.
941 	 */
942 	hw->config_base = event->attr.config1;
943 
944 	/*
945 	 * We must NOT create groups containing mixed PMUs, although software
946 	 * events are acceptable
947 	 */
948 	if (event->group_leader->pmu != event->pmu &&
949 			!is_software_event(event->group_leader))
950 		return -EINVAL;
951 
952 	for_each_sibling_event(sibling, event->group_leader) {
953 		if (sibling->pmu != event->pmu &&
954 				!is_software_event(sibling))
955 			return -EINVAL;
956 	}
957 
958 	return 0;
959 }
960 
961 static void xgene_perf_enable_event(struct perf_event *event)
962 {
963 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
964 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
965 
966 	xgene_pmu->ops->write_evttype(pmu_dev, GET_CNTR(event),
967 				      GET_EVENTID(event));
968 	xgene_pmu->ops->write_agentmsk(pmu_dev, ~((u32)GET_AGENTID(event)));
969 	if (pmu_dev->inf->type == PMU_TYPE_IOB)
970 		xgene_pmu->ops->write_agent1msk(pmu_dev,
971 						~((u32)GET_AGENT1ID(event)));
972 
973 	xgene_pmu->ops->enable_counter(pmu_dev, GET_CNTR(event));
974 	xgene_pmu->ops->enable_counter_int(pmu_dev, GET_CNTR(event));
975 }
976 
977 static void xgene_perf_disable_event(struct perf_event *event)
978 {
979 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
980 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
981 
982 	xgene_pmu->ops->disable_counter(pmu_dev, GET_CNTR(event));
983 	xgene_pmu->ops->disable_counter_int(pmu_dev, GET_CNTR(event));
984 }
985 
986 static void xgene_perf_event_set_period(struct perf_event *event)
987 {
988 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
989 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
990 	struct hw_perf_event *hw = &event->hw;
991 	/*
992 	 * For 32 bit counter, it has a period of 2^32. To account for the
993 	 * possibility of extreme interrupt latency we program for a period of
994 	 * half that. Hopefully, we can handle the interrupt before another 2^31
995 	 * events occur and the counter overtakes its previous value.
996 	 * For 64 bit counter, we don't expect it overflow.
997 	 */
998 	u64 val = 1ULL << 31;
999 
1000 	local64_set(&hw->prev_count, val);
1001 	xgene_pmu->ops->write_counter(pmu_dev, hw->idx, val);
1002 }
1003 
1004 static void xgene_perf_event_update(struct perf_event *event)
1005 {
1006 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
1007 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
1008 	struct hw_perf_event *hw = &event->hw;
1009 	u64 delta, prev_raw_count, new_raw_count;
1010 
1011 again:
1012 	prev_raw_count = local64_read(&hw->prev_count);
1013 	new_raw_count = xgene_pmu->ops->read_counter(pmu_dev, GET_CNTR(event));
1014 
1015 	if (local64_cmpxchg(&hw->prev_count, prev_raw_count,
1016 			    new_raw_count) != prev_raw_count)
1017 		goto again;
1018 
1019 	delta = (new_raw_count - prev_raw_count) & pmu_dev->max_period;
1020 
1021 	local64_add(delta, &event->count);
1022 }
1023 
1024 static void xgene_perf_read(struct perf_event *event)
1025 {
1026 	xgene_perf_event_update(event);
1027 }
1028 
1029 static void xgene_perf_start(struct perf_event *event, int flags)
1030 {
1031 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
1032 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
1033 	struct hw_perf_event *hw = &event->hw;
1034 
1035 	if (WARN_ON_ONCE(!(hw->state & PERF_HES_STOPPED)))
1036 		return;
1037 
1038 	WARN_ON_ONCE(!(hw->state & PERF_HES_UPTODATE));
1039 	hw->state = 0;
1040 
1041 	xgene_perf_event_set_period(event);
1042 
1043 	if (flags & PERF_EF_RELOAD) {
1044 		u64 prev_raw_count =  local64_read(&hw->prev_count);
1045 
1046 		xgene_pmu->ops->write_counter(pmu_dev, GET_CNTR(event),
1047 					      prev_raw_count);
1048 	}
1049 
1050 	xgene_perf_enable_event(event);
1051 	perf_event_update_userpage(event);
1052 }
1053 
1054 static void xgene_perf_stop(struct perf_event *event, int flags)
1055 {
1056 	struct hw_perf_event *hw = &event->hw;
1057 	u64 config;
1058 
1059 	if (hw->state & PERF_HES_UPTODATE)
1060 		return;
1061 
1062 	xgene_perf_disable_event(event);
1063 	WARN_ON_ONCE(hw->state & PERF_HES_STOPPED);
1064 	hw->state |= PERF_HES_STOPPED;
1065 
1066 	if (hw->state & PERF_HES_UPTODATE)
1067 		return;
1068 
1069 	config = hw->config;
1070 	xgene_perf_read(event);
1071 	hw->state |= PERF_HES_UPTODATE;
1072 }
1073 
1074 static int xgene_perf_add(struct perf_event *event, int flags)
1075 {
1076 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
1077 	struct hw_perf_event *hw = &event->hw;
1078 
1079 	hw->state = PERF_HES_UPTODATE | PERF_HES_STOPPED;
1080 
1081 	/* Allocate an event counter */
1082 	hw->idx = get_next_avail_cntr(pmu_dev);
1083 	if (hw->idx < 0)
1084 		return -EAGAIN;
1085 
1086 	/* Update counter event pointer for Interrupt handler */
1087 	pmu_dev->pmu_counter_event[hw->idx] = event;
1088 
1089 	if (flags & PERF_EF_START)
1090 		xgene_perf_start(event, PERF_EF_RELOAD);
1091 
1092 	return 0;
1093 }
1094 
1095 static void xgene_perf_del(struct perf_event *event, int flags)
1096 {
1097 	struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu);
1098 	struct hw_perf_event *hw = &event->hw;
1099 
1100 	xgene_perf_stop(event, PERF_EF_UPDATE);
1101 
1102 	/* clear the assigned counter */
1103 	clear_avail_cntr(pmu_dev, GET_CNTR(event));
1104 
1105 	perf_event_update_userpage(event);
1106 	pmu_dev->pmu_counter_event[hw->idx] = NULL;
1107 }
1108 
1109 static int xgene_init_perf(struct xgene_pmu_dev *pmu_dev, char *name)
1110 {
1111 	struct xgene_pmu *xgene_pmu;
1112 
1113 	if (pmu_dev->parent->version == PCP_PMU_V3)
1114 		pmu_dev->max_period = PMU_V3_CNT_MAX_PERIOD;
1115 	else
1116 		pmu_dev->max_period = PMU_CNT_MAX_PERIOD;
1117 	/* First version PMU supports only single event counter */
1118 	xgene_pmu = pmu_dev->parent;
1119 	if (xgene_pmu->version == PCP_PMU_V1)
1120 		pmu_dev->max_counters = 1;
1121 	else
1122 		pmu_dev->max_counters = PMU_MAX_COUNTERS;
1123 
1124 	/* Perf driver registration */
1125 	pmu_dev->pmu = (struct pmu) {
1126 		.attr_groups	= pmu_dev->attr_groups,
1127 		.task_ctx_nr	= perf_invalid_context,
1128 		.pmu_enable	= xgene_perf_pmu_enable,
1129 		.pmu_disable	= xgene_perf_pmu_disable,
1130 		.event_init	= xgene_perf_event_init,
1131 		.add		= xgene_perf_add,
1132 		.del		= xgene_perf_del,
1133 		.start		= xgene_perf_start,
1134 		.stop		= xgene_perf_stop,
1135 		.read		= xgene_perf_read,
1136 	};
1137 
1138 	/* Hardware counter init */
1139 	xgene_pmu->ops->stop_counters(pmu_dev);
1140 	xgene_pmu->ops->reset_counters(pmu_dev);
1141 
1142 	return perf_pmu_register(&pmu_dev->pmu, name, -1);
1143 }
1144 
1145 static int
1146 xgene_pmu_dev_add(struct xgene_pmu *xgene_pmu, struct xgene_pmu_dev_ctx *ctx)
1147 {
1148 	struct device *dev = xgene_pmu->dev;
1149 	struct xgene_pmu_dev *pmu;
1150 
1151 	pmu = devm_kzalloc(dev, sizeof(*pmu), GFP_KERNEL);
1152 	if (!pmu)
1153 		return -ENOMEM;
1154 	pmu->parent = xgene_pmu;
1155 	pmu->inf = &ctx->inf;
1156 	ctx->pmu_dev = pmu;
1157 
1158 	switch (pmu->inf->type) {
1159 	case PMU_TYPE_L3C:
1160 		if (!(xgene_pmu->l3c_active_mask & pmu->inf->enable_mask))
1161 			return -ENODEV;
1162 		if (xgene_pmu->version == PCP_PMU_V3)
1163 			pmu->attr_groups = l3c_pmu_v3_attr_groups;
1164 		else
1165 			pmu->attr_groups = l3c_pmu_attr_groups;
1166 		break;
1167 	case PMU_TYPE_IOB:
1168 		if (xgene_pmu->version == PCP_PMU_V3)
1169 			pmu->attr_groups = iob_fast_pmu_v3_attr_groups;
1170 		else
1171 			pmu->attr_groups = iob_pmu_attr_groups;
1172 		break;
1173 	case PMU_TYPE_IOB_SLOW:
1174 		if (xgene_pmu->version == PCP_PMU_V3)
1175 			pmu->attr_groups = iob_slow_pmu_v3_attr_groups;
1176 		break;
1177 	case PMU_TYPE_MCB:
1178 		if (!(xgene_pmu->mcb_active_mask & pmu->inf->enable_mask))
1179 			return -ENODEV;
1180 		if (xgene_pmu->version == PCP_PMU_V3)
1181 			pmu->attr_groups = mcb_pmu_v3_attr_groups;
1182 		else
1183 			pmu->attr_groups = mcb_pmu_attr_groups;
1184 		break;
1185 	case PMU_TYPE_MC:
1186 		if (!(xgene_pmu->mc_active_mask & pmu->inf->enable_mask))
1187 			return -ENODEV;
1188 		if (xgene_pmu->version == PCP_PMU_V3)
1189 			pmu->attr_groups = mc_pmu_v3_attr_groups;
1190 		else
1191 			pmu->attr_groups = mc_pmu_attr_groups;
1192 		break;
1193 	default:
1194 		return -EINVAL;
1195 	}
1196 
1197 	if (xgene_init_perf(pmu, ctx->name)) {
1198 		dev_err(dev, "%s PMU: Failed to init perf driver\n", ctx->name);
1199 		return -ENODEV;
1200 	}
1201 
1202 	dev_info(dev, "%s PMU registered\n", ctx->name);
1203 
1204 	return 0;
1205 }
1206 
1207 static void _xgene_pmu_isr(int irq, struct xgene_pmu_dev *pmu_dev)
1208 {
1209 	struct xgene_pmu *xgene_pmu = pmu_dev->parent;
1210 	void __iomem *csr = pmu_dev->inf->csr;
1211 	u32 pmovsr;
1212 	int idx;
1213 
1214 	xgene_pmu->ops->stop_counters(pmu_dev);
1215 
1216 	if (xgene_pmu->version == PCP_PMU_V3)
1217 		pmovsr = readl(csr + PMU_PMOVSSET) & PMU_OVERFLOW_MASK;
1218 	else
1219 		pmovsr = readl(csr + PMU_PMOVSR) & PMU_OVERFLOW_MASK;
1220 
1221 	if (!pmovsr)
1222 		goto out;
1223 
1224 	/* Clear interrupt flag */
1225 	if (xgene_pmu->version == PCP_PMU_V1)
1226 		writel(0x0, csr + PMU_PMOVSR);
1227 	else if (xgene_pmu->version == PCP_PMU_V2)
1228 		writel(pmovsr, csr + PMU_PMOVSR);
1229 	else
1230 		writel(pmovsr, csr + PMU_PMOVSCLR);
1231 
1232 	for (idx = 0; idx < PMU_MAX_COUNTERS; idx++) {
1233 		struct perf_event *event = pmu_dev->pmu_counter_event[idx];
1234 		int overflowed = pmovsr & BIT(idx);
1235 
1236 		/* Ignore if we don't have an event. */
1237 		if (!event || !overflowed)
1238 			continue;
1239 		xgene_perf_event_update(event);
1240 		xgene_perf_event_set_period(event);
1241 	}
1242 
1243 out:
1244 	xgene_pmu->ops->start_counters(pmu_dev);
1245 }
1246 
1247 static irqreturn_t xgene_pmu_isr(int irq, void *dev_id)
1248 {
1249 	u32 intr_mcu, intr_mcb, intr_l3c, intr_iob;
1250 	struct xgene_pmu_dev_ctx *ctx;
1251 	struct xgene_pmu *xgene_pmu = dev_id;
1252 	unsigned long flags;
1253 	u32 val;
1254 
1255 	raw_spin_lock_irqsave(&xgene_pmu->lock, flags);
1256 
1257 	/* Get Interrupt PMU source */
1258 	val = readl(xgene_pmu->pcppmu_csr + PCPPMU_INTSTATUS_REG);
1259 	if (xgene_pmu->version == PCP_PMU_V3) {
1260 		intr_mcu = PCPPMU_V3_INT_MCU;
1261 		intr_mcb = PCPPMU_V3_INT_MCB;
1262 		intr_l3c = PCPPMU_V3_INT_L3C;
1263 		intr_iob = PCPPMU_V3_INT_IOB;
1264 	} else {
1265 		intr_mcu = PCPPMU_INT_MCU;
1266 		intr_mcb = PCPPMU_INT_MCB;
1267 		intr_l3c = PCPPMU_INT_L3C;
1268 		intr_iob = PCPPMU_INT_IOB;
1269 	}
1270 	if (val & intr_mcu) {
1271 		list_for_each_entry(ctx, &xgene_pmu->mcpmus, next) {
1272 			_xgene_pmu_isr(irq, ctx->pmu_dev);
1273 		}
1274 	}
1275 	if (val & intr_mcb) {
1276 		list_for_each_entry(ctx, &xgene_pmu->mcbpmus, next) {
1277 			_xgene_pmu_isr(irq, ctx->pmu_dev);
1278 		}
1279 	}
1280 	if (val & intr_l3c) {
1281 		list_for_each_entry(ctx, &xgene_pmu->l3cpmus, next) {
1282 			_xgene_pmu_isr(irq, ctx->pmu_dev);
1283 		}
1284 	}
1285 	if (val & intr_iob) {
1286 		list_for_each_entry(ctx, &xgene_pmu->iobpmus, next) {
1287 			_xgene_pmu_isr(irq, ctx->pmu_dev);
1288 		}
1289 	}
1290 
1291 	raw_spin_unlock_irqrestore(&xgene_pmu->lock, flags);
1292 
1293 	return IRQ_HANDLED;
1294 }
1295 
1296 static int acpi_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
1297 					     struct platform_device *pdev)
1298 {
1299 	void __iomem *csw_csr, *mcba_csr, *mcbb_csr;
1300 	struct resource *res;
1301 	unsigned int reg;
1302 
1303 	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1304 	csw_csr = devm_ioremap_resource(&pdev->dev, res);
1305 	if (IS_ERR(csw_csr)) {
1306 		dev_err(&pdev->dev, "ioremap failed for CSW CSR resource\n");
1307 		return PTR_ERR(csw_csr);
1308 	}
1309 
1310 	res = platform_get_resource(pdev, IORESOURCE_MEM, 2);
1311 	mcba_csr = devm_ioremap_resource(&pdev->dev, res);
1312 	if (IS_ERR(mcba_csr)) {
1313 		dev_err(&pdev->dev, "ioremap failed for MCBA CSR resource\n");
1314 		return PTR_ERR(mcba_csr);
1315 	}
1316 
1317 	res = platform_get_resource(pdev, IORESOURCE_MEM, 3);
1318 	mcbb_csr = devm_ioremap_resource(&pdev->dev, res);
1319 	if (IS_ERR(mcbb_csr)) {
1320 		dev_err(&pdev->dev, "ioremap failed for MCBB CSR resource\n");
1321 		return PTR_ERR(mcbb_csr);
1322 	}
1323 
1324 	xgene_pmu->l3c_active_mask = 0x1;
1325 
1326 	reg = readl(csw_csr + CSW_CSWCR);
1327 	if (reg & CSW_CSWCR_DUALMCB_MASK) {
1328 		/* Dual MCB active */
1329 		xgene_pmu->mcb_active_mask = 0x3;
1330 		/* Probe all active MC(s) */
1331 		reg = readl(mcbb_csr + CSW_CSWCR);
1332 		xgene_pmu->mc_active_mask =
1333 			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0xF : 0x5;
1334 	} else {
1335 		/* Single MCB active */
1336 		xgene_pmu->mcb_active_mask = 0x1;
1337 		/* Probe all active MC(s) */
1338 		reg = readl(mcba_csr + CSW_CSWCR);
1339 		xgene_pmu->mc_active_mask =
1340 			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0x3 : 0x1;
1341 	}
1342 
1343 	return 0;
1344 }
1345 
1346 static int acpi_pmu_v3_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
1347 						struct platform_device *pdev)
1348 {
1349 	void __iomem *csw_csr;
1350 	struct resource *res;
1351 	unsigned int reg;
1352 	u32 mcb0routing;
1353 	u32 mcb1routing;
1354 
1355 	res = platform_get_resource(pdev, IORESOURCE_MEM, 1);
1356 	csw_csr = devm_ioremap_resource(&pdev->dev, res);
1357 	if (IS_ERR(csw_csr)) {
1358 		dev_err(&pdev->dev, "ioremap failed for CSW CSR resource\n");
1359 		return PTR_ERR(csw_csr);
1360 	}
1361 
1362 	reg = readl(csw_csr + CSW_CSWCR);
1363 	mcb0routing = CSW_CSWCR_MCB0_ROUTING(reg);
1364 	mcb1routing = CSW_CSWCR_MCB1_ROUTING(reg);
1365 	if (reg & CSW_CSWCR_DUALMCB_MASK) {
1366 		/* Dual MCB active */
1367 		xgene_pmu->mcb_active_mask = 0x3;
1368 		/* Probe all active L3C(s), maximum is 8 */
1369 		xgene_pmu->l3c_active_mask = 0xFF;
1370 		/* Probe all active MC(s), maximum is 8 */
1371 		if ((mcb0routing == 0x2) && (mcb1routing == 0x2))
1372 			xgene_pmu->mc_active_mask = 0xFF;
1373 		else if ((mcb0routing == 0x1) && (mcb1routing == 0x1))
1374 			xgene_pmu->mc_active_mask =  0x33;
1375 		else
1376 			xgene_pmu->mc_active_mask =  0x11;
1377 	} else {
1378 		/* Single MCB active */
1379 		xgene_pmu->mcb_active_mask = 0x1;
1380 		/* Probe all active L3C(s), maximum is 4 */
1381 		xgene_pmu->l3c_active_mask = 0x0F;
1382 		/* Probe all active MC(s), maximum is 4 */
1383 		if (mcb0routing == 0x2)
1384 			xgene_pmu->mc_active_mask = 0x0F;
1385 		else if (mcb0routing == 0x1)
1386 			xgene_pmu->mc_active_mask =  0x03;
1387 		else
1388 			xgene_pmu->mc_active_mask =  0x01;
1389 	}
1390 
1391 	return 0;
1392 }
1393 
1394 static int fdt_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
1395 					    struct platform_device *pdev)
1396 {
1397 	struct regmap *csw_map, *mcba_map, *mcbb_map;
1398 	struct device_node *np = pdev->dev.of_node;
1399 	unsigned int reg;
1400 
1401 	csw_map = syscon_regmap_lookup_by_phandle(np, "regmap-csw");
1402 	if (IS_ERR(csw_map)) {
1403 		dev_err(&pdev->dev, "unable to get syscon regmap csw\n");
1404 		return PTR_ERR(csw_map);
1405 	}
1406 
1407 	mcba_map = syscon_regmap_lookup_by_phandle(np, "regmap-mcba");
1408 	if (IS_ERR(mcba_map)) {
1409 		dev_err(&pdev->dev, "unable to get syscon regmap mcba\n");
1410 		return PTR_ERR(mcba_map);
1411 	}
1412 
1413 	mcbb_map = syscon_regmap_lookup_by_phandle(np, "regmap-mcbb");
1414 	if (IS_ERR(mcbb_map)) {
1415 		dev_err(&pdev->dev, "unable to get syscon regmap mcbb\n");
1416 		return PTR_ERR(mcbb_map);
1417 	}
1418 
1419 	xgene_pmu->l3c_active_mask = 0x1;
1420 	if (regmap_read(csw_map, CSW_CSWCR, &reg))
1421 		return -EINVAL;
1422 
1423 	if (reg & CSW_CSWCR_DUALMCB_MASK) {
1424 		/* Dual MCB active */
1425 		xgene_pmu->mcb_active_mask = 0x3;
1426 		/* Probe all active MC(s) */
1427 		if (regmap_read(mcbb_map, MCBADDRMR, &reg))
1428 			return 0;
1429 		xgene_pmu->mc_active_mask =
1430 			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0xF : 0x5;
1431 	} else {
1432 		/* Single MCB active */
1433 		xgene_pmu->mcb_active_mask = 0x1;
1434 		/* Probe all active MC(s) */
1435 		if (regmap_read(mcba_map, MCBADDRMR, &reg))
1436 			return 0;
1437 		xgene_pmu->mc_active_mask =
1438 			(reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0x3 : 0x1;
1439 	}
1440 
1441 	return 0;
1442 }
1443 
1444 static int xgene_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu,
1445 					      struct platform_device *pdev)
1446 {
1447 	if (has_acpi_companion(&pdev->dev)) {
1448 		if (xgene_pmu->version == PCP_PMU_V3)
1449 			return acpi_pmu_v3_probe_active_mcb_mcu_l3c(xgene_pmu,
1450 								    pdev);
1451 		else
1452 			return acpi_pmu_probe_active_mcb_mcu_l3c(xgene_pmu,
1453 								 pdev);
1454 	}
1455 	return fdt_pmu_probe_active_mcb_mcu_l3c(xgene_pmu, pdev);
1456 }
1457 
1458 static char *xgene_pmu_dev_name(struct device *dev, u32 type, int id)
1459 {
1460 	switch (type) {
1461 	case PMU_TYPE_L3C:
1462 		return devm_kasprintf(dev, GFP_KERNEL, "l3c%d", id);
1463 	case PMU_TYPE_IOB:
1464 		return devm_kasprintf(dev, GFP_KERNEL, "iob%d", id);
1465 	case PMU_TYPE_IOB_SLOW:
1466 		return devm_kasprintf(dev, GFP_KERNEL, "iob_slow%d", id);
1467 	case PMU_TYPE_MCB:
1468 		return devm_kasprintf(dev, GFP_KERNEL, "mcb%d", id);
1469 	case PMU_TYPE_MC:
1470 		return devm_kasprintf(dev, GFP_KERNEL, "mc%d", id);
1471 	default:
1472 		return devm_kasprintf(dev, GFP_KERNEL, "unknown");
1473 	}
1474 }
1475 
1476 #if defined(CONFIG_ACPI)
1477 static int acpi_pmu_dev_add_resource(struct acpi_resource *ares, void *data)
1478 {
1479 	struct resource *res = data;
1480 
1481 	if (ares->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32)
1482 		acpi_dev_resource_memory(ares, res);
1483 
1484 	/* Always tell the ACPI core to skip this resource */
1485 	return 1;
1486 }
1487 
1488 static struct
1489 xgene_pmu_dev_ctx *acpi_get_pmu_hw_inf(struct xgene_pmu *xgene_pmu,
1490 				       struct acpi_device *adev, u32 type)
1491 {
1492 	struct device *dev = xgene_pmu->dev;
1493 	struct list_head resource_list;
1494 	struct xgene_pmu_dev_ctx *ctx;
1495 	const union acpi_object *obj;
1496 	struct hw_pmu_info *inf;
1497 	void __iomem *dev_csr;
1498 	struct resource res;
1499 	int enable_bit;
1500 	int rc;
1501 
1502 	ctx = devm_kzalloc(dev, sizeof(*ctx), GFP_KERNEL);
1503 	if (!ctx)
1504 		return NULL;
1505 
1506 	INIT_LIST_HEAD(&resource_list);
1507 	rc = acpi_dev_get_resources(adev, &resource_list,
1508 				    acpi_pmu_dev_add_resource, &res);
1509 	acpi_dev_free_resource_list(&resource_list);
1510 	if (rc < 0) {
1511 		dev_err(dev, "PMU type %d: No resource address found\n", type);
1512 		return NULL;
1513 	}
1514 
1515 	dev_csr = devm_ioremap_resource(dev, &res);
1516 	if (IS_ERR(dev_csr)) {
1517 		dev_err(dev, "PMU type %d: Fail to map resource\n", type);
1518 		return NULL;
1519 	}
1520 
1521 	/* A PMU device node without enable-bit-index is always enabled */
1522 	rc = acpi_dev_get_property(adev, "enable-bit-index",
1523 				   ACPI_TYPE_INTEGER, &obj);
1524 	if (rc < 0)
1525 		enable_bit = 0;
1526 	else
1527 		enable_bit = (int) obj->integer.value;
1528 
1529 	ctx->name = xgene_pmu_dev_name(dev, type, enable_bit);
1530 	if (!ctx->name) {
1531 		dev_err(dev, "PMU type %d: Fail to get device name\n", type);
1532 		return NULL;
1533 	}
1534 	inf = &ctx->inf;
1535 	inf->type = type;
1536 	inf->csr = dev_csr;
1537 	inf->enable_mask = 1 << enable_bit;
1538 
1539 	return ctx;
1540 }
1541 
1542 static const struct acpi_device_id xgene_pmu_acpi_type_match[] = {
1543 	{"APMC0D5D", PMU_TYPE_L3C},
1544 	{"APMC0D5E", PMU_TYPE_IOB},
1545 	{"APMC0D5F", PMU_TYPE_MCB},
1546 	{"APMC0D60", PMU_TYPE_MC},
1547 	{"APMC0D84", PMU_TYPE_L3C},
1548 	{"APMC0D85", PMU_TYPE_IOB},
1549 	{"APMC0D86", PMU_TYPE_IOB_SLOW},
1550 	{"APMC0D87", PMU_TYPE_MCB},
1551 	{"APMC0D88", PMU_TYPE_MC},
1552 	{},
1553 };
1554 
1555 static const struct acpi_device_id *xgene_pmu_acpi_match_type(
1556 					const struct acpi_device_id *ids,
1557 					struct acpi_device *adev)
1558 {
1559 	const struct acpi_device_id *match_id = NULL;
1560 	const struct acpi_device_id *id;
1561 
1562 	for (id = ids; id->id[0] || id->cls; id++) {
1563 		if (!acpi_match_device_ids(adev, id))
1564 			match_id = id;
1565 		else if (match_id)
1566 			break;
1567 	}
1568 
1569 	return match_id;
1570 }
1571 
1572 static acpi_status acpi_pmu_dev_add(acpi_handle handle, u32 level,
1573 				    void *data, void **return_value)
1574 {
1575 	const struct acpi_device_id *acpi_id;
1576 	struct xgene_pmu *xgene_pmu = data;
1577 	struct xgene_pmu_dev_ctx *ctx;
1578 	struct acpi_device *adev;
1579 
1580 	if (acpi_bus_get_device(handle, &adev))
1581 		return AE_OK;
1582 	if (acpi_bus_get_status(adev) || !adev->status.present)
1583 		return AE_OK;
1584 
1585 	acpi_id = xgene_pmu_acpi_match_type(xgene_pmu_acpi_type_match, adev);
1586 	if (!acpi_id)
1587 		return AE_OK;
1588 
1589 	ctx = acpi_get_pmu_hw_inf(xgene_pmu, adev, (u32)acpi_id->driver_data);
1590 	if (!ctx)
1591 		return AE_OK;
1592 
1593 	if (xgene_pmu_dev_add(xgene_pmu, ctx)) {
1594 		/* Can't add the PMU device, skip it */
1595 		devm_kfree(xgene_pmu->dev, ctx);
1596 		return AE_OK;
1597 	}
1598 
1599 	switch (ctx->inf.type) {
1600 	case PMU_TYPE_L3C:
1601 		list_add(&ctx->next, &xgene_pmu->l3cpmus);
1602 		break;
1603 	case PMU_TYPE_IOB:
1604 		list_add(&ctx->next, &xgene_pmu->iobpmus);
1605 		break;
1606 	case PMU_TYPE_IOB_SLOW:
1607 		list_add(&ctx->next, &xgene_pmu->iobpmus);
1608 		break;
1609 	case PMU_TYPE_MCB:
1610 		list_add(&ctx->next, &xgene_pmu->mcbpmus);
1611 		break;
1612 	case PMU_TYPE_MC:
1613 		list_add(&ctx->next, &xgene_pmu->mcpmus);
1614 		break;
1615 	}
1616 	return AE_OK;
1617 }
1618 
1619 static int acpi_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
1620 				  struct platform_device *pdev)
1621 {
1622 	struct device *dev = xgene_pmu->dev;
1623 	acpi_handle handle;
1624 	acpi_status status;
1625 
1626 	handle = ACPI_HANDLE(dev);
1627 	if (!handle)
1628 		return -EINVAL;
1629 
1630 	status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle, 1,
1631 				     acpi_pmu_dev_add, NULL, xgene_pmu, NULL);
1632 	if (ACPI_FAILURE(status)) {
1633 		dev_err(dev, "failed to probe PMU devices\n");
1634 		return -ENODEV;
1635 	}
1636 
1637 	return 0;
1638 }
1639 #else
1640 static int acpi_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
1641 				  struct platform_device *pdev)
1642 {
1643 	return 0;
1644 }
1645 #endif
1646 
1647 static struct
1648 xgene_pmu_dev_ctx *fdt_get_pmu_hw_inf(struct xgene_pmu *xgene_pmu,
1649 				      struct device_node *np, u32 type)
1650 {
1651 	struct device *dev = xgene_pmu->dev;
1652 	struct xgene_pmu_dev_ctx *ctx;
1653 	struct hw_pmu_info *inf;
1654 	void __iomem *dev_csr;
1655 	struct resource res;
1656 	int enable_bit;
1657 
1658 	ctx = devm_kzalloc(dev, sizeof(*ctx), GFP_KERNEL);
1659 	if (!ctx)
1660 		return NULL;
1661 
1662 	if (of_address_to_resource(np, 0, &res) < 0) {
1663 		dev_err(dev, "PMU type %d: No resource address found\n", type);
1664 		return NULL;
1665 	}
1666 
1667 	dev_csr = devm_ioremap_resource(dev, &res);
1668 	if (IS_ERR(dev_csr)) {
1669 		dev_err(dev, "PMU type %d: Fail to map resource\n", type);
1670 		return NULL;
1671 	}
1672 
1673 	/* A PMU device node without enable-bit-index is always enabled */
1674 	if (of_property_read_u32(np, "enable-bit-index", &enable_bit))
1675 		enable_bit = 0;
1676 
1677 	ctx->name = xgene_pmu_dev_name(dev, type, enable_bit);
1678 	if (!ctx->name) {
1679 		dev_err(dev, "PMU type %d: Fail to get device name\n", type);
1680 		return NULL;
1681 	}
1682 
1683 	inf = &ctx->inf;
1684 	inf->type = type;
1685 	inf->csr = dev_csr;
1686 	inf->enable_mask = 1 << enable_bit;
1687 
1688 	return ctx;
1689 }
1690 
1691 static int fdt_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
1692 				 struct platform_device *pdev)
1693 {
1694 	struct xgene_pmu_dev_ctx *ctx;
1695 	struct device_node *np;
1696 
1697 	for_each_child_of_node(pdev->dev.of_node, np) {
1698 		if (!of_device_is_available(np))
1699 			continue;
1700 
1701 		if (of_device_is_compatible(np, "apm,xgene-pmu-l3c"))
1702 			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_L3C);
1703 		else if (of_device_is_compatible(np, "apm,xgene-pmu-iob"))
1704 			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_IOB);
1705 		else if (of_device_is_compatible(np, "apm,xgene-pmu-mcb"))
1706 			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_MCB);
1707 		else if (of_device_is_compatible(np, "apm,xgene-pmu-mc"))
1708 			ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_MC);
1709 		else
1710 			ctx = NULL;
1711 
1712 		if (!ctx)
1713 			continue;
1714 
1715 		if (xgene_pmu_dev_add(xgene_pmu, ctx)) {
1716 			/* Can't add the PMU device, skip it */
1717 			devm_kfree(xgene_pmu->dev, ctx);
1718 			continue;
1719 		}
1720 
1721 		switch (ctx->inf.type) {
1722 		case PMU_TYPE_L3C:
1723 			list_add(&ctx->next, &xgene_pmu->l3cpmus);
1724 			break;
1725 		case PMU_TYPE_IOB:
1726 			list_add(&ctx->next, &xgene_pmu->iobpmus);
1727 			break;
1728 		case PMU_TYPE_IOB_SLOW:
1729 			list_add(&ctx->next, &xgene_pmu->iobpmus);
1730 			break;
1731 		case PMU_TYPE_MCB:
1732 			list_add(&ctx->next, &xgene_pmu->mcbpmus);
1733 			break;
1734 		case PMU_TYPE_MC:
1735 			list_add(&ctx->next, &xgene_pmu->mcpmus);
1736 			break;
1737 		}
1738 	}
1739 
1740 	return 0;
1741 }
1742 
1743 static int xgene_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu,
1744 				   struct platform_device *pdev)
1745 {
1746 	if (has_acpi_companion(&pdev->dev))
1747 		return acpi_pmu_probe_pmu_dev(xgene_pmu, pdev);
1748 	return fdt_pmu_probe_pmu_dev(xgene_pmu, pdev);
1749 }
1750 
1751 static const struct xgene_pmu_data xgene_pmu_data = {
1752 	.id   = PCP_PMU_V1,
1753 };
1754 
1755 static const struct xgene_pmu_data xgene_pmu_v2_data = {
1756 	.id   = PCP_PMU_V2,
1757 };
1758 
1759 static const struct xgene_pmu_ops xgene_pmu_ops = {
1760 	.mask_int = xgene_pmu_mask_int,
1761 	.unmask_int = xgene_pmu_unmask_int,
1762 	.read_counter = xgene_pmu_read_counter32,
1763 	.write_counter = xgene_pmu_write_counter32,
1764 	.write_evttype = xgene_pmu_write_evttype,
1765 	.write_agentmsk = xgene_pmu_write_agentmsk,
1766 	.write_agent1msk = xgene_pmu_write_agent1msk,
1767 	.enable_counter = xgene_pmu_enable_counter,
1768 	.disable_counter = xgene_pmu_disable_counter,
1769 	.enable_counter_int = xgene_pmu_enable_counter_int,
1770 	.disable_counter_int = xgene_pmu_disable_counter_int,
1771 	.reset_counters = xgene_pmu_reset_counters,
1772 	.start_counters = xgene_pmu_start_counters,
1773 	.stop_counters = xgene_pmu_stop_counters,
1774 };
1775 
1776 static const struct xgene_pmu_ops xgene_pmu_v3_ops = {
1777 	.mask_int = xgene_pmu_v3_mask_int,
1778 	.unmask_int = xgene_pmu_v3_unmask_int,
1779 	.read_counter = xgene_pmu_read_counter64,
1780 	.write_counter = xgene_pmu_write_counter64,
1781 	.write_evttype = xgene_pmu_write_evttype,
1782 	.write_agentmsk = xgene_pmu_v3_write_agentmsk,
1783 	.write_agent1msk = xgene_pmu_v3_write_agent1msk,
1784 	.enable_counter = xgene_pmu_enable_counter,
1785 	.disable_counter = xgene_pmu_disable_counter,
1786 	.enable_counter_int = xgene_pmu_enable_counter_int,
1787 	.disable_counter_int = xgene_pmu_disable_counter_int,
1788 	.reset_counters = xgene_pmu_reset_counters,
1789 	.start_counters = xgene_pmu_start_counters,
1790 	.stop_counters = xgene_pmu_stop_counters,
1791 };
1792 
1793 static const struct of_device_id xgene_pmu_of_match[] = {
1794 	{ .compatible	= "apm,xgene-pmu",	.data = &xgene_pmu_data },
1795 	{ .compatible	= "apm,xgene-pmu-v2",	.data = &xgene_pmu_v2_data },
1796 	{},
1797 };
1798 MODULE_DEVICE_TABLE(of, xgene_pmu_of_match);
1799 #ifdef CONFIG_ACPI
1800 static const struct acpi_device_id xgene_pmu_acpi_match[] = {
1801 	{"APMC0D5B", PCP_PMU_V1},
1802 	{"APMC0D5C", PCP_PMU_V2},
1803 	{"APMC0D83", PCP_PMU_V3},
1804 	{},
1805 };
1806 MODULE_DEVICE_TABLE(acpi, xgene_pmu_acpi_match);
1807 #endif
1808 
1809 static int xgene_pmu_probe(struct platform_device *pdev)
1810 {
1811 	const struct xgene_pmu_data *dev_data;
1812 	const struct of_device_id *of_id;
1813 	struct xgene_pmu *xgene_pmu;
1814 	struct resource *res;
1815 	int irq, rc;
1816 	int version;
1817 
1818 	xgene_pmu = devm_kzalloc(&pdev->dev, sizeof(*xgene_pmu), GFP_KERNEL);
1819 	if (!xgene_pmu)
1820 		return -ENOMEM;
1821 	xgene_pmu->dev = &pdev->dev;
1822 	platform_set_drvdata(pdev, xgene_pmu);
1823 
1824 	version = -EINVAL;
1825 	of_id = of_match_device(xgene_pmu_of_match, &pdev->dev);
1826 	if (of_id) {
1827 		dev_data = (const struct xgene_pmu_data *) of_id->data;
1828 		version = dev_data->id;
1829 	}
1830 
1831 #ifdef CONFIG_ACPI
1832 	if (ACPI_COMPANION(&pdev->dev)) {
1833 		const struct acpi_device_id *acpi_id;
1834 
1835 		acpi_id = acpi_match_device(xgene_pmu_acpi_match, &pdev->dev);
1836 		if (acpi_id)
1837 			version = (int) acpi_id->driver_data;
1838 	}
1839 #endif
1840 	if (version < 0)
1841 		return -ENODEV;
1842 
1843 	if (version == PCP_PMU_V3)
1844 		xgene_pmu->ops = &xgene_pmu_v3_ops;
1845 	else
1846 		xgene_pmu->ops = &xgene_pmu_ops;
1847 
1848 	INIT_LIST_HEAD(&xgene_pmu->l3cpmus);
1849 	INIT_LIST_HEAD(&xgene_pmu->iobpmus);
1850 	INIT_LIST_HEAD(&xgene_pmu->mcbpmus);
1851 	INIT_LIST_HEAD(&xgene_pmu->mcpmus);
1852 
1853 	xgene_pmu->version = version;
1854 	dev_info(&pdev->dev, "X-Gene PMU version %d\n", xgene_pmu->version);
1855 
1856 	res = platform_get_resource(pdev, IORESOURCE_MEM, 0);
1857 	xgene_pmu->pcppmu_csr = devm_ioremap_resource(&pdev->dev, res);
1858 	if (IS_ERR(xgene_pmu->pcppmu_csr)) {
1859 		dev_err(&pdev->dev, "ioremap failed for PCP PMU resource\n");
1860 		return PTR_ERR(xgene_pmu->pcppmu_csr);
1861 	}
1862 
1863 	irq = platform_get_irq(pdev, 0);
1864 	if (irq < 0) {
1865 		dev_err(&pdev->dev, "No IRQ resource\n");
1866 		return -EINVAL;
1867 	}
1868 	rc = devm_request_irq(&pdev->dev, irq, xgene_pmu_isr,
1869 				IRQF_NOBALANCING | IRQF_NO_THREAD,
1870 				dev_name(&pdev->dev), xgene_pmu);
1871 	if (rc) {
1872 		dev_err(&pdev->dev, "Could not request IRQ %d\n", irq);
1873 		return rc;
1874 	}
1875 
1876 	raw_spin_lock_init(&xgene_pmu->lock);
1877 
1878 	/* Check for active MCBs and MCUs */
1879 	rc = xgene_pmu_probe_active_mcb_mcu_l3c(xgene_pmu, pdev);
1880 	if (rc) {
1881 		dev_warn(&pdev->dev, "Unknown MCB/MCU active status\n");
1882 		xgene_pmu->mcb_active_mask = 0x1;
1883 		xgene_pmu->mc_active_mask = 0x1;
1884 	}
1885 
1886 	/* Pick one core to use for cpumask attributes */
1887 	cpumask_set_cpu(smp_processor_id(), &xgene_pmu->cpu);
1888 
1889 	/* Make sure that the overflow interrupt is handled by this CPU */
1890 	rc = irq_set_affinity(irq, &xgene_pmu->cpu);
1891 	if (rc) {
1892 		dev_err(&pdev->dev, "Failed to set interrupt affinity!\n");
1893 		return rc;
1894 	}
1895 
1896 	/* Walk through the tree for all PMU perf devices */
1897 	rc = xgene_pmu_probe_pmu_dev(xgene_pmu, pdev);
1898 	if (rc) {
1899 		dev_err(&pdev->dev, "No PMU perf devices found!\n");
1900 		return rc;
1901 	}
1902 
1903 	/* Enable interrupt */
1904 	xgene_pmu->ops->unmask_int(xgene_pmu);
1905 
1906 	return 0;
1907 }
1908 
1909 static void
1910 xgene_pmu_dev_cleanup(struct xgene_pmu *xgene_pmu, struct list_head *pmus)
1911 {
1912 	struct xgene_pmu_dev_ctx *ctx;
1913 
1914 	list_for_each_entry(ctx, pmus, next) {
1915 		perf_pmu_unregister(&ctx->pmu_dev->pmu);
1916 	}
1917 }
1918 
1919 static int xgene_pmu_remove(struct platform_device *pdev)
1920 {
1921 	struct xgene_pmu *xgene_pmu = dev_get_drvdata(&pdev->dev);
1922 
1923 	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->l3cpmus);
1924 	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->iobpmus);
1925 	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->mcbpmus);
1926 	xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->mcpmus);
1927 
1928 	return 0;
1929 }
1930 
1931 static struct platform_driver xgene_pmu_driver = {
1932 	.probe = xgene_pmu_probe,
1933 	.remove = xgene_pmu_remove,
1934 	.driver = {
1935 		.name		= "xgene-pmu",
1936 		.of_match_table = xgene_pmu_of_match,
1937 		.acpi_match_table = ACPI_PTR(xgene_pmu_acpi_match),
1938 	},
1939 };
1940 
1941 builtin_platform_driver(xgene_pmu_driver);
1942