power8-pmu.c 16 KB

123456789101112131415161718192021222324252627282930313233343536373839404142434445464748495051525354555657585960616263646566676869707172737475767778798081828384858687888990919293949596979899100101102103104105106107108109110111112113114115116117118119120121122123124125126127128129130131132133134135136137138139140141142143144145146147148149150151152153154155156157158159160161162163164165166167168169170171172173174175176177178179180181182183184185186187188189190191192193194195196197198199200201202203204205206207208209210211212213214215216217218219220221222223224225226227228229230231232233234235236237238239240241242243244245246247248249250251252253254255256257258259260261262263264265266267268269270271272273274275276277278279280281282283284285286287288289290291292293294295296297298299300301302303304305306307308309310311312313314315316317318319320321322323324325326327328329330331332333334335336337338339340341342343344345346347348349350351352353354355356357358359360361362363364365366367368369370371372373374375376377378379380381382383384385386387388389390391392393394395396397398399400401402403404405406407408409410411412413414415416417418419420421422423424425426427428429430431432433434435436437438439440441442443444445446447448449450451452453454455456457458459460461462463464465466467468469470471472473474475476477478479480481482483484485486487488489490491492493494495496497498499500501502503504505506507508509510511512513514515516517518519520521522523524525526527528529530531532533534535536537
  1. /*
  2. * Performance counter support for POWER8 processors.
  3. *
  4. * Copyright 2009 Paul Mackerras, IBM Corporation.
  5. * Copyright 2013 Michael Ellerman, IBM Corporation.
  6. *
  7. * This program is free software; you can redistribute it and/or
  8. * modify it under the terms of the GNU General Public License
  9. * as published by the Free Software Foundation; either version
  10. * 2 of the License, or (at your option) any later version.
  11. */
  12. #include <linux/kernel.h>
  13. #include <linux/perf_event.h>
  14. #include <asm/firmware.h>
  15. /*
  16. * Some power8 event codes.
  17. */
  18. #define PM_CYC 0x0001e
  19. #define PM_GCT_NOSLOT_CYC 0x100f8
  20. #define PM_CMPLU_STALL 0x4000a
  21. #define PM_INST_CMPL 0x00002
  22. #define PM_BRU_FIN 0x10068
  23. #define PM_BR_MPRED_CMPL 0x400f6
  24. /*
  25. * Raw event encoding for POWER8:
  26. *
  27. * 60 56 52 48 44 40 36 32
  28. * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
  29. * [ thresh_cmp ] [ thresh_ctl ]
  30. * |
  31. * thresh start/stop OR FAB match -*
  32. *
  33. * 28 24 20 16 12 8 4 0
  34. * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
  35. * [ ] [ sample ] [cache] [ pmc ] [unit ] c m [ pmcxsel ]
  36. * | | | | |
  37. * | | | | *- mark
  38. * | | *- L1/L2/L3 cache_sel |
  39. * | | |
  40. * | *- sampling mode for marked events *- combine
  41. * |
  42. * *- thresh_sel
  43. *
  44. * Below uses IBM bit numbering.
  45. *
  46. * MMCR1[x:y] = unit (PMCxUNIT)
  47. * MMCR1[x] = combine (PMCxCOMB)
  48. *
  49. * if pmc == 3 and unit == 0 and pmcxsel[0:6] == 0b0101011
  50. * # PM_MRK_FAB_RSP_MATCH
  51. * MMCR1[20:27] = thresh_ctl (FAB_CRESP_MATCH / FAB_TYPE_MATCH)
  52. * else if pmc == 4 and unit == 0xf and pmcxsel[0:6] == 0b0101001
  53. * # PM_MRK_FAB_RSP_MATCH_CYC
  54. * MMCR1[20:27] = thresh_ctl (FAB_CRESP_MATCH / FAB_TYPE_MATCH)
  55. * else
  56. * MMCRA[48:55] = thresh_ctl (THRESH START/END)
  57. *
  58. * if thresh_sel:
  59. * MMCRA[45:47] = thresh_sel
  60. *
  61. * if thresh_cmp:
  62. * MMCRA[22:24] = thresh_cmp[0:2]
  63. * MMCRA[25:31] = thresh_cmp[3:9]
  64. *
  65. * if unit == 6 or unit == 7
  66. * MMCRC[53:55] = cache_sel[1:3] (L2EVENT_SEL)
  67. * else if unit == 8 or unit == 9:
  68. * if cache_sel[0] == 0: # L3 bank
  69. * MMCRC[47:49] = cache_sel[1:3] (L3EVENT_SEL0)
  70. * else if cache_sel[0] == 1:
  71. * MMCRC[50:51] = cache_sel[2:3] (L3EVENT_SEL1)
  72. * else if cache_sel[1]: # L1 event
  73. * MMCR1[16] = cache_sel[2]
  74.  * MMCR1[17] = cache_sel[3]
  75. *
  76. * if mark:
  77. * MMCRA[63] = 1 (SAMPLE_ENABLE)
  78. * MMCRA[57:59] = sample[0:2] (RAND_SAMP_ELIG)
  79.  * MMCRA[61:62] = sample[3:4] (RAND_SAMP_MODE)
  80. *
  81. */
  82. #define EVENT_THR_CMP_SHIFT 40 /* Threshold CMP value */
  83. #define EVENT_THR_CMP_MASK 0x3ff
  84. #define EVENT_THR_CTL_SHIFT 32 /* Threshold control value (start/stop) */
  85. #define EVENT_THR_CTL_MASK 0xffull
  86. #define EVENT_THR_SEL_SHIFT 29 /* Threshold select value */
  87. #define EVENT_THR_SEL_MASK 0x7
  88. #define EVENT_THRESH_SHIFT 29 /* All threshold bits */
  89. #define EVENT_THRESH_MASK 0x1fffffull
  90. #define EVENT_SAMPLE_SHIFT 24 /* Sampling mode & eligibility */
  91. #define EVENT_SAMPLE_MASK 0x1f
  92. #define EVENT_CACHE_SEL_SHIFT 20 /* L2/L3 cache select */
  93. #define EVENT_CACHE_SEL_MASK 0xf
  94. #define EVENT_IS_L1 (4 << EVENT_CACHE_SEL_SHIFT)
  95. #define EVENT_PMC_SHIFT 16 /* PMC number (1-based) */
  96. #define EVENT_PMC_MASK 0xf
  97. #define EVENT_UNIT_SHIFT 12 /* Unit */
  98. #define EVENT_UNIT_MASK 0xf
  99. #define EVENT_COMBINE_SHIFT 11 /* Combine bit */
  100. #define EVENT_COMBINE_MASK 0x1
  101. #define EVENT_MARKED_SHIFT 8 /* Marked bit */
  102. #define EVENT_MARKED_MASK 0x1
  103. #define EVENT_IS_MARKED (EVENT_MARKED_MASK << EVENT_MARKED_SHIFT)
  104. #define EVENT_PSEL_MASK 0xff /* PMCxSEL value */
  105. /*
  106. * Layout of constraint bits:
  107. *
  108. * 60 56 52 48 44 40 36 32
  109. * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
  110. * [ fab_match ] [ thresh_cmp ] [ thresh_ctl ] [ ]
  111. * |
  112. * thresh_sel -*
  113. *
  114. * 28 24 20 16 12 8 4 0
  115. * | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - | - - - - |
  116. * [ ] [ sample ] [ ] [6] [5] [4] [3] [2] [1]
  117. * | |
  118. * L1 I/D qualifier -* | Count of events for each PMC.
  119. * | p1, p2, p3, p4, p5, p6.
  120. * nc - number of counters -*
  121. *
  122. * The PMC fields P1..P6, and NC, are adder fields. As we accumulate constraints
  123. * we want the low bit of each field to be added to any existing value.
  124. *
  125. * Everything else is a value field.
  126. */
  127. #define CNST_FAB_MATCH_VAL(v) (((v) & EVENT_THR_CTL_MASK) << 56)
  128. #define CNST_FAB_MATCH_MASK CNST_FAB_MATCH_VAL(EVENT_THR_CTL_MASK)
  129. /* We just throw all the threshold bits into the constraint */
  130. #define CNST_THRESH_VAL(v) (((v) & EVENT_THRESH_MASK) << 32)
  131. #define CNST_THRESH_MASK CNST_THRESH_VAL(EVENT_THRESH_MASK)
  132. #define CNST_L1_QUAL_VAL(v) (((v) & 3) << 22)
  133. #define CNST_L1_QUAL_MASK CNST_L1_QUAL_VAL(3)
  134. #define CNST_SAMPLE_VAL(v) (((v) & EVENT_SAMPLE_MASK) << 16)
  135. #define CNST_SAMPLE_MASK CNST_SAMPLE_VAL(EVENT_SAMPLE_MASK)
  136. /*
  137. * For NC we are counting up to 4 events. This requires three bits, and we need
  138. * the fifth event to overflow and set the 4th bit. To achieve that we bias the
  139. * fields by 3 in test_adder.
  140. */
  141. #define CNST_NC_SHIFT 12
  142. #define CNST_NC_VAL (1 << CNST_NC_SHIFT)
  143. #define CNST_NC_MASK (8 << CNST_NC_SHIFT)
  144. #define POWER8_TEST_ADDER (3 << CNST_NC_SHIFT)
  145. /*
  146. * For the per-PMC fields we have two bits. The low bit is added, so if two
  147. * events ask for the same PMC the sum will overflow, setting the high bit,
  148. * indicating an error. So our mask sets the high bit.
  149. */
  150. #define CNST_PMC_SHIFT(pmc) ((pmc - 1) * 2)
  151. #define CNST_PMC_VAL(pmc) (1 << CNST_PMC_SHIFT(pmc))
  152. #define CNST_PMC_MASK(pmc) (2 << CNST_PMC_SHIFT(pmc))
  153. /* Our add_fields is defined as: */
  154. #define POWER8_ADD_FIELDS \
  155. CNST_PMC_VAL(1) | CNST_PMC_VAL(2) | CNST_PMC_VAL(3) | \
  156. CNST_PMC_VAL(4) | CNST_PMC_VAL(5) | CNST_PMC_VAL(6) | CNST_NC_VAL
  157. /* Bits in MMCR1 for POWER8 */
  158. #define MMCR1_UNIT_SHIFT(pmc) (60 - (4 * ((pmc) - 1)))
  159. #define MMCR1_COMBINE_SHIFT(pmc) (35 - ((pmc) - 1))
  160. #define MMCR1_PMCSEL_SHIFT(pmc) (24 - (((pmc) - 1)) * 8)
  161. #define MMCR1_DC_QUAL_SHIFT 47
  162. #define MMCR1_IC_QUAL_SHIFT 46
  163. /* Bits in MMCRA for POWER8 */
  164. #define MMCRA_SAMP_MODE_SHIFT 1
  165. #define MMCRA_SAMP_ELIG_SHIFT 4
  166. #define MMCRA_THR_CTL_SHIFT 8
  167. #define MMCRA_THR_SEL_SHIFT 16
  168. #define MMCRA_THR_CMP_SHIFT 32
  169. #define MMCRA_SDAR_MODE_TLB (1ull << 42)
  170. static inline bool event_is_fab_match(u64 event)
  171. {
  172. /* Only check pmc, unit and pmcxsel, ignore the edge bit (0) */
  173. event &= 0xff0fe;
  174. /* PM_MRK_FAB_RSP_MATCH & PM_MRK_FAB_RSP_MATCH_CYC */
  175. return (event == 0x30056 || event == 0x4f052);
  176. }
  177. static int power8_get_constraint(u64 event, unsigned long *maskp, unsigned long *valp)
  178. {
  179. unsigned int unit, pmc, cache;
  180. unsigned long mask, value;
  181. mask = value = 0;
  182. pmc = (event >> EVENT_PMC_SHIFT) & EVENT_PMC_MASK;
  183. unit = (event >> EVENT_UNIT_SHIFT) & EVENT_UNIT_MASK;
  184. cache = (event >> EVENT_CACHE_SEL_SHIFT) & EVENT_CACHE_SEL_MASK;
  185. if (pmc) {
  186. if (pmc > 6)
  187. return -1;
  188. mask |= CNST_PMC_MASK(pmc);
  189. value |= CNST_PMC_VAL(pmc);
  190. if (pmc >= 5 && event != 0x500fa && event != 0x600f4)
  191. return -1;
  192. }
  193. if (pmc <= 4) {
  194. /*
  195. * Add to number of counters in use. Note this includes events with
  196. * a PMC of 0 - they still need a PMC, it's just assigned later.
  197. * Don't count events on PMC 5 & 6, there is only one valid event
  198. * on each of those counters, and they are handled above.
  199. */
  200. mask |= CNST_NC_MASK;
  201. value |= CNST_NC_VAL;
  202. }
  203. if (unit >= 6 && unit <= 9) {
  204. /*
  205. * L2/L3 events contain a cache selector field, which is
  206. * supposed to be programmed into MMCRC. However MMCRC is only
  207. * HV writable, and there is no API for guest kernels to modify
  208. * it. The solution is for the hypervisor to initialise the
  209. * field to zeroes, and for us to only ever allow events that
  210. * have a cache selector of zero.
  211. */
  212. if (cache)
  213. return -1;
  214. } else if (event & EVENT_IS_L1) {
  215. mask |= CNST_L1_QUAL_MASK;
  216. value |= CNST_L1_QUAL_VAL(cache);
  217. }
  218. if (event & EVENT_IS_MARKED) {
  219. mask |= CNST_SAMPLE_MASK;
  220. value |= CNST_SAMPLE_VAL(event >> EVENT_SAMPLE_SHIFT);
  221. }
  222. /*
  223. * Special case for PM_MRK_FAB_RSP_MATCH and PM_MRK_FAB_RSP_MATCH_CYC,
  224. * the threshold control bits are used for the match value.
  225. */
  226. if (event_is_fab_match(event)) {
  227. mask |= CNST_FAB_MATCH_MASK;
  228. value |= CNST_FAB_MATCH_VAL(event >> EVENT_THR_CTL_SHIFT);
  229. } else {
  230. /*
  231. * Check the mantissa upper two bits are not zero, unless the
  232. * exponent is also zero. See the THRESH_CMP_MANTISSA doc.
  233. */
  234. unsigned int cmp, exp;
  235. cmp = (event >> EVENT_THR_CMP_SHIFT) & EVENT_THR_CMP_MASK;
  236. exp = cmp >> 7;
  237. if (exp && (cmp & 0x60) == 0)
  238. return -1;
  239. mask |= CNST_THRESH_MASK;
  240. value |= CNST_THRESH_VAL(event >> EVENT_THRESH_SHIFT);
  241. }
  242. *maskp = mask;
  243. *valp = value;
  244. return 0;
  245. }
  246. static int power8_compute_mmcr(u64 event[], int n_ev,
  247. unsigned int hwc[], unsigned long mmcr[])
  248. {
  249. unsigned long mmcra, mmcr1, unit, combine, psel, cache, val;
  250. unsigned int pmc, pmc_inuse;
  251. int i;
  252. pmc_inuse = 0;
  253. /* First pass to count resource use */
  254. for (i = 0; i < n_ev; ++i) {
  255. pmc = (event[i] >> EVENT_PMC_SHIFT) & EVENT_PMC_MASK;
  256. if (pmc)
  257. pmc_inuse |= 1 << pmc;
  258. }
  259. /* In continous sampling mode, update SDAR on TLB miss */
  260. mmcra = MMCRA_SDAR_MODE_TLB;
  261. mmcr1 = 0;
  262. /* Second pass: assign PMCs, set all MMCR1 fields */
  263. for (i = 0; i < n_ev; ++i) {
  264. pmc = (event[i] >> EVENT_PMC_SHIFT) & EVENT_PMC_MASK;
  265. unit = (event[i] >> EVENT_UNIT_SHIFT) & EVENT_UNIT_MASK;
  266. combine = (event[i] >> EVENT_COMBINE_SHIFT) & EVENT_COMBINE_MASK;
  267. psel = event[i] & EVENT_PSEL_MASK;
  268. if (!pmc) {
  269. for (pmc = 1; pmc <= 4; ++pmc) {
  270. if (!(pmc_inuse & (1 << pmc)))
  271. break;
  272. }
  273. pmc_inuse |= 1 << pmc;
  274. }
  275. if (pmc <= 4) {
  276. mmcr1 |= unit << MMCR1_UNIT_SHIFT(pmc);
  277. mmcr1 |= combine << MMCR1_COMBINE_SHIFT(pmc);
  278. mmcr1 |= psel << MMCR1_PMCSEL_SHIFT(pmc);
  279. }
  280. if (event[i] & EVENT_IS_L1) {
  281. cache = event[i] >> EVENT_CACHE_SEL_SHIFT;
  282. mmcr1 |= (cache & 1) << MMCR1_IC_QUAL_SHIFT;
  283. cache >>= 1;
  284. mmcr1 |= (cache & 1) << MMCR1_DC_QUAL_SHIFT;
  285. }
  286. if (event[i] & EVENT_IS_MARKED) {
  287. mmcra |= MMCRA_SAMPLE_ENABLE;
  288. val = (event[i] >> EVENT_SAMPLE_SHIFT) & EVENT_SAMPLE_MASK;
  289. if (val) {
  290. mmcra |= (val & 3) << MMCRA_SAMP_MODE_SHIFT;
  291. mmcra |= (val >> 2) << MMCRA_SAMP_ELIG_SHIFT;
  292. }
  293. }
  294. /*
  295. * PM_MRK_FAB_RSP_MATCH and PM_MRK_FAB_RSP_MATCH_CYC,
  296. * the threshold bits are used for the match value.
  297. */
  298. if (event_is_fab_match(event[i])) {
  299. mmcr1 |= (event[i] >> EVENT_THR_CTL_SHIFT) &
  300. EVENT_THR_CTL_MASK;
  301. } else {
  302. val = (event[i] >> EVENT_THR_CTL_SHIFT) & EVENT_THR_CTL_MASK;
  303. mmcra |= val << MMCRA_THR_CTL_SHIFT;
  304. val = (event[i] >> EVENT_THR_SEL_SHIFT) & EVENT_THR_SEL_MASK;
  305. mmcra |= val << MMCRA_THR_SEL_SHIFT;
  306. val = (event[i] >> EVENT_THR_CMP_SHIFT) & EVENT_THR_CMP_MASK;
  307. mmcra |= val << MMCRA_THR_CMP_SHIFT;
  308. }
  309. hwc[i] = pmc - 1;
  310. }
  311. /* Return MMCRx values */
  312. mmcr[0] = 0;
  313. /* pmc_inuse is 1-based */
  314. if (pmc_inuse & 2)
  315. mmcr[0] = MMCR0_PMC1CE;
  316. if (pmc_inuse & 0x7c)
  317. mmcr[0] |= MMCR0_PMCjCE;
  318. mmcr[1] = mmcr1;
  319. mmcr[2] = mmcra;
  320. return 0;
  321. }
  322. #define MAX_ALT 2
  323. /* Table of alternatives, sorted by column 0 */
  324. static const unsigned int event_alternatives[][MAX_ALT] = {
  325. { 0x10134, 0x301e2 }, /* PM_MRK_ST_CMPL */
  326. { 0x10138, 0x40138 }, /* PM_BR_MRK_2PATH */
  327. { 0x18082, 0x3e05e }, /* PM_L3_CO_MEPF */
  328. { 0x1d14e, 0x401e8 }, /* PM_MRK_DATA_FROM_L2MISS */
  329. { 0x1e054, 0x4000a }, /* PM_CMPLU_STALL */
  330. { 0x20036, 0x40036 }, /* PM_BR_2PATH */
  331. { 0x200f2, 0x300f2 }, /* PM_INST_DISP */
  332. { 0x200f4, 0x600f4 }, /* PM_RUN_CYC */
  333. { 0x2013c, 0x3012e }, /* PM_MRK_FILT_MATCH */
  334. { 0x3e054, 0x400f0 }, /* PM_LD_MISS_L1 */
  335. { 0x400fa, 0x500fa }, /* PM_RUN_INST_CMPL */
  336. };
  337. /*
  338. * Scan the alternatives table for a match and return the
  339. * index into the alternatives table if found, else -1.
  340. */
  341. static int find_alternative(u64 event)
  342. {
  343. int i, j;
  344. for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) {
  345. if (event < event_alternatives[i][0])
  346. break;
  347. for (j = 0; j < MAX_ALT && event_alternatives[i][j]; ++j)
  348. if (event == event_alternatives[i][j])
  349. return i;
  350. }
  351. return -1;
  352. }
  353. static int power8_get_alternatives(u64 event, unsigned int flags, u64 alt[])
  354. {
  355. int i, j, num_alt = 0;
  356. u64 alt_event;
  357. alt[num_alt++] = event;
  358. i = find_alternative(event);
  359. if (i >= 0) {
  360. /* Filter out the original event, it's already in alt[0] */
  361. for (j = 0; j < MAX_ALT; ++j) {
  362. alt_event = event_alternatives[i][j];
  363. if (alt_event && alt_event != event)
  364. alt[num_alt++] = alt_event;
  365. }
  366. }
  367. if (flags & PPMU_ONLY_COUNT_RUN) {
  368. /*
  369. * We're only counting in RUN state, so PM_CYC is equivalent to
  370. * PM_RUN_CYC and PM_INST_CMPL === PM_RUN_INST_CMPL.
  371. */
  372. j = num_alt;
  373. for (i = 0; i < num_alt; ++i) {
  374. switch (alt[i]) {
  375. case 0x1e: /* PM_CYC */
  376. alt[j++] = 0x600f4; /* PM_RUN_CYC */
  377. break;
  378. case 0x600f4: /* PM_RUN_CYC */
  379. alt[j++] = 0x1e;
  380. break;
  381. case 0x2: /* PM_PPC_CMPL */
  382. alt[j++] = 0x500fa; /* PM_RUN_INST_CMPL */
  383. break;
  384. case 0x500fa: /* PM_RUN_INST_CMPL */
  385. alt[j++] = 0x2; /* PM_PPC_CMPL */
  386. break;
  387. }
  388. }
  389. num_alt = j;
  390. }
  391. return num_alt;
  392. }
  393. static void power8_disable_pmc(unsigned int pmc, unsigned long mmcr[])
  394. {
  395. if (pmc <= 3)
  396. mmcr[1] &= ~(0xffUL << MMCR1_PMCSEL_SHIFT(pmc + 1));
  397. }
  398. PMU_FORMAT_ATTR(event, "config:0-49");
  399. PMU_FORMAT_ATTR(pmcxsel, "config:0-7");
  400. PMU_FORMAT_ATTR(mark, "config:8");
  401. PMU_FORMAT_ATTR(combine, "config:11");
  402. PMU_FORMAT_ATTR(unit, "config:12-15");
  403. PMU_FORMAT_ATTR(pmc, "config:16-19");
  404. PMU_FORMAT_ATTR(cache_sel, "config:20-23");
  405. PMU_FORMAT_ATTR(sample_mode, "config:24-28");
  406. PMU_FORMAT_ATTR(thresh_sel, "config:29-31");
  407. PMU_FORMAT_ATTR(thresh_stop, "config:32-35");
  408. PMU_FORMAT_ATTR(thresh_start, "config:36-39");
  409. PMU_FORMAT_ATTR(thresh_cmp, "config:40-49");
  410. static struct attribute *power8_pmu_format_attr[] = {
  411. &format_attr_event.attr,
  412. &format_attr_pmcxsel.attr,
  413. &format_attr_mark.attr,
  414. &format_attr_combine.attr,
  415. &format_attr_unit.attr,
  416. &format_attr_pmc.attr,
  417. &format_attr_cache_sel.attr,
  418. &format_attr_sample_mode.attr,
  419. &format_attr_thresh_sel.attr,
  420. &format_attr_thresh_stop.attr,
  421. &format_attr_thresh_start.attr,
  422. &format_attr_thresh_cmp.attr,
  423. NULL,
  424. };
  425. struct attribute_group power8_pmu_format_group = {
  426. .name = "format",
  427. .attrs = power8_pmu_format_attr,
  428. };
  429. static const struct attribute_group *power8_pmu_attr_groups[] = {
  430. &power8_pmu_format_group,
  431. NULL,
  432. };
  433. static int power8_generic_events[] = {
  434. [PERF_COUNT_HW_CPU_CYCLES] = PM_CYC,
  435. [PERF_COUNT_HW_STALLED_CYCLES_FRONTEND] = PM_GCT_NOSLOT_CYC,
  436. [PERF_COUNT_HW_STALLED_CYCLES_BACKEND] = PM_CMPLU_STALL,
  437. [PERF_COUNT_HW_INSTRUCTIONS] = PM_INST_CMPL,
  438. [PERF_COUNT_HW_BRANCH_INSTRUCTIONS] = PM_BRU_FIN,
  439. [PERF_COUNT_HW_BRANCH_MISSES] = PM_BR_MPRED_CMPL,
  440. };
  441. static struct power_pmu power8_pmu = {
  442. .name = "POWER8",
  443. .n_counter = 6,
  444. .max_alternatives = MAX_ALT + 1,
  445. .add_fields = POWER8_ADD_FIELDS,
  446. .test_adder = POWER8_TEST_ADDER,
  447. .compute_mmcr = power8_compute_mmcr,
  448. .get_constraint = power8_get_constraint,
  449. .get_alternatives = power8_get_alternatives,
  450. .disable_pmc = power8_disable_pmc,
  451. .flags = PPMU_HAS_SSLOT | PPMU_HAS_SIER,
  452. .n_generic = ARRAY_SIZE(power8_generic_events),
  453. .generic_events = power8_generic_events,
  454. .attr_groups = power8_pmu_attr_groups,
  455. };
  456. static int __init init_power8_pmu(void)
  457. {
  458. if (!cur_cpu_spec->oprofile_cpu_type ||
  459. strcmp(cur_cpu_spec->oprofile_cpu_type, "ppc64/power8"))
  460. return -ENODEV;
  461. return register_power_pmu(&power8_pmu);
  462. }
  463. early_initcall(init_power8_pmu);