Commit | Line | Data |
---|---|---|
f7862837 PM |
1 | /* |
2 | * Performance counter support for POWER6 processors. | |
3 | * | |
4 | * Copyright 2008-2009 Paul Mackerras, IBM Corporation. | |
5 | * | |
6 | * This program is free software; you can redistribute it and/or | |
7 | * modify it under the terms of the GNU General Public License | |
8 | * as published by the Free Software Foundation; either version | |
9 | * 2 of the License, or (at your option) any later version. | |
10 | */ | |
11 | #include <linux/kernel.h> | |
12 | #include <linux/perf_counter.h> | |
13 | #include <asm/reg.h> | |
14 | ||
15 | /* | |
16 | * Bits in event code for POWER6 | |
17 | */ | |
18 | #define PM_PMC_SH 20 /* PMC number (1-based) for direct events */ | |
19 | #define PM_PMC_MSK 0x7 | |
20 | #define PM_PMC_MSKS (PM_PMC_MSK << PM_PMC_SH) | |
21 | #define PM_UNIT_SH 16 /* Unit event comes (TTMxSEL encoding) */ | |
22 | #define PM_UNIT_MSK 0xf | |
23 | #define PM_UNIT_MSKS (PM_UNIT_MSK << PM_UNIT_SH) | |
24 | #define PM_LLAV 0x8000 /* Load lookahead match value */ | |
25 | #define PM_LLA 0x4000 /* Load lookahead match enable */ | |
26 | #define PM_BYTE_SH 12 /* Byte of event bus to use */ | |
27 | #define PM_BYTE_MSK 3 | |
28 | #define PM_SUBUNIT_SH 8 /* Subunit event comes from (NEST_SEL enc.) */ | |
29 | #define PM_SUBUNIT_MSK 7 | |
30 | #define PM_SUBUNIT_MSKS (PM_SUBUNIT_MSK << PM_SUBUNIT_SH) | |
31 | #define PM_PMCSEL_MSK 0xff /* PMCxSEL value */ | |
32 | #define PM_BUSEVENT_MSK 0xf3700 | |
33 | ||
34 | /* | |
35 | * Bits in MMCR1 for POWER6 | |
36 | */ | |
37 | #define MMCR1_TTM0SEL_SH 60 | |
38 | #define MMCR1_TTMSEL_SH(n) (MMCR1_TTM0SEL_SH - (n) * 4) | |
39 | #define MMCR1_TTMSEL_MSK 0xf | |
40 | #define MMCR1_TTMSEL(m, n) (((m) >> MMCR1_TTMSEL_SH(n)) & MMCR1_TTMSEL_MSK) | |
41 | #define MMCR1_NESTSEL_SH 45 | |
42 | #define MMCR1_NESTSEL_MSK 0x7 | |
43 | #define MMCR1_NESTSEL(m) (((m) >> MMCR1_NESTSEL_SH) & MMCR1_NESTSEL_MSK) | |
44 | #define MMCR1_PMC1_LLA ((u64)1 << 44) | |
45 | #define MMCR1_PMC1_LLA_VALUE ((u64)1 << 39) | |
46 | #define MMCR1_PMC1_ADDR_SEL ((u64)1 << 35) | |
47 | #define MMCR1_PMC1SEL_SH 24 | |
48 | #define MMCR1_PMCSEL_SH(n) (MMCR1_PMC1SEL_SH - (n) * 8) | |
49 | #define MMCR1_PMCSEL_MSK 0xff | |
50 | ||
f708223d PM |
51 | /* |
52 | * Map of which direct events on which PMCs are marked instruction events. | |
53 | * Indexed by PMCSEL value >> 1. | |
54 | * Bottom 4 bits are a map of which PMCs are interesting, | |
55 | * top 4 bits say what sort of event: | |
56 | * 0 = direct marked event, | |
57 | * 1 = byte decode event, | |
58 | * 4 = add/and event (PMC1 -> bits 0 & 4), | |
59 | * 5 = add/and event (PMC1 -> bits 1 & 5), | |
60 | * 6 = add/and event (PMC1 -> bits 2 & 6), | |
61 | * 7 = add/and event (PMC1 -> bits 3 & 7). | |
62 | */ | |
63 | static unsigned char direct_event_is_marked[0x60 >> 1] = { | |
64 | 0, /* 00 */ | |
65 | 0, /* 02 */ | |
66 | 0, /* 04 */ | |
67 | 0x07, /* 06 PM_MRK_ST_CMPL, PM_MRK_ST_GPS, PM_MRK_ST_CMPL_INT */ | |
68 | 0x04, /* 08 PM_MRK_DFU_FIN */ | |
69 | 0x06, /* 0a PM_MRK_IFU_FIN, PM_MRK_INST_FIN */ | |
70 | 0, /* 0c */ | |
71 | 0, /* 0e */ | |
72 | 0x02, /* 10 PM_MRK_INST_DISP */ | |
73 | 0x08, /* 12 PM_MRK_LSU_DERAT_MISS */ | |
74 | 0, /* 14 */ | |
75 | 0, /* 16 */ | |
76 | 0x0c, /* 18 PM_THRESH_TIMEO, PM_MRK_INST_FIN */ | |
77 | 0x0f, /* 1a PM_MRK_INST_DISP, PM_MRK_{FXU,FPU,LSU}_FIN */ | |
78 | 0x01, /* 1c PM_MRK_INST_ISSUED */ | |
79 | 0, /* 1e */ | |
80 | 0, /* 20 */ | |
81 | 0, /* 22 */ | |
82 | 0, /* 24 */ | |
83 | 0, /* 26 */ | |
84 | 0x15, /* 28 PM_MRK_DATA_FROM_L2MISS, PM_MRK_DATA_FROM_L3MISS */ | |
85 | 0, /* 2a */ | |
86 | 0, /* 2c */ | |
87 | 0, /* 2e */ | |
88 | 0x4f, /* 30 */ | |
89 | 0x7f, /* 32 */ | |
90 | 0x4f, /* 34 */ | |
91 | 0x5f, /* 36 */ | |
92 | 0x6f, /* 38 */ | |
93 | 0x4f, /* 3a */ | |
94 | 0, /* 3c */ | |
95 | 0x08, /* 3e PM_MRK_INST_TIMEO */ | |
96 | 0x1f, /* 40 */ | |
97 | 0x1f, /* 42 */ | |
98 | 0x1f, /* 44 */ | |
99 | 0x1f, /* 46 */ | |
100 | 0x1f, /* 48 */ | |
101 | 0x1f, /* 4a */ | |
102 | 0x1f, /* 4c */ | |
103 | 0x1f, /* 4e */ | |
104 | 0, /* 50 */ | |
105 | 0x05, /* 52 PM_MRK_BR_TAKEN, PM_MRK_BR_MPRED */ | |
106 | 0x1c, /* 54 PM_MRK_PTEG_FROM_L3MISS, PM_MRK_PTEG_FROM_L2MISS */ | |
107 | 0x02, /* 56 PM_MRK_LD_MISS_L1 */ | |
108 | 0, /* 58 */ | |
109 | 0, /* 5a */ | |
110 | 0, /* 5c */ | |
111 | 0, /* 5e */ | |
112 | }; | |
113 | ||
114 | /* | |
115 | * Masks showing for each unit which bits are marked events. | |
116 | * These masks are in LE order, i.e. 0x00000001 is byte 0, bit 0. | |
117 | */ | |
118 | static u32 marked_bus_events[16] = { | |
119 | 0x01000000, /* direct events set 1: byte 3 bit 0 */ | |
120 | 0x00010000, /* direct events set 2: byte 2 bit 0 */ | |
121 | 0, 0, 0, 0, /* IDU, IFU, nest: nothing */ | |
122 | 0x00000088, /* VMX set 1: byte 0 bits 3, 7 */ | |
123 | 0x000000c0, /* VMX set 2: byte 0 bits 4-7 */ | |
124 | 0x04010000, /* LSU set 1: byte 2 bit 0, byte 3 bit 2 */ | |
125 | 0xff010000u, /* LSU set 2: byte 2 bit 0, all of byte 3 */ | |
126 | 0, /* LSU set 3 */ | |
127 | 0x00000010, /* VMX set 3: byte 0 bit 4 */ | |
128 | 0, /* BFP set 1 */ | |
129 | 0x00000022, /* BFP set 2: byte 0 bits 1, 5 */ | |
130 | 0, 0 | |
131 | }; | |
132 | ||
133 | /* | |
134 | * Returns 1 if event counts things relating to marked instructions | |
135 | * and thus needs the MMCRA_SAMPLE_ENABLE bit set, or 0 if not. | |
136 | */ | |
137 | static int power6_marked_instr_event(unsigned int event) | |
138 | { | |
139 | int pmc, psel, ptype; | |
140 | int bit, byte, unit; | |
141 | u32 mask; | |
142 | ||
143 | pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; | |
144 | psel = (event & PM_PMCSEL_MSK) >> 1; /* drop edge/level bit */ | |
145 | if (pmc >= 5) | |
146 | return 0; | |
147 | ||
148 | bit = -1; | |
149 | if (psel < sizeof(direct_event_is_marked)) { | |
150 | ptype = direct_event_is_marked[psel]; | |
151 | if (pmc == 0 || !(ptype & (1 << (pmc - 1)))) | |
152 | return 0; | |
153 | ptype >>= 4; | |
154 | if (ptype == 0) | |
155 | return 1; | |
156 | if (ptype == 1) | |
157 | bit = 0; | |
158 | else | |
159 | bit = ptype ^ (pmc - 1); | |
160 | } else if ((psel & 0x48) == 0x40) | |
161 | bit = psel & 7; | |
162 | ||
163 | if (!(event & PM_BUSEVENT_MSK) || bit == -1) | |
164 | return 0; | |
165 | ||
166 | byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK; | |
167 | unit = (event >> PM_UNIT_SH) & PM_UNIT_MSK; | |
168 | mask = marked_bus_events[unit]; | |
169 | return (mask >> (byte * 8 + bit)) & 1; | |
170 | } | |
171 | ||
f7862837 PM |
172 | /* |
173 | * Assign PMC numbers and compute MMCR1 value for a set of events | |
174 | */ | |
175 | static int p6_compute_mmcr(unsigned int event[], int n_ev, | |
176 | unsigned int hwc[], u64 mmcr[]) | |
177 | { | |
178 | u64 mmcr1 = 0; | |
f708223d | 179 | u64 mmcra = 0; |
f7862837 PM |
180 | int i; |
181 | unsigned int pmc, ev, b, u, s, psel; | |
182 | unsigned int ttmset = 0; | |
183 | unsigned int pmc_inuse = 0; | |
184 | ||
185 | if (n_ev > 4) | |
186 | return -1; | |
187 | for (i = 0; i < n_ev; ++i) { | |
188 | pmc = (event[i] >> PM_PMC_SH) & PM_PMC_MSK; | |
189 | if (pmc) { | |
190 | if (pmc_inuse & (1 << (pmc - 1))) | |
191 | return -1; /* collision! */ | |
192 | pmc_inuse |= 1 << (pmc - 1); | |
193 | } | |
194 | } | |
195 | for (i = 0; i < n_ev; ++i) { | |
196 | ev = event[i]; | |
197 | pmc = (ev >> PM_PMC_SH) & PM_PMC_MSK; | |
198 | if (pmc) { | |
199 | --pmc; | |
200 | } else { | |
201 | /* can go on any PMC; find a free one */ | |
202 | for (pmc = 0; pmc < 4; ++pmc) | |
203 | if (!(pmc_inuse & (1 << pmc))) | |
204 | break; | |
205 | pmc_inuse |= 1 << pmc; | |
206 | } | |
207 | hwc[i] = pmc; | |
208 | psel = ev & PM_PMCSEL_MSK; | |
209 | if (ev & PM_BUSEVENT_MSK) { | |
210 | /* this event uses the event bus */ | |
211 | b = (ev >> PM_BYTE_SH) & PM_BYTE_MSK; | |
212 | u = (ev >> PM_UNIT_SH) & PM_UNIT_MSK; | |
213 | /* check for conflict on this byte of event bus */ | |
214 | if ((ttmset & (1 << b)) && MMCR1_TTMSEL(mmcr1, b) != u) | |
215 | return -1; | |
216 | mmcr1 |= (u64)u << MMCR1_TTMSEL_SH(b); | |
217 | ttmset |= 1 << b; | |
218 | if (u == 5) { | |
219 | /* Nest events have a further mux */ | |
220 | s = (ev >> PM_SUBUNIT_SH) & PM_SUBUNIT_MSK; | |
221 | if ((ttmset & 0x10) && | |
222 | MMCR1_NESTSEL(mmcr1) != s) | |
223 | return -1; | |
224 | ttmset |= 0x10; | |
225 | mmcr1 |= (u64)s << MMCR1_NESTSEL_SH; | |
226 | } | |
227 | if (0x30 <= psel && psel <= 0x3d) { | |
228 | /* these need the PMCx_ADDR_SEL bits */ | |
229 | if (b >= 2) | |
230 | mmcr1 |= MMCR1_PMC1_ADDR_SEL >> pmc; | |
231 | } | |
232 | /* bus select values are different for PMC3/4 */ | |
233 | if (pmc >= 2 && (psel & 0x90) == 0x80) | |
234 | psel ^= 0x20; | |
235 | } | |
236 | if (ev & PM_LLA) { | |
237 | mmcr1 |= MMCR1_PMC1_LLA >> pmc; | |
238 | if (ev & PM_LLAV) | |
239 | mmcr1 |= MMCR1_PMC1_LLA_VALUE >> pmc; | |
240 | } | |
f708223d PM |
241 | if (power6_marked_instr_event(event[i])) |
242 | mmcra |= MMCRA_SAMPLE_ENABLE; | |
f7862837 PM |
243 | mmcr1 |= (u64)psel << MMCR1_PMCSEL_SH(pmc); |
244 | } | |
245 | mmcr[0] = 0; | |
246 | if (pmc_inuse & 1) | |
247 | mmcr[0] = MMCR0_PMC1CE; | |
248 | if (pmc_inuse & 0xe) | |
249 | mmcr[0] |= MMCR0_PMCjCE; | |
250 | mmcr[1] = mmcr1; | |
f708223d | 251 | mmcr[2] = mmcra; |
f7862837 PM |
252 | return 0; |
253 | } | |
254 | ||
255 | /* | |
256 | * Layout of constraint bits: | |
257 | * | |
258 | * 0-1 add field: number of uses of PMC1 (max 1) | |
259 | * 2-3, 4-5, 6-7: ditto for PMC2, 3, 4 | |
260 | * 8-10 select field: nest (subunit) event selector | |
261 | * 16-19 select field: unit on byte 0 of event bus | |
262 | * 20-23, 24-27, 28-31 ditto for bytes 1, 2, 3 | |
263 | */ | |
264 | static int p6_get_constraint(unsigned int event, u64 *maskp, u64 *valp) | |
265 | { | |
266 | int pmc, byte, sh; | |
267 | unsigned int mask = 0, value = 0; | |
268 | ||
269 | pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; | |
270 | if (pmc) { | |
271 | if (pmc > 4) | |
272 | return -1; | |
273 | sh = (pmc - 1) * 2; | |
274 | mask |= 2 << sh; | |
275 | value |= 1 << sh; | |
276 | } | |
277 | if (event & PM_BUSEVENT_MSK) { | |
278 | byte = (event >> PM_BYTE_SH) & PM_BYTE_MSK; | |
279 | sh = byte * 4; | |
280 | mask |= PM_UNIT_MSKS << sh; | |
281 | value |= (event & PM_UNIT_MSKS) << sh; | |
282 | if ((event & PM_UNIT_MSKS) == (5 << PM_UNIT_SH)) { | |
283 | mask |= PM_SUBUNIT_MSKS; | |
284 | value |= event & PM_SUBUNIT_MSKS; | |
285 | } | |
286 | } | |
287 | *maskp = mask; | |
288 | *valp = value; | |
289 | return 0; | |
290 | } | |
291 | ||
292 | #define MAX_ALT 4 /* at most 4 alternatives for any event */ | |
293 | ||
294 | static const unsigned int event_alternatives[][MAX_ALT] = { | |
295 | { 0x0130e8, 0x2000f6, 0x3000fc }, /* PM_PTEG_RELOAD_VALID */ | |
296 | { 0x080080, 0x10000d, 0x30000c, 0x4000f0 }, /* PM_LD_MISS_L1 */ | |
297 | { 0x080088, 0x200054, 0x3000f0 }, /* PM_ST_MISS_L1 */ | |
298 | { 0x10000a, 0x2000f4 }, /* PM_RUN_CYC */ | |
299 | { 0x10000b, 0x2000f5 }, /* PM_RUN_COUNT */ | |
300 | { 0x10000e, 0x400010 }, /* PM_PURR */ | |
301 | { 0x100010, 0x4000f8 }, /* PM_FLUSH */ | |
302 | { 0x10001a, 0x200010 }, /* PM_MRK_INST_DISP */ | |
303 | { 0x100026, 0x3000f8 }, /* PM_TB_BIT_TRANS */ | |
304 | { 0x100054, 0x2000f0 }, /* PM_ST_FIN */ | |
305 | { 0x100056, 0x2000fc }, /* PM_L1_ICACHE_MISS */ | |
306 | { 0x1000f0, 0x40000a }, /* PM_INST_IMC_MATCH_CMPL */ | |
307 | { 0x1000f8, 0x200008 }, /* PM_GCT_EMPTY_CYC */ | |
308 | { 0x1000fc, 0x400006 }, /* PM_LSU_DERAT_MISS_CYC */ | |
309 | { 0x20000e, 0x400007 }, /* PM_LSU_DERAT_MISS */ | |
310 | { 0x200012, 0x300012 }, /* PM_INST_DISP */ | |
311 | { 0x2000f2, 0x3000f2 }, /* PM_INST_DISP */ | |
312 | { 0x2000f8, 0x300010 }, /* PM_EXT_INT */ | |
313 | { 0x2000fe, 0x300056 }, /* PM_DATA_FROM_L2MISS */ | |
314 | { 0x2d0030, 0x30001a }, /* PM_MRK_FPU_FIN */ | |
315 | { 0x30000a, 0x400018 }, /* PM_MRK_INST_FIN */ | |
316 | { 0x3000f6, 0x40000e }, /* PM_L1_DCACHE_RELOAD_VALID */ | |
317 | { 0x3000fe, 0x400056 }, /* PM_DATA_FROM_L3MISS */ | |
318 | }; | |
319 | ||
320 | /* | |
321 | * This could be made more efficient with a binary search on | |
322 | * a presorted list, if necessary | |
323 | */ | |
324 | static int find_alternatives_list(unsigned int event) | |
325 | { | |
326 | int i, j; | |
327 | unsigned int alt; | |
328 | ||
329 | for (i = 0; i < ARRAY_SIZE(event_alternatives); ++i) { | |
330 | if (event < event_alternatives[i][0]) | |
331 | return -1; | |
332 | for (j = 0; j < MAX_ALT; ++j) { | |
333 | alt = event_alternatives[i][j]; | |
334 | if (!alt || event < alt) | |
335 | break; | |
336 | if (event == alt) | |
337 | return i; | |
338 | } | |
339 | } | |
340 | return -1; | |
341 | } | |
342 | ||
343 | static int p6_get_alternatives(unsigned int event, unsigned int alt[]) | |
344 | { | |
345 | int i, j; | |
346 | unsigned int aevent, psel, pmc; | |
347 | unsigned int nalt = 1; | |
348 | ||
349 | alt[0] = event; | |
350 | ||
351 | /* check the alternatives table */ | |
352 | i = find_alternatives_list(event); | |
353 | if (i >= 0) { | |
354 | /* copy out alternatives from list */ | |
355 | for (j = 0; j < MAX_ALT; ++j) { | |
356 | aevent = event_alternatives[i][j]; | |
357 | if (!aevent) | |
358 | break; | |
359 | if (aevent != event) | |
360 | alt[nalt++] = aevent; | |
361 | } | |
362 | ||
363 | } else { | |
364 | /* Check for alternative ways of computing sum events */ | |
365 | /* PMCSEL 0x32 counter N == PMCSEL 0x34 counter 5-N */ | |
366 | psel = event & (PM_PMCSEL_MSK & ~1); /* ignore edge bit */ | |
367 | pmc = (event >> PM_PMC_SH) & PM_PMC_MSK; | |
368 | if (pmc && (psel == 0x32 || psel == 0x34)) | |
369 | alt[nalt++] = ((event ^ 0x6) & ~PM_PMC_MSKS) | | |
370 | ((5 - pmc) << PM_PMC_SH); | |
371 | ||
372 | /* PMCSEL 0x38 counter N == PMCSEL 0x3a counter N+/-2 */ | |
373 | if (pmc && (psel == 0x38 || psel == 0x3a)) | |
374 | alt[nalt++] = ((event ^ 0x2) & ~PM_PMC_MSKS) | | |
375 | ((pmc > 2? pmc - 2: pmc + 2) << PM_PMC_SH); | |
376 | } | |
377 | ||
378 | return nalt; | |
379 | } | |
380 | ||
381 | static void p6_disable_pmc(unsigned int pmc, u64 mmcr[]) | |
382 | { | |
383 | /* Set PMCxSEL to 0 to disable PMCx */ | |
384 | mmcr[1] &= ~(0xffUL << MMCR1_PMCSEL_SH(pmc)); | |
385 | } | |
386 | ||
387 | static int power6_generic_events[] = { | |
388 | [PERF_COUNT_CPU_CYCLES] = 0x1e, | |
389 | [PERF_COUNT_INSTRUCTIONS] = 2, | |
390 | [PERF_COUNT_CACHE_REFERENCES] = 0x280030, /* LD_REF_L1 */ | |
391 | [PERF_COUNT_CACHE_MISSES] = 0x30000c, /* LD_MISS_L1 */ | |
392 | [PERF_COUNT_BRANCH_INSTRUCTIONS] = 0x410a0, /* BR_PRED */ | |
393 | [PERF_COUNT_BRANCH_MISSES] = 0x400052, /* BR_MPRED */ | |
394 | }; | |
395 | ||
396 | struct power_pmu power6_pmu = { | |
397 | .n_counter = 4, | |
398 | .max_alternatives = MAX_ALT, | |
399 | .add_fields = 0x55, | |
400 | .test_adder = 0, | |
401 | .compute_mmcr = p6_compute_mmcr, | |
402 | .get_constraint = p6_get_constraint, | |
403 | .get_alternatives = p6_get_alternatives, | |
404 | .disable_pmc = p6_disable_pmc, | |
405 | .n_generic = ARRAY_SIZE(power6_generic_events), | |
406 | .generic_events = power6_generic_events, | |
407 | }; |