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/cpuhotplug.h> 25 #include <linux/cpumask.h> 26 #include <linux/interrupt.h> 27 #include <linux/io.h> 28 #include <linux/mfd/syscon.h> 29 #include <linux/module.h> 30 #include <linux/of_address.h> 31 #include <linux/of_fdt.h> 32 #include <linux/of_irq.h> 33 #include <linux/of_platform.h> 34 #include <linux/perf_event.h> 35 #include <linux/platform_device.h> 36 #include <linux/regmap.h> 37 #include <linux/slab.h> 38 39 #define CSW_CSWCR 0x0000 40 #define CSW_CSWCR_DUALMCB_MASK BIT(0) 41 #define CSW_CSWCR_MCB0_ROUTING(x) (((x) & 0x0C) >> 2) 42 #define CSW_CSWCR_MCB1_ROUTING(x) (((x) & 0x30) >> 4) 43 #define MCBADDRMR 0x0000 44 #define MCBADDRMR_DUALMCU_MODE_MASK BIT(2) 45 46 #define PCPPMU_INTSTATUS_REG 0x000 47 #define PCPPMU_INTMASK_REG 0x004 48 #define PCPPMU_INTMASK 0x0000000F 49 #define PCPPMU_INTENMASK 0xFFFFFFFF 50 #define PCPPMU_INTCLRMASK 0xFFFFFFF0 51 #define PCPPMU_INT_MCU BIT(0) 52 #define PCPPMU_INT_MCB BIT(1) 53 #define PCPPMU_INT_L3C BIT(2) 54 #define PCPPMU_INT_IOB BIT(3) 55 56 #define PCPPMU_V3_INTMASK 0x00FF33FF 57 #define PCPPMU_V3_INTENMASK 0xFFFFFFFF 58 #define PCPPMU_V3_INTCLRMASK 0xFF00CC00 59 #define PCPPMU_V3_INT_MCU 0x000000FF 60 #define PCPPMU_V3_INT_MCB 0x00000300 61 #define PCPPMU_V3_INT_L3C 0x00FF0000 62 #define PCPPMU_V3_INT_IOB 0x00003000 63 64 #define PMU_MAX_COUNTERS 4 65 #define PMU_CNT_MAX_PERIOD 0xFFFFFFFFULL 66 #define PMU_V3_CNT_MAX_PERIOD 0xFFFFFFFFFFFFFFFFULL 67 #define PMU_OVERFLOW_MASK 0xF 68 #define PMU_PMCR_E BIT(0) 69 #define PMU_PMCR_P BIT(1) 70 71 #define PMU_PMEVCNTR0 0x000 72 #define PMU_PMEVCNTR1 0x004 73 #define PMU_PMEVCNTR2 0x008 74 #define PMU_PMEVCNTR3 0x00C 75 #define PMU_PMEVTYPER0 0x400 76 #define PMU_PMEVTYPER1 0x404 77 #define PMU_PMEVTYPER2 0x408 78 #define PMU_PMEVTYPER3 0x40C 79 #define PMU_PMAMR0 0xA00 80 #define PMU_PMAMR1 0xA04 81 #define PMU_PMCNTENSET 0xC00 82 #define PMU_PMCNTENCLR 0xC20 83 #define PMU_PMINTENSET 0xC40 84 #define PMU_PMINTENCLR 0xC60 85 #define PMU_PMOVSR 0xC80 86 #define PMU_PMCR 0xE04 87 88 /* PMU registers for V3 */ 89 #define PMU_PMOVSCLR 0xC80 90 #define PMU_PMOVSSET 0xCC0 91 92 #define to_pmu_dev(p) container_of(p, struct xgene_pmu_dev, pmu) 93 #define GET_CNTR(ev) (ev->hw.idx) 94 #define GET_EVENTID(ev) (ev->hw.config & 0xFFULL) 95 #define GET_AGENTID(ev) (ev->hw.config_base & 0xFFFFFFFFUL) 96 #define GET_AGENT1ID(ev) ((ev->hw.config_base >> 32) & 0xFFFFFFFFUL) 97 98 struct hw_pmu_info { 99 u32 type; 100 u32 enable_mask; 101 void __iomem *csr; 102 }; 103 104 struct xgene_pmu_dev { 105 struct hw_pmu_info *inf; 106 struct xgene_pmu *parent; 107 struct pmu pmu; 108 u8 max_counters; 109 DECLARE_BITMAP(cntr_assign_mask, PMU_MAX_COUNTERS); 110 u64 max_period; 111 const struct attribute_group **attr_groups; 112 struct perf_event *pmu_counter_event[PMU_MAX_COUNTERS]; 113 }; 114 115 struct xgene_pmu_ops { 116 void (*mask_int)(struct xgene_pmu *pmu); 117 void (*unmask_int)(struct xgene_pmu *pmu); 118 u64 (*read_counter)(struct xgene_pmu_dev *pmu, int idx); 119 void (*write_counter)(struct xgene_pmu_dev *pmu, int idx, u64 val); 120 void (*write_evttype)(struct xgene_pmu_dev *pmu_dev, int idx, u32 val); 121 void (*write_agentmsk)(struct xgene_pmu_dev *pmu_dev, u32 val); 122 void (*write_agent1msk)(struct xgene_pmu_dev *pmu_dev, u32 val); 123 void (*enable_counter)(struct xgene_pmu_dev *pmu_dev, int idx); 124 void (*disable_counter)(struct xgene_pmu_dev *pmu_dev, int idx); 125 void (*enable_counter_int)(struct xgene_pmu_dev *pmu_dev, int idx); 126 void (*disable_counter_int)(struct xgene_pmu_dev *pmu_dev, int idx); 127 void (*reset_counters)(struct xgene_pmu_dev *pmu_dev); 128 void (*start_counters)(struct xgene_pmu_dev *pmu_dev); 129 void (*stop_counters)(struct xgene_pmu_dev *pmu_dev); 130 }; 131 132 struct xgene_pmu { 133 struct device *dev; 134 struct hlist_node node; 135 int version; 136 void __iomem *pcppmu_csr; 137 u32 mcb_active_mask; 138 u32 mc_active_mask; 139 u32 l3c_active_mask; 140 cpumask_t cpu; 141 int irq; 142 raw_spinlock_t lock; 143 const struct xgene_pmu_ops *ops; 144 struct list_head l3cpmus; 145 struct list_head iobpmus; 146 struct list_head mcbpmus; 147 struct list_head mcpmus; 148 }; 149 150 struct xgene_pmu_dev_ctx { 151 char *name; 152 struct list_head next; 153 struct xgene_pmu_dev *pmu_dev; 154 struct hw_pmu_info inf; 155 }; 156 157 struct xgene_pmu_data { 158 int id; 159 u32 data; 160 }; 161 162 enum xgene_pmu_version { 163 PCP_PMU_V1 = 1, 164 PCP_PMU_V2, 165 PCP_PMU_V3, 166 }; 167 168 enum xgene_pmu_dev_type { 169 PMU_TYPE_L3C = 0, 170 PMU_TYPE_IOB, 171 PMU_TYPE_IOB_SLOW, 172 PMU_TYPE_MCB, 173 PMU_TYPE_MC, 174 }; 175 176 /* 177 * sysfs format attributes 178 */ 179 static ssize_t xgene_pmu_format_show(struct device *dev, 180 struct device_attribute *attr, char *buf) 181 { 182 struct dev_ext_attribute *eattr; 183 184 eattr = container_of(attr, struct dev_ext_attribute, attr); 185 return sprintf(buf, "%s\n", (char *) eattr->var); 186 } 187 188 #define XGENE_PMU_FORMAT_ATTR(_name, _config) \ 189 (&((struct dev_ext_attribute[]) { \ 190 { .attr = __ATTR(_name, S_IRUGO, xgene_pmu_format_show, NULL), \ 191 .var = (void *) _config, } \ 192 })[0].attr.attr) 193 194 static struct attribute *l3c_pmu_format_attrs[] = { 195 XGENE_PMU_FORMAT_ATTR(l3c_eventid, "config:0-7"), 196 XGENE_PMU_FORMAT_ATTR(l3c_agentid, "config1:0-9"), 197 NULL, 198 }; 199 200 static struct attribute *iob_pmu_format_attrs[] = { 201 XGENE_PMU_FORMAT_ATTR(iob_eventid, "config:0-7"), 202 XGENE_PMU_FORMAT_ATTR(iob_agentid, "config1:0-63"), 203 NULL, 204 }; 205 206 static struct attribute *mcb_pmu_format_attrs[] = { 207 XGENE_PMU_FORMAT_ATTR(mcb_eventid, "config:0-5"), 208 XGENE_PMU_FORMAT_ATTR(mcb_agentid, "config1:0-9"), 209 NULL, 210 }; 211 212 static struct attribute *mc_pmu_format_attrs[] = { 213 XGENE_PMU_FORMAT_ATTR(mc_eventid, "config:0-28"), 214 NULL, 215 }; 216 217 static const struct attribute_group l3c_pmu_format_attr_group = { 218 .name = "format", 219 .attrs = l3c_pmu_format_attrs, 220 }; 221 222 static const struct attribute_group iob_pmu_format_attr_group = { 223 .name = "format", 224 .attrs = iob_pmu_format_attrs, 225 }; 226 227 static const struct attribute_group mcb_pmu_format_attr_group = { 228 .name = "format", 229 .attrs = mcb_pmu_format_attrs, 230 }; 231 232 static const struct attribute_group mc_pmu_format_attr_group = { 233 .name = "format", 234 .attrs = mc_pmu_format_attrs, 235 }; 236 237 static struct attribute *l3c_pmu_v3_format_attrs[] = { 238 XGENE_PMU_FORMAT_ATTR(l3c_eventid, "config:0-39"), 239 NULL, 240 }; 241 242 static struct attribute *iob_pmu_v3_format_attrs[] = { 243 XGENE_PMU_FORMAT_ATTR(iob_eventid, "config:0-47"), 244 NULL, 245 }; 246 247 static struct attribute *iob_slow_pmu_v3_format_attrs[] = { 248 XGENE_PMU_FORMAT_ATTR(iob_slow_eventid, "config:0-16"), 249 NULL, 250 }; 251 252 static struct attribute *mcb_pmu_v3_format_attrs[] = { 253 XGENE_PMU_FORMAT_ATTR(mcb_eventid, "config:0-35"), 254 NULL, 255 }; 256 257 static struct attribute *mc_pmu_v3_format_attrs[] = { 258 XGENE_PMU_FORMAT_ATTR(mc_eventid, "config:0-44"), 259 NULL, 260 }; 261 262 static const struct attribute_group l3c_pmu_v3_format_attr_group = { 263 .name = "format", 264 .attrs = l3c_pmu_v3_format_attrs, 265 }; 266 267 static const struct attribute_group iob_pmu_v3_format_attr_group = { 268 .name = "format", 269 .attrs = iob_pmu_v3_format_attrs, 270 }; 271 272 static const struct attribute_group iob_slow_pmu_v3_format_attr_group = { 273 .name = "format", 274 .attrs = iob_slow_pmu_v3_format_attrs, 275 }; 276 277 static const struct attribute_group mcb_pmu_v3_format_attr_group = { 278 .name = "format", 279 .attrs = mcb_pmu_v3_format_attrs, 280 }; 281 282 static const struct attribute_group mc_pmu_v3_format_attr_group = { 283 .name = "format", 284 .attrs = mc_pmu_v3_format_attrs, 285 }; 286 287 /* 288 * sysfs event attributes 289 */ 290 static ssize_t xgene_pmu_event_show(struct device *dev, 291 struct device_attribute *attr, char *buf) 292 { 293 struct dev_ext_attribute *eattr; 294 295 eattr = container_of(attr, struct dev_ext_attribute, attr); 296 return sprintf(buf, "config=0x%lx\n", (unsigned long) eattr->var); 297 } 298 299 #define XGENE_PMU_EVENT_ATTR(_name, _config) \ 300 (&((struct dev_ext_attribute[]) { \ 301 { .attr = __ATTR(_name, S_IRUGO, xgene_pmu_event_show, NULL), \ 302 .var = (void *) _config, } \ 303 })[0].attr.attr) 304 305 static struct attribute *l3c_pmu_events_attrs[] = { 306 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 307 XGENE_PMU_EVENT_ATTR(cycle-count-div-64, 0x01), 308 XGENE_PMU_EVENT_ATTR(read-hit, 0x02), 309 XGENE_PMU_EVENT_ATTR(read-miss, 0x03), 310 XGENE_PMU_EVENT_ATTR(write-need-replacement, 0x06), 311 XGENE_PMU_EVENT_ATTR(write-not-need-replacement, 0x07), 312 XGENE_PMU_EVENT_ATTR(tq-full, 0x08), 313 XGENE_PMU_EVENT_ATTR(ackq-full, 0x09), 314 XGENE_PMU_EVENT_ATTR(wdb-full, 0x0a), 315 XGENE_PMU_EVENT_ATTR(bank-fifo-full, 0x0b), 316 XGENE_PMU_EVENT_ATTR(odb-full, 0x0c), 317 XGENE_PMU_EVENT_ATTR(wbq-full, 0x0d), 318 XGENE_PMU_EVENT_ATTR(bank-conflict-fifo-issue, 0x0e), 319 XGENE_PMU_EVENT_ATTR(bank-fifo-issue, 0x0f), 320 NULL, 321 }; 322 323 static struct attribute *iob_pmu_events_attrs[] = { 324 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 325 XGENE_PMU_EVENT_ATTR(cycle-count-div-64, 0x01), 326 XGENE_PMU_EVENT_ATTR(axi0-read, 0x02), 327 XGENE_PMU_EVENT_ATTR(axi0-read-partial, 0x03), 328 XGENE_PMU_EVENT_ATTR(axi1-read, 0x04), 329 XGENE_PMU_EVENT_ATTR(axi1-read-partial, 0x05), 330 XGENE_PMU_EVENT_ATTR(csw-read-block, 0x06), 331 XGENE_PMU_EVENT_ATTR(csw-read-partial, 0x07), 332 XGENE_PMU_EVENT_ATTR(axi0-write, 0x10), 333 XGENE_PMU_EVENT_ATTR(axi0-write-partial, 0x11), 334 XGENE_PMU_EVENT_ATTR(axi1-write, 0x13), 335 XGENE_PMU_EVENT_ATTR(axi1-write-partial, 0x14), 336 XGENE_PMU_EVENT_ATTR(csw-inbound-dirty, 0x16), 337 NULL, 338 }; 339 340 static struct attribute *mcb_pmu_events_attrs[] = { 341 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 342 XGENE_PMU_EVENT_ATTR(cycle-count-div-64, 0x01), 343 XGENE_PMU_EVENT_ATTR(csw-read, 0x02), 344 XGENE_PMU_EVENT_ATTR(csw-write-request, 0x03), 345 XGENE_PMU_EVENT_ATTR(mcb-csw-stall, 0x04), 346 XGENE_PMU_EVENT_ATTR(cancel-read-gack, 0x05), 347 NULL, 348 }; 349 350 static struct attribute *mc_pmu_events_attrs[] = { 351 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 352 XGENE_PMU_EVENT_ATTR(cycle-count-div-64, 0x01), 353 XGENE_PMU_EVENT_ATTR(act-cmd-sent, 0x02), 354 XGENE_PMU_EVENT_ATTR(pre-cmd-sent, 0x03), 355 XGENE_PMU_EVENT_ATTR(rd-cmd-sent, 0x04), 356 XGENE_PMU_EVENT_ATTR(rda-cmd-sent, 0x05), 357 XGENE_PMU_EVENT_ATTR(wr-cmd-sent, 0x06), 358 XGENE_PMU_EVENT_ATTR(wra-cmd-sent, 0x07), 359 XGENE_PMU_EVENT_ATTR(pde-cmd-sent, 0x08), 360 XGENE_PMU_EVENT_ATTR(sre-cmd-sent, 0x09), 361 XGENE_PMU_EVENT_ATTR(prea-cmd-sent, 0x0a), 362 XGENE_PMU_EVENT_ATTR(ref-cmd-sent, 0x0b), 363 XGENE_PMU_EVENT_ATTR(rd-rda-cmd-sent, 0x0c), 364 XGENE_PMU_EVENT_ATTR(wr-wra-cmd-sent, 0x0d), 365 XGENE_PMU_EVENT_ATTR(in-rd-collision, 0x0e), 366 XGENE_PMU_EVENT_ATTR(in-wr-collision, 0x0f), 367 XGENE_PMU_EVENT_ATTR(collision-queue-not-empty, 0x10), 368 XGENE_PMU_EVENT_ATTR(collision-queue-full, 0x11), 369 XGENE_PMU_EVENT_ATTR(mcu-request, 0x12), 370 XGENE_PMU_EVENT_ATTR(mcu-rd-request, 0x13), 371 XGENE_PMU_EVENT_ATTR(mcu-hp-rd-request, 0x14), 372 XGENE_PMU_EVENT_ATTR(mcu-wr-request, 0x15), 373 XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-all, 0x16), 374 XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-cancel, 0x17), 375 XGENE_PMU_EVENT_ATTR(mcu-rd-response, 0x18), 376 XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-speculative-all, 0x19), 377 XGENE_PMU_EVENT_ATTR(mcu-rd-proceed-speculative-cancel, 0x1a), 378 XGENE_PMU_EVENT_ATTR(mcu-wr-proceed-all, 0x1b), 379 XGENE_PMU_EVENT_ATTR(mcu-wr-proceed-cancel, 0x1c), 380 NULL, 381 }; 382 383 static const struct attribute_group l3c_pmu_events_attr_group = { 384 .name = "events", 385 .attrs = l3c_pmu_events_attrs, 386 }; 387 388 static const struct attribute_group iob_pmu_events_attr_group = { 389 .name = "events", 390 .attrs = iob_pmu_events_attrs, 391 }; 392 393 static const struct attribute_group mcb_pmu_events_attr_group = { 394 .name = "events", 395 .attrs = mcb_pmu_events_attrs, 396 }; 397 398 static const struct attribute_group mc_pmu_events_attr_group = { 399 .name = "events", 400 .attrs = mc_pmu_events_attrs, 401 }; 402 403 static struct attribute *l3c_pmu_v3_events_attrs[] = { 404 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 405 XGENE_PMU_EVENT_ATTR(read-hit, 0x01), 406 XGENE_PMU_EVENT_ATTR(read-miss, 0x02), 407 XGENE_PMU_EVENT_ATTR(index-flush-eviction, 0x03), 408 XGENE_PMU_EVENT_ATTR(write-caused-replacement, 0x04), 409 XGENE_PMU_EVENT_ATTR(write-not-caused-replacement, 0x05), 410 XGENE_PMU_EVENT_ATTR(clean-eviction, 0x06), 411 XGENE_PMU_EVENT_ATTR(dirty-eviction, 0x07), 412 XGENE_PMU_EVENT_ATTR(read, 0x08), 413 XGENE_PMU_EVENT_ATTR(write, 0x09), 414 XGENE_PMU_EVENT_ATTR(request, 0x0a), 415 XGENE_PMU_EVENT_ATTR(tq-bank-conflict-issue-stall, 0x0b), 416 XGENE_PMU_EVENT_ATTR(tq-full, 0x0c), 417 XGENE_PMU_EVENT_ATTR(ackq-full, 0x0d), 418 XGENE_PMU_EVENT_ATTR(wdb-full, 0x0e), 419 XGENE_PMU_EVENT_ATTR(odb-full, 0x10), 420 XGENE_PMU_EVENT_ATTR(wbq-full, 0x11), 421 XGENE_PMU_EVENT_ATTR(input-req-async-fifo-stall, 0x12), 422 XGENE_PMU_EVENT_ATTR(output-req-async-fifo-stall, 0x13), 423 XGENE_PMU_EVENT_ATTR(output-data-async-fifo-stall, 0x14), 424 XGENE_PMU_EVENT_ATTR(total-insertion, 0x15), 425 XGENE_PMU_EVENT_ATTR(sip-insertions-r-set, 0x16), 426 XGENE_PMU_EVENT_ATTR(sip-insertions-r-clear, 0x17), 427 XGENE_PMU_EVENT_ATTR(dip-insertions-r-set, 0x18), 428 XGENE_PMU_EVENT_ATTR(dip-insertions-r-clear, 0x19), 429 XGENE_PMU_EVENT_ATTR(dip-insertions-force-r-set, 0x1a), 430 XGENE_PMU_EVENT_ATTR(egression, 0x1b), 431 XGENE_PMU_EVENT_ATTR(replacement, 0x1c), 432 XGENE_PMU_EVENT_ATTR(old-replacement, 0x1d), 433 XGENE_PMU_EVENT_ATTR(young-replacement, 0x1e), 434 XGENE_PMU_EVENT_ATTR(r-set-replacement, 0x1f), 435 XGENE_PMU_EVENT_ATTR(r-clear-replacement, 0x20), 436 XGENE_PMU_EVENT_ATTR(old-r-replacement, 0x21), 437 XGENE_PMU_EVENT_ATTR(old-nr-replacement, 0x22), 438 XGENE_PMU_EVENT_ATTR(young-r-replacement, 0x23), 439 XGENE_PMU_EVENT_ATTR(young-nr-replacement, 0x24), 440 XGENE_PMU_EVENT_ATTR(bloomfilter-clearing, 0x25), 441 XGENE_PMU_EVENT_ATTR(generation-flip, 0x26), 442 XGENE_PMU_EVENT_ATTR(vcc-droop-detected, 0x27), 443 NULL, 444 }; 445 446 static struct attribute *iob_fast_pmu_v3_events_attrs[] = { 447 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 448 XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-all, 0x01), 449 XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-rd, 0x02), 450 XGENE_PMU_EVENT_ATTR(pa-req-buf-alloc-wr, 0x03), 451 XGENE_PMU_EVENT_ATTR(pa-all-cp-req, 0x04), 452 XGENE_PMU_EVENT_ATTR(pa-cp-blk-req, 0x05), 453 XGENE_PMU_EVENT_ATTR(pa-cp-ptl-req, 0x06), 454 XGENE_PMU_EVENT_ATTR(pa-cp-rd-req, 0x07), 455 XGENE_PMU_EVENT_ATTR(pa-cp-wr-req, 0x08), 456 XGENE_PMU_EVENT_ATTR(ba-all-req, 0x09), 457 XGENE_PMU_EVENT_ATTR(ba-rd-req, 0x0a), 458 XGENE_PMU_EVENT_ATTR(ba-wr-req, 0x0b), 459 XGENE_PMU_EVENT_ATTR(pa-rd-shared-req-issued, 0x10), 460 XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-req-issued, 0x11), 461 XGENE_PMU_EVENT_ATTR(pa-wr-invalidate-req-issued-stashable, 0x12), 462 XGENE_PMU_EVENT_ATTR(pa-wr-invalidate-req-issued-nonstashable, 0x13), 463 XGENE_PMU_EVENT_ATTR(pa-wr-back-req-issued-stashable, 0x14), 464 XGENE_PMU_EVENT_ATTR(pa-wr-back-req-issued-nonstashable, 0x15), 465 XGENE_PMU_EVENT_ATTR(pa-ptl-wr-req, 0x16), 466 XGENE_PMU_EVENT_ATTR(pa-ptl-rd-req, 0x17), 467 XGENE_PMU_EVENT_ATTR(pa-wr-back-clean-data, 0x18), 468 XGENE_PMU_EVENT_ATTR(pa-wr-back-cancelled-on-SS, 0x1b), 469 XGENE_PMU_EVENT_ATTR(pa-barrier-occurrence, 0x1c), 470 XGENE_PMU_EVENT_ATTR(pa-barrier-cycles, 0x1d), 471 XGENE_PMU_EVENT_ATTR(pa-total-cp-snoops, 0x20), 472 XGENE_PMU_EVENT_ATTR(pa-rd-shared-snoop, 0x21), 473 XGENE_PMU_EVENT_ATTR(pa-rd-shared-snoop-hit, 0x22), 474 XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-snoop, 0x23), 475 XGENE_PMU_EVENT_ATTR(pa-rd-exclusive-snoop-hit, 0x24), 476 XGENE_PMU_EVENT_ATTR(pa-rd-wr-invalid-snoop, 0x25), 477 XGENE_PMU_EVENT_ATTR(pa-rd-wr-invalid-snoop-hit, 0x26), 478 XGENE_PMU_EVENT_ATTR(pa-req-buffer-full, 0x28), 479 XGENE_PMU_EVENT_ATTR(cswlf-outbound-req-fifo-full, 0x29), 480 XGENE_PMU_EVENT_ATTR(cswlf-inbound-snoop-fifo-backpressure, 0x2a), 481 XGENE_PMU_EVENT_ATTR(cswlf-outbound-lack-fifo-full, 0x2b), 482 XGENE_PMU_EVENT_ATTR(cswlf-inbound-gack-fifo-backpressure, 0x2c), 483 XGENE_PMU_EVENT_ATTR(cswlf-outbound-data-fifo-full, 0x2d), 484 XGENE_PMU_EVENT_ATTR(cswlf-inbound-data-fifo-backpressure, 0x2e), 485 XGENE_PMU_EVENT_ATTR(cswlf-inbound-req-backpressure, 0x2f), 486 NULL, 487 }; 488 489 static struct attribute *iob_slow_pmu_v3_events_attrs[] = { 490 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 491 XGENE_PMU_EVENT_ATTR(pa-axi0-rd-req, 0x01), 492 XGENE_PMU_EVENT_ATTR(pa-axi0-wr-req, 0x02), 493 XGENE_PMU_EVENT_ATTR(pa-axi1-rd-req, 0x03), 494 XGENE_PMU_EVENT_ATTR(pa-axi1-wr-req, 0x04), 495 XGENE_PMU_EVENT_ATTR(ba-all-axi-req, 0x07), 496 XGENE_PMU_EVENT_ATTR(ba-axi-rd-req, 0x08), 497 XGENE_PMU_EVENT_ATTR(ba-axi-wr-req, 0x09), 498 XGENE_PMU_EVENT_ATTR(ba-free-list-empty, 0x10), 499 NULL, 500 }; 501 502 static struct attribute *mcb_pmu_v3_events_attrs[] = { 503 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 504 XGENE_PMU_EVENT_ATTR(req-receive, 0x01), 505 XGENE_PMU_EVENT_ATTR(rd-req-recv, 0x02), 506 XGENE_PMU_EVENT_ATTR(rd-req-recv-2, 0x03), 507 XGENE_PMU_EVENT_ATTR(wr-req-recv, 0x04), 508 XGENE_PMU_EVENT_ATTR(wr-req-recv-2, 0x05), 509 XGENE_PMU_EVENT_ATTR(rd-req-sent-to-mcu, 0x06), 510 XGENE_PMU_EVENT_ATTR(rd-req-sent-to-mcu-2, 0x07), 511 XGENE_PMU_EVENT_ATTR(rd-req-sent-to-spec-mcu, 0x08), 512 XGENE_PMU_EVENT_ATTR(rd-req-sent-to-spec-mcu-2, 0x09), 513 XGENE_PMU_EVENT_ATTR(glbl-ack-recv-for-rd-sent-to-spec-mcu, 0x0a), 514 XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-for-rd-sent-to-spec-mcu, 0x0b), 515 XGENE_PMU_EVENT_ATTR(glbl-ack-nogo-recv-for-rd-sent-to-spec-mcu, 0x0c), 516 XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-any-rd-req, 0x0d), 517 XGENE_PMU_EVENT_ATTR(glbl-ack-go-recv-any-rd-req-2, 0x0e), 518 XGENE_PMU_EVENT_ATTR(wr-req-sent-to-mcu, 0x0f), 519 XGENE_PMU_EVENT_ATTR(gack-recv, 0x10), 520 XGENE_PMU_EVENT_ATTR(rd-gack-recv, 0x11), 521 XGENE_PMU_EVENT_ATTR(wr-gack-recv, 0x12), 522 XGENE_PMU_EVENT_ATTR(cancel-rd-gack, 0x13), 523 XGENE_PMU_EVENT_ATTR(cancel-wr-gack, 0x14), 524 XGENE_PMU_EVENT_ATTR(mcb-csw-req-stall, 0x15), 525 XGENE_PMU_EVENT_ATTR(mcu-req-intf-blocked, 0x16), 526 XGENE_PMU_EVENT_ATTR(mcb-mcu-rd-intf-stall, 0x17), 527 XGENE_PMU_EVENT_ATTR(csw-rd-intf-blocked, 0x18), 528 XGENE_PMU_EVENT_ATTR(csw-local-ack-intf-blocked, 0x19), 529 XGENE_PMU_EVENT_ATTR(mcu-req-table-full, 0x1a), 530 XGENE_PMU_EVENT_ATTR(mcu-stat-table-full, 0x1b), 531 XGENE_PMU_EVENT_ATTR(mcu-wr-table-full, 0x1c), 532 XGENE_PMU_EVENT_ATTR(mcu-rdreceipt-resp, 0x1d), 533 XGENE_PMU_EVENT_ATTR(mcu-wrcomplete-resp, 0x1e), 534 XGENE_PMU_EVENT_ATTR(mcu-retryack-resp, 0x1f), 535 XGENE_PMU_EVENT_ATTR(mcu-pcrdgrant-resp, 0x20), 536 XGENE_PMU_EVENT_ATTR(mcu-req-from-lastload, 0x21), 537 XGENE_PMU_EVENT_ATTR(mcu-req-from-bypass, 0x22), 538 XGENE_PMU_EVENT_ATTR(volt-droop-detect, 0x23), 539 NULL, 540 }; 541 542 static struct attribute *mc_pmu_v3_events_attrs[] = { 543 XGENE_PMU_EVENT_ATTR(cycle-count, 0x00), 544 XGENE_PMU_EVENT_ATTR(act-sent, 0x01), 545 XGENE_PMU_EVENT_ATTR(pre-sent, 0x02), 546 XGENE_PMU_EVENT_ATTR(rd-sent, 0x03), 547 XGENE_PMU_EVENT_ATTR(rda-sent, 0x04), 548 XGENE_PMU_EVENT_ATTR(wr-sent, 0x05), 549 XGENE_PMU_EVENT_ATTR(wra-sent, 0x06), 550 XGENE_PMU_EVENT_ATTR(pd-entry-vld, 0x07), 551 XGENE_PMU_EVENT_ATTR(sref-entry-vld, 0x08), 552 XGENE_PMU_EVENT_ATTR(prea-sent, 0x09), 553 XGENE_PMU_EVENT_ATTR(ref-sent, 0x0a), 554 XGENE_PMU_EVENT_ATTR(rd-rda-sent, 0x0b), 555 XGENE_PMU_EVENT_ATTR(wr-wra-sent, 0x0c), 556 XGENE_PMU_EVENT_ATTR(raw-hazard, 0x0d), 557 XGENE_PMU_EVENT_ATTR(war-hazard, 0x0e), 558 XGENE_PMU_EVENT_ATTR(waw-hazard, 0x0f), 559 XGENE_PMU_EVENT_ATTR(rar-hazard, 0x10), 560 XGENE_PMU_EVENT_ATTR(raw-war-waw-hazard, 0x11), 561 XGENE_PMU_EVENT_ATTR(hprd-lprd-wr-req-vld, 0x12), 562 XGENE_PMU_EVENT_ATTR(lprd-req-vld, 0x13), 563 XGENE_PMU_EVENT_ATTR(hprd-req-vld, 0x14), 564 XGENE_PMU_EVENT_ATTR(hprd-lprd-req-vld, 0x15), 565 XGENE_PMU_EVENT_ATTR(wr-req-vld, 0x16), 566 XGENE_PMU_EVENT_ATTR(partial-wr-req-vld, 0x17), 567 XGENE_PMU_EVENT_ATTR(rd-retry, 0x18), 568 XGENE_PMU_EVENT_ATTR(wr-retry, 0x19), 569 XGENE_PMU_EVENT_ATTR(retry-gnt, 0x1a), 570 XGENE_PMU_EVENT_ATTR(rank-change, 0x1b), 571 XGENE_PMU_EVENT_ATTR(dir-change, 0x1c), 572 XGENE_PMU_EVENT_ATTR(rank-dir-change, 0x1d), 573 XGENE_PMU_EVENT_ATTR(rank-active, 0x1e), 574 XGENE_PMU_EVENT_ATTR(rank-idle, 0x1f), 575 XGENE_PMU_EVENT_ATTR(rank-pd, 0x20), 576 XGENE_PMU_EVENT_ATTR(rank-sref, 0x21), 577 XGENE_PMU_EVENT_ATTR(queue-fill-gt-thresh, 0x22), 578 XGENE_PMU_EVENT_ATTR(queue-rds-gt-thresh, 0x23), 579 XGENE_PMU_EVENT_ATTR(queue-wrs-gt-thresh, 0x24), 580 XGENE_PMU_EVENT_ATTR(phy-updt-complt, 0x25), 581 XGENE_PMU_EVENT_ATTR(tz-fail, 0x26), 582 XGENE_PMU_EVENT_ATTR(dram-errc, 0x27), 583 XGENE_PMU_EVENT_ATTR(dram-errd, 0x28), 584 XGENE_PMU_EVENT_ATTR(rd-enq, 0x29), 585 XGENE_PMU_EVENT_ATTR(wr-enq, 0x2a), 586 XGENE_PMU_EVENT_ATTR(tmac-limit-reached, 0x2b), 587 XGENE_PMU_EVENT_ATTR(tmaw-tracker-full, 0x2c), 588 NULL, 589 }; 590 591 static const struct attribute_group l3c_pmu_v3_events_attr_group = { 592 .name = "events", 593 .attrs = l3c_pmu_v3_events_attrs, 594 }; 595 596 static const struct attribute_group iob_fast_pmu_v3_events_attr_group = { 597 .name = "events", 598 .attrs = iob_fast_pmu_v3_events_attrs, 599 }; 600 601 static const struct attribute_group iob_slow_pmu_v3_events_attr_group = { 602 .name = "events", 603 .attrs = iob_slow_pmu_v3_events_attrs, 604 }; 605 606 static const struct attribute_group mcb_pmu_v3_events_attr_group = { 607 .name = "events", 608 .attrs = mcb_pmu_v3_events_attrs, 609 }; 610 611 static const struct attribute_group mc_pmu_v3_events_attr_group = { 612 .name = "events", 613 .attrs = mc_pmu_v3_events_attrs, 614 }; 615 616 /* 617 * sysfs cpumask attributes 618 */ 619 static ssize_t xgene_pmu_cpumask_show(struct device *dev, 620 struct device_attribute *attr, char *buf) 621 { 622 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(dev_get_drvdata(dev)); 623 624 return cpumap_print_to_pagebuf(true, buf, &pmu_dev->parent->cpu); 625 } 626 627 static DEVICE_ATTR(cpumask, S_IRUGO, xgene_pmu_cpumask_show, NULL); 628 629 static struct attribute *xgene_pmu_cpumask_attrs[] = { 630 &dev_attr_cpumask.attr, 631 NULL, 632 }; 633 634 static const struct attribute_group pmu_cpumask_attr_group = { 635 .attrs = xgene_pmu_cpumask_attrs, 636 }; 637 638 /* 639 * Per PMU device attribute groups of PMU v1 and v2 640 */ 641 static const struct attribute_group *l3c_pmu_attr_groups[] = { 642 &l3c_pmu_format_attr_group, 643 &pmu_cpumask_attr_group, 644 &l3c_pmu_events_attr_group, 645 NULL 646 }; 647 648 static const struct attribute_group *iob_pmu_attr_groups[] = { 649 &iob_pmu_format_attr_group, 650 &pmu_cpumask_attr_group, 651 &iob_pmu_events_attr_group, 652 NULL 653 }; 654 655 static const struct attribute_group *mcb_pmu_attr_groups[] = { 656 &mcb_pmu_format_attr_group, 657 &pmu_cpumask_attr_group, 658 &mcb_pmu_events_attr_group, 659 NULL 660 }; 661 662 static const struct attribute_group *mc_pmu_attr_groups[] = { 663 &mc_pmu_format_attr_group, 664 &pmu_cpumask_attr_group, 665 &mc_pmu_events_attr_group, 666 NULL 667 }; 668 669 /* 670 * Per PMU device attribute groups of PMU v3 671 */ 672 static const struct attribute_group *l3c_pmu_v3_attr_groups[] = { 673 &l3c_pmu_v3_format_attr_group, 674 &pmu_cpumask_attr_group, 675 &l3c_pmu_v3_events_attr_group, 676 NULL 677 }; 678 679 static const struct attribute_group *iob_fast_pmu_v3_attr_groups[] = { 680 &iob_pmu_v3_format_attr_group, 681 &pmu_cpumask_attr_group, 682 &iob_fast_pmu_v3_events_attr_group, 683 NULL 684 }; 685 686 static const struct attribute_group *iob_slow_pmu_v3_attr_groups[] = { 687 &iob_slow_pmu_v3_format_attr_group, 688 &pmu_cpumask_attr_group, 689 &iob_slow_pmu_v3_events_attr_group, 690 NULL 691 }; 692 693 static const struct attribute_group *mcb_pmu_v3_attr_groups[] = { 694 &mcb_pmu_v3_format_attr_group, 695 &pmu_cpumask_attr_group, 696 &mcb_pmu_v3_events_attr_group, 697 NULL 698 }; 699 700 static const struct attribute_group *mc_pmu_v3_attr_groups[] = { 701 &mc_pmu_v3_format_attr_group, 702 &pmu_cpumask_attr_group, 703 &mc_pmu_v3_events_attr_group, 704 NULL 705 }; 706 707 static int get_next_avail_cntr(struct xgene_pmu_dev *pmu_dev) 708 { 709 int cntr; 710 711 cntr = find_first_zero_bit(pmu_dev->cntr_assign_mask, 712 pmu_dev->max_counters); 713 if (cntr == pmu_dev->max_counters) 714 return -ENOSPC; 715 set_bit(cntr, pmu_dev->cntr_assign_mask); 716 717 return cntr; 718 } 719 720 static void clear_avail_cntr(struct xgene_pmu_dev *pmu_dev, int cntr) 721 { 722 clear_bit(cntr, pmu_dev->cntr_assign_mask); 723 } 724 725 static inline void xgene_pmu_mask_int(struct xgene_pmu *xgene_pmu) 726 { 727 writel(PCPPMU_INTENMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG); 728 } 729 730 static inline void xgene_pmu_v3_mask_int(struct xgene_pmu *xgene_pmu) 731 { 732 writel(PCPPMU_V3_INTENMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG); 733 } 734 735 static inline void xgene_pmu_unmask_int(struct xgene_pmu *xgene_pmu) 736 { 737 writel(PCPPMU_INTCLRMASK, xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG); 738 } 739 740 static inline void xgene_pmu_v3_unmask_int(struct xgene_pmu *xgene_pmu) 741 { 742 writel(PCPPMU_V3_INTCLRMASK, 743 xgene_pmu->pcppmu_csr + PCPPMU_INTMASK_REG); 744 } 745 746 static inline u64 xgene_pmu_read_counter32(struct xgene_pmu_dev *pmu_dev, 747 int idx) 748 { 749 return readl(pmu_dev->inf->csr + PMU_PMEVCNTR0 + (4 * idx)); 750 } 751 752 static inline u64 xgene_pmu_read_counter64(struct xgene_pmu_dev *pmu_dev, 753 int idx) 754 { 755 u32 lo, hi; 756 757 /* 758 * v3 has 64-bit counter registers composed by 2 32-bit registers 759 * This can be a problem if the counter increases and carries 760 * out of bit [31] between 2 reads. The extra reads would help 761 * to prevent this issue. 762 */ 763 do { 764 hi = xgene_pmu_read_counter32(pmu_dev, 2 * idx + 1); 765 lo = xgene_pmu_read_counter32(pmu_dev, 2 * idx); 766 } while (hi != xgene_pmu_read_counter32(pmu_dev, 2 * idx + 1)); 767 768 return (((u64)hi << 32) | lo); 769 } 770 771 static inline void 772 xgene_pmu_write_counter32(struct xgene_pmu_dev *pmu_dev, int idx, u64 val) 773 { 774 writel(val, pmu_dev->inf->csr + PMU_PMEVCNTR0 + (4 * idx)); 775 } 776 777 static inline void 778 xgene_pmu_write_counter64(struct xgene_pmu_dev *pmu_dev, int idx, u64 val) 779 { 780 u32 cnt_lo, cnt_hi; 781 782 cnt_hi = upper_32_bits(val); 783 cnt_lo = lower_32_bits(val); 784 785 /* v3 has 64-bit counter registers composed by 2 32-bit registers */ 786 xgene_pmu_write_counter32(pmu_dev, 2 * idx, cnt_lo); 787 xgene_pmu_write_counter32(pmu_dev, 2 * idx + 1, cnt_hi); 788 } 789 790 static inline void 791 xgene_pmu_write_evttype(struct xgene_pmu_dev *pmu_dev, int idx, u32 val) 792 { 793 writel(val, pmu_dev->inf->csr + PMU_PMEVTYPER0 + (4 * idx)); 794 } 795 796 static inline void 797 xgene_pmu_write_agentmsk(struct xgene_pmu_dev *pmu_dev, u32 val) 798 { 799 writel(val, pmu_dev->inf->csr + PMU_PMAMR0); 800 } 801 802 static inline void 803 xgene_pmu_v3_write_agentmsk(struct xgene_pmu_dev *pmu_dev, u32 val) { } 804 805 static inline void 806 xgene_pmu_write_agent1msk(struct xgene_pmu_dev *pmu_dev, u32 val) 807 { 808 writel(val, pmu_dev->inf->csr + PMU_PMAMR1); 809 } 810 811 static inline void 812 xgene_pmu_v3_write_agent1msk(struct xgene_pmu_dev *pmu_dev, u32 val) { } 813 814 static inline void 815 xgene_pmu_enable_counter(struct xgene_pmu_dev *pmu_dev, int idx) 816 { 817 u32 val; 818 819 val = readl(pmu_dev->inf->csr + PMU_PMCNTENSET); 820 val |= 1 << idx; 821 writel(val, pmu_dev->inf->csr + PMU_PMCNTENSET); 822 } 823 824 static inline void 825 xgene_pmu_disable_counter(struct xgene_pmu_dev *pmu_dev, int idx) 826 { 827 u32 val; 828 829 val = readl(pmu_dev->inf->csr + PMU_PMCNTENCLR); 830 val |= 1 << idx; 831 writel(val, pmu_dev->inf->csr + PMU_PMCNTENCLR); 832 } 833 834 static inline void 835 xgene_pmu_enable_counter_int(struct xgene_pmu_dev *pmu_dev, int idx) 836 { 837 u32 val; 838 839 val = readl(pmu_dev->inf->csr + PMU_PMINTENSET); 840 val |= 1 << idx; 841 writel(val, pmu_dev->inf->csr + PMU_PMINTENSET); 842 } 843 844 static inline void 845 xgene_pmu_disable_counter_int(struct xgene_pmu_dev *pmu_dev, int idx) 846 { 847 u32 val; 848 849 val = readl(pmu_dev->inf->csr + PMU_PMINTENCLR); 850 val |= 1 << idx; 851 writel(val, pmu_dev->inf->csr + PMU_PMINTENCLR); 852 } 853 854 static inline void xgene_pmu_reset_counters(struct xgene_pmu_dev *pmu_dev) 855 { 856 u32 val; 857 858 val = readl(pmu_dev->inf->csr + PMU_PMCR); 859 val |= PMU_PMCR_P; 860 writel(val, pmu_dev->inf->csr + PMU_PMCR); 861 } 862 863 static inline void xgene_pmu_start_counters(struct xgene_pmu_dev *pmu_dev) 864 { 865 u32 val; 866 867 val = readl(pmu_dev->inf->csr + PMU_PMCR); 868 val |= PMU_PMCR_E; 869 writel(val, pmu_dev->inf->csr + PMU_PMCR); 870 } 871 872 static inline void xgene_pmu_stop_counters(struct xgene_pmu_dev *pmu_dev) 873 { 874 u32 val; 875 876 val = readl(pmu_dev->inf->csr + PMU_PMCR); 877 val &= ~PMU_PMCR_E; 878 writel(val, pmu_dev->inf->csr + PMU_PMCR); 879 } 880 881 static void xgene_perf_pmu_enable(struct pmu *pmu) 882 { 883 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(pmu); 884 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 885 int enabled = bitmap_weight(pmu_dev->cntr_assign_mask, 886 pmu_dev->max_counters); 887 888 if (!enabled) 889 return; 890 891 xgene_pmu->ops->start_counters(pmu_dev); 892 } 893 894 static void xgene_perf_pmu_disable(struct pmu *pmu) 895 { 896 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(pmu); 897 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 898 899 xgene_pmu->ops->stop_counters(pmu_dev); 900 } 901 902 static int xgene_perf_event_init(struct perf_event *event) 903 { 904 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 905 struct hw_perf_event *hw = &event->hw; 906 struct perf_event *sibling; 907 908 /* Test the event attr type check for PMU enumeration */ 909 if (event->attr.type != event->pmu->type) 910 return -ENOENT; 911 912 /* 913 * SOC PMU counters are shared across all cores. 914 * Therefore, it does not support per-process mode. 915 * Also, it does not support event sampling mode. 916 */ 917 if (is_sampling_event(event) || event->attach_state & PERF_ATTACH_TASK) 918 return -EINVAL; 919 920 if (event->cpu < 0) 921 return -EINVAL; 922 /* 923 * Many perf core operations (eg. events rotation) operate on a 924 * single CPU context. This is obvious for CPU PMUs, where one 925 * expects the same sets of events being observed on all CPUs, 926 * but can lead to issues for off-core PMUs, where each 927 * event could be theoretically assigned to a different CPU. To 928 * mitigate this, we enforce CPU assignment to one, selected 929 * processor (the one described in the "cpumask" attribute). 930 */ 931 event->cpu = cpumask_first(&pmu_dev->parent->cpu); 932 933 hw->config = event->attr.config; 934 /* 935 * Each bit of the config1 field represents an agent from which the 936 * request of the event come. The event is counted only if it's caused 937 * by a request of an agent has the bit cleared. 938 * By default, the event is counted for all agents. 939 */ 940 hw->config_base = event->attr.config1; 941 942 /* 943 * We must NOT create groups containing mixed PMUs, although software 944 * events are acceptable 945 */ 946 if (event->group_leader->pmu != event->pmu && 947 !is_software_event(event->group_leader)) 948 return -EINVAL; 949 950 for_each_sibling_event(sibling, event->group_leader) { 951 if (sibling->pmu != event->pmu && 952 !is_software_event(sibling)) 953 return -EINVAL; 954 } 955 956 return 0; 957 } 958 959 static void xgene_perf_enable_event(struct perf_event *event) 960 { 961 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 962 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 963 964 xgene_pmu->ops->write_evttype(pmu_dev, GET_CNTR(event), 965 GET_EVENTID(event)); 966 xgene_pmu->ops->write_agentmsk(pmu_dev, ~((u32)GET_AGENTID(event))); 967 if (pmu_dev->inf->type == PMU_TYPE_IOB) 968 xgene_pmu->ops->write_agent1msk(pmu_dev, 969 ~((u32)GET_AGENT1ID(event))); 970 971 xgene_pmu->ops->enable_counter(pmu_dev, GET_CNTR(event)); 972 xgene_pmu->ops->enable_counter_int(pmu_dev, GET_CNTR(event)); 973 } 974 975 static void xgene_perf_disable_event(struct perf_event *event) 976 { 977 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 978 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 979 980 xgene_pmu->ops->disable_counter(pmu_dev, GET_CNTR(event)); 981 xgene_pmu->ops->disable_counter_int(pmu_dev, GET_CNTR(event)); 982 } 983 984 static void xgene_perf_event_set_period(struct perf_event *event) 985 { 986 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 987 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 988 struct hw_perf_event *hw = &event->hw; 989 /* 990 * For 32 bit counter, it has a period of 2^32. To account for the 991 * possibility of extreme interrupt latency we program for a period of 992 * half that. Hopefully, we can handle the interrupt before another 2^31 993 * events occur and the counter overtakes its previous value. 994 * For 64 bit counter, we don't expect it overflow. 995 */ 996 u64 val = 1ULL << 31; 997 998 local64_set(&hw->prev_count, val); 999 xgene_pmu->ops->write_counter(pmu_dev, hw->idx, val); 1000 } 1001 1002 static void xgene_perf_event_update(struct perf_event *event) 1003 { 1004 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 1005 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 1006 struct hw_perf_event *hw = &event->hw; 1007 u64 delta, prev_raw_count, new_raw_count; 1008 1009 again: 1010 prev_raw_count = local64_read(&hw->prev_count); 1011 new_raw_count = xgene_pmu->ops->read_counter(pmu_dev, GET_CNTR(event)); 1012 1013 if (local64_cmpxchg(&hw->prev_count, prev_raw_count, 1014 new_raw_count) != prev_raw_count) 1015 goto again; 1016 1017 delta = (new_raw_count - prev_raw_count) & pmu_dev->max_period; 1018 1019 local64_add(delta, &event->count); 1020 } 1021 1022 static void xgene_perf_read(struct perf_event *event) 1023 { 1024 xgene_perf_event_update(event); 1025 } 1026 1027 static void xgene_perf_start(struct perf_event *event, int flags) 1028 { 1029 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 1030 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 1031 struct hw_perf_event *hw = &event->hw; 1032 1033 if (WARN_ON_ONCE(!(hw->state & PERF_HES_STOPPED))) 1034 return; 1035 1036 WARN_ON_ONCE(!(hw->state & PERF_HES_UPTODATE)); 1037 hw->state = 0; 1038 1039 xgene_perf_event_set_period(event); 1040 1041 if (flags & PERF_EF_RELOAD) { 1042 u64 prev_raw_count = local64_read(&hw->prev_count); 1043 1044 xgene_pmu->ops->write_counter(pmu_dev, GET_CNTR(event), 1045 prev_raw_count); 1046 } 1047 1048 xgene_perf_enable_event(event); 1049 perf_event_update_userpage(event); 1050 } 1051 1052 static void xgene_perf_stop(struct perf_event *event, int flags) 1053 { 1054 struct hw_perf_event *hw = &event->hw; 1055 u64 config; 1056 1057 if (hw->state & PERF_HES_UPTODATE) 1058 return; 1059 1060 xgene_perf_disable_event(event); 1061 WARN_ON_ONCE(hw->state & PERF_HES_STOPPED); 1062 hw->state |= PERF_HES_STOPPED; 1063 1064 if (hw->state & PERF_HES_UPTODATE) 1065 return; 1066 1067 config = hw->config; 1068 xgene_perf_read(event); 1069 hw->state |= PERF_HES_UPTODATE; 1070 } 1071 1072 static int xgene_perf_add(struct perf_event *event, int flags) 1073 { 1074 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 1075 struct hw_perf_event *hw = &event->hw; 1076 1077 hw->state = PERF_HES_UPTODATE | PERF_HES_STOPPED; 1078 1079 /* Allocate an event counter */ 1080 hw->idx = get_next_avail_cntr(pmu_dev); 1081 if (hw->idx < 0) 1082 return -EAGAIN; 1083 1084 /* Update counter event pointer for Interrupt handler */ 1085 pmu_dev->pmu_counter_event[hw->idx] = event; 1086 1087 if (flags & PERF_EF_START) 1088 xgene_perf_start(event, PERF_EF_RELOAD); 1089 1090 return 0; 1091 } 1092 1093 static void xgene_perf_del(struct perf_event *event, int flags) 1094 { 1095 struct xgene_pmu_dev *pmu_dev = to_pmu_dev(event->pmu); 1096 struct hw_perf_event *hw = &event->hw; 1097 1098 xgene_perf_stop(event, PERF_EF_UPDATE); 1099 1100 /* clear the assigned counter */ 1101 clear_avail_cntr(pmu_dev, GET_CNTR(event)); 1102 1103 perf_event_update_userpage(event); 1104 pmu_dev->pmu_counter_event[hw->idx] = NULL; 1105 } 1106 1107 static int xgene_init_perf(struct xgene_pmu_dev *pmu_dev, char *name) 1108 { 1109 struct xgene_pmu *xgene_pmu; 1110 1111 if (pmu_dev->parent->version == PCP_PMU_V3) 1112 pmu_dev->max_period = PMU_V3_CNT_MAX_PERIOD; 1113 else 1114 pmu_dev->max_period = PMU_CNT_MAX_PERIOD; 1115 /* First version PMU supports only single event counter */ 1116 xgene_pmu = pmu_dev->parent; 1117 if (xgene_pmu->version == PCP_PMU_V1) 1118 pmu_dev->max_counters = 1; 1119 else 1120 pmu_dev->max_counters = PMU_MAX_COUNTERS; 1121 1122 /* Perf driver registration */ 1123 pmu_dev->pmu = (struct pmu) { 1124 .attr_groups = pmu_dev->attr_groups, 1125 .task_ctx_nr = perf_invalid_context, 1126 .pmu_enable = xgene_perf_pmu_enable, 1127 .pmu_disable = xgene_perf_pmu_disable, 1128 .event_init = xgene_perf_event_init, 1129 .add = xgene_perf_add, 1130 .del = xgene_perf_del, 1131 .start = xgene_perf_start, 1132 .stop = xgene_perf_stop, 1133 .read = xgene_perf_read, 1134 .capabilities = PERF_PMU_CAP_NO_EXCLUDE, 1135 }; 1136 1137 /* Hardware counter init */ 1138 xgene_pmu->ops->stop_counters(pmu_dev); 1139 xgene_pmu->ops->reset_counters(pmu_dev); 1140 1141 return perf_pmu_register(&pmu_dev->pmu, name, -1); 1142 } 1143 1144 static int 1145 xgene_pmu_dev_add(struct xgene_pmu *xgene_pmu, struct xgene_pmu_dev_ctx *ctx) 1146 { 1147 struct device *dev = xgene_pmu->dev; 1148 struct xgene_pmu_dev *pmu; 1149 1150 pmu = devm_kzalloc(dev, sizeof(*pmu), GFP_KERNEL); 1151 if (!pmu) 1152 return -ENOMEM; 1153 pmu->parent = xgene_pmu; 1154 pmu->inf = &ctx->inf; 1155 ctx->pmu_dev = pmu; 1156 1157 switch (pmu->inf->type) { 1158 case PMU_TYPE_L3C: 1159 if (!(xgene_pmu->l3c_active_mask & pmu->inf->enable_mask)) 1160 return -ENODEV; 1161 if (xgene_pmu->version == PCP_PMU_V3) 1162 pmu->attr_groups = l3c_pmu_v3_attr_groups; 1163 else 1164 pmu->attr_groups = l3c_pmu_attr_groups; 1165 break; 1166 case PMU_TYPE_IOB: 1167 if (xgene_pmu->version == PCP_PMU_V3) 1168 pmu->attr_groups = iob_fast_pmu_v3_attr_groups; 1169 else 1170 pmu->attr_groups = iob_pmu_attr_groups; 1171 break; 1172 case PMU_TYPE_IOB_SLOW: 1173 if (xgene_pmu->version == PCP_PMU_V3) 1174 pmu->attr_groups = iob_slow_pmu_v3_attr_groups; 1175 break; 1176 case PMU_TYPE_MCB: 1177 if (!(xgene_pmu->mcb_active_mask & pmu->inf->enable_mask)) 1178 return -ENODEV; 1179 if (xgene_pmu->version == PCP_PMU_V3) 1180 pmu->attr_groups = mcb_pmu_v3_attr_groups; 1181 else 1182 pmu->attr_groups = mcb_pmu_attr_groups; 1183 break; 1184 case PMU_TYPE_MC: 1185 if (!(xgene_pmu->mc_active_mask & pmu->inf->enable_mask)) 1186 return -ENODEV; 1187 if (xgene_pmu->version == PCP_PMU_V3) 1188 pmu->attr_groups = mc_pmu_v3_attr_groups; 1189 else 1190 pmu->attr_groups = mc_pmu_attr_groups; 1191 break; 1192 default: 1193 return -EINVAL; 1194 } 1195 1196 if (xgene_init_perf(pmu, ctx->name)) { 1197 dev_err(dev, "%s PMU: Failed to init perf driver\n", ctx->name); 1198 return -ENODEV; 1199 } 1200 1201 dev_info(dev, "%s PMU registered\n", ctx->name); 1202 1203 return 0; 1204 } 1205 1206 static void _xgene_pmu_isr(int irq, struct xgene_pmu_dev *pmu_dev) 1207 { 1208 struct xgene_pmu *xgene_pmu = pmu_dev->parent; 1209 void __iomem *csr = pmu_dev->inf->csr; 1210 u32 pmovsr; 1211 int idx; 1212 1213 xgene_pmu->ops->stop_counters(pmu_dev); 1214 1215 if (xgene_pmu->version == PCP_PMU_V3) 1216 pmovsr = readl(csr + PMU_PMOVSSET) & PMU_OVERFLOW_MASK; 1217 else 1218 pmovsr = readl(csr + PMU_PMOVSR) & PMU_OVERFLOW_MASK; 1219 1220 if (!pmovsr) 1221 goto out; 1222 1223 /* Clear interrupt flag */ 1224 if (xgene_pmu->version == PCP_PMU_V1) 1225 writel(0x0, csr + PMU_PMOVSR); 1226 else if (xgene_pmu->version == PCP_PMU_V2) 1227 writel(pmovsr, csr + PMU_PMOVSR); 1228 else 1229 writel(pmovsr, csr + PMU_PMOVSCLR); 1230 1231 for (idx = 0; idx < PMU_MAX_COUNTERS; idx++) { 1232 struct perf_event *event = pmu_dev->pmu_counter_event[idx]; 1233 int overflowed = pmovsr & BIT(idx); 1234 1235 /* Ignore if we don't have an event. */ 1236 if (!event || !overflowed) 1237 continue; 1238 xgene_perf_event_update(event); 1239 xgene_perf_event_set_period(event); 1240 } 1241 1242 out: 1243 xgene_pmu->ops->start_counters(pmu_dev); 1244 } 1245 1246 static irqreturn_t xgene_pmu_isr(int irq, void *dev_id) 1247 { 1248 u32 intr_mcu, intr_mcb, intr_l3c, intr_iob; 1249 struct xgene_pmu_dev_ctx *ctx; 1250 struct xgene_pmu *xgene_pmu = dev_id; 1251 unsigned long flags; 1252 u32 val; 1253 1254 raw_spin_lock_irqsave(&xgene_pmu->lock, flags); 1255 1256 /* Get Interrupt PMU source */ 1257 val = readl(xgene_pmu->pcppmu_csr + PCPPMU_INTSTATUS_REG); 1258 if (xgene_pmu->version == PCP_PMU_V3) { 1259 intr_mcu = PCPPMU_V3_INT_MCU; 1260 intr_mcb = PCPPMU_V3_INT_MCB; 1261 intr_l3c = PCPPMU_V3_INT_L3C; 1262 intr_iob = PCPPMU_V3_INT_IOB; 1263 } else { 1264 intr_mcu = PCPPMU_INT_MCU; 1265 intr_mcb = PCPPMU_INT_MCB; 1266 intr_l3c = PCPPMU_INT_L3C; 1267 intr_iob = PCPPMU_INT_IOB; 1268 } 1269 if (val & intr_mcu) { 1270 list_for_each_entry(ctx, &xgene_pmu->mcpmus, next) { 1271 _xgene_pmu_isr(irq, ctx->pmu_dev); 1272 } 1273 } 1274 if (val & intr_mcb) { 1275 list_for_each_entry(ctx, &xgene_pmu->mcbpmus, next) { 1276 _xgene_pmu_isr(irq, ctx->pmu_dev); 1277 } 1278 } 1279 if (val & intr_l3c) { 1280 list_for_each_entry(ctx, &xgene_pmu->l3cpmus, next) { 1281 _xgene_pmu_isr(irq, ctx->pmu_dev); 1282 } 1283 } 1284 if (val & intr_iob) { 1285 list_for_each_entry(ctx, &xgene_pmu->iobpmus, next) { 1286 _xgene_pmu_isr(irq, ctx->pmu_dev); 1287 } 1288 } 1289 1290 raw_spin_unlock_irqrestore(&xgene_pmu->lock, flags); 1291 1292 return IRQ_HANDLED; 1293 } 1294 1295 static int acpi_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu, 1296 struct platform_device *pdev) 1297 { 1298 void __iomem *csw_csr, *mcba_csr, *mcbb_csr; 1299 struct resource *res; 1300 unsigned int reg; 1301 1302 res = platform_get_resource(pdev, IORESOURCE_MEM, 1); 1303 csw_csr = devm_ioremap_resource(&pdev->dev, res); 1304 if (IS_ERR(csw_csr)) { 1305 dev_err(&pdev->dev, "ioremap failed for CSW CSR resource\n"); 1306 return PTR_ERR(csw_csr); 1307 } 1308 1309 res = platform_get_resource(pdev, IORESOURCE_MEM, 2); 1310 mcba_csr = devm_ioremap_resource(&pdev->dev, res); 1311 if (IS_ERR(mcba_csr)) { 1312 dev_err(&pdev->dev, "ioremap failed for MCBA CSR resource\n"); 1313 return PTR_ERR(mcba_csr); 1314 } 1315 1316 res = platform_get_resource(pdev, IORESOURCE_MEM, 3); 1317 mcbb_csr = devm_ioremap_resource(&pdev->dev, res); 1318 if (IS_ERR(mcbb_csr)) { 1319 dev_err(&pdev->dev, "ioremap failed for MCBB CSR resource\n"); 1320 return PTR_ERR(mcbb_csr); 1321 } 1322 1323 xgene_pmu->l3c_active_mask = 0x1; 1324 1325 reg = readl(csw_csr + CSW_CSWCR); 1326 if (reg & CSW_CSWCR_DUALMCB_MASK) { 1327 /* Dual MCB active */ 1328 xgene_pmu->mcb_active_mask = 0x3; 1329 /* Probe all active MC(s) */ 1330 reg = readl(mcbb_csr + CSW_CSWCR); 1331 xgene_pmu->mc_active_mask = 1332 (reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0xF : 0x5; 1333 } else { 1334 /* Single MCB active */ 1335 xgene_pmu->mcb_active_mask = 0x1; 1336 /* Probe all active MC(s) */ 1337 reg = readl(mcba_csr + CSW_CSWCR); 1338 xgene_pmu->mc_active_mask = 1339 (reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0x3 : 0x1; 1340 } 1341 1342 return 0; 1343 } 1344 1345 static int acpi_pmu_v3_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu, 1346 struct platform_device *pdev) 1347 { 1348 void __iomem *csw_csr; 1349 struct resource *res; 1350 unsigned int reg; 1351 u32 mcb0routing; 1352 u32 mcb1routing; 1353 1354 res = platform_get_resource(pdev, IORESOURCE_MEM, 1); 1355 csw_csr = devm_ioremap_resource(&pdev->dev, res); 1356 if (IS_ERR(csw_csr)) { 1357 dev_err(&pdev->dev, "ioremap failed for CSW CSR resource\n"); 1358 return PTR_ERR(csw_csr); 1359 } 1360 1361 reg = readl(csw_csr + CSW_CSWCR); 1362 mcb0routing = CSW_CSWCR_MCB0_ROUTING(reg); 1363 mcb1routing = CSW_CSWCR_MCB1_ROUTING(reg); 1364 if (reg & CSW_CSWCR_DUALMCB_MASK) { 1365 /* Dual MCB active */ 1366 xgene_pmu->mcb_active_mask = 0x3; 1367 /* Probe all active L3C(s), maximum is 8 */ 1368 xgene_pmu->l3c_active_mask = 0xFF; 1369 /* Probe all active MC(s), maximum is 8 */ 1370 if ((mcb0routing == 0x2) && (mcb1routing == 0x2)) 1371 xgene_pmu->mc_active_mask = 0xFF; 1372 else if ((mcb0routing == 0x1) && (mcb1routing == 0x1)) 1373 xgene_pmu->mc_active_mask = 0x33; 1374 else 1375 xgene_pmu->mc_active_mask = 0x11; 1376 } else { 1377 /* Single MCB active */ 1378 xgene_pmu->mcb_active_mask = 0x1; 1379 /* Probe all active L3C(s), maximum is 4 */ 1380 xgene_pmu->l3c_active_mask = 0x0F; 1381 /* Probe all active MC(s), maximum is 4 */ 1382 if (mcb0routing == 0x2) 1383 xgene_pmu->mc_active_mask = 0x0F; 1384 else if (mcb0routing == 0x1) 1385 xgene_pmu->mc_active_mask = 0x03; 1386 else 1387 xgene_pmu->mc_active_mask = 0x01; 1388 } 1389 1390 return 0; 1391 } 1392 1393 static int fdt_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu, 1394 struct platform_device *pdev) 1395 { 1396 struct regmap *csw_map, *mcba_map, *mcbb_map; 1397 struct device_node *np = pdev->dev.of_node; 1398 unsigned int reg; 1399 1400 csw_map = syscon_regmap_lookup_by_phandle(np, "regmap-csw"); 1401 if (IS_ERR(csw_map)) { 1402 dev_err(&pdev->dev, "unable to get syscon regmap csw\n"); 1403 return PTR_ERR(csw_map); 1404 } 1405 1406 mcba_map = syscon_regmap_lookup_by_phandle(np, "regmap-mcba"); 1407 if (IS_ERR(mcba_map)) { 1408 dev_err(&pdev->dev, "unable to get syscon regmap mcba\n"); 1409 return PTR_ERR(mcba_map); 1410 } 1411 1412 mcbb_map = syscon_regmap_lookup_by_phandle(np, "regmap-mcbb"); 1413 if (IS_ERR(mcbb_map)) { 1414 dev_err(&pdev->dev, "unable to get syscon regmap mcbb\n"); 1415 return PTR_ERR(mcbb_map); 1416 } 1417 1418 xgene_pmu->l3c_active_mask = 0x1; 1419 if (regmap_read(csw_map, CSW_CSWCR, ®)) 1420 return -EINVAL; 1421 1422 if (reg & CSW_CSWCR_DUALMCB_MASK) { 1423 /* Dual MCB active */ 1424 xgene_pmu->mcb_active_mask = 0x3; 1425 /* Probe all active MC(s) */ 1426 if (regmap_read(mcbb_map, MCBADDRMR, ®)) 1427 return 0; 1428 xgene_pmu->mc_active_mask = 1429 (reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0xF : 0x5; 1430 } else { 1431 /* Single MCB active */ 1432 xgene_pmu->mcb_active_mask = 0x1; 1433 /* Probe all active MC(s) */ 1434 if (regmap_read(mcba_map, MCBADDRMR, ®)) 1435 return 0; 1436 xgene_pmu->mc_active_mask = 1437 (reg & MCBADDRMR_DUALMCU_MODE_MASK) ? 0x3 : 0x1; 1438 } 1439 1440 return 0; 1441 } 1442 1443 static int xgene_pmu_probe_active_mcb_mcu_l3c(struct xgene_pmu *xgene_pmu, 1444 struct platform_device *pdev) 1445 { 1446 if (has_acpi_companion(&pdev->dev)) { 1447 if (xgene_pmu->version == PCP_PMU_V3) 1448 return acpi_pmu_v3_probe_active_mcb_mcu_l3c(xgene_pmu, 1449 pdev); 1450 else 1451 return acpi_pmu_probe_active_mcb_mcu_l3c(xgene_pmu, 1452 pdev); 1453 } 1454 return fdt_pmu_probe_active_mcb_mcu_l3c(xgene_pmu, pdev); 1455 } 1456 1457 static char *xgene_pmu_dev_name(struct device *dev, u32 type, int id) 1458 { 1459 switch (type) { 1460 case PMU_TYPE_L3C: 1461 return devm_kasprintf(dev, GFP_KERNEL, "l3c%d", id); 1462 case PMU_TYPE_IOB: 1463 return devm_kasprintf(dev, GFP_KERNEL, "iob%d", id); 1464 case PMU_TYPE_IOB_SLOW: 1465 return devm_kasprintf(dev, GFP_KERNEL, "iob_slow%d", id); 1466 case PMU_TYPE_MCB: 1467 return devm_kasprintf(dev, GFP_KERNEL, "mcb%d", id); 1468 case PMU_TYPE_MC: 1469 return devm_kasprintf(dev, GFP_KERNEL, "mc%d", id); 1470 default: 1471 return devm_kasprintf(dev, GFP_KERNEL, "unknown"); 1472 } 1473 } 1474 1475 #if defined(CONFIG_ACPI) 1476 static int acpi_pmu_dev_add_resource(struct acpi_resource *ares, void *data) 1477 { 1478 struct resource *res = data; 1479 1480 if (ares->type == ACPI_RESOURCE_TYPE_FIXED_MEMORY32) 1481 acpi_dev_resource_memory(ares, res); 1482 1483 /* Always tell the ACPI core to skip this resource */ 1484 return 1; 1485 } 1486 1487 static struct 1488 xgene_pmu_dev_ctx *acpi_get_pmu_hw_inf(struct xgene_pmu *xgene_pmu, 1489 struct acpi_device *adev, u32 type) 1490 { 1491 struct device *dev = xgene_pmu->dev; 1492 struct list_head resource_list; 1493 struct xgene_pmu_dev_ctx *ctx; 1494 const union acpi_object *obj; 1495 struct hw_pmu_info *inf; 1496 void __iomem *dev_csr; 1497 struct resource res; 1498 int enable_bit; 1499 int rc; 1500 1501 ctx = devm_kzalloc(dev, sizeof(*ctx), GFP_KERNEL); 1502 if (!ctx) 1503 return NULL; 1504 1505 INIT_LIST_HEAD(&resource_list); 1506 rc = acpi_dev_get_resources(adev, &resource_list, 1507 acpi_pmu_dev_add_resource, &res); 1508 acpi_dev_free_resource_list(&resource_list); 1509 if (rc < 0) { 1510 dev_err(dev, "PMU type %d: No resource address found\n", type); 1511 return NULL; 1512 } 1513 1514 dev_csr = devm_ioremap_resource(dev, &res); 1515 if (IS_ERR(dev_csr)) { 1516 dev_err(dev, "PMU type %d: Fail to map resource\n", type); 1517 return NULL; 1518 } 1519 1520 /* A PMU device node without enable-bit-index is always enabled */ 1521 rc = acpi_dev_get_property(adev, "enable-bit-index", 1522 ACPI_TYPE_INTEGER, &obj); 1523 if (rc < 0) 1524 enable_bit = 0; 1525 else 1526 enable_bit = (int) obj->integer.value; 1527 1528 ctx->name = xgene_pmu_dev_name(dev, type, enable_bit); 1529 if (!ctx->name) { 1530 dev_err(dev, "PMU type %d: Fail to get device name\n", type); 1531 return NULL; 1532 } 1533 inf = &ctx->inf; 1534 inf->type = type; 1535 inf->csr = dev_csr; 1536 inf->enable_mask = 1 << enable_bit; 1537 1538 return ctx; 1539 } 1540 1541 static const struct acpi_device_id xgene_pmu_acpi_type_match[] = { 1542 {"APMC0D5D", PMU_TYPE_L3C}, 1543 {"APMC0D5E", PMU_TYPE_IOB}, 1544 {"APMC0D5F", PMU_TYPE_MCB}, 1545 {"APMC0D60", PMU_TYPE_MC}, 1546 {"APMC0D84", PMU_TYPE_L3C}, 1547 {"APMC0D85", PMU_TYPE_IOB}, 1548 {"APMC0D86", PMU_TYPE_IOB_SLOW}, 1549 {"APMC0D87", PMU_TYPE_MCB}, 1550 {"APMC0D88", PMU_TYPE_MC}, 1551 {}, 1552 }; 1553 1554 static const struct acpi_device_id *xgene_pmu_acpi_match_type( 1555 const struct acpi_device_id *ids, 1556 struct acpi_device *adev) 1557 { 1558 const struct acpi_device_id *match_id = NULL; 1559 const struct acpi_device_id *id; 1560 1561 for (id = ids; id->id[0] || id->cls; id++) { 1562 if (!acpi_match_device_ids(adev, id)) 1563 match_id = id; 1564 else if (match_id) 1565 break; 1566 } 1567 1568 return match_id; 1569 } 1570 1571 static acpi_status acpi_pmu_dev_add(acpi_handle handle, u32 level, 1572 void *data, void **return_value) 1573 { 1574 const struct acpi_device_id *acpi_id; 1575 struct xgene_pmu *xgene_pmu = data; 1576 struct xgene_pmu_dev_ctx *ctx; 1577 struct acpi_device *adev; 1578 1579 if (acpi_bus_get_device(handle, &adev)) 1580 return AE_OK; 1581 if (acpi_bus_get_status(adev) || !adev->status.present) 1582 return AE_OK; 1583 1584 acpi_id = xgene_pmu_acpi_match_type(xgene_pmu_acpi_type_match, adev); 1585 if (!acpi_id) 1586 return AE_OK; 1587 1588 ctx = acpi_get_pmu_hw_inf(xgene_pmu, adev, (u32)acpi_id->driver_data); 1589 if (!ctx) 1590 return AE_OK; 1591 1592 if (xgene_pmu_dev_add(xgene_pmu, ctx)) { 1593 /* Can't add the PMU device, skip it */ 1594 devm_kfree(xgene_pmu->dev, ctx); 1595 return AE_OK; 1596 } 1597 1598 switch (ctx->inf.type) { 1599 case PMU_TYPE_L3C: 1600 list_add(&ctx->next, &xgene_pmu->l3cpmus); 1601 break; 1602 case PMU_TYPE_IOB: 1603 list_add(&ctx->next, &xgene_pmu->iobpmus); 1604 break; 1605 case PMU_TYPE_IOB_SLOW: 1606 list_add(&ctx->next, &xgene_pmu->iobpmus); 1607 break; 1608 case PMU_TYPE_MCB: 1609 list_add(&ctx->next, &xgene_pmu->mcbpmus); 1610 break; 1611 case PMU_TYPE_MC: 1612 list_add(&ctx->next, &xgene_pmu->mcpmus); 1613 break; 1614 } 1615 return AE_OK; 1616 } 1617 1618 static int acpi_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu, 1619 struct platform_device *pdev) 1620 { 1621 struct device *dev = xgene_pmu->dev; 1622 acpi_handle handle; 1623 acpi_status status; 1624 1625 handle = ACPI_HANDLE(dev); 1626 if (!handle) 1627 return -EINVAL; 1628 1629 status = acpi_walk_namespace(ACPI_TYPE_DEVICE, handle, 1, 1630 acpi_pmu_dev_add, NULL, xgene_pmu, NULL); 1631 if (ACPI_FAILURE(status)) { 1632 dev_err(dev, "failed to probe PMU devices\n"); 1633 return -ENODEV; 1634 } 1635 1636 return 0; 1637 } 1638 #else 1639 static int acpi_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu, 1640 struct platform_device *pdev) 1641 { 1642 return 0; 1643 } 1644 #endif 1645 1646 static struct 1647 xgene_pmu_dev_ctx *fdt_get_pmu_hw_inf(struct xgene_pmu *xgene_pmu, 1648 struct device_node *np, u32 type) 1649 { 1650 struct device *dev = xgene_pmu->dev; 1651 struct xgene_pmu_dev_ctx *ctx; 1652 struct hw_pmu_info *inf; 1653 void __iomem *dev_csr; 1654 struct resource res; 1655 int enable_bit; 1656 1657 ctx = devm_kzalloc(dev, sizeof(*ctx), GFP_KERNEL); 1658 if (!ctx) 1659 return NULL; 1660 1661 if (of_address_to_resource(np, 0, &res) < 0) { 1662 dev_err(dev, "PMU type %d: No resource address found\n", type); 1663 return NULL; 1664 } 1665 1666 dev_csr = devm_ioremap_resource(dev, &res); 1667 if (IS_ERR(dev_csr)) { 1668 dev_err(dev, "PMU type %d: Fail to map resource\n", type); 1669 return NULL; 1670 } 1671 1672 /* A PMU device node without enable-bit-index is always enabled */ 1673 if (of_property_read_u32(np, "enable-bit-index", &enable_bit)) 1674 enable_bit = 0; 1675 1676 ctx->name = xgene_pmu_dev_name(dev, type, enable_bit); 1677 if (!ctx->name) { 1678 dev_err(dev, "PMU type %d: Fail to get device name\n", type); 1679 return NULL; 1680 } 1681 1682 inf = &ctx->inf; 1683 inf->type = type; 1684 inf->csr = dev_csr; 1685 inf->enable_mask = 1 << enable_bit; 1686 1687 return ctx; 1688 } 1689 1690 static int fdt_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu, 1691 struct platform_device *pdev) 1692 { 1693 struct xgene_pmu_dev_ctx *ctx; 1694 struct device_node *np; 1695 1696 for_each_child_of_node(pdev->dev.of_node, np) { 1697 if (!of_device_is_available(np)) 1698 continue; 1699 1700 if (of_device_is_compatible(np, "apm,xgene-pmu-l3c")) 1701 ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_L3C); 1702 else if (of_device_is_compatible(np, "apm,xgene-pmu-iob")) 1703 ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_IOB); 1704 else if (of_device_is_compatible(np, "apm,xgene-pmu-mcb")) 1705 ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_MCB); 1706 else if (of_device_is_compatible(np, "apm,xgene-pmu-mc")) 1707 ctx = fdt_get_pmu_hw_inf(xgene_pmu, np, PMU_TYPE_MC); 1708 else 1709 ctx = NULL; 1710 1711 if (!ctx) 1712 continue; 1713 1714 if (xgene_pmu_dev_add(xgene_pmu, ctx)) { 1715 /* Can't add the PMU device, skip it */ 1716 devm_kfree(xgene_pmu->dev, ctx); 1717 continue; 1718 } 1719 1720 switch (ctx->inf.type) { 1721 case PMU_TYPE_L3C: 1722 list_add(&ctx->next, &xgene_pmu->l3cpmus); 1723 break; 1724 case PMU_TYPE_IOB: 1725 list_add(&ctx->next, &xgene_pmu->iobpmus); 1726 break; 1727 case PMU_TYPE_IOB_SLOW: 1728 list_add(&ctx->next, &xgene_pmu->iobpmus); 1729 break; 1730 case PMU_TYPE_MCB: 1731 list_add(&ctx->next, &xgene_pmu->mcbpmus); 1732 break; 1733 case PMU_TYPE_MC: 1734 list_add(&ctx->next, &xgene_pmu->mcpmus); 1735 break; 1736 } 1737 } 1738 1739 return 0; 1740 } 1741 1742 static int xgene_pmu_probe_pmu_dev(struct xgene_pmu *xgene_pmu, 1743 struct platform_device *pdev) 1744 { 1745 if (has_acpi_companion(&pdev->dev)) 1746 return acpi_pmu_probe_pmu_dev(xgene_pmu, pdev); 1747 return fdt_pmu_probe_pmu_dev(xgene_pmu, pdev); 1748 } 1749 1750 static const struct xgene_pmu_data xgene_pmu_data = { 1751 .id = PCP_PMU_V1, 1752 }; 1753 1754 static const struct xgene_pmu_data xgene_pmu_v2_data = { 1755 .id = PCP_PMU_V2, 1756 }; 1757 1758 static const struct xgene_pmu_ops xgene_pmu_ops = { 1759 .mask_int = xgene_pmu_mask_int, 1760 .unmask_int = xgene_pmu_unmask_int, 1761 .read_counter = xgene_pmu_read_counter32, 1762 .write_counter = xgene_pmu_write_counter32, 1763 .write_evttype = xgene_pmu_write_evttype, 1764 .write_agentmsk = xgene_pmu_write_agentmsk, 1765 .write_agent1msk = xgene_pmu_write_agent1msk, 1766 .enable_counter = xgene_pmu_enable_counter, 1767 .disable_counter = xgene_pmu_disable_counter, 1768 .enable_counter_int = xgene_pmu_enable_counter_int, 1769 .disable_counter_int = xgene_pmu_disable_counter_int, 1770 .reset_counters = xgene_pmu_reset_counters, 1771 .start_counters = xgene_pmu_start_counters, 1772 .stop_counters = xgene_pmu_stop_counters, 1773 }; 1774 1775 static const struct xgene_pmu_ops xgene_pmu_v3_ops = { 1776 .mask_int = xgene_pmu_v3_mask_int, 1777 .unmask_int = xgene_pmu_v3_unmask_int, 1778 .read_counter = xgene_pmu_read_counter64, 1779 .write_counter = xgene_pmu_write_counter64, 1780 .write_evttype = xgene_pmu_write_evttype, 1781 .write_agentmsk = xgene_pmu_v3_write_agentmsk, 1782 .write_agent1msk = xgene_pmu_v3_write_agent1msk, 1783 .enable_counter = xgene_pmu_enable_counter, 1784 .disable_counter = xgene_pmu_disable_counter, 1785 .enable_counter_int = xgene_pmu_enable_counter_int, 1786 .disable_counter_int = xgene_pmu_disable_counter_int, 1787 .reset_counters = xgene_pmu_reset_counters, 1788 .start_counters = xgene_pmu_start_counters, 1789 .stop_counters = xgene_pmu_stop_counters, 1790 }; 1791 1792 static const struct of_device_id xgene_pmu_of_match[] = { 1793 { .compatible = "apm,xgene-pmu", .data = &xgene_pmu_data }, 1794 { .compatible = "apm,xgene-pmu-v2", .data = &xgene_pmu_v2_data }, 1795 {}, 1796 }; 1797 MODULE_DEVICE_TABLE(of, xgene_pmu_of_match); 1798 #ifdef CONFIG_ACPI 1799 static const struct acpi_device_id xgene_pmu_acpi_match[] = { 1800 {"APMC0D5B", PCP_PMU_V1}, 1801 {"APMC0D5C", PCP_PMU_V2}, 1802 {"APMC0D83", PCP_PMU_V3}, 1803 {}, 1804 }; 1805 MODULE_DEVICE_TABLE(acpi, xgene_pmu_acpi_match); 1806 #endif 1807 1808 static int xgene_pmu_online_cpu(unsigned int cpu, struct hlist_node *node) 1809 { 1810 struct xgene_pmu *xgene_pmu = hlist_entry_safe(node, struct xgene_pmu, 1811 node); 1812 1813 if (cpumask_empty(&xgene_pmu->cpu)) 1814 cpumask_set_cpu(cpu, &xgene_pmu->cpu); 1815 1816 /* Overflow interrupt also should use the same CPU */ 1817 WARN_ON(irq_set_affinity(xgene_pmu->irq, &xgene_pmu->cpu)); 1818 1819 return 0; 1820 } 1821 1822 static int xgene_pmu_offline_cpu(unsigned int cpu, struct hlist_node *node) 1823 { 1824 struct xgene_pmu *xgene_pmu = hlist_entry_safe(node, struct xgene_pmu, 1825 node); 1826 struct xgene_pmu_dev_ctx *ctx; 1827 unsigned int target; 1828 1829 if (!cpumask_test_and_clear_cpu(cpu, &xgene_pmu->cpu)) 1830 return 0; 1831 target = cpumask_any_but(cpu_online_mask, cpu); 1832 if (target >= nr_cpu_ids) 1833 return 0; 1834 1835 list_for_each_entry(ctx, &xgene_pmu->mcpmus, next) { 1836 perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target); 1837 } 1838 list_for_each_entry(ctx, &xgene_pmu->mcbpmus, next) { 1839 perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target); 1840 } 1841 list_for_each_entry(ctx, &xgene_pmu->l3cpmus, next) { 1842 perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target); 1843 } 1844 list_for_each_entry(ctx, &xgene_pmu->iobpmus, next) { 1845 perf_pmu_migrate_context(&ctx->pmu_dev->pmu, cpu, target); 1846 } 1847 1848 cpumask_set_cpu(target, &xgene_pmu->cpu); 1849 /* Overflow interrupt also should use the same CPU */ 1850 WARN_ON(irq_set_affinity(xgene_pmu->irq, &xgene_pmu->cpu)); 1851 1852 return 0; 1853 } 1854 1855 static int xgene_pmu_probe(struct platform_device *pdev) 1856 { 1857 const struct xgene_pmu_data *dev_data; 1858 const struct of_device_id *of_id; 1859 struct xgene_pmu *xgene_pmu; 1860 struct resource *res; 1861 int irq, rc; 1862 int version; 1863 1864 /* Install a hook to update the reader CPU in case it goes offline */ 1865 rc = cpuhp_setup_state_multi(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE, 1866 "CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE", 1867 xgene_pmu_online_cpu, 1868 xgene_pmu_offline_cpu); 1869 if (rc) 1870 return rc; 1871 1872 xgene_pmu = devm_kzalloc(&pdev->dev, sizeof(*xgene_pmu), GFP_KERNEL); 1873 if (!xgene_pmu) 1874 return -ENOMEM; 1875 xgene_pmu->dev = &pdev->dev; 1876 platform_set_drvdata(pdev, xgene_pmu); 1877 1878 version = -EINVAL; 1879 of_id = of_match_device(xgene_pmu_of_match, &pdev->dev); 1880 if (of_id) { 1881 dev_data = (const struct xgene_pmu_data *) of_id->data; 1882 version = dev_data->id; 1883 } 1884 1885 #ifdef CONFIG_ACPI 1886 if (ACPI_COMPANION(&pdev->dev)) { 1887 const struct acpi_device_id *acpi_id; 1888 1889 acpi_id = acpi_match_device(xgene_pmu_acpi_match, &pdev->dev); 1890 if (acpi_id) 1891 version = (int) acpi_id->driver_data; 1892 } 1893 #endif 1894 if (version < 0) 1895 return -ENODEV; 1896 1897 if (version == PCP_PMU_V3) 1898 xgene_pmu->ops = &xgene_pmu_v3_ops; 1899 else 1900 xgene_pmu->ops = &xgene_pmu_ops; 1901 1902 INIT_LIST_HEAD(&xgene_pmu->l3cpmus); 1903 INIT_LIST_HEAD(&xgene_pmu->iobpmus); 1904 INIT_LIST_HEAD(&xgene_pmu->mcbpmus); 1905 INIT_LIST_HEAD(&xgene_pmu->mcpmus); 1906 1907 xgene_pmu->version = version; 1908 dev_info(&pdev->dev, "X-Gene PMU version %d\n", xgene_pmu->version); 1909 1910 res = platform_get_resource(pdev, IORESOURCE_MEM, 0); 1911 xgene_pmu->pcppmu_csr = devm_ioremap_resource(&pdev->dev, res); 1912 if (IS_ERR(xgene_pmu->pcppmu_csr)) { 1913 dev_err(&pdev->dev, "ioremap failed for PCP PMU resource\n"); 1914 return PTR_ERR(xgene_pmu->pcppmu_csr); 1915 } 1916 1917 irq = platform_get_irq(pdev, 0); 1918 if (irq < 0) { 1919 dev_err(&pdev->dev, "No IRQ resource\n"); 1920 return -EINVAL; 1921 } 1922 1923 rc = devm_request_irq(&pdev->dev, irq, xgene_pmu_isr, 1924 IRQF_NOBALANCING | IRQF_NO_THREAD, 1925 dev_name(&pdev->dev), xgene_pmu); 1926 if (rc) { 1927 dev_err(&pdev->dev, "Could not request IRQ %d\n", irq); 1928 return rc; 1929 } 1930 1931 xgene_pmu->irq = irq; 1932 1933 raw_spin_lock_init(&xgene_pmu->lock); 1934 1935 /* Check for active MCBs and MCUs */ 1936 rc = xgene_pmu_probe_active_mcb_mcu_l3c(xgene_pmu, pdev); 1937 if (rc) { 1938 dev_warn(&pdev->dev, "Unknown MCB/MCU active status\n"); 1939 xgene_pmu->mcb_active_mask = 0x1; 1940 xgene_pmu->mc_active_mask = 0x1; 1941 } 1942 1943 /* Add this instance to the list used by the hotplug callback */ 1944 rc = cpuhp_state_add_instance(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE, 1945 &xgene_pmu->node); 1946 if (rc) { 1947 dev_err(&pdev->dev, "Error %d registering hotplug", rc); 1948 return rc; 1949 } 1950 1951 /* Walk through the tree for all PMU perf devices */ 1952 rc = xgene_pmu_probe_pmu_dev(xgene_pmu, pdev); 1953 if (rc) { 1954 dev_err(&pdev->dev, "No PMU perf devices found!\n"); 1955 goto out_unregister; 1956 } 1957 1958 /* Enable interrupt */ 1959 xgene_pmu->ops->unmask_int(xgene_pmu); 1960 1961 return 0; 1962 1963 out_unregister: 1964 cpuhp_state_remove_instance(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE, 1965 &xgene_pmu->node); 1966 return rc; 1967 } 1968 1969 static void 1970 xgene_pmu_dev_cleanup(struct xgene_pmu *xgene_pmu, struct list_head *pmus) 1971 { 1972 struct xgene_pmu_dev_ctx *ctx; 1973 1974 list_for_each_entry(ctx, pmus, next) { 1975 perf_pmu_unregister(&ctx->pmu_dev->pmu); 1976 } 1977 } 1978 1979 static int xgene_pmu_remove(struct platform_device *pdev) 1980 { 1981 struct xgene_pmu *xgene_pmu = dev_get_drvdata(&pdev->dev); 1982 1983 xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->l3cpmus); 1984 xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->iobpmus); 1985 xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->mcbpmus); 1986 xgene_pmu_dev_cleanup(xgene_pmu, &xgene_pmu->mcpmus); 1987 cpuhp_state_remove_instance(CPUHP_AP_PERF_ARM_APM_XGENE_ONLINE, 1988 &xgene_pmu->node); 1989 1990 return 0; 1991 } 1992 1993 static struct platform_driver xgene_pmu_driver = { 1994 .probe = xgene_pmu_probe, 1995 .remove = xgene_pmu_remove, 1996 .driver = { 1997 .name = "xgene-pmu", 1998 .of_match_table = xgene_pmu_of_match, 1999 .acpi_match_table = ACPI_PTR(xgene_pmu_acpi_match), 2000 }, 2001 }; 2002 2003 builtin_platform_driver(xgene_pmu_driver); 2004